Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
DARTMOUTH MAGNETIC EVOLUTIONARY STELLAR TRACKS AND RELATIONS A Thesis Submitted to the Faculty in partial fulfillment of the requirements for the degree of Doctor of Philosophy in Physics and Astronomy by Gregory Alexander Feiden DARTMOUTH COLLEGE Hanover, New Hampshire July 11, 2013 Examining Committee: Brian C. Chaboyer (Chair) Robert A. Fesen John R. Thorstensen F. Jon Kull Dean of Graduate Studies Sarbani Basu Copyright © 2013 by Gregory A. Feiden This work is licensed under the Creative Commons Attribution 3.0 Unported License. To view a copy of this license, visit http://creativecommons.org/licenses/by/3.0/. Abstract Strong evidence exists showing that stellar evolution models are unable to accurately predict the fundamental properties of low-mass stars. Observations of low-mass stars in detached eclipsing binaries (DEBs) indicate that stellar models under-predict real stellar radii by 5 – 10% and predict effective temperatures that are 3 – 5% too hot. This dissertation provides a careful examination of this problem using the Dartmouth stellar evolution code. Accurate models of the three stars in KOI-126 are presented. These models represent the first successful stellar evolution models of fully convective stars. I then introduce a novel method for estimating the ages of young, low-mass DEBs. The method takes advantage of apsidal motion to enable the use of stellar interior structure to predict ages instead of stellar surface properties, which are prone to significant uncertainty. Next, a reanalysis of the magnitude of the mass-radius discrepancies is performed with models that account for realistic metallicity and age variation. Results suggest that discrepancies are about a factor of two smaller than previously believed, although the problem is not entirely resolved. Lastly, I describe the development of a new one-dimensional stellar evolution code that includes effects of a globally pervasive magnetic field. This is done within the framework of the existing Dartmouth code. I find that model radius and effective temperature discrepancies can be reconciled with a magnetic field in stars with a radiative core. The predictions from these models can be observationally tested. Fully convective stars appear insensitive to the influence of magnetic fields, in contradiction with previous studies. I suggest that deficiencies in fully convective stars may instead be related to metallicity. ii Looking back to all that has occurred to me since that eventful day, I am scarcely able to believe in the reality of my adventures. They were truly so wonderful that even now I am bewildered when I think of them. — Jules Verne Preface My time at Dartmouth truly has been an adventure. The adventure would not have been successful, however, without the support of a great many people. There is much talk of standing on the shoulders of giants, but I would be remiss if I did not acknowledge those who provided the boost necessary to reach those shoulders. The Dartmouth stellar evolution code is the product of the hard work of successive generations of faculty, postdocs, and graduate students from Yale University and Dartmouth College. This thesis is an extension of their work and is a testament to how far the field of stellar evolution has come. Without the untiring dedication of those before me, this project would not have been possible. To all of my predecessors, you are the giants. Even if we have not met, thank you. On a more personal level, there are many people who have helped support me over the past five years. To all of the hockey folks in the Upper Valley, I will miss the competition on the ice and the pub visits that inevitably followed. This is especially true of my teammates on K2, the past four years have been a blast and I am truly grateful to have been a part of the franchise. To my graduate school friends and colleagues, past and present, it has been an absolute pleasure to have gotten to know you and work with you. I have cherished all of the conversations and discussions, from late night homework sessions to thought-provoking conversations about your research and mine. And who could ever forget the powerhouse that is the Absolute Zeros hockey team? Without a doubt you have all been an integral part iii of the completion of this thesis. Thank you for the boost. It’s been one hell of a ride. The faculty and staff at Dartmouth have been wonderful. The guidance and mentoring I have received was crucial to my development as a scientist. This is especially true of the astronomers: Gary Wegner, Ryan Hickox, Rob Fesen, John Thorstensen, and Brian Chaboyer. You have all been an inspiration. I can only hope to one day be as good of a scientist as you all are. Special thanks is owed to Rob and John for serving on my thesis committee and reading through the pages that are contained within. I also have to thank Sarbani Basu, from Yale, who made the trip up to Hanover to sit as the external committee member and who provided excellent feedback on my science. Of course, so many thanks are owed to my adviser, Brian, for his continuing support, guidance, and excellent defense. I always knew the front of the net would be well guarded when he was on the ice. I must also mention those others that helped me along my way. Bruce Zellar, Aaron Dotter, the faculty during my tenure at SUNY Oswego, and the astronomers in the Physics and Astronomy Department at Uppsala University. Swedish hospitality is second to none. To my family, you have always been my strongest advocates. Mom and Dad, this thesis is a testament to your success at raising me. Finally, to my wife, Meghan. No words in a preface can express how grateful I am to have you in my life. On a less personal level, I’d like to thank the William H. Neukom 1964 Institute for Computational Science for their generous financial support. The science in this thesis has also been supported by the National Science Foundation (NSF) grant AST-0908345. This research has made use of NASA’s Astrophysics Data System (ADS), the SIMBAD database, operated at CDS, Strasbourg, France, and the ROSAT data archive tools hosted by the High Energy Astrophysics Science Archive Research Center (HEASARC) at NASA’s Goddard Space Flight Center. iv Contents Abstract ii Preface iii Contents v List of Tables xi List of Figures xiii 1 Introduction 1 1.1 Low-Mass Stars . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 1.1.1 Properties of Low-Mass Stars . . . . . . . . . . . . . . . . . . . . . . . 3 1.1.2 Low-Mass Stars as Hosts for Exoplanets . . . . . . . . . . . . . . . . . 6 1.2 Detached Eclipsing Binaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 1.3 The Mass–Radius(–Teff ) Problem . . . . . . . . . . . . . . . . . . . . . . . . . . 10 1.3.1 Metallicity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 1.3.2 Opacity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 1.3.3 Convection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18 1.3.4 Magnetic Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19 Magnetic Fields in Stellar Evolution . . . . . . . . . . . . . . . . . . . . . . . . 22 1.4.1 22 1.4 Modified Adiabatic Gradient . . . . . . . . . . . . . . . . . . . . . . . . v 1.5 1.6 2 3 Reduced Mixing Length . . . . . . . . . . . . . . . . . . . . . . . . . . 23 1.4.3 Star Spots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24 1.4.4 A Need for New Models? . . . . . . . . . . . . . . . . . . . . . . . . . . 26 The Dartmouth Stellar Evolution Code . . . . . . . . . . . . . . . . . . . . . . 27 1.5.1 Physics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28 1.5.2 Solar Calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32 Thesis Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 Accurate Low-Mass Stellar Models of KOI-126 36 2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 2.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 2.2.1 Stellar Age . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 2.2.2 Mass-Radius Relation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40 2.2.3 Relative Fluxes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43 2.2.4 Apsidal Motion Constant . . . . . . . . . . . . . . . . . . . . . . . . . . 46 2.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47 2.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50 Using the Interior Structure Constants as an Age Diagnostic 52 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52 3.2 Interior Structure Constants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54 3.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58 3.3.1 Single Stars . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58 3.3.2 Binary Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64 3.4.1 Observational Considerations . . . . . . . . . . . . . . . . . . . . . . . 66 3.4.2 Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68 3.4 4 1.4.2 The Mass-Radius Relation for Low-Mass, Main-Sequence Stars vi 70 4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70 4.2 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74 4.3 Isochrone Fitting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77 4.3.1 Isochrone Grid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77 4.3.2 Model Rotation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79 4.3.3 Fitting Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79 4.3.4 Age & Metallicity Priors . . . . . . . . . . . . . . . . . . . . . . . . . . 82 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86 4.4.1 Standard Stellar Models . . . . . . . . . . . . . . . . . . . . . . . . . . . 86 4.4.2 Variable Mixing Length Models . . . . . . . . . . . . . . . . . . . . . . 93 4.4.3 Peculiar Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98 4.4 4.5 4.6 5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100 4.5.1 Radius Deviations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100 4.5.2 Radius-Rotation-Activity Correlations . . . . . . . . . . . . . . . . . . 103 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114 Magnetic Perturbation within the Framework of DSEP 116 5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116 5.2 Magnetic Perturbation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119 5.3 5.4 5.2.1 Magnetic Field Characterization . . . . . . . . . . . . . . . . . . . . . . 119 5.2.2 Stellar Structure Perturbations . . . . . . . . . . . . . . . . . . . . . . . 124 5.2.3 Thermodynamic Structure . . . . . . . . . . . . . . . . . . . . . . . . . 126 5.2.4 Magnetic Mixing Length Theory . . . . . . . . . . . . . . . . . . . . . 136 5.2.5 The Parameter f and the Frozen Flux Condition . . . . . . . . . . . . 154 Implementation in the Dartmouth Code . . . . . . . . . . . . . . . . . . . . . . 156 5.3.1 Magnetic Field Strength Distribution . . . . . . . . . . . . . . . . . . . 156 5.3.2 Numerical Implementation . . . . . . . . . . . . . . . . . . . . . . . . . 159 Testing and Validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160 vii 5.5 5.6 6 5.4.1 Numerical Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160 5.4.2 Solar Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161 5.4.3 Ramped Versus Single Perturbation . . . . . . . . . . . . . . . . . . . . 162 Case Study: EF Aquarii . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165 5.5.1 Standard Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165 5.5.2 Magnetic Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169 5.6.1 Field Strengths . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169 5.6.2 Implications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175 Magnetic Models of Low-Mass Stars with a Radiative Core 178 6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178 6.2 Analysis of Individual DEB Systems . . . . . . . . . . . . . . . . . . . . . . . . 182 6.3 6.4 6.5 6.2.1 UV Piscium . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182 6.2.2 YY Geminorum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191 6.2.3 CU Cancri . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198 Magnetic Field Strengths . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205 6.3.1 Surface Field Strengths . . . . . . . . . . . . . . . . . . . . . . . . . . . 205 6.3.2 Interior Field Strengths . . . . . . . . . . . . . . . . . . . . . . . . . . . 212 Reducing the Magnetic Field Strengths . . . . . . . . . . . . . . . . . . . . . . 213 6.4.1 Magnetic Field Radial Profile . . . . . . . . . . . . . . . . . . . . . . . 214 6.4.2 Finite Electrical Conductivity . . . . . . . . . . . . . . . . . . . . . . . 218 6.4.3 Presence of E-fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220 6.4.4 Dynamo Energy Source . . . . . . . . . . . . . . . . . . . . . . . . . . . 221 Further Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226 6.5.1 Interior Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226 6.5.2 Comparison to Previous Work . . . . . . . . . . . . . . . . . . . . . . . 228 6.5.3 Starspots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231 viii 6.6 7 Implications for Asteroseismology . . . . . . . . . . . . . . . . . . . . 232 6.5.5 Exoplanets & the Circumstellar Habitable Zone . . . . . . . . . . . . 235 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239 Magnetic Models of Low-Mass, Fully Convective Stars 246 7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246 7.2 Magnetic Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 248 7.3 7.4 7.5 8 6.5.4 7.2.1 Dipole Radial Profile . . . . . . . . . . . . . . . . . . . . . . . . . . . . 248 7.2.2 Gaussian Radial Profile . . . . . . . . . . . . . . . . . . . . . . . . . . . 248 7.2.3 Constant Λ Turbulent Dynamo . . . . . . . . . . . . . . . . . . . . . . 249 Analysis of Individual DEB Systems . . . . . . . . . . . . . . . . . . . . . . . . 250 7.3.1 Kepler-16 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250 7.3.2 CM Draconis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 260 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 266 7.4.1 Magnetic Field Radial Profiles . . . . . . . . . . . . . . . . . . . . . . . 266 7.4.2 Surface Magnetic Field Strengths . . . . . . . . . . . . . . . . . . . . . 267 7.4.3 Comparison to Previous Studies . . . . . . . . . . . . . . . . . . . . . . 269 7.4.4 Alternatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 274 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 278 Conclusions & Future Investigations 280 8.1 Thesis Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 280 8.2 Future Investigations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283 A DMESTAR User’s Guide 286 B Additional Model Physics 295 B.1 Turbulent Mixing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295 B.2 Internal Structure Constant . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 301 ix B.3 Convective Overturn Timescale . . . . . . . . . . . . . . . . . . . . . . . . . . . 301 B.4 Nuclear Reaction Rates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302 B.5 Gough & Taylor Magnetic Inhibition . . . . . . . . . . . . . . . . . . . . . . . . 302 B.6 Electron Conduction Opacities . . . . . . . . . . . . . . . . . . . . . . . . . . . 303 B.7 τ=100 Boundary Condition Fit Point . . . . . . . . . . . . . . . . . . . . . . . . 304 Bibliography 305 x List of Tables 1.1 Solar calibration parameters. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32 2.1 Properties of the KOI-126 system. . . . . . . . . . . . . . . . . . . . . . . . . . . 38 2.2 Influence of the adopted EOS on low-mass stellar radii. . . . . . . . . . . . . . 42 2.3 Apsidal motion constants (k 2 ) for fully convective stars. . . . . . . . . . . . . 46 4.1 DEBs with at least one low-mass component with precise masses and radii. . 75 4.2 Age and [Fe/H] priors for low-mass DEBs. . . . . . . . . . . . . . . . . . . . . 82 4.3 Best fit isochrone using a solar calibrated mixing-length. . . . . . . . . . . . . 90 4.4 Best fit isochrone using a mass dependent convective mixing length. . . . . . 95 5.1 Fundamental stellar properties of EF Aquarii. . . . . . . . . . . . . . . . . . . . 118 5.2 Comparison of estimated magnetic field strengths (in G). . . . . . . . . . . . . 173 6.1 Sample of DEBs whose stars possess a radiative core. . . . . . . . . . . . . . . 181 6.2 X-ray properties for the three DEB systems. . . . . . . . . . . . . . . . . . . . 206 6.3 Surface magnetic field properties of the stars in UV Psc, YY Gem, & CU Cnc 210 6.4 Peak interior magnetic field strengths. . . . . . . . . . . . . . . . . . . . . . . . 213 6.5 Low-mass stars with direct magnetic field measurements. . . . . . . . . . . . 242 7.1 Fundamental properties of Kepler-16. . . . . . . . . . . . . . . . . . . . . . . . 251 7.2 Fundamental properties of CM Draconis. . . . . . . . . . . . . . . . . . . . . . 261 xi A.1 Magnetic namelist variables. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 288 B.1 Updated DSEP nuclear reaction rates. . . . . . . . . . . . . . . . . . . . . . . . 303 xii List of Figures 1.1 The noted mass-radius and mass-Teff discrepancies with the BCAH98 models. 12 1.2 The influence of αMLT on the BCAH98 models. . . . . . . . . . . . . . . . . . . 1.3 Figure 4 from López-Morales (2007) plotting radius deviations against [Fe/H]. 16 2.1 Age determination of KOI-126 A. . . . . . . . . . . . . . . . . . . . . . . . . . . 40 2.2 Model Fit to KOI-126 B & C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41 2.3 Comparison of three popular equations of state. . . . . . . . . . . . . . . . . . 43 2.4 The Kepler filter transmission profile. . . . . . . . . . . . . . . . . . . . . . . . 44 3.1 Time evolution of k 2 for different stellar masses. . . . . . . . . . . . . . . . . . 59 3.2 The influence of model inputs on the predited k 2 . . . . . . . . . . . . . . . . . 61 3.3 The evolution of the theoretical k 2 for a binary system. . . . . . . . . . . . . . 63 4.1 Comparison of solar-calibrated and variable mixing length isochrones. . . . 80 4.2 Relative radius errors between known DEBs and DSEP . . . . . . . . . . . . . 87 4.3 Relative radius errors between known DEBs and DSEP with variable αMLT . . 94 4.4 Correlation between radius deviations and P orb for standard models. . . . . 104 4.5 Theoretical Rossby Number versus Relative Radius Error . . . . . . . . . . . . 105 4.6 Correlation between radius deviations and P orb for variable αMLT models. . 108 4.7 Rossby number versus relative radius error for variable αMLT models. . . . . 109 4.8 X-ray Luminosity of Low-Mass DEB Systems . . . . . . . . . . . . . . . . . . 111 xiii 13 5.1 The assumed radial profile of the magnetic field strength. . . . . . . . . . . . 157 5.2 Numerical stability tests for the magnetic Dartmouth code. . . . . . . . . . . 160 5.3 The influence of ramping the magnetic field perturbation. . . . . . . . . . . . 163 5.4 Effect of altering the perturbation age on magnetic models. . . . . . . . . . . 165 5.5 Mass tracks for EF Aqaurii in the age–radius plane. . . . . . . . . . . . . . . . 167 5.6 Mass tracks for EF Aqaurii in the Teff –radius plane. . . . . . . . . . . . . . . . 168 6.1 Standard DSEP mass tracks of UV Psc at two different metallicities. . . . . . 185 6.2 Magnetic mass track of UV Psc B with a 4.0 kG surface magnetic field. . . . 186 6.3 Magnetic mass track of UV Psc B with a reduced αMLT . . . . . . . . . . . . . . 187 6.4 Magnetic mass tracks of UV Psc A and UV Psc B with [Fe/H] = −0.3. . . . . 190 6.5 Mass tracks of YY Gem in the age-radius plane with [Fe/H] = −0.1. . . . . . 192 6.6 Mass tracks of YY Gem in the Teff -radius plane with [Fe/H] = −0.1. . . . . . 193 6.7 Mass tracks of YY Gem in the age-radius plane with [Fe/H] = −0.2. . . . . . 196 6.8 Mass tracks of YY Gem in the Teff -radius plane with [Fe/H] = −0.2. . . . . . 197 6.9 Standard DSEP mass tracks for CU Cnc. . . . . . . . . . . . . . . . . . . . . . . 201 6.10 Magnetic stellar evolution mass tracks for CU Cnc. . . . . . . . . . . . . . . . 202 6.11 Temperature profile within a model of CU Cnc A. . . . . . . . . . . . . . . . . 203 6.12 Surface magnetic flux Φ versus X-ray luminosity. . . . . . . . . . . . . . . . . 208 6.13 Theoretical surface magnetic flux Φ versus X-ray luminosity. . . . . . . . . . 211 6.14 Dipole versus Gaussian field profile. . . . . . . . . . . . . . . . . . . . . . . . . 215 6.15 Dependence of the stellar model radius evolution on the radial field profile. . 216 6.16 (∇s − ∇ad ) as a function of the logarithmic density. . . . . . . . . . . . . . . . 217 6.17 Influence of the parameter f on model radius predictions. . . . . . . . . . . . 220 6.18 Effect on the model radius predictions assuming a turbulent dynamo. . . . . 224 6.19 Interior density profile using different magnetic field prescriptions. . . . . . . 227 6.20 Interior density profile of a magnetic and reduced αMLT model. . . . . . . . . 230 6.21 Sound speed profile for a magnetic and non-magnetic model of equal mass. xiv 233 6.22 Habitable zone boundaries for magnetic models. . . . . . . . . . . . . . . . . . 237 7.1 Standard DSEP models for Kepler-16 A. . . . . . . . . . . . . . . . . . . . . . . 252 7.2 Standard DSEP models for Kepler-16 B . . . . . . . . . . . . . . . . . . . . . . 255 7.3 Magnetic models of Kepler-16 B w/ D11 mass and a dipole radial field. . . . 258 7.4 Magnetic models of Kepler-16 B w/ D11 mass and a Gaussian radial field. . . 258 7.5 Magnetic model of Kepler-16 B w/ D11 mass and a constant Λ profile. . . . . 259 7.6 Magnetic models of Kepler-16 B w/ B12 mass and a Gaussian radial field. . . 259 7.7 Standard stellar evolution models of CM Dra A and B. . . . . . . . . . . . . . 262 7.8 Magnetic models CM Dra with a dipole radial profile. . . . . . . . . . . . . . 263 7.9 Magnetic models CM Dra with a Gaussian radial profile. . . . . . . . . . . . . 265 7.10 Constant Λ models for CM Dra A and B. . . . . . . . . . . . . . . . . . . . . . 265 7.11 (∇s − ∇ad ) as a function of the logarithmic density. . . . . . . . . . . . . . . . 267 7.12 Radius residuals for fully convective stars as a function of stellar properties. 276 7.13 Radius residuals for fully convective stars versus metallicity with fixed-Y . . 278 xv Chapter 1 Introduction It was a dark and stormy night, so he decided to be a theorist. There are approximately 100 billion stars in the Milky Way Galaxy. Of those 100 billion stars, nearly 90% are thought to have a mass equal to or less than the mass of the Sun (Research Consortium On Nearby Stars, RECONS;1 Henry et al. 2006). Furthermore, nearly 75% of the stars in the Galaxy, or at least in the local galactic neighborhood, belong to a particular class of low-mass stars known as M-dwarfs (also called red dwarfs; Henry et al. 2006; Lépine & Gaidos 2011). Indeed, stars less massive than the Sun dominate the stellar population of the Milky Way. Low-mass stars appear to also dominate the total stellar population of large, elliptical galaxies, suggesting that low-mass stars are the most common type of star in the Universe (van Dokkum & Conroy 2010; Conroy & van Dokkum 2012). The sheer number of low-mass stars makes them worthy of intensive study, but significant interest in low-mass stars has grown in recent years for other reasons. A resurgence of interest in these stars has been motivated by the search for habitable extra-solar planets. Low-mass stars turn out to make better targets when searching for small, rocky planets orbiting in the “habitable zone,” or the region of space surrounding a star where life, as we 1 Number counts taken from their January 1, 2012 census. See http://www.recons.org/. 1 know it, has a chance to exist (Charbonneau 2009; Gillon et al. 2010). Low-mass stars also play host to a wide range of interesting physical processes. This is due to the extreme physical conditions that exist in and around these stars, conditions which are nearly impossible to achieve within an Earth-based laboratory. Complex physics such as the condensation of molecules and dust grains (see, e.g., Allard et al. 1997), non-ideal gas thermodynamics at high pressure and low temperature (see, e.g., Saumon et al. 1995), turbulent convection (Wende et al. 2009; Trampedach & Stein 2011), and multiple magnetic dynamo mechanisms (see, e.g., Parker 1979; Mestel 1999) are thought to be associated with low-mass stars. It is for these reasons that make investigations of low-mass stars worthwhile and rewarding. The information that can be gathered from observations of low-mass stars, however, is rather limited. Low-mass stars are intrinsically faints objects, as we will see, which makes them difficult to study. Investigations involving low-mass stars often have to rely on the construction of theoretical models to aid in the interpretation of the observations. The models allow us to predict the physical properties of low-mass stars, which in turn allows for a complete characterization of the star. Given that studies of low-mass stars often require a model dependent characterization of the stars, it is crucial that the models be accurate. The thesis that follows assesses our present ability to construct accurate theoretical models of low-mass stars and proceeds to introduce new physics that may improve model predictions. Before we delve into investigations involving low-mass stars, we would be remiss if we did not first explore the fundamental concepts and themes that will be present throughout the text. 1.1 Low-Mass Stars The definition of what constitutes a low-mass star typically depends on whom you ask. However, it is only the upper mass limit that can be debated. The lower boundary is “fixed” 2 to the lowest possible mass where a self-gravitating body can fuse hydrogen in its core (∼ 0.08M⊙ ; Kumar 1963; Baraffe et al. 1998). Below that critical mass, electron degeneracy pressure prevents the star from collapsing prior to the onset of hydrogen fusion.2 While the precise numerical value defining the hydrogen burning mass limit may vary slightly, it is safe to say that a star near the hydrogen burning mass limit is low-mass. To avoid any ambiguity or personal preference that might exist about what defines the upper mass limit, we shall define a low-mass star as one that has a mass less than 80% that of the Sun (0.80M⊙ ). While our motivation may not be clear now, the choice of 0.80M⊙ as an upper mass limit will be revealed in later sections. If there needs to be any more reason for selecting 0.80M⊙ at the moment, let it be for purely aesthetic purposes. After all, 0.80 is an order of magnitude larger than 0.08. Much to astronomers’ chagrin, observations of single low-mass stars cannot reveal their mass in a model-independent fashion. Therefore, defining the class of stars we are interested in studying by their mass is not very beneficial. How will we know when we have found one? Instead, we take advantage of the fact that the low-mass that characterizes these stars directly influences their observable properties. 1.1.1 Properties of Low-Mass Stars Stars that fuse hydrogen to helium in their core belong to a class of stars known as main sequence stars, or “dwarfs.”3 The properties of main sequence stars are largely governed by their mass. In fact, one can derive simple scaling relations, either from theory or from empirical data (Weiss et al. 2004; Andersen 1991). A low-mass star’s radius is approximately 2 Deuterium burning can, however, occur in sub-stellar objects. The use of the word dwarfs is the correct lexical construction for the plural of “dwarf.” Some may be tempted to invoke “dwarves” as the plural form, but this appears to be an alternative construction popularized by Tolkein in his books. See http://grammarist.com/usage/dwarfs-dwarves/ for a further discussion. 3 3 proportional to its mass, M R ∝ , R⊙ M⊙ (1.1) where R ⊙ is the radius of the Sun, or 6.9598 × 1010 cm (Neckel 1995; Bahcall et al. 2005). Using this relation, we infer that low-mass stars have typical radii between 0.10R ⊙ and 0.80R ⊙ , for a 0.10M ⊙ and 0.80M ⊙ star, respectively. Additionally, low-mass stars are found to have a luminosity, or intrinsic brightness, that varies with mass according to the relation µ ¶ M 3 L ∝ , L⊙ M⊙ (1.2) with L ⊙ being the luminosity of the Sun, here defined as 3.8418×1033 erg s−1 (Bahcall et al. 2005). We see that low-mass stars have luminosities between approximately 10−3 L ⊙ and 0.5L ⊙ , for a 0.10M ⊙ and 0.80M ⊙ star, respectively. Invoking the Stefan-Boltzmann equation relating a star’s luminosity to its radius and effective (surface) temperature, 4 , L = 4πσR 2 Teff (1.3) we find that the surface temperature of low-mass stars is approximately µ ¶ Teff M 1/4 ∝ , Teff,⊙ M⊙ (1.4) where Teff,⊙ = 5 776 K is the effective temperature of the Sun from the Stefan-Boltzmann equation and our definitions of R ⊙ and L ⊙ . Low-mass stars are then expected to have effective temperatures in the range of ∼ 3 250 K to ∼ 5 500 K. In reality, the lower bound on the effective temperature is below 3 200 K, but the scaling relation is approximately valid. These effective temperatures imply that the spectral types of low-mass stars are in the range of mid-G to mid-M. The low temperatures and luminosities associated with low-mass stars permit a variety of 4 unique observational characteristics. Photometrically, low-mass stars appear fainter and redder than stars like the Sun. Therefore, identifying low-mass stars can be as simple as identifying faint, red stars. However, very distant red giant stars may have similar effective temperatures (and thus colors) as low-mass main sequence stars. Red giants can therefore masquerade as nearby low-mass stars. Identifying low-mass stars by color alone is not sufficient, at least without a distance estimate to provide information regarding the intrinsic brightness of the star. Instead, photometric information must be combined with spectroscopic data. Spectra of low-mass stars are dominated by the presence of broad absorption features created by the dense forest of available molecular transitions (Kirkpatrick et al. 1991; Allard et al. 1997). That is, for stars of sufficiently low-mass. Stars with effective temperatures between about 4 000 K and 5 500 K (spectral types late-K to mid-G) have spectra similar to the Sun. Titanium oxide (TiO) bands begin to appear around 4 000 K (late-K) and grow stronger with decreasing effective temperature before saturating around 3 000 K (mid-M). While TiO begins to saturate at the coolest effective temperatures, vanadium oxide (VO) bands begin to form and grow stronger with decreasing temperature (Kirkpatrick et al. 1991). In the nearinfrared (NIR), the dominant source of opacity is due to water (H2 O), creating the “steam bands.” Additional contributions in the optical and NIR from metal-hydrides such as calcium hydride (CaH), magnesium hydride (MgH), and iron hydride (FeH) are also present in low-mass stellar spectra owing to their cool effective temperatures. Ambiguity may occur when comparing the spectra of two cool stars with different luminosities (e.g., giants versus dwarfs). To discern between the two, optical and NIR atomic features are generally used (Kirkpatrick et al. 1991). Neutral sodium doublets (Na i; λ = 5890, 5896 Å and 8183, 8195 Å), neutral potassium doublets (K i; λ = 7665, 7699 Å), and neutral calcium (Ca i; λ = 4227 Å) optical absorption lines are gravity- (or luminosity-) sensitive features that appear very strong in the spectra of dwarf stars, but not giant stars. The optical Ca 5 i line is less favorable than Na i as there is considerably less flux for an M-dwarf short of 6000 Å. Additionally, the NIR Ca ii triplet and Na i doublet may be used as gravity diagnos- tics, where the Ca ii triplet (λ = 8498, 8542, 8662 Å) is weakest in dwarf spectra and the Na i doublet is strongest for dwarfs (Kirkpatrick et al. 1991). Owing to their low mass, low-mass stars fuse hydrogen at a relatively slow rate. The internal temperature profile is proportional to the internal pressure gradient. It is the latter that supports the star against gravity. Less mass means that less of a pressure gradient is required, leading to cooler temperatures throughout the star. The proton-proton (p-p) chain produces energy at a rate proportional to the temperature to the fourth-power, ǫp−p ∝ T 4 . Due to the cooler temperatures inside low-mass stars, they primarily burn hydrogen via the first channel of the p-p chain (Chabrier & Baraffe 1997), and they do so slowly. Low-mass stars thus have main sequence lifetimes that range from 12 Gyr for an 0.80M⊙ star to over 3 Tyr for a 0.1M⊙ star (Laughlin et al. 1997). After their main sequence life, most low-mass stars will progress up the red giant branch. However, below a given (model-dependent) mass between 0.1M⊙ and 0.2M⊙ , low-mass stars never ascend the red giant branch and are doomed to slowly fade away. This makes models of low-mass stars possible tools to study why stars become red giants (Laughlin et al. 1997). 1.1.2 Low-Mass Stars as Hosts for Exoplanets Interest in low-mass stars has seen a revival due to interest in discovering extra-solar planets (henceforth exoplanets). However, it is not just any planet that researchers are interested in finding. The ultimate goal is to discover the first Earth-sized rocky planet in the habitable zone of its host star. Why, though, should low-mass stars be of interest to those searching for the next Earth? We know that at least one Earth exists around an early G-dwarf. Would not other solar-type stars provide the best opportunities based on our existing knowledge? The advantages provided by low-mass stars over solar-type stars are numerous. First, M6 dwarfs are the most common type of star and thus provide a large number of potential exoplanet hosts. Within 10 pc of the Sun, there are 248 M-dwarfs and only 20 solar-type G-dwarfs (RECONS4 ). Other advantages rely on the lower mass and cooler effective temperatures of low-mass stars compared to solar-type stars. Consider the two primary techniques for detecting exoplanets—radial velocity and transit methods. Let us imagine two stars, one an M-dwarf and one a G-dwarf, each with an identical Earth-sized planet orbiting in the circumstellar habitable zone (HZ). It is important to mention that the exoplanet will be orbiting closer to the M-dwarf than the G-dwarf. The low luminosity and cool effective temperature of the M-dwarf requires an Earth-like planet to orbit closer to the star if it is to receive enough incident flux to support liquid water and a suitable greenhouse effect (Kasting et al. 1993). Measuring the radial velocity induced on the two stars by their respective planets depends primarily on the semi-major axis of the planet’s orbit and the planet’s mass relative to the star (we assume the inclination to be i = 90◦ ). The semi-amplitude of the radial velocity signal is (assuming a circular orbit), 2πG K= P orb µ ¶1/3 Mp ¡ M p + M⋆ ¢2/3 , (1.5) where P orb is the orbital period, M p is the planet mass, and M⋆ is the stellar mass. The closer the planet and higher the planet-to-star mass ratio, the larger the radial velocity signal. In both situations, M-dwarfs provide an advantage over G-dwarfs. For a similar mass planet around an M-dwarf compared to a G-dwarf, the ratio of the semi-amplitude is ¶ µ ¶ µ A g 1/2 M g 2/3 Km , = Kg Am Mm (1.6) where the subscripts m and g refer to the values for an M-dwarf and G-dwarf, respectively. Here, A is the semi-major axis of the planet orbit and M is the mass. For a planet orbiting in 4 http://www.chara.gsu.edu/RECONS/census.posted.htm 7 the habitable zone, K m /K g ∼ 2 to 15 for the range of M-dwarf masses. M-dwarfs do suffer from significant radial velocity noise (Saar & Donahue 1997), decreasing the radial velocity sensitivity, but there is considerable effort to mitigate this problem (see, e.g., Barnes et al. 2011; Muirhead et al. 2011). Detecting an Earth-like planet via the transit method requires precise photometry and the fortuitous alignment of a planet orbiting across the face of the star from the observer’s point-of-view. Precise photometry is necessary because the dip in the flux received from a star due to the transit of a planet is proportional to the ratio of the projected planet surface area to the projected stellar surface area—assuming the planet contributes negligible flux and can be considered dark. Explicitly, the change in flux is ∆F ∝ µ Rp R⋆ ¶2 . (1.7) In other words, the key variable is the square of the ratio of the planet-to-star radius. Therefore, an Earth-like planet transiting an M-dwarf will create a change in flux that is a factor of 4 to 100 times larger compared to transiting a G-dwarf, depending on the size of the M-dwarf. The second requirement is actually observing a transit. There is a finite probability of detecting a planet transit if we assume that the inclination of the planet orbit from the observer’s point-of-view is completely random. This geometric probability is proportional to the sum of the stellar and planetary radius and inversely proportional to the orbital semimajor axis (Borucki & Summers 1984). The probability can be simplified such that p∝ R⋆ , A (1.8) where p is the probability. The planetary radius can usually be ignored because it is a small fraction of the stellar radius. The proximity of the HZ to low-mass stars compensates for 8 the small stellar radius, giving an observer a greater geometric probability, by up to a factor of 2, of detecting the transit of an Earth-like planet around a low-mass star. We can now see why low-mass stars are considered preferable when it comes to exoplanet target selection. But, there is one glaring caveat to locating exoplanets around low-mass stars: the stellar properties are required to characterize the planet. This work largely falls on the shoulders of stellar evolution theory. There are serious issues with low-mass stellar evolution theory that have arisen due to the study of eclipsing binary systems. Throughout the rest of this chapter we will introduce these topics and the known issues with low-mass stellar evolution theory. 1.2 Detached Eclipsing Binaries Eclipsing binary systems (EBs) are binary stellar systems in which the two components are observed to pass in front of—or eclipse—one another. Here, we are concerned only with detached eclipsing binaries (DEBs). EBs are considered detached when the two stars have a binary separation (or semi-major axis, A ), that is large enough to prevent the stars from undergoing mass exchange or from having undergone mass exchange in the past. This condition also suggests that the individual stars are not strongly distorted by tidal interactions with the companion (see, also, Chapter 3). What makes DEBs important for the present work is that the characteristics of these systems allow for the masses and radii of the individual stars to be determined with very high precision (see reviews by Popper 1980; Andersen 1991; Torres et al. 2010). Typical random uncertainties of well-studied systems are routinely below 2% in both quantities. Furthermore, the mass and radius determinations are very nearly model-independent. This makes DEBs phenomenal tools for testing stellar structure and evolution theory. Effective temperatures may also be derived for these systems, but the results are less robust as additional 9 assumptions, such as the temperature scale, must be made (Torres et al. 2010). Rigorous discussions of the data quality needed to obtain precise (and accurate) stellar properties from DEBs are presented in the literature (Popper 1980, 1984; Andersen 1991; Torres et al. 2010). For our purposes, we are more concerned with using DEBs as tools to test the validity of stellar evolution models and to assess the current state of low-mass stellar evolution theory. We therefore choose to avoid a discussion of the actual data products and refer the reader to the referenced works for further information. The important information to take away is that DEBs provide a powerful, direct test of the predictions of stellar evolution models. 1.3 The Mass–Radius(–Teff) Problem Historically, the development of sophisticated low-mass stellar evolution models (Osterbrock 1953; Limber 1958a,b)5 has been accompanied by discussions that the observed radius and luminosity of stars in DEBs do not compare well to the predictions of stellar evolution theory. Although stellar models continued to become more physically realistic by including updated opacity sources, detailed equations of state, and rigorous atmosphere calculations (Hoxie 1970), it was recognized that observational errors and model uncertainties were too large to provide a meaningful comparison (Hoxie 1973). Additionally, prior to the discovery and characterization of CM Draconis (Zwicky 1966; Lacy 1977), most comparisons were made to only a single low-mass DEB (YY Geminorum; Kron 1952). Until more recently, the state of the low-mass stellar evolution field was rather stagnant. Investigations over the past two decades have made it clear that stellar evolution models are unable to accurately reproduce the properties of low-mass stars. This fact has been the result of significant reductions in observational uncertainties (Andersen 1991; Torres et al. 2010) 5 Here we refer to the inclusion of basic nuclear reaction networks, opacity effects, and the inclusion of convective envelopes. 10 and the development of sophisticated stellar models (Baraffe et al. 1998; Dotter et al. 2008). The primary line of evidence stems from studies of DEB systems, where it has been found that stellar evolution models systematically under-predict stellar radii by 5% – 10% and over-predict stellar effective temperatures by 3% – 5% at a given mass (e.g., Metcalfe et al. 1996; Popper 1997; Torres & Ribas 2002; López-Morales & Ribas 2005; López-Morales et al. 2006; Bayless & Orosz 2006; López-Morales & Shaw 2007; Morales et al. 2009b,a; Torres et al. 2010). Further evidence has been supplied by the direct measurement of stellar radii using interferometry (Berger et al. 2006), although the uncertainty in the quoted stellar masses makes comparisons with models tenuous. Figures 1.1(a) and 1.1(b) demonstrate the discrepancies observed between models and observations in the mass-radius and mass-Teff planes. These figures represent the cutting-edge data and models prior to the work for this dissertation (i.e., prior to 2011). The data shown are only those DEB systems with masses and radii measured to better than 3% and that passed the screening process for inclusion in the Torres et al. (2010) review.6 The models are those from the seminal work of Baraffe et al. (1998). A full description of the physics used in the models is presented elsewhere (Chabrier & Baraffe 1997; Baraffe et al. 1997; Baraffe et al. 1998), but it is worth remarking that they represent the most physically realistic low-mass models that had undergone rigorous evaluation in the literature prior to 2011. The models include a highly detailed equation of state that accounts for numerous non-ideal gas effects (Saumon et al. 1995), they use the latest low-temperature opacities that account for complicated atomic and molecular species, and they prescribe the surface boundary conditions using non-gray model atmospheres that rigorously solve the equations of radiative transfer (Allard & Hauschildt 1995; Allard et al. 1997). It is clear from Figures 1.1(a) and 1.1(b) that the models under-predict the stellar radii and over-predict the Teff at a given mass. This result is independent of the adopted isochrone 6 The total number of DEBs included in the Torres et al. (2010) review is only a fraction of the total number of DEBs identified and characterized. 11 0.8 BCAH98 Isochrones 0.7 1 Gyr 5 Gyr 4400 0.6 Teff (K) Radius (R⊙) BCAH98 Isochrones 4800 1 Gyr 5 Gyr 0.5 0.4 4000 3600 0.3 3200 [M/H] = 0.0 0.2 0.2 0.3 0.4 0.5 0.6 Mass (M⊙) 0.7 0.8 [M/H] = 0.0 0.2 0.3 0.4 0.5 0.6 Mass (M⊙) 0.7 0.8 Figure 1.1: (a) The mass-radius relation as defined by DEBs with precise masses and radii (grey points) and by the Baraffe et al. (1998) low-mass stellar models (solid lines). The models have a solar metallicity and are shown at two ages: 1 Gyr (maroon) and 5 Gyr (lightblue). (b) Mass-Teff relation for the same set of DEBs and isochrones. age. The data above 0.8M⊙ , on the other hand, appears to be better fit by the models—one star even has a radius smaller than model predictions. However, the BCAH98 models shown in Figure 1.1 adopted a convective mixing length of αMLT = 1.0. The reduced mixing length disproportionately affects models with masses above ∼ 0.6M⊙ , making the models appear significantly larger than when a solar calibrated αMLT is adopted. Figure 1.2 shows the difference in model radius predictions when a solar calibrated mixing length is adopted. The relative insensitivity of models to αMLT is the primary reason why we define the upper limit of “low-mass” to be 0.8M⊙ . There are additional benefits to this selection, including relative insensitivity to age and metallicity, but these are largely secondary to the influence of αMLT . To understand the implications of the discrepancies in radius and Teff , consider the example of an exoplanet discovered transiting a low-mass star. The properties of the exoplanet depend on the properties inferred for the low-mass star, which are typically model dependent (see e.g., Charbonneau et al. 2009). Of particular importance is the stellar radius, which sets the radius and average density of the planet (Seager & Mallén-Ornelas 12 BCAH98 Isochrones 0.7 αMLT = 1.9 αMLT = 1.0 BCAH98 Isochrones 4800 αMLT = 1.9 αMLT = 1.0 4400 0.6 Teff (K) Radius (R⊙) 0.8 0.5 0.4 4000 3600 0.3 3200 [M/H] = 0.0 0.2 0.2 0.3 0.4 0.5 0.6 Mass (M⊙) 0.7 0.8 [M/H] = 0.0 0.2 0.3 0.4 0.5 0.6 Mass (M⊙) 0.7 0.8 Figure 1.2: The influence of αMLT on the Baraffe et al. (1998) models above 0.6M⊙ . The isochrones were calculated at 1 Gyr with [M/H] = 0.0 and have αMLT = 1.0 (solid line) and αMLT = 1.9 (dashed line). (a) The mass-radius relation. (b) The mass-Teff relation. 2003; Charbonneau et al. 2009). The latter permits an estimate of the planet composition. Therefore, a 10% underestimate of the stellar radius (as we see for low-mass stellar models) translates into a 10% under estimate of the planet radius (R p ) and a 40% overestimate of the average density (ρ ∝ R p−3 ). Both effects create a situation where a planet will be characterized as more “Earth-like.” To avoid such deceptions, models must be made more reliable. The relative insensitivity of low-mass stellar evolution models to various modeling parameters (e.g., age and αMLT ) compared to solar-type stars is both a curse and a blessing. On the one hand, the predictions from low-mass stellar evolution models are robust. This suggests the model predictions are precise. However, this also means that the radius and Teff deviations observed in Figures 1.1 and 1.2 are robust, calling into question the accuracy of the models. The high precisions with which the measurements are quoted exacerbate the situation. There must be another reason for the appearance of the deviations. Several hypotheses have been advanced, which we shall now describe. 13 1.3.1 Metallicity Stellar composition is the second most important property, after mass, governing the evolution of a main sequence star. The precise breakdown of the chemical composition—the mass fractions of hydrogen, X , helium, Y , and other “metals,”7 Z —can affect the evolution of a star in multiple ways. The two primary channels are through altering the nuclear reaction network and through altering the plasma’s opacity. Therefore, stellar composition is a key physical ingredient that must be specified in stellar models. It is sufficient to specify only two of the mass fractions listed above knowing that X + Y + Z = 1. In practice, Y and Z are specified. Some simple assumptions are made regarding the composition of helium: it is either fixed to a pre-determined value or it is allowed to scale with the metal abundance. The latter approach is typically used to cut down on the number of models required to compare theory to observation and to reduce the specified mass fractions to just Z . The standard approach is to allow Y to scale linearly with Z , Y = Yprim + dY ∆Z , dZ (1.9) where Yprim is the primordial helium abundance adopted from Big Bang Nucleosynthesis calculations (Alpher et al. 1948) or determined semi-empirically (Peimbert et al. 2007), ∆Z = Z − Zprim = Z ( Zprim = 0), and the slope of the linear relation is determined empirically (see e.g., Casagrande et al. 2007). Unfortunately, metallicities of low-mass stars are also notoriously difficult to measure (Woolf & Wallerstein 2005). The same complex molecular bands that allow for the classification of low-mass stars hamper metallicity and temperature analyses, especially at optical wavelengths where TiO and VO bands dominate stellar spectra (Kirkpatrick et al. 1991; Reid et al. 1995). Current photometric (Johnson & Apps 2009; Schlaufman & Laughlin 2010) and spectroscopic 7 All elements heavier than helium are known as “metals”. 14 (Woolf & Wallerstein 2005; Bonfils et al. 2005; Bean et al. 2006; Rojas-Ayala et al. 2010) techniques for estimating the metallicity of low-mass stars are beginning to converge, but those are only valid for single stars. DEB systems add the complication that the contributions of the two stars must be accurately decomposed, unless we are lucky enough to find that one of the eclipses is total. All told, there does not exist any robust metallicity determinations for low-mass DEBs. Not only are metallicities difficult to measure, but previous state-of-the-art low-mass stellar evolution models were only computed for a limited number of metallicities (Chabrier & Baraffe 1997; Baraffe et al. 1997; Baraffe et al. 1998). It was argued that variations in the stellar composition would affect the properties of low-mass stars at the few percent level and that uncertainties at that level could be tolerated (Chabrier & Baraffe 1997). This view was acceptable, at the time. However, the availability of high quality data and improved methods of analyzing DEB systems dramatically increased in the decade following Baraffe et al. (1998). This meant that the precision with which stellar properties could be determined was considerably higher (see Section 1.2). Therefore, it was becoming increasingly essential that stellar models used in comparisons with observations account for variations in metallicity (Burrows et al. 2011). The neglect of metallicity variations in stellar models became a natural candidate to explain the observed radius and Teff deviations. The metallicity hypothesis was advanced by Berger et al. (2006). They used the CHARA8 Array on Mount Wilson to measure the radii of six single M-dwarfs with interferometry. The key result was that for the objects they measured, the radius deviations with the BCAH98 stellar models correlated with metallicity (Berger et al. 2006). Shortly after the Berger et al. (2006) study, López-Morales (2007) investigated the origin of the radius deviations in single and DEB stars. She found that radius deviations correlate 8 Center for High Angular Resolution Astronomy 15 Figure 1.3: Figure 4 from López-Morales (2007) plotting the deviations between models and observations against metallicity. The models had an age of 1 Gyr with [Fe/H] = 0.0. (top) Single stars with radii determined using interferometry. (bottom) Binary stars. Used by permission of M. López-Morales. with metallicity for single field stars, but the two properties do not correlate for DEBs. However, the single stars that showed radius discrepancies with stellar models were all from Berger et al. (2006), which may explain the supporting evidence. Regardless, Figure 1.3 demonstrates that there appears to be no correlation with metallicity and radius deviations in DEB systems (López-Morales 2007). Although no correlation is present in Figure 1.3, it is difficult to provide a definitive interpretation for two reasons: (1) all of the stars were compared to a 1 Gyr, solar metallicity isochrone, and (2) there are only a few DEBs with metallicity estimates, all of which are highly uncertain. Metallicity should therefore still be considered a plausible explanation, though it may not be able to induce radius variation in the models at the 10% level. We address metallicity issues specifically in Chapters 2 and 4. 16 1.3.2 Opacity Closely related to the aforementioned metallicity hypothesis is the idea that there are important sources of opacity missing from stellar interior and atmosphere models (Baraffe et al. 1998). These missing sources of opacity are presumed to belong to molecular species that are able to condense in the near-surface and atmosphere regions of low-mass stars. The molecules typically singled out as possible candidates for reevaluation are TiO and VO in the optical and H2 O in the NIR. The motivation for discussion about opacity, however, does not stem from the discrepancies observed in the mass-radius plane. Instead, they are largely motivated by discrepancies observed in the mass-magnitude and color-magnitude diagrams. Converting model predictions for stellar temperatures and luminosities to observable properties (i.e., colors and magnitudes) requires accurate non-gray atmosphere models, such as those described in Section 1.5 (Allard & Hauschildt 1995; Allard et al. 1997; Hauschildt et al. 1999a). The accuracy of the resulting colors and magnitudes are dependent on the ability for the atmosphere model to correctly determine the distribution of flux across the electromagnetic spectrum. Several studies have shown that the theoretical atmosphere models perform rather poorly when it comes to predicting fluxes in the optical. However, they perform adequately in the NIR (Baraffe et al. 1997; Baraffe et al. 1998; Delfosse et al. 1998). The consensus is that the treatment of TiO opacity likely needs improvement while VO may be adequate. In the NIR, the predicted colors, and thus the broad distribution of flux, appears to be accurate, but the modeling of individual H2 O line profiles needs to be improved (Baraffe et al. 1998; Delfosse et al. 1998). Improvements in the theoretical atmosphere models are ongoing, but a large factor in the model accuracy may have to do with the solar abundance adopted (Allard et al. 2011). We elect to avoid pursuing opacity as a solution for the mass-radius problem. This is because studies discussing opacity have been largely focused on problems with theoretical colors 17 and magnitudes, and only indirectly on stellar radii. Our decision is reinforced by the acknowledgement that opacity sources are unlikely to significantly alter the atmosphere structure (Baraffe et al. 1998; Allard et al. 2011). It is the latter feature of atmosphere models that would have the greatest impact on stellar interior models. 1.3.3 Convection One of the greatest unsolved problems in (astro)physics is the theory of convection. This is particularly true in the context of stellar evolution, where a one-dimensional prescription of convection is required. While there are several different local (e.g., Böhm-Vitense 1958) and non-local (e.g., Canuto & Mazzitelli 1991) theories of convection, they still largely rely on the mixing length parameter, αMLT = ℓm /HP , relating the radial extent of a convective eddy, ℓm , to the pressure scale height, HP .9 The notion of a convective mixing length (ℓm ) is to set a distance over which a convecting bubble travels before mixing into its surroundings. Thus, if a convecting bubble travels a large distance, it is efficiently transporting energy from one point in a fluid to the other. Conversely, if the bubble travels a small distance such that multiple bubbles are required to transfer excess heat the same distance as our first example, it can be deemed inefficient. In that sense, αMLT is a measure of the convective efficiency. The precise numerical value of the mixing length parameter is the subject of considerable debate. Standard practice is to avoid making an arbitrary choice by calibrating a 1.0M⊙ stellar evolution model to the Sun, as we will do later in Section 1.5.2. However, it has long been recognized that there is no a priori reason to leave αMLT set to the solar calibrated value. Stars with a mass different than that of the Sun may convect with greater or less efficiency. A lower convective efficiency, interpreted as a lower αMLT , would act to inflate a low-mass 9 The distance over which the pressure within a star changes by a factor of e . 18 star. Reducing the flux of heat due to convection forces the star to compensate by developing a steeper radiative temperature gradient. Therefore, the star will inflate to conserve flux. The effects of this were shown in Figure 1.2. Since the radii of low-mass stars appear to be inflated compared to theoretical models, one might expect that convection is a naturally inefficient process in low-mass stars. This idea motivated Baraffe et al. (1998) to adopt αMLT = 1.0 for all of their models below 0.6M ⊙ . Testing the inefficient convection hypothesis is difficult, as we do not yet have the ability to peer inside low-mass stars. Furthermore, the current sample of low-mass stars in well-studied DEBs totals 8 stars (Torres et al. 2010). This can be noted in Figure 1.1. A considerably larger sample is required to develop an understanding of how radius inflation might correlate with mass and other intrinsic properties, such as metallicity. If naturally inefficient convection is the culprit leading to inflated radii, then two stars of similar mass and metallicity part of two different DEB systems should have similar radii and Teff s. For further discussion of this matter, see Chapter 4. Inefficient convection may also arise for another reason: the presence of a magnetic field. This leads us to the final and most prominent hypothesis. 1.3.4 Magnetic Fields Magnetic fields and magnetic activity are currently the leading explanation for the inflation of stellar radii and suppressed Teff s. The hypothesis was advanced by Ribas (2006) after he presented evidence that radius and temperature disagreements were largely apparent in systems where the stellar parameters were determined with exquisite precision. All of those well-characterized DEB systems were also found to have orbital periods under three days. It was theorized that tidal synchronization of the components could be driving strong magnetic fields, and thus the altering of observable stellar properties. Mutual tidal interactions between the two components would act to both circularize the orbit and synchronize their 19 rotation periods (e.g., Zahn 1977). Considering that the orbital periods were relatively short, the stellar rotational periods would also be short, forcing the stars to rotate more rapidly than if they were isolated single stars. Stellar magnetism is thought to be governed by the dynamo mechanism (Parker 1955, 1979; Mestel 1999). At the most fundamental level, the stellar dynamo is theorized to source the energy for a magnetic field from the Coriolis force acting on convecting fluid motions in a rotating star and also from shear generated at the boundary between the convective envelope and radiative core (Parker 1955, 1979; Mestel 1999). Conventional wisdom dictates that the more rapid the rotation, the stronger the dynamo action and, as a result, the stronger the magnetic field.10 Therefore, tidal synchronization is theorized to help maintain a strong dynamo within the stars in DEB systems. The existence of strong magnetic fields is thought to affect stellar structure through two primary channels: (1) the suppression of thermal convection (Thompson 1951; Chandrasekhar 1961; Gough & Tayler 1966) and (2) by the emergence of surface spots (Hale 1908; Spruit 1982a,b). Ultimately, the energy flux across a given surface within the star is reduced, forcing the star to inflate and cool to maintain the necessary energy flux required for global hydrodynamic stability (Spruit 1982a,b; Mullan & MacDonald 2001; Chabrier et al. 2007). Magnetic fields therefore appear to provide an attractive solution. However, the reduction in flux caused by magnetic fields is only part of the picture. Observations of total solar irradiance (TSI)—the total flux of solar radiation received at Earth— show that it is variable at the 0.1% level on two timescales: (1) the timescale of the solar rotation period, and (2) on the timescale of the 11 year solar cycle (see review by Foukal et al. 2006, and references therein). Reductions in flux caused by spot explain the variation over the rotational period of the Sun. Spots rotating into view will decrease the TSI by a fraction related to their projected surface coverage and their temperature contrast 10 A fluid composed of mostly neutral species can be highly resistive, allowing for rapid dissipation of magnetic energy inhibiting the generation of favorable current networks (Mohanty et al. 2002). 20 with the non-active photosphere (Spruit 1982a,b). The second variability timescale is related to the appearance of faculae, which are brighter than the non-magnetic photosphere (Spruit 1977). Observations reveal that the TSI increases with the number of sun spots and therefore increases with magnetic activity. Whether the changes in TSI translate to changes in the fundamental solar properties (radius, Teff , and luminosity) is still a point of contention (Li & Sofia 2001; Foukal et al. 2006). Observationally, there is evidence to support the hypothesis that magnetic fields may be at the origin of the discrepancies. Studies have suggested that correlations exist between observed radius and Teff discrepancies and the intensity of particular emission features (López-Morales 2007; Morales et al. 2008). The emission features, collectively referred to as “magnetic activity,” are thought to be the result of magnetic energy being dissipated in the stellar atmosphere. This energy heats the very tenuous atmospheric layers. The chromosphere, heated to nearly 104 K, produces strong Hα (Young et al. 1989) and Ca ii H & K emission (Skumanich et al. 1975), while the stellar corona is heated to temperatures in excess of 106 K, generating significant emission at X-ray wavelengths. Radio emission is observed as well, correlated with the X-ray emission (Güdel et al. 1993). Additional, indirect evidence for existence of strong magnetic fields has also been observed. Low-mass stars are known to undergo energetic flares across the entire spectrum (see, e.g., Osten et al. 2005). These events are associated with magnetic activity on the Sun. Lowmass stars have also been monitored for periodic amplitude fluctuations in their light curves, thought to betray the presence of star spots. As the star rotates, different spot configurations on the stellar disc rotate into and out of view. Theoretically testing the magnetic hypothesis is difficult, owing to the inherently threedimensional nature of magnetic fields. Stellar evolution codes, being calculated in only a single dimension, therefore require a substantially simplified approach. Nevertheless, attempts have been made to model the effects of magnetic fields on low-mass stellar structure 21 (Lydon & Sofia 1995; D’Antona et al. 2000; Mullan & MacDonald 2001; Chabrier et al. 2007). We will review these approaches in the next section. 1.4 Magnetic Fields in Stellar Evolution The identification of the magnetic hypothesis made the inclusion of a magnetic field in lowmass stellar evolution calculations desirable. Two groups have developed different methods for including magnetic fields and the effects on convection (Mullan & MacDonald 2001; Chabrier et al. 2007). An additional ingredient was also proposed by Chabrier et al. (2007) to include the effects of photospheric spots. Here, we will briefly introduce the various methods so that they may be familiar to the reader. 1.4.1 Modified Adiabatic Gradient Placing a vertical magnetic field in a plasma helps to stabilize the plasma against convective instability (Thompson 1951; Chandrasekhar 1961). Knowing this, Gough & Tayler (1966) derived modifications to the Schwarzschild criterion that may be used in stellar evolution models to mimic the presence of a magnetic field. The Schwarzschild criterion is a local condition that stellar evolution models use to determine whether a location within the model transports energy primarily through radiation or convection. In a non-magnetic fluid the condition states that a fluid is stable against convection if ∇s − ∇ad < 0, (1.10) where ∇s is the temperature gradient (d ln T /d ln P ) of the fluid and ∇ad is the adiabatic temperature gradient. If a magnetic field is included, however, the right-hand-side of the 22 equation is greater than zero, ∇s − ∇ad < B2 , B 2 + 4πγP gas (1.11) where B is the magnetic induction, γ is the ratio of specific heats c P /cV , and P gas is the pressure due to the fluid or gas particles. This is one of several conditions that can be formulated depending on various assumptions (Gough & Tayler 1966). Hereafter, B will be referred to as the magnetic field strength. The new stability criterion advanced by Gough & Tayler (1966) has since been implemented in stellar evolution models of low-mass stars on the pre-main-sequence (D’Antona et al. 2000) and main sequence (Mullan & MacDonald 2001). The authors of the latter study have since used their models in a variety of situations, including testing if magnetic fields can prevent stars from being fully convective (Mullan & MacDonald 2001), looking at the influence of magnetic field on the lithium abundances of stars in β-Pic (MacDonald & Mullan 2010), and investigating how magnetic fields affect the structure of brown dwarfs (MacDonald & Mullan 2009; Mullan & MacDonald 2010). Additionally, the authors have attempted to quantify how finite electrical conductivity might alter the approach, which is crucial for work on brown dwarfs (MacDonald & Mullan 2009). 1.4.2 Reduced Mixing Length The second method to include magnetic field effects was proposed by Chabrier et al. (2007). Their approach is qualitatively different from that of Mullan & MacDonald (2001). Instead of looking at the stabilization of convection, Chabrier et al. (2007) attempted to mimic the development of cooling flows along magnetic flux tubes. These cooling flows would transport energy away from a convecting bubble, reducing the efficiency of convection (see Section 1.3.3). To mimic the reduction of convective efficiency the authors reduced the convec- 23 tive mixing length, αMLT . The method has been used by Chabrier et al. (2007) and Morales et al. (2010) to explore the effect of a magnetic field on models of low-mass stars in DEBs. Significant reductions to αMLT were required, but the authors in both papers were able to reconcile their models with the observations. The problem with the approach, however, is that it has no predictive power. There is no means of calibrating a reduction in αMLT to magnetic field strengths. Consequently, it is not possible to test the validity of these models. The ambiguity of whether the reduction in αMLT is the result of magnetic fields or naturally inefficient convection only further complicates the matter. 1.4.3 Star Spots In addition to the reductions in αMLT , Chabrier et al. (2007) introduced a means of accounting for the emergence of photospheric spots. They did this by reducing the total bolometric flux at the stellar surface (Spruit 1982a,b; Spruit & Weiss 1986). Spots were included by imagining a given fraction of the total surface flux is blocked by spots with some temperature contrast relative to an immaculate photosphere. Specifically, ¶ ¸ · µ Ss Ts 4 β= 1− S Teff (1.12) where S s /S is the fractional surface coverage by spots, Ts is the temperature within the active region, and Teff is the effective temperature of the immaculate (i.e., spot free) photosphere. The bolometric flux is then modified such that ¡ ¢ F = 1 − β F⋆ (1.13) with F and F⋆ representing the total flux and the immaculate photospheric flux, respectively. By combining the above star spot correction and a reduced αMLT formalism, 24 Chabrier et al. (2007) and Morales et al. (2010) were able to effectively inflate low-mass stellar radii and suppress effective temperatures. There are observational techniques developed to measure spot filling factors and temperature contrasts (O’Neal et al. 1998, 2004; O’Neal 2006; Catalano et al. 2002). Unfortunately, the large majority of stars in their samples are evolved stars. Two very active K-dwarfs were observed (O’Neal et al. 1998, 2004), but the extraction of spot properties requires accurate knowledge of the stellar reference spectra, in particular the Teff of their M-dwarf calibrators (O’Neal et al. 1998). Photometric color indices were used to assign M-dwarf Teff s (Bessell 1991) and appear to be several hundred degrees too cool. We determined the latter by cross-referencing their M-dwarf calibration stars with stars that have had their temperature determined using interferometry (Berger et al. 2006). Therefore, caution must be exhibited when benchmarking spot properties to the results of these studies. Including star spots in this fashion also raises several issues. First, by including spots with other mechanisms for convective suppression (either reduced αMLT or δMM ), there is an inherent risk that one might “double count” the effects of magnetic fields. Star spots are the physical manifestation of localized vertical magnetic fields suppressing convection. Therefore, methods that include the suppression or inhibition of convection and star spots conflate the actual ability of magnetic fields to suppress convection. Ultimately, star spots have a physical origin in the suppression of convection. This leads us to question why reductions in flux must be accounted for in “spots,” when methods like those described above ought to perform the same task. This leads into the second issue, which is that it is difficult to verify to what degree the total flux of a star is being bottled up by the presence of spots. Stellar evolution models are computed in 1-dimension, a severe restriction. Accounting for the 2- or 3-dimensional flow of heat around or near spots and active regions is extremely difficult. Thermal modeling of the regions near spots does suggest that a spot creates a bottleneck and traps flux in the 25 immediate region below it (Foukal et al. 1983). It appears that heat does not necessarily flow efficiently around spots. However, such reductions in flux observed at the surface of a star may be counteracted by the appearance of bright faculae, such as those observed on the Sun (Foukal et al. 2006). Small depressions near the stellar photosphere allow flux from deeper in the star to escape causing those small regions to appear brighter (Spruit 1977). This has been cited as the reason for the increase in TSI with sun spot number (Foukal et al. 2006). It is not clear if the solar-stellar comparison is completely valid, but sun spot data and the effects of solar activity on the solar properties should not be neglected. 1.4.4 A Need for New Models? With the availability of the magnetic field models described above, do we need another set of models using yet another approach? We feel that there is a need, predicated on the desire for accurate low-mass stellar models with strong predictive power. The example we have used throughout this chapter is that of exoplanet host star characterization. Currently available magnetic stellar models do not provide a complete package necessary for accurate exoplanet characterization. The models of Baraffe et al. (1998) are commonly used to predict the properties of low-mass stars under the assumption that magnetic fields are not required. Correcting for magnetic fields with the Baraffe et al. (1998) models has been explored, but only through the use of a reduced αMLT and the star spot parameterization (Chabrier et al. 2007; Morales et al. 2010). However, there is currently no means of assuming what value of αMLT is required, since no observable has been posited to correlate with the necessary αMLT . The variables needed in the star spot approach can be verified observationally, but must be used in combination with the reduced αMLT approach. On the other hand, the magnetic models of Mullan & MacDonald (2001) can be verified observationally. Their magnetic inhibition parameter, δMM , has the value of the magnetic 26 field strength, which is defined at the surface. Observations of stellar magnetic fields can then be used to validate the proposed magnetic field strengths (Reiners & Basri 2007, 2009; Donati & Landstreet 2009). However, the models advanced by Mullan & MacDonald (2001) have not been benchmarked against observations of low-mass stars without their magnetic perturbation, nor have they been compared to the BCAH98 standard models. Thus, the changes in radius as a function of magnetic field strength should be used with caution to correct the radius predictions of other model sets. We see a need for a set of magnetic models that can be trusted as accurate in all situations— magnetic and non-magnetic—across multiple mass regimes. The goal is to create a set of robust and versatile low-mass stellar models. To achieve this goal, we have opted to pursue the development of magnetic models within the framework of the Dartmouth stellar evolution code. 1.5 The Dartmouth Stellar Evolution Code The Dartmouth Stellar Evolution Program (DSEP),11 is one of the latest progeny of the Yale Rotating Evolution Code (Guenther et al. 1992). Since the code’s inception, the physics used in the models have been updated and improved. These additions and updates have been described extensively in the literature (Chaboyer & Kim 1995; Chaboyer et al. 2001; Bjork & Chaboyer 2006; Dotter et al. 2007, 2008). Below, we will summarize the physics used in the code that are pertinent for the present work and also discuss the calibration of the models. 11 http://stellar.dartmouth.edu/models/ 27 1.5.1 Physics The conditions in the optically thin atmosphere of low-mass stars precludes the use of gray atmosphere approximations and the use of radiative T(τ) relations (Allard & Hauschildt 1995; Chabrier & Baraffe 2000, and references therein). Instead, non-gray radiative transfer calculations are required to accurately describe the thermal structure of the atmosphere. We therefore define the surface boundary conditions of our models using the phoenix amescond theoretical atmospheres (Hauschildt et al. 1999a,b). While the validity of gray atmosphere approximations increases as the model mass increases, all of our models use the phoenix model atmospheres for consistency. The surface boundary condition is defined by the gas pressure at a pre-selected optical depth. The surface gas pressure is found by interpolating in pre-compiled tables that list the gas pressure as a function of log g and Teff at different metallicities. These tables are compiled using the atmosphere structures generated by the phoenix code. The interpolation quantities (log g , Teff , and metallicity) are defined by the interior structure model (i.e., DSEP). The choice of where to attach the model atmosphere to the interior model—the pre-selected optical depth that was alluded to—is an important one. Our experience indicates that attaching the atmosphere at T = Teff is reasonable for M & 0.2M⊙ (Dotter 2007). The optical depth is then set by the phoenix structures, which compute the optical depth at which T = Teff . Below that threshold mass it is imperative to attach the model atmosphere deeper into the interior (say τ = 100) because the temperature gradient in the outer layers is significantly non-adiabatic (Chabrier & Baraffe 2000; Dotter 2007). Arguably, the most crucial piece of physics in stellar evolution codes is the equation of state (EOS). In general, the EOS is mass dependent, with the level of non-ideal contributions gaining importance towards lower masses. There are two primary EOSs used by DSEP: an analytical EOS and the FreeEOS, a numerical EOS created by Alan Irwin.12 12 http://freeeos.sourceforge.net 28 Model comparisons have shown that models of main seqeuence stars with masses above 0.8M ⊙ can be accurately calculated using the analytical EOS (Prather 1976; Dotter 2007). This analytical EOS takes into account the particular ionization state of the stellar plasma using the Saha ionization equation. Furthermore, ion-charge shielding is included via a Debye-Hückel correction (Chaboyer & Kim 1995). The EOS for stars below 0.8M⊙ must include a variety of non-ideal contributions. For this job, DSEP uses the FreeEOS in the EOS4 configuration (Dotter 2007; Dotter et al. 2008). We selected the FreeEOS for three primary reasons. First, it includes several non-ideal contributions to the EOS, such as Coulomb interactions and pressure ionization, that become important in the dense plasma of low-mass stars. Second, the FreeEOS calculates the EOS for hydrogen, helium, and eighteen heavier elements as opposed to only calculating the EOS for hydrogen and helium. One can therefore alter the specific solar abundance reference data as well as the particular stellar metallicity. Finally, the FreeEOS may be called directly from within the stellar evolution code. This feature avoids the need to interpolate within EOS tables, thereby minimizing numerical errors. Opacities are an integral component within stellar evolution models. Radiative opacity governs, almost entirely, the size of the convective envelope in low-mass stars. The efficiency of energy transport, in turn, largely determines the star’s physical properties. Radiative opacities used by DSEP are temperature dependent. The OPAL opacities are used at temperatures above 104 K (Iglesias & Rogers 1996) while the Ferguson et al. (2005) opacities are used at lower temperatures. Contributions to the opacity also arise from the conduction of heat through the mobility of free electrons in the stellar plasma (Hubbard & Lampe 1969; Canuto 1970; Sweigart 1973). Updates to the standard electron thermal conduction calculations were presented by Cassisi et al. (2007) and have been included (see Appendix B). However, they a negligible effect on the main-sequence evolution of low-mass stars. As stellar mass increases there is a decrease in the size of the surface convective enve- 29 lope. Accompanying this decrease is an increase in the importance of diffusion physics, which occurs in the radiative zone below the convection zone boundary (Michaud et al. 1984). Diffusion physics governs the separation of chemical species through processes such as gravitational settling and thermal diffusion. DSEP follows the diffusion of helium and heavy elements following the formulation of Thoul et al. (1994), unless the star is fully convective. Fully convective stars are assumed to be completely and homogeneously mixed throughout since the convective timescale dominates the diffusion timescale (Michaud et al. 1984). We note that an extra diffusion term due to turbulent mixing is also included (Proffitt & Michaud 1991; Richard et al. 2005). This extra mixing is described in Appendix B and replaces the method of partially inhibiting diffusion (Chaboyer et al. 2001). Many of the stars in low-mass DEBs are likely to have their rotation periods synchronized with their orbital periods. This means they are likely to be rotating rapidly (v sin i > 5 km s−1 ). It is then natural to wonder whether we should consider the effects of rotational deformation in our models. As a first approximation, we applied Chandrasekhar’s analysis of slowly rotating polytropes to our DEB sample. Chandrasekhar (1933) derived an expression for the stellar oblateness analytically. He defined slowly rotating to be when χ≡ ω2 ≪1 2πGρ c (1.14) where ω is the stellar angular velocity and ρ c is the central mass density. As an example we will assume rather extreme values for a 0.80M⊙ star under consideration. With a rotational velocity of 102 km s−1 and a radius of 0.80R ⊙ , this star would have ω ≈ 10−4 . If the star has ρ c ∼ 50 g cm−3 , it will have χ ∼ 10−3 . Based on the assumed polytropic index, n , Chandrasekhar derived that the deviation from 30 sphericity (oblateness) could be approximated as F≡ r eq − r pole r eq 5.79χ = 9.82χ 41.81χ for n = 1.5 for n = 2.0 for n = 3.0 where F is the oblateness and r eq and r pole are the equatorial and polar radius, respectively. We approximated the polytropic index for each star using the interior density profile predicted by DSEP. The density profile for low-mass stars only slightly deviates from the polytrope prediction over the inner 98% of the star (by radius). Specifically, we found that below 0.4M⊙ , the models were best represented by an n = 1.5 polytrope. Above 0.4M⊙ but below 0.65M⊙ , an n = 2.0 polytrope was appropriate. For all masses greater than 0.65M⊙ , we defaulted to assuming an n = 3.0 polytropic index. The deviation from sphericity for our imagined case above would then be around 5%. Note that these conditions are exaggerated. We’ll see in Chapter 4 that for a main sequence star, χ < 10−4 is more typical. We therefore feel justified in neglecting rotation. At times we will return to this calculation to re-justify our assumption with more specific stellar properties. The final pieces of physics worth mentioning are the updated nuclear reaction rates and the introduction of the convective overturn time. We have updated the previous set of nuclear reaction rates (Adelberger et al. 1998; Dotter 2007) with the latest nuclear reaction rate recommendations (Adelberger et al. 2011). Although modifications have been made to the p-p chain reaction rates, they have a negligible impact on low-mass main sequence evolution. Further discussion of the updated rates are provided in Appendix B. Lastly, DSEP now computes the “local” convective overturn time at each model time step (Kim & Demarque 1996). As with other physics that have been added or updated, details are located in Appendix B. 31 Table 1.1 Solar calibration parameters. Parameter Adopted Model Age (Gyr) 4.57 ··· M ⊙ (g) 1.989 × 1033 R ⊙ (cm) 6.9598 × 1010 R bcz /R ⊙ 0.713 ± 0.001 L ⊙ (erg s−1 ) (Z /X )surf Y⊙, surf † ··· Reference Bahcall et al. (2005) IAU 2009† log(R) = 8 × 10−5 Neckel (1995); Bahcall et al. (2005) 3.8418 × 1033 log(L ) = 2 × 10−4 0.714 Basu & Antia (1997); Basu (1998) 0.0231 0.0230 Grevesse & Sauval (1998) 0.2485 ± 0.0034 0.2455 Basu & Antia (2004) Bahcall et al. (2005) http://maia.usno.navy.mil/NSFA/NSFA_cbe.html 1.5.2 Solar Calibration The primary input variables for stellar evolution models are defined relative to comparable solar values. These input variables include the stellar mass, the initial mass fractions of helium and heavy elements (Yi and Zi , respectively), and the convective mixing length parameter, αMLT . The latter defines the length scale of a turbulent convective eddy relative to the pressure scale height. Since they are all defined relative to the solar values, we must first define what constitutes the Sun for the model setup. To do this, we require a 1 M⊙ model to reproduce the solar radius, luminosity, radius to the base of the convection zone, and the solar photospheric (Z /X ) at the solar age (4.57 Gyr; Bahcall et al. 2005). These values are presented in Table 1.1. We use an iterative procedure to find the optimal combination of model input variables that yield the solar properties at the solar age. First, we supply the model with an initial guess for X i , Zi , and αMLT . Four models are then automatically generated. One of these is the “base model” where the user-supplied values are used. The three additional models are run with one of the input variables perturbed slightly. Second, we calculate the difference between the actual solar radius, luminosity, and photospheric (Z /X ) and those predicted by the base model. Since the radius and luminosity 32 of the base model are quoted in solar units, this is straightforward. Note that the solar photospheric (Z /X ) is dependent on the heavy element composition adopted. This process yields δR , δL , and δ(Z /X ), where we have defined R = R/R⊙ and L = L/L ⊙ . Ultimately, we seek to identify the set of X , Z , and αMLT that produces a model where δR , δL , and δ(Z /X ) are all within some pre-defined tolerance of zero. This tolerance was set to 4×10−4 in the appropriate units. It is unlikely that the first base model will yield a correct solar calibration. We therefore want to improve upon our guess, which we do using the three additional models with perturbed input variables. The three perturbed models allow us to compute how the predicted radius, luminosity, and photospheric (Z /X ) change as a function of the individual input variables. For each perturbed model, we compute the difference between the model radius, luminosity, and photospheric (Z /X ) at the solar age relative to the base model. We then setup a set of three linearly independent equations: ¶ µ ¶ µ ¶ ∂R ∂R ∂R δX + δZ + δαMLT δR = ∂X ∂Z ∂αMLT µ (1.15) µ ¶ ¶ µ ¶ µ ∂L ∂L ∂L δL = δαMLT δX + δZ + ∂X ∂Z ∂αMLT (1.16) ¶ µ ¶ µ ¶ ∂(Z /X ) ∂(Z /X ) ∂(Z /X ) δX + δZ + δαMLT , δ(Z /X ) = ∂X ∂Z ∂αMLT (1.17) µ where the partial differentials in the above equations are determined by comparing the perturbed models to the base model, as described above. We then solve the set of linear 33 equations in matrix form ³ ´ ³ ´ ³ ´ −1 ∂R ∂R δX ∂R ∂Z ∂αMLT ∂X ³ ´ ³ ´ ³ ´ ∂L ∂L δY = ∂L ∂Z ∂αMLT ∂X ³ ´ ´ ³ ´ ³ ∂(Z /X ) ∂(Z /X ) ∂(Z /X ) δαMLT ∂X ∂Z ∂α MLT δR · δL . δ(Z /X ) (1.18) Solving this system of equations yields δX , δZ , and δαMLT . These values are used to iterate on the values for our initial guess. This procedure is repeated until the code converges on a solution for δR , δL , and δ(Z /X ) that are within the defined tolerance. The final model is considered calibrated and the code prints the final values of X i , Zi , and αMLT . Results of the calibration procedure using the model physics described in Section 1.5.1 are give in Table 1.1. The final set of input variables was X i = 0.70625, Yi = 0.27491, Zi = 0.01884, and αMLT = 1.938. The adopted heavy element abundance used throughout this work is the solar photospheric abundance determined by Grevesse & Sauval (1998). 1.6 Thesis Outlook The body of work contained in this thesis represents our analysis and approach to addressing the low-mass stellar mass-radius problem. Our aim from the outset was to develop a set of self-consistent magnetic stellar evolution models using methods independent from previous investigations (Mullan & MacDonald 2001; Chabrier et al. 2007). However, along the way we discovered that other factors had not been adequately addressed in previous studies. This realization had its genesis after the results from Kepler began to trickle in with the announcement of a new DEB KOI-126 (Carter et al. 2011). We present the theoretical modeling effort of KOI-126 in Chapter 2. Our investigation in Chapter 2 led to an interest in the application of stellar interior structure constants in stellar evolution, as seen in Chapter 3. 34 Shortly after the announcement of KOI-126, several more low-mass DEBs were discovered and characterized (e.g., Kraus et al. 2011; Doyle et al. 2011). Chapter 4 concerns using the enlarged DEB data set to look at the different hypotheses suggested for the mass-radius problem. The following three chapters present the development, testing, and initial results of a new set of magnetic stellar evolution models. Our models are introduced and tested in Chapter 5. A detailed look at modeling DEB stars is presented in Chapters 6 and 7. We not only study whether or not magnetic models can sufficiently inflate stellar radii, but also discuss how physically realistic our models are so that we may attempt to enhance their predictive power. Finally, in Chapter 8 we present the current state of the low-mass star field and ideas for future projects involving our magnetic models. We also include two appendices that do not present any significant scientific content. Appendix A provides a short user’s guide for the Dartmouth Magnetic Evolutionary Stellar Tracks And Relations, DMESTAR. Appendix B gives a summary of additional model physics introduced to DSEP during the course of the development of DMESTAR. The intention of this appendix is to provide a reference for new model input variables. 35 Chapter 2 Accurate Low-Mass Stellar Models of KOI-126 This chapter presents and expands on material that appeared in The Astrophysical Journal (2012, 740, L25) in an article titled “Accurate Low-Mass Stellar Models of KOI-126.” 2.1 Introduction Prior to 2011, only nine low-mass, detached eclipsing binary (DEB) systems with precisely measured masses and radii were known (Popper 1997; Torres & Ribas 2002; Ribas 2003; López-Morales & Ribas 2005; Bayless & Orosz 2006; López-Morales et al. 2006; López-Morales & Shaw 2007; Morales et al. 2009b,a). Of those nine, only CM Draconis (Lacy 1977) contains a star thought to lie below the fully convective boundary (M . 0.35M⊙ ). In fact, it contains two. CM Draconis displays radii that are discrepant with stellar evolution models (Morales et al. 2009a). However, because it is the only system for which we have precisely measured masses and radii, it is unclear whether the models are insufficient or whether CM Draconis is peculiar. This scenario changed with the launch of the NASA’s Kepler Space Telescope (hereafter 36 Kepler). Kepler was launched in May of 2009 with the express purpose of constraining the the frequency of Earth-sized planets orbiting solar-type stars in the habitable zone (Borucki et al. 2010). To acheive this goal, Kepler was designed to continuously stare at the same galactic location—in the direction of the constellation Cygnus—and photometrically monitor the brightness of over 150,000 stars for transiting exoplanets. High precision photometry of each star is acquired with a 30 minute cadence, 1 providing extremely detailed light curves that are continuous over nearly the entire lifetime of the mission. The Kepler mission created a revolution in identifying DEB candidates—or any stars showing variability, for that matter—resulting in a catalog of 1 879 DEB candidates, 231 of which are potentially low-mass, from only 44 days of data (Prša et al. 2011; Coughlin et al. 2011). Among these candidates was a system that displayed a highly irregular pattern of eclipses. The system, KOI-126, was announced by Carter et al. (2011) as a triply eclipsing, hierarchical triple system with two low-mass, fully convective stars. KOI-126 has a G star primary2 with two low-mass, fully convective companions (Carter et al. 2011). The two low-mass stars (KOI-126 B and C) are in a tight 1.77 day orbit that is orbiting the more massive primary star (KOI-126 A) every 33.92 days on a fairly eccentric path (e = 0.3043 ± 0.0024). The authors were able to derive fundamental properties for all three stars using a photometric-dynamical model that finds the most probable stellar and orbital properties by reconstructing the detailed light curve (see supplementary material for Carter et al. 2011). They found that M A = 1.347 ± 0.032M⊙ , R A = 2.0254 ± 0.0098R ⊙ , M B = 0.2413 ± 0.0030M ⊙ , R B = 0.2543 ± 0.0014R ⊙ , MC = 0.2127 ± 0.0026M ⊙ , and RC = 0.2318 ± 0.0013R ⊙ . Spectroscopy of the more massive primary reveiled it has an effective temperature of 5 875±100 K with a slightly super-solar metallicity of [Fe/H] = +0.15±0.08. Lastly, a non-zero eccentricity of the (B, C) system (e = 0.02234) allowed the authors to 1 2 Short cadence observations of 1 minute are also available for select targets. In truth, the G star is actually the tertiary, but was deemed the primary based on its relative brightness. 37 Table 2.1 Properties of the KOI-126 hierarchical triple system. Property Mass (M⊙ ) Radius (R⊙ ) Teff (K) Relative Flux [Fe/H] (dex) KOI-126 A KOI-126 B KOI-126 C 1.347 ± 0.032 0.2413 ± 0.0030 0.2127 ± 0.0026 2.0254 ± 0.0098 0.2543 ± 0.0014 0.2318 ± 0.0013 1.0 (3.26 ± 0.24) × 10−4 (2.24 ± 0.48) × 10−4 5875 ± 100 ··· +0.15 ± 0.08 ··· Properties of the (A, (B, C)) binary orbit Period (day) 33.9214 ± 0.0013 Eccentricity 0.3043 ± 0.0024 Semi-major axis (R ⊙ ) 54.63 ± 0.37 Properties of the (B, C) binary orbit Period (day) 1.76713 ± 0.00019 Eccentricity 0.02234 ± 0.00036 Semi-major axis (R ⊙ ) 4.726 ± 0.019 place a weak constraint on the apsidal motion constant for the two low-mass stars. They determined that it is below 0.6 at the 95% confidence level. The properties of the individual stars and the orbits are summarized in Table 2.1. Carter et al. (2011) derived an age for the system of 4 ± 1 Gyr by fitting the properties of KOI-126 A to theoretical Y2 isochrones (Demarque et al. 2004). They then compared theoretical isochrones from Baraffe et al. (1998, henceforth BCAH98)3 to observations of the low-mass stars in the mass-radius plane to investigate the low-mass, stellar mass-radius relationship. The models were seen to underpredict the radii of the two low-mass stars by 2% – 5%, even though several ages (1, 2, 4, and 5 Gyr) were adopted. It is important to note that no super-solar metallicity isochrones are available for the BCAH98 models. This is in part due to limitations of their equation of state. Carter et al. (2011) suggested that 3 Two separate isochrone sets were used due to specialization of each set in the different mass regimes. 38 the observed radius discrepancy could be the result of both the super-solar metallicity and possible magnetic activity of the system. Here, we test the hypothesis proposed by Carter et al. (2011) that metallicity underlies the observed discrepancies between the BCAH98 models and the observations of the two lowmass stars. Results are presented in Section 2.2 of theoretical stellar modeling of the KOI126 components using the Dartmouth Stellar Evolution Program, which was described in Chapter 1.5. If the metallicity hypothesis is confirmed, it would represent the first accurate modeling of low-mass, fully convective stars using standard stellar evolution models. In Section 2.2.1 we estimate the age of the stellar system using the Dartmouth models to provide consistency between the properties of the primary and the two low-mass stars. We then compare models of the two low-mass stars to the observations in the mass-radius plane in Section 2.2.2. To further improve the comparison between the models and observations we compute the relative fluxes expected for the theoretical models in the Kepler bandpass in Section 2.2.3 and derive the theoretical apsidal motion constants in Section 2.2.4. Finally, in Section 2.3, we discuss the implications of this study with regards to the only other well known low-mass, fully convective, detached eclipsing binary, CM Draconis. 2.2 Results 2.2.1 Stellar Age The age of the system was constrained by evolving a 1.347M⊙ model with a metallicity of [Fe/H] = +0.15 and a solar calibrated αMLT and matching the model radius with the observed radius of KOI-126 A. This is illustrated in Figure 2.1 by the black, solid line. We derive an age for KOI-126 of 4.1 ± 0.6 Gyr, consistent with the age derived by Carter et al. (2011). The uncertainty in the age is dominated by the uncertainty in the observed mass and metal- 39 2.4 1.347 M⊙, [Fe/H] = +0.15 1.315 M⊙, [Fe/H] = +0.23 1.379 M⊙, [Fe/H] = +0.07 Radius (R⊙) 2.2 2.0 1.8 1.6 1.4 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 Age (Gyr) Figure 2.1: Stellar evolution mass tracks used to constrain the age of KOI-126 A. The lightblue horizontal band represents the observed radius with the associated 1σ uncertainty. licity of the system. Figure 2.1 shows two additional mass tracks with mass-metallicity combinations chosen to produce the largest variations in the possible age of the primary while remaining with the observational constraints. The observational uncertainty in the primary’s radius contributes little to the overall uncertainty budget as the star appears to be at an age where its radius is undergoing rapid evolution. We also examined the effect of changing the reference temperature for turbulent diffusion, which alters the chemical abudance profile within the star. It was found to play a negligible role in the age determination, even when turbulent diffusion was completely removed. 2.2.2 Mass-Radius Relation The primary results of this chapter are demonstrated in Figure 2.2. DSEP accurately reproduces the observed radius for each of the low-mass stars at 4.1 Gyr with [Fe/H] = +0.15. 40 0.26 Radius (R⊙) 0.25 0.24 0.23 DSEP, [Fe/H] = +0.00 DSEP, [Fe/H] = +0.15 BCAH98, [Fe/H] = +0.00 0.22 0.20 0.21 0.22 0.23 0.24 0.25 Mass (M⊙) Figure 2.2: Mass-radius relationship as defined by CM Draconis (light gray – points) and KOI-126 (maroon – points). Overlaid are 4.1 Gyr theoretical isochrones from DSEP with [Fe/H] = 0.0 (black – solid) and [Fe/H] = +0.15 (light blue – dashed) and from BCAH98 with [Fe/H] = 0.0 (gray – dash-dotted). We find that for model masses4 M2 = 0.2410M⊙ and M3 = 0.2130M⊙ the predicted radii from DSEP are R2 = 0.2544R ⊙ and R 3 = 0.2312R ⊙ . The relative error between the model and observed radii is less than 0.3%. At solar metallicity, the models predict radii that are approximately 1% smaller than those predicted by the super-solar metallicity models. Figure 2.2 shows this 1% radius deflation caused by lowering the model metallicity to a solar value is enough to destroy the agreement between the models and the observations. The predicted model radii appear to be robust. Artificially reducing the mixing length (αMLT = 1.00), as is typically done for low-mass stellar models (Chabrier & Baraffe 1997; Baraffe et al. 1997; Baraffe et al. 1998), introduces radius variations on the order of 0.5%. Similarly, fitting the atmosphere to a deeper point in the stellar envelope (at τ = 100 instead 4 Observational component indicies j = {A, B, C } map directly to i = {1, 2, 3}, which are the equivalent model component indicies. 41 of where T = Teff ) changes the model radius by less than 0.5%. The 1% reduction in the model radius caused by reducing the metallicity from [Fe/H] = +0.5 to [Fe/H] = 0.0 is not as large as the 2% – 5% difference noted by Carter et al.. This is because low-mass, solar metallicity DSEP models are approximately 1% larger than those of BCAH98, at a given mass. The difference is likely a consequence of the equation of state (EOS) adopted by each group. Whereas the EOS used by BCAH98 is calculated for a pure hydrogen/helium plasma, FreeEOS calculates the EOS for an arbitrary metal abundance. Therefore, our models can be more reliably calculated above solar metallicity. This effect of metals on the EOS was estimated by Chabrier & Baraffe (1997) to create radius changes of about 1% – 2%, consistent with our simulations. To explicitly test the influence of metals on the EOS, and on the model radii, we generated DSEP models with the Saumon et al. (1995, hereafter SCVH95) EOS in an attempt to reproduce the BCAH98 model radii predictions. In addition, models were computed using the OPAL 2001 EOS (Rogers & Nayfonov 2002). The OPAL equation of state accounts for heavy elements, but using a different numerical technique and different physics than FreeEOS. Results are compiled in Table 2.2 and a detailed look at the internal density profile produced by each EOS is shown in Figure 2.3. Models using the SCVH95 EOS produced radii within 0.5% of the BCAH98 models at the same mass and composition. Those models using the OPAL EOS had radii intermediate between the FreeEOS predictions and the SCVH95 EOS. Table 2.2 Radii (R ⊙ ) of low-mass stars derived using DSEP and BCAH98 at 4.1 Gyr with [Fe/H] = 0.0. DSEP BCAH98 Mass FreeEOS OPAL SCVH SCVH 0.20 M⊙ 0.2180 0.2174 0.2155 0.2168 0.2591 0.2588 0.2578 0.2567 0.25 M⊙ 42 7.0 OPAL ’01 6.5 SCVH ’95 FreeEOS log10(T) [K] 6.0 0.25 M⊙ 5.5 0.15 M⊙ 5.0 4.5 4.0 3.5 -6 -4 -2 0 -3 log10(ρ) [g cm ] 2 Figure 2.3: Comparison of three popular equations of state. 2.2.3 Relative Fluxes Stellar effective temperature is another property against which to compare the predictions of stellar models. Measuring the effective temperature of an M-dwarfs, however, is a nontrivial task beset by numerous complications and uncertainties. The molecule-ridden spectrum presented by most M-dwarfs masks the presence of the background continuum and any line features present in the spectrum are typically blends of multiple atomic and molecular species. It is not surprising, then, that while Carter et al. (2011) could determine the effective temperature of KOI-126 A to be 5 875±100 K from spectroscopy they were unable to measure the effective temperature of the two M-dwarfs. However, the dynamical/photometric model used by Carter et al. (2011) did yield the flux of each M-dwarf relative to the primary. In fact, this is a requirement to be able to reconstruct the precise lightcurve eclipse profiles. Their results showed that f B / f A = (3.26±0.24)×10−4 and fC / f A = (2.24 ± 0.48) × 10−4 . 43 Normalized Flux 1.0 Kepler Filter Transmission Curve 0.8 0.6 0.4 0.2 0.0 4000 5000 6000 7000 8000 9000 10000 7000 8000 9000 10000 Normalized Flux 1.0 0.8 Kepler Filter Transmission Curve 0.6 0.4 0.2 0.0 4000 5000 6000 Wavelength (Å) Figure 2.4: The Kepler filter transmission profile overplotted on synthetic spectra from the phoenix ames-cond model set. The synthetic spectra are for a typical A star (log g = 4.1, Teff = 9 600 K, and [M /H] = 0.0; blue, solid line) and an mid-M-dwarf that is similar to KOI-126 B and C (log g = 5.0, Teff = 3 500 K, and [M /H] = +0.15; maroon solid line). To compare our stellar models, we derived a synthetic color-Teff transformation for the Kepler bandpass. This was performed by convolving the Kepler spectral response function provided in the Kepler Instrument Handbook5 (Van Cleve & Caldwell 2009) with synthetic spectra created using the phoenix ames-cond model atmosphere code. The Kepler filter profile is illustrated in Figure 2.4 with two synthetic spectra computed using the phoenix online simulator.6 The two synthetic spectra are reprsentative of a typical A star (log g = 4.1, Teff = 9 600 K, and [M /H] = 0.0) and a mid-M-dwarf similar to KOI-126 B and C (log g = 5.0, 3 500 K, and [M /H] = +0.15). Given the properties ([Fe/H], luminosity, Teff ) of the interior structure models for each KOI126 star, we were able to predict the flux ratio for each of the two low-mass stars relative to KOI-126 A. We found f 2 / f 1 = 4.77 × 10−4 and f 3 / f 1 = 3.71 × 10−4 . These results are differ5 6 See also http://keplergo.arc.nasa.gov/CalibrationResponse.shtml http://phoenix.ens-lyon.fr/simulator/index.faces 44 ent from the observations at the 6σ and 3σ level, respectively. In addition to transforming directly from model properties to the fluxes in the Kepler bandpass, we used an alternative, semi-empirical method. The semi-empirical method relied on transforming the model properties into fluxes in the Sloan Digital Sky Survey (SDSS) g and r bandpasses and then using an empirical relation to convert to the flux in the Kepler bandpass. The empirical relation is available on the Kepler Guest Observer website.7 SDSS fluxes for our models were calculated using a synthetic color-Teff transformation (Dotter et al. 2008). The semi-empirical transformation yielded flux ratios of f 2 / f 1 = 3.07×10−4 and f 3 / f 1 = 2.36×10−4 . Curiously, these flux ratios are consistent with those measured by Carter et al. (2011). Diagnosing why the flux ratios are discrepant for the purely theoretical transformation is difficult. There could be two possible reasons for the errors: (1) our models over predict the M-dwarf fluxes, meaning the Teff s we derive are too hot, or (2) the flux of the primary is underpredicted. Addressing the first reason would likely require us to adopt a new set of theoretical atmospheres, such as the updated phoenix bt-settl models (Allard et al. 2001, 2011) and is beyond the scope of this chapter. Instead, we elect to examine the possibility that we underpredict the flux of the primary. In an effort to increase the flux of the primary, we varied several properties of the primary star. Specifically, we considered changes in the stellar mass, metallicity, and the amount of convective core overshoot (CCO). Note that we still required the models to be consistent with the observational data. When considering the largest possible mass (1.379M⊙ ), lowest metallicity (+0.07), and a relatively high amount of CCO, the flux ratios were reduced to within 2σ of the accepted values. As with the age determination, the tight observational constraint imposed on the radius of the primary made the range of relevant radii insignificant in deriving our results. 7 See http://keplergo.arc.nasa.gov/CalibrationZeropoint.shtml 45 Table 2.3 Apsidal motion constants (k 2 ) for fully convective stars with [Fe/H] = -0.50, 0.0, and +0.15. Values are quoted at 7 different ages (in Gyr). Mass (M⊙ ) 0.15 .. . .. . 0.20 .. . .. . 0.25 .. . .. . 0.30 .. . .. . [Fe/H] 0.5 1.0 2.0 3.0 4.0 5.0 10.0 -0.50 0.1561 0.1561 0.1561 0.1561 0.1560 0.1560 0.1560 0.00 0.1560 0.1560 0.1560 0.1560 0.1560 0.1560 0.1559 +0.15 0.1557 0.1557 0.1557 0.1557 0.1556 0.1556 0.1556 -0.50 0.1523 0.1523 0.1522 0.1522 0.1522 0.1522 0.1521 0.00 0.1521 0.1521 0.1521 0.1521 0.1521 0.1520 0.1519 +0.15 0.1518 0.1517 0.1517 0.1517 0.1517 0.1517 0.1516 -0.50 0.1499 0.1498 0.1498 0.1498 0.1498 0.1497 0.1496 0.00 0.1496 0.1496 0.1496 0.1496 0.1495 0.1495 0.1494 +0.15 0.1496 0.1496 0.1496 0.1495 0.1495 0.1495 0.1493 -0.50 0.1482 0.1482 0.1481 0.1481 0.1481 0.1480 0.1479 0.00 0.1480 0.1480 0.1479 0.1479 0.1478 0.1478 0.1477 +0.15 0.1479 0.1479 0.1479 0.1478 0.1478 0.1477 0.1476 2.2.4 Apsidal Motion Constant By the end of the nominal Kepler mission, Carter et al. (2011) predict they will know the mean apsidal motion constant of the two low-mass stars with a relative precision of nearly 1%. From our low-mass stellar models, we can predict an apsidal motion constant for each star, quantifying the degree to which the mass within the star is centrally concentrated. To do this we followed the prescription of Kopal (1978) and solved Radau’s equation with j = 2, a ¡ ¢ ¢ d η 2 6ρ(a) ¡ η2 + 1 + η2 η2 − 1 = 6 + 〈ρ〉 da (2.1) where η 2 ≡ η 2 (a) = 46 a d ǫ2 ǫ2 d a (2.2) is the value of a particular solution to Radau’s equation, which is related to the stellar deviation from sphericity, ǫ2 . Additionally, ρ(a) is the density of the plasma and 〈ρ〉 is the average density. In the above equations, a is the radial distance to an equipotential surface (a = r when the surface is spherically symmetric). We used a fourth-order Runge-Kutta integration method to solve Radau’s equation to obtain a particular solution at the surface of each star. Having determined η2 (R), we were able to determine the apsidal motion constant since k2 = 3 − η 2 (R) 4 + 2η 2 (R) (2.3) As a check on our apsidal motion constant integrator, we generated polytropic models characterized by the polytropic constant n = 1.0 and 1.5 and compared the results of our code with the results of Brooker & Olle (1955), who tabulated a large number of apsidal motion constants for polytropes with various n values. The apsidal motion constants derived from the interior structure of our models for KOI-126 B and C are k 2, 2 = 0.1499 and k 2, 3 = 0.1512. This suggests our models are slightly less centrally condensed than an n = 1.5 polytrope, which is characterized by k 2 = 0.1433. In fact, the run of density and pressure for our models indicate our models are best described by a polytrope with n ∼ 1.45. A set of theoretical apsidal motion constants for fully convective stars, generated using DSEP, is given in Table 2.3 and may be compared to future observations. Finally, we note the effects of rotation on the derived k 2 values are negligible. Employing the formulae of Stothers (1974), we find that rotation affects our k 2 values at the 0.02% level for stars rotating with a period of 1.7 days. 2.3 Discussion Low-mass stars below the fully convective boundary are a wonderful tool to test fundamental physics. Their low mass affords theorists a stable, long lived (τage > 1011 yr) laboratory 47 with which to test the physics incorporated in the models. These fully convective stars have relatively simple structures and uncertainties in the opacities, surface boundary conditions, and the treatment of convection have relatively small effects on the predicted properties of the models (see Section 3.2 of Dotter 2007). Modelers are relieved of having to specify a free parameter since the internal structure is largely insensitive to the prescribed mixing length. The age of these stars is also of little consequence, as the radii of low-mass stars is nearly constant over their main sequence lifetime. These effects imply a unique mass-radius relation for stars of a given composition below the fully convective boundary. The discovery of two new data points in this regime indicate the mass-radius relation is not unique and that there is significant dispersion amongst the data. Although we find agreement between our models and the stars in KOI-126, discrepancies are still seen between our models and the components of CM Draconis (hereafter CM Dra; Morales et al. 2009a). This fact is visible in Figure 2.2. Comparing KOI-126 and CM Dra, we notice that they are strikingly similar. Both systems have low-mass components with nearly identical masses, estimated ages, orbital periods (KOI-126: 1.77 d; CM Dra: 1.27 d), and both systems are thought to be tidally synchronized, though not necessarily circularized. Their similar characteristics suggest the M-dwarfs in both systems should be very much alike. Apart from the differences caused by chemical composition, the stars should have nearly identical radii. Not only do they not, but we observe the opposite relationship of what is theoretically expected. CM Dra has a sub-solar metallicity (Viti et al. 1997, 2002; Morales et al. 2009a) and should therefore possess smaller radii than KOI-126. Figure 2.2 indicates that the stars in CM Dra actually have larger radii than those in KOI-126. It is possible that the KOI-126 system has been caught in a period of inactivity, similar to a solar minimum. CM Dra appears to have undergone such a period of quiescence in 2000. Morales et al. noted that no corrections due to starspots were needed in the analysis of the light curve data from that year. If KOI-126 B and C are in a magnetically quiescent state, it 48 is possible that their radii would not show signs of starspots or inflated radii. We note that starspots would likely be of negligible importance since variations due to spots would be reduced to noise amongst the signal of KOI-126 A. We find it encouraging that DSEP is able to predict the mass-radius relation suggested by KOI-126 B and C. Although, more data points are needed to allow for a more complete understanding of the reliance of the mass-radius relation on physics (standard and nonstandard) incorporated in current models. It is clear from this work that no standard stellar evolution model can simultaneously fit both CM Dra and KOI-126 and that progress on non-standard stellar evolution models will be required to fit CM Dra. As suggested by numerous authors, magnetic activity is likely the culprit and must be incorporated into the next generation of models. However, the effects of a magnetic field on the interior structure of stars has previously been considered as a reduction in the prescribed mixing length, mimicking the reduction in convective efficiency that should accompany the presence of a magnetic field (Chabrier et al. 2007). With this and the discussion presented at the beginning of this section in mind, it is not clear how magnetic activity would affect stars below the fully convective boundary. Self-consistent magnetic stellar models should help lend insight into the discrepancies with CM Dra. Concerning the relative flux discrepancies observed between the purely theoretical transformation and the photometric models of Carter et al. (2011), the color-Teff transformations are naturally suspect. The theoretical transformations might not provide a fully accurate transformation to the observational plane. In the low-mass regime, opacities are complicated by the formation of molecules, which have presented previous problems for low-mass model atmospheres (Delfosse et al. 2000). This is particularly relevant since the Kepler bandpass is very broad and covers a large portion of the optical spectrum, where molecular opacities are the most uncertain for M-dwarfs. Though the SDSS g and r filters provide nearly the same wavelength coverage, the fact that they are two narrow bands instead of single broad band 49 may explain why we find better agreement using the semi-empirical method. However, we must be cautious with this result as the systematic errors are not well constrained for the transformation from SDSS magnitudes to a Kepler magnitude. It must also be noted that the flux of the primary star is sensitive to the details of CCO. We observed that increasing the amount of CCO brought the purely theoretical fluxes closer to the observational values. More investigation will be required to accurately diagnose the discrepancies. Finally, we note that the determination of the apsidal motion constant will provide a crucial test of low-mass stellar evolution models. Morales et al. found k 2 ≈ 0.11 using the BCAH98 models, whereas our models predict a larger value of approximately 0.15. The difference between these two values is directly attributable to the EOS, which determines the run of density within a stellar model. If Carter et al. are able to accurately derive the apsidal motion constant to within 1%, it will provide a stringent benchmark against which to test the interior physics of low-mass stellar evolution models. 2.4 Summary In their discovery paper, Carter et al. reported that the triply eclipsing hierarchical triple KOI-126 appeared to support the mounting evidence that current standard low-mass stellar models are unable to reproduce the observed mass-radius relation. However, we have generated stellar models and theoretical isochrone tracks using the Dartmouth Stellar Evolution Program and find our model radii agree with the observations. Combining the KOI-126 measurements with previous observations of the low-mass binary system CM Dra, we find that the dispersion in the observed fully convective mass-radius relation is significant and stands in contrast to theoretical predictions. The fact that CM Dra, a system with sub-solar metallicity, lies on the super-solar side of the theoretical mass-radius relation is indicative of physics currently not incorporated in standard stellar models. We predict the apsidal 50 motion constant for each of the KOI-126 low-mass stars and find k 2 ≃ 0.15. Carter et al. postulate that they will be able to determine the apsidal motion constant with a relative precision of 1% by the end of the nominal Kepler mission. This will provide a crucial test for our models and will provide a stringent constraint against which to test all current and future low-mass standard stellar evolution models. 51 Chapter 3 Using the Interior Structure Constants as an Age Diagnostic The following chapter appears in the Astrophysical Journal (2013, 765, 86) under the title “The Interior Structure Constants as an Age Diagnostic for Low-Mass, Pre-Main-Sequence Detached Eclipsing Binary Stars.” It appears chronologically out of sequence, but in sequence from a conceptual standpoint. 3.1 Introduction In the study of the structure and evolution of low-mass stars, there are a variety of different methods capable of yielding reasonably accurate age estimates; for a thorough discussion of the different methods, and their strengths and weaknesses, see the review by Soderblom (2010). One such method uses detached eclipsing binaries (DEBs) to assign an age. DEBs are fantastic systems for studying stellar evolution. Observations can provide precise masses and radii for the component stars that are nearly model independent (see reviews by Andersen 1991; Torres et al. 2010). Tight constraints on the stellar masses and 52 radii allow for stringent tests of stellar evolution models. Furthermore, the age of a DEB system can be derived if stellar models can predict the radius of each star in the binary at a common age and with a single chemical composition. Deriving an age estimate from a DEB is straightforward once precise masses and radii are extracted from the data. However, the reliability of the age estimate is contingent upon the accuracy of the stellar models. Recently, pre-main-sequence (pre-MS) and MS models of low-mass stars (< 0.8M⊙ ) have received substantial criticism for not accurately predicting the radii of stars in DEBs. As the number of DEBs with precisely measured masses and radii has increased, it has become clear that stellar models under-predict the radii of stars in DEBs by upward of 10% (see, for example, Chapter 4; Mathieu et al. 2007; Jackson et al. 2009; Torres et al. 2010; Feiden & Chaboyer 2012a). The discrepancies between model and observed radii have been largely attributed to the effects of magnetic fields and magnetic activity (Ribas 2006; Chabrier et al. 2007; Jackson et al. 2009; Morales et al. 2010). When present, the discrepancy between observations and model predictions severely limits the use of stellar evolution models to derive the age of individual DEB systems. We propose a novel method to use low-mass, pre-MS DEBs to estimate the ages of young stellar systems. Instead of comparing individual stellar surface properties to stellar evolution models, we propose to use the dynamics of the DEB system. That is, comparing the observed rate of apsidal motion to stellar model predictions computed using the interior structure constants. This technique is less sensitive to the surface effects of magnetic fields than are methods that only invoke the stellar radius or effective temperature. Our method has the potential to provide a more reliable age estimate. In this chapter, we outline the computation of the interior structure constants in Section 3.2 and present results concerning their time evolution (Section 3.3). The analysis covers mass-dependency, as well as the associated uncertainties due to stellar model inputs. We conclude with a discussion on the observational considerations and the limitations of the 53 present study in Section 3.4. 3.2 Interior Structure Constants The distribution of mass within a star in a close binary system is influenced by the star’s rotation and by tidal interaction with its companion. Imagine two stars, A and B. The rotation of star A and the tidal interaction of star B with star A distorts the shape of star A. Instead of remaining spherically symmetric, the equilibrium configuration of star A will become ellipsoidal. Subsequently, the gravitational potential of star A will also become ellipsoidal. The same can be said from the perspective of star B. If the binary orbit is elliptical, the distorted gravitational potential will cause the orbit to precess. This precession may be likened to the precession of Mercury’s orbit about the Sun, where Mercury’s precession also requires general relativity and not just classical gravitational interactions. The precession of the binary orbit is known as apsidal motion. The rate of apsidal motion (ω̇; measured in degrees per cycle), or the rate at which the orbit precesses, is governed by the shape of the gravitational potential, which may be deformed as discussed above. Therefore, ω̇ depends on the properties of the stars and of the orbit. Explicitly, ¢ ¡ ω̇ = 360 c 2, 1 + c 2, 2 k 2 , (3.1) where c 2, i = ·µ Ωi ΩK ¸ µ ¶5 ¶2 µ ¶ Ri m 3−i 15m 3−i g (e) 1+ f (e) + . mi mi A (3.2) In the above equation, ΩK is the mean orbital angular velocity, Ωi , m i , and R i are the rotational velocity, the mass, and the radius of the i -th component in the binary, respectively. 54 Additionally, A is the semi-major axis of the orbit, f (e) = (1 − e 2 )−2 , (3.3) and g (e) = (8 + 12e 2 + e 4 ) f (e)5/2 , 8 (3.4) with e being the eccentricity of the orbit. Finally, the last term in Equation (3.1), k 2 , is the weighted interior structure constant observed for the two binary stars. In general, k2 = c 2, 1 k 2, 1 + c 2, 2 k 2, 2 . c 2, 1 + c 2, 2 (3.5) Here, k 2, 1 and k 2, 2 are the interior structure constants for each star. Equations (3.1)–(3.5) are derived from a j -th order solid harmonic expansion of the gravitational potential (see, for example, Kopal 1978). The interior structure constant for a given star, k 2 , is the second-order term from a more general set of expansion coefficients, k j . These second-order coefficients quantify the central concentration—or radial distribution— of mass within a star (Kopal 1978). Lower values of k 2 correspond to a higher level of central mass concentration. Point sources, for example, have k 2 = 0. Observationally, we can not solve for the individual k 2, i values. However, in the above equations, every variable is a direct observable, except for k 2 . The latter must be inferred from observational determinations of ω̇ and the c 2, i coefficients. Since the individual k 2, i values depend on the stellar density distribution, they can provide deep insight into the validity of stellar evolution models. To bypass the restriction that only k 2 can be inferred from observations, we use k 2 values from stellar evolution models in combination with the observed c 2, i coefficients to derive a theoretical k 2 . DSEP is equipped to calculate the general interior structure constants, k j , for a star at every 55 evolutionary time step. This is achieved by solving Radau’s equation after each model iteration. Following the formalism outlined by Kopal (1978), a dηj da + ¡ ¢ ¢ 6ρ(a) ¡ η j + 1 + η j η j − 1 = j ( j + 1) 〈ρ〉 (3.6) with j ∈ {2, 3, 4, . . .} being the order of the solid harmonic, a is the radius of an equipotential surface (a = r when the surfaces are spherically symmetric), and η j (a) = a dǫj . ǫj d a (3.7) In the above equation, ǫ j is the stellar deviation from sphericity. We also introduced ρ(a), the density of the stellar plasma at radius a , and 〈ρ〉, the volume averaged density at each radius, 3 〈ρ〉 = 3 a Za 0 ρ(a ′ )a ′2 d a ′ . (3.8) If we assume that tides and rotation do not significantly alter the shape of a star, we are permitted to use a spherically symmetric model to compute the interior structure constant. Our code employs a 4th -order Runge-Kutta integration scheme to obtain a particular solution of Radau’s equation at the stellar photosphere. Interior structure constants are then directly related to the particular solutions at the surface, η j (R), through j + 1 − η j (R) ¤. kj = £ 2 j + η j (R) (3.9) But, is it reasonable to assume that the stellar mass is not significantly redistributed due to rotation and tides? The effect of rotation on the central mass concentration of stars with a total mass greater than 0.8M⊙ was investigated in several previous studies (Stothers 1974; Claret & Giménez 1993; Claret 1999). For stars less massive than 0.8M⊙ , the effect of rotation should be neg56 ligible because they have a higher mean density compared to solar-type stars. To test this assumption, we used Chandrasekhar’s analysis of slowly rotating polytropes to estimate the amount of oblateness—or deviation from spherical symmetry—rotating low-mass stars may be expected to have. Chandrasekhar (1933) derived an analytical expression for the stellar oblateness for slowly rotating polytropes. The oblateness was defined to be the relative difference between the equatorial radius and the polar radius, F≡ r eq − r pole r eq , (3.10) with r eq and r pole being the equatorial and polar radius, respectively. Polytropes were considered slowly rotating when χ≡ Ω2 ≪ 1, 2πGρ c (3.11) where Ω is the stellar angular velocity, G is the gravitational constant, and ρ c is the mass density in the stellar core. Does this criterion apply to real low-mass stars? If we assume a rotation period of 1.0 day and a core density of 10 g cm−3 (realistic for pre-MS low-mass stars), then χ ≈ 10−3 . The result of Chandrasekhar’s analysis was a relation between the stellar oblateness and χ for different values of the polytropic index, n , 5.79χ F = 9.82χ 41.8χ for n = 1.5 for n = 2.0 . (3.12) for n = 3.0 Stars with masses below 0.65M⊙ are best represented by a polytrope with 1.5 < n < 2.0. Assuming n = 2.0, a low-mass star will have F ∼ 0.01. This treatment indicates that the 57 effect of rotation on the sphericity of low-mass stars is a 1% effect; rotation will not be addressed in this study. Assessing the influence of tides is difficult. We expect spherically symmetric models to provide accurate estimates of the mass distribution if R∗ ≪ R roche , where R roche is the Roche lobe radius. We also caution that the configuration of the binary must be considered. If the rotational axes are not aligned, as found with DI Her (Albrecht et al. 2009), then the validity of the assumptions required to derive of k 2 no longer hold. See Section 3.4 for a further discussion. General relativistic distortion of the gravitational potential also plays a role in determining the rate of apsidal motion (Giménez 1985). This contribution to the apsidal motion rate can be added to the classical contribution (i.e., Equation 3.1) ω̇tot = ω̇N + ω̇GR , (3.13) where ω̇tot , ω̇N , and ω̇GR are the total apsidal motion rate, classical apsidal motion rate, and the rate predicted from general relativity. The general relativistic contribution does not depend on the mass distribution of the stars. Thus, general relativity does not affect the theoretical derivation of the interior structure constants, but must be accounted for prior to comparing the theoretical and observational determinations of k 2 . 3.3 Results 3.3.1 Single Stars Individual, solar-metallicity evolutionary tracks were computed for several masses ranging from the fully-convective regime (0.25M⊙ ) up to masses where stars have thin convective 58 0.25 M⊙ 0.35 M⊙ 0.55 M⊙ 0.65 M⊙ 0.27 M⊙ 0.29 M⊙ 0.33 M⊙ 0.35 M⊙ 0.45 M⊙ 0.75 M⊙ 0.31 M⊙ 0.37 M⊙ 0.15 0.15 [Fe/H] = 0.0 k2 (b) 0.10 0.14 0.05 (a) [Fe/H] = 0.0 1 10 0.13 100 1000 Age (Myr) 1 10 100 1000 Age (Myr) Figure 3.1: The time evolution of the interior structure constant, k 2 , for stars of various masses. Stars that develop a radiative core during the pre-main-sequence exhibit rapidly decreasing k 2 values between roughly 10 and 100 Myr. (a) Total range of masses considered in this study shown in increments of 0.05M⊙ . (b) A detailed view of the transition to fullyconvective interiors. Only the 0.37M⊙ model does not ultimately become fully-convective. envelopes (0.75M⊙ ).1 Near the boundary where the transition from a radiative core to a fully-convective interior is expected (∼ 0.35M⊙ ), a finer grid of mass tracks was generated to allow for further exploration. The evolution of k 2 with age for each mass track is presented in Figure 3.1. The full collection of mass tracks (including analysis routines) used hereafter have been made available on online2 . Stellar models that develop a radiative core have a rapidly-changing interior structure constant between the age of 10 and 100 Myr. This can be observed in Figure 3.1(a). As the convection zone recedes, the central regions of the collapsing pre-MS star create a more centrally-concentrated mass profile, lowering the derived value of k 2 . The result is that the value of k 2 decreases by about 5% – 10% every 10 Myr for masses above 0.45M⊙ . Masses below approximately 0.45M⊙ undergo variations up to about 5%. 1 Pre-MS models were initialized using a simple polytrope model that was relaxed onto the Hyashi track. Uncertainties in the initial seed model largely effect the model properties younger than about 1 Myr. 2 https://github.com/gfeiden/k2age/ 59 This period of rapid contraction continues until a small convective core develops, producing a star with three energy transfer zones: a convective core, a radiative shell, and a convective outer envelope. The star settles onto the MS once the small convective core subsides with the equilibration of 3 He burning. In Figure 3.1, this process is manifested by the upward turn of the mass tracks near 100 Myr, followed by a flattening of k 2 as the star enters the MS. After first developing a small radiative core on the pre-MS, stars with masses between 0.29M ⊙ and 0.35M ⊙ eventually maintain a fully-convective interior (see Figure 3.1(b)). These small radiative cores manifest themselves as small dips in the evolution of k 2 , just as for stars that maintain a radiative core on the MS. While the rate of change in k 2 is rapid during the core contraction, the relative change in k 2 with stellar age is small (< 5%). It is evident from Figure 3.1(b) that near the fully-convective transition, stars that eventually end up with fully-convective interiors exhibit a degeneracy in the age-k 2 plane. There is no differentiating between their pre-MS core contraction and the eventual reduction of the radiative shell using k 2 alone. Below 0.29M⊙ , a radiative core does not develop on the pre-MS. Additional models were generated to investigate the effect of specific stellar properties on the predicted values of k 2 . In Figures 3.2(a)–(f) we illustrate the results of changing the scaled-solar stellar metallicity, helium abundance (Y ), and the effects of artificially inflating and deflating the stellar radius. For this exercise, two masses were selected to study the effects on the two broad categories of low-mass stars: fully-convective stars and stars with radiative cores. It is apparent from Figure 3.2 that changes to the chemical composition have the largest effect on the value of the interior structure constant on the pre-MS. Variations in metallicity of 0.2 dex translate into a 5% difference in the calculated k 2 for a 0.55M⊙ star at a given pre-MS age. Similarly, large variations in the helium abundance have the ability to produce 60 Metallicity He Abundance Stellar Radius 0.15 k2 M = 0.55 M⊙ (a) M = 0.55 M⊙ (b) M = 0.55 M⊙ (c) 0.10 +20% Psurf -20% Psurf 0.05 k2 0.15 M = 0.25 M⊙ (d) M = 0.25 M⊙ (e) M = 0.25 M⊙ (f) 0.14 αMLT = 0.50 αMLT = 1.00 αMLT = 1.94 αMLT = 3.00 [Fe/H] = -0.5 [Fe/H] = -0.3 0.13 [Fe/H] = 0.0 [Fe/H] = +0.2 1 10 100 Age (Myr) Y = 0.27 Y = 0.33 1 10 100 Age (Myr) 1 10 100 Age (Myr) Figure 3.2: The influence of various model properties on the predicted evolution of k 2 for two different stellar masses. From left to right, the properties investigated are: scaled-solar metallicity in (a) and (d), helium mass fraction (Y ) in (b) and (e), and artificial radius changes in (c) and (f). Models in the top panels ((a)–(c)) show a 0.55M⊙ star while the bottom series ((d)–(f)) show a fully-convective 0.25M⊙ star. Note, the legend for panels (c) and (f) is split between the top and bottom panels due to space restrictions. The separate tracks indicated by both legends are presented for each mass. changes in k 2 at the 5% level. Without some prior knowledge of the stellar composition, the effects of such variations may be confused with a difference in stellar age. The effects on k 2 are lessened as mass decreases, until variations nearly vanish in the fully-convective regime (panels (d) and (e) of Figure 3.2). Note that a change in Y pushes k 2 in the same direction in both mass regimes, whereas a change in metallicity produces a change in one direction for the radiative core case but the opposite direction in the fully-convective case. While this behavior does not lift the age-composition degeneracy entirely, it could prove a useful diagnostic in DEB systems whose components straddle the fully-convective boundary. 61 We attempted to mimic the possible effects of magnetic fields on the structure of our models by computing models at solar-metallicity with αMLT = 0.5, 1, and 3. The modified mixinglength represents magnetic suppression of convection in the deep interior. Additionally, models were run where we artificially changed the surface pressure by ±20%. The altered surface pressure represents the possible influence of star spots on the stellar photosphere. The magnetic Dartmouth models (Chapter 5; Feiden & Chaboyer 2012b) were not used as they have yet to be evaluated for stars on the pre-MS. Figures 3.2(c) and 3.2(f) show how k 2 varies with changes to αMLT and the surface pressure. During the pre-MS contraction, a 20% change in the surface pressure results in a ∼ 2% change to the stellar radius. The accompanying change in k 2 was found to be 0.2% and 1.0%, for a positive and negative change to the surface pressure, respectively. In the 0.55M⊙ models, increasing αMLT to 3 yielded a stellar radius that was 2% smaller than our solarcalibrated model with k 2 variations under 1% throughout the star’s pre-MS contraction. Decreasing αMLT , however, produced larger variations in k 2 . At an age of 60 Myr the model radii appeared inflated by 5% and 15% for αMLT = 1.0 and 0.5, respectively. The corresponding changes in k 2 were, respectively, 4% and 12%. Panel (f) of Figure 3.2 indicates that the 0.25M⊙ , fully-convective models experience the greatest differences at the youngest age and that these differences diminish until the different tracks converge on the MS. These changes to αMLT and surface pressure are for illustrative purposes only. It is not at all clear that simply reducing αMLT is a suitable approximation for the presence of an interior magnetic field nor that altering the surface pressure is a good approximation for star spots at the surface. 3.3.2 Binary Systems The rapid evolution of k 2 for a single star can provide an accurate age estimate for that star, but what about for a binary system? We stated in Section 3.2 that observations are only 62 0.55 M⊙ + 0.25 M⊙ Binary — k2 0.14 0.12 0.10 [Fe/H] = 0.0 1 10 100 1000 Age (Myr) Figure 3.3: The evolution of the theoretical k 2 for a binary having a 0.55M⊙ primary and a 0.25M⊙ secondary. The “observed” stellar radii are fixed at 0.62R⊙ and 0.41R ⊙ for the primary and secondary, respectively (see Section 3.3.2). The orbit was chosen to have an eccentricity e = 0.2. able to provide the weighted mean value of k 2 for two stars in a binary. To derive an age estimate for a binary system, we must find the theoretical k 2 . Temporal evolution of the theoretical k 2 values can be obtained by combining two k 2 mass tracks using Equation (3.5). Computation of the c 2, i coefficients requires precise knowledge of the stellar masses, radii, and the orbital eccentricity (see Equation (3.2) and note that A becomes irrelevant). The angular velocity term in Equation (3.2) can either be measured or approximated using the orbital eccentricity, assuming pseudo-synchronization (Kopal 1978), Ω2i = (1 + e) (1 − e)3 Ω2K . (3.14) Hence our reason for focusing on DEBs: they can yield precise estimates of the stellar and orbital properties. The evolution of k 2 is simplest for an equal mass binary. In this case, both stars contribute 63 equally to k 2 , meaning the k 2 track is exactly equal to the two individual k 2 tracks. The discussion from Section 3.3.1 on single star tracks would then apply to the binary system. A binary with unequal mass components is not so simple. Two mass tracks are required— one computed for the mass of each star—and must be combined as a single track using Equation (3.5). How does this effect our ability to extract an age estimate? We have provided an example of a weighted k 2 evolutionary track in Figure 3.3. The masses selected are those of the two stars presented in Figure 3.2, M1 = 0.55M⊙ and M2 = 0.25M⊙ . For this example, we arbitrarily adopted an orbit with an eccentricity e = 0.2. Weighting of the theoretical k 2 values is insensitive to the semi-major axis. The stellar radii (R1 = 0.62R ⊙ and R 2 = 0.41R⊙ ) were selected from a solar metallicity mass track at an age of 40 Myr. To simulate the type of data available to an observer, we elected to fix the radius, and thus, fix the c 2, i coefficients (used to weight the value of k 2 ) at each age in Figure 3.3. The rapid evolution of k 2 that is taking place in the 0.55M⊙ star largely dominates the relatively slow k 2 evolution of the 0.25M⊙ star. While the weighted k 2 value for the binary does not evolve as rapidly as for a single 0.55M⊙ star, the weighted value still changes by about 5% every 10 Myr. This is comparable to models of single stars and should not significantly hinder any age analysis. 3.4 Discussion DEBs provide an excellent laboratory for testing stellar structure and evolution theory in different mass and evolutionary regimes. The mass-radius plane is the strictest test of stellar models because these two quantities are best constrained by the observations (Torres et al. 2010). Results from studies performing such comparisons have led to the consensus that standard stellar evolution models are unable to accurately reproduce the observed stellar radii for masses below ∼ 0.8M⊙ (e.g., Chapter 4, Torres et al. 2010; Feiden & Chaboyer 64 2012a). The model inaccuracies are particularly evident among pre-MS binaries (Mathieu et al. 2007; Jackson et al. 2009). Radius discrepancies of approximately 10 – 15% are routinely quoted between pre-MS DEBs and models. This makes it difficult to derive an age with less than 50% uncertainty from stellar positions in the mass-radius plane. Magnetic effects, particularly surface spots, are thought to belie the observed radius deviations. Canonical stellar evolution models are non-magnetic and are therefore unable to properly account for magnetic modifications to convective energy transport and for the presence of magnetic spots on the stellar photosphere. We therefore advocate the inclusion of the interior structure constant, k 2 , to overcome these age determination inaccuracies whenever possible. While this study does not lead to us to conclude that individual k 2 values (and thus k 2 ) are entirely insensitive to magnetic activity in low-mass stars, it is evident that k 2 has the potential to be a better diagnostic of the pre-MS evolutionary state than the stellar radius when surface magnetic activity is present. For instance, setting αMLT = 0.5 increases the radius and k 2 of a single star by 15% and 12%, respectively. Fixing the radius to determine an age leads to an age that is upward of 180% greater than if we assume a solar calibrated αMLT . On the other hand, fixing k 2 leads to only a 20% greater age. The age errors are then compounded when we consider both stars in the binary. The decreased sensitivity of the individual k 2 values makes the mean k 2 a superior choice compared to surface quantities like the radius and effective temperature. The age precision returned from observational determinations of k 2 is dependent on the precision with which the observational k 2 and the system’s metallicity are known. For example, given an equal mass binary with 0.5M⊙ stars, knowing k 2 with 5% uncertainty and the metallicity to ±0.2 dex yields a pre-MS age with an uncertainty of approximately 33%. Constraining the metallicity uncertainty to ±0.1 dex improves this age uncertainty to 20%. Furthermore, to obtain an age with 5% precision would require k 2 to be measured with 65 near 1% precision for a metallicity known to within 0.1 dex. 3.4.1 Observational Considerations The results presented and discussed above show that it is possible to precisely derive the age of a binary system from measurements of apsidal motion. However, obtaining precise observations of apsidal motion and measuring k 2 with 5% precision is a painstaking task. The binary system must meet several criteria and the data must be of high quality. Foremost is that the binary must be a double-lined, eclipsing system. While this has been alluded to, we have yet to illuminate precisely why this is so. The reason for requiring a DEB stems from the need for extremely accurate and precise stellar and orbital properties. Equations (3.1) and (3.2) reveal that derivation of k 2 requires exquisite knowledge of the stellar mass ratio, the stellar radii, and the orbital properties (eccentricity and semi-major axis). Only data from DEBs can provide these quantities in a (nearly) model-independent fashion (Andersen 1991; Torres et al. 2010). A meticulous examination of the binary light curve and radial velocity curve is of the utmost importance. The light curve must be assembled from multi-epoch, time-series differential photometry and must provide nearly complete phase coverage. This latter feature is essential. Not only must the primary and secondary eclipses be captured, but also the behavior of the light curve out of eclipse. Such data is becoming increasingly available with the latest generation of photometric surveys (e.g., Kepler, CoRoT3 , SuperWASP4 ) and those to come (e.g., LSST5 , BRITE6 Constellation Mission). Similar to the photometry, a large number of high dispersion, high signal-to-noise spectra are needed to construct a detailed radial velocity curve. Attempts must be made to 3 Convection, Rotation, & planetary Transits Angle Search for Planets 5 Large Synoptic Survey Telescope 6 Bright Target Explorer 4 Wide 66 provide adequate phase coverage (see Andersen 1991) and uncover deviations due to the Rossiter-McLaughlin effect (Rossiter 1924; McLaughlin 1924) for reasons we shall discuss momentarily. Torres (2013) points out that the spectroscopy is still the limiting factor of quality in DEB analyses. Only with the quality of data described above and within Andersen (1991) and Torres et al. (2010) can a truly adequate analysis of a DEB be performed. However, acquisition of data of such quality is very rewarding and allows for a rigorous examination and characterization of the stellar system. It would permit the measurement of the stellar masses and radii with extreme precision (below 2%). This is imperative considering that Equation (3.2) depends upon the fractional radii (R/A) to the 5th power. Additionally, actual measurement of ω̇ requires careful monitoring of eclipse times of minimum. This can only be performed if one has a densely populated light curve. The data would also permit measurement of the binary eccentricity and semi-major axis with high precision. These properties affect both Equation (3.2) and the general relativistic contribution discussed in Section 3.2 (Giménez 1985). Recall that the contribution from the latter must be removed from the total apsidal motion rate to derive the classical contribution given in Equation (3.1). Detailed radial velocity curves not only provide accurate mass and eccentricity estimates, but may also be used to investigate the inclination of the system. The theory presented in Section 3.2 relies on the assumption that the rotational axes of the two stars be parallel to one another and perpendicular to the orbital plane. It is possible to evaluate this restriction by using the Rossiter-McLaughlin effect in a manner similar to that presented for DI Her (Albrecht et al. 2009). A detailed radial velocity curve may, additionally, betray the presence of a third body. The presence of a tertiary may affect the binary orbit, altering the derived apsidal motion. One added benefit is that lengthy observations may help to identify—and thus correct for— the impact of star spots and magnetic activity on the eclipse profiles. Star spots have the 67 ability to distort the light curve, which can diminish the accuracy of the derived stellar properties (Windmiller et al. 2010). Removing the effects of spots is critical to obtain not only precise but also accurate stellar properties. The easiest means of obtaining detailed time-series photometry is through space-based satellites, such as CoRoT and Kepler. However, long-term ground-based observational efforts are beginning to produce apsidal motion detections (e.g., Zasche 2012), demonstrating that it is feasible to carry out the necessary observations using ground-based telescopes. Finally, by acquiring a large number of quality spectra, it may be possible to extract the projected rotational velocities (v sin i ) and a modest estimate of the chemical composition. Spectroscopic determinations of cool star metallicities is notoriously complicated, but most pre-MS DEBs will likely reside near or in a cluster from which metallicity estimates can be extracted using the higher mass stellar population. In the event there is no known association from which to draw a metallicity, techniques based on low- and medium-resolution spectra are encouraging (see, e.g., Rojas-Ayala et al. 2012). While the validity of such techniques along the pre-MS is unclear, they provide a viable starting point and typically produce metallicities with uncertainties below 0.2 dex, the limit we recommend. 3.4.2 Limitations The usefulness of k 2 as an age estimator is limited to the pre-MS, in particular, during the evolutionary phase where the radiative core is rapidly contracting. This typically corresponds to an age between 10 Myr and 100 Myr (see Figure 3.1). At the 5 – 10 % measurement level, this technique is also restricted to binaries where one of the stars has a mass above ∼0.40 M ⊙ . This restriction ensures the radiative core contraction and relative change in k 2 for the more massive star is rapid enough to dominate the theoretical k 2 evolution. Reducing the observational uncertainty in k 2 below 5% enables a more accurate age estimate and would allow for DEBs with lower mass components to be reliably analyzed. 68 Precise measures of apsidal motion and metallicity are challenging to obtain, but are already feasible and should only improve over time. Instead of observational limitations, the greatest limiting factor for using k 2 as an age indicator is the circularization of the DEB orbit. Mutual tidal interactions will circularize binary orbits over time (Zahn 1977). Apsidal motion requires an elliptical orbit. If an equal-mass binary is to maintain an elliptical orbit for the duration of its pre-MS contraction, the orbital period must be at least 2.4 days (Zahn 1977). The probability of discovering an elliptical binary decreases with time, but this provides an additional consistency check. The age suggested by k 2 should not be significantly older than the orbital circularization timescale. Finally, the age estimate is only as accurate as the stellar models. Verifying that a given stellar model produces the proper mass distribution, hence k 2 , may at first seem rather unreasonable. However, at least one known system, with the possibility of a second (KIC 002856960; Lee et al. 2013), is capable of providing validation of the physics incorporated in low-mass stellar evolution models. Carter et al. (2011) have indicated that by the end of the nominal Kepler mission, they will know the interior structure constants of KOI-126 B and C with about 1% precision. Interior structure constants known with this precision can place stringent constraints on the equation of state of the stellar plasma (Chapter 2; Feiden et al. 2011). The veracity of low-mass models, and therefore the validity of their predicted interior structure constants, may be assessed according to the results from KOI126 and similar systems. 69 Chapter 4 The Mass-Radius Relation for Low-Mass, Main-Sequence Stars The content of this chapter is largely based on material from the paper entitled “Reevaluating the Mass-Radius Relation for Low-Mass, Main-Sequence Stars,” published in the Astrophysical Journal (2012, 757, 42). 4.1 Introduction We saw in Chapter 2 that low-mass stellar evolution models accurately reproduced the stellar properties of all three stars in the KOI-126 triple system. The inclusion of the super-solar metallicity within the stellar evolutionary calculations permitted the accurate modeling. The agreement between stellar models and observations of a low-mass system based on metallicity forced us to reconsider whether the routinely quoted 5 – 15% disagreement is representative of true radius discrepancies or whether there are other factors contributing to the derivation of such large radius errors. One such factor derives from the fact that previous studies focusing on the comparison 70 between models and observations have generally applied a limited sample of isochrones to their data. Largely, these sets are comprised of 1 Gyr and 5 Gyr, solar metallicity isochrones. This is predominantly a consequence of the limited age and metallicity range of currently available low-mass stellar models. Age and metallicity effects are less important in the lowmass regime, but the stringent uncertainties quoted by observational efforts preclude the use of such a limited set of isochrones. For example, Burrows et al. (2011) discovered nonnegligible radius variations in brown dwarfs and very-low-mass stars (< 0.1M⊙ ) when allowing for a more comprehensive set of metallicities. Isochrones with metallicities spanning a range characteristic of the local galactic neighborhood are therefore essential to accurately assess the validity of stellar evolution models. Furthermore, one must also consider that the population of well-characterized DEBs has, until recently, consisted of eight systems. While unlikely, it is not unimaginable that those eight systems were more the exception than the rule in terms of their lack of consistency with stellar evolution models. Since publication of the Torres et al. (2010) review, the population of well-characterized, low-mass DEBs has more than doubled. The availability of this new data allows for a more accurate statistical characterization of the agreement (or lack of) between the MR relationship defined by models and observations. When discrepancies are observed, they are typically attributed to the effects of a large scale magnetic field (e.g., Ribas 2006; López-Morales 2007; Morales et al. 2008, 2009a; Torres et al. 2010; Kraus et al. 2011) as DEBs are often found in tight, short-period orbits with periods under three days. Tidal interactions and angular momentum conservation act to synchronize the orbital and rotational periods of the components, increasing the rotational velocity of each star in the process. The dynamo mechanism, thought to be responsible for generating and sustaining stellar magnetic fields, is amplified as a result of the rotational spin-up and enhances the efficiency of magnetic field generation within the star. Each component in the binary system is then more able to produce and maintain a strong, large-scale magnetic 71 field than a comparable single field star. The effects of a large-scale magnetic field are thought to be two-fold: convective motions within the star are suppressed and the total surface coverage of starspots is increased. In both cases, a reduction in the total energy flux across a given surface within the star occurs, forcing the stellar radius to inflate in order to conserve flux (Gough & Tayler 1966). Recent attempts at modeling these effects have indicated that an enhanced magnetic field is a plausible explanation, although the primary physical mechanism affecting the structure of the star is still debated (Mullan & MacDonald 2001; Chabrier et al. 2007; MacDonald & Mullan 2012). Regardless of the precise physical mechanism, magnetic fields should betray their presence through the generation of magnetic activity in the stellar atmosphere. If magnetism is responsible for the observed inflated stellar radii, then correlations should be expected between individual stellar radius deviations and magnetic activity indicators (i.e., chromospheric Hα and Ca ii H & K emission, coronal x-ray emission, etc.). Tantalizing evidence of such correlations has been reported previously by López-Morales (2007) and Morales et al. (2008). However, recent evidence appears to stand in contrast with the current theory. Two systems, LSPM J1112+7626 (Irwin et al. 2011) and Kepler-16 (Doyle et al. 2011), were discovered that have wide orbits with approximately forty-one day periods. Despite this, both appear to display discrepancies with stellar evolution models. In these systems, the component stars should be evolving individually with mutual tidal interactions playing a negligible role in the overall angular momentum evolution. The stars should be spinning down over time due to magnetic breaking processes (Skumanich 1972), meaning the stars should not be as magnetically active compared with short period binary systems. The contrast is particularly evident for LSPM J1112+7626, where a rotation period of sixty-five days was detected via starspot modulation in the out-of-eclipse light curve. Gyrochronology suggests that the sys- 72 tem has an age of approximately 9 Gyr (Barnes 2010) and implies further that the secondary is likely slowly rotating and should, therefore, not shown signs of strong magnetic activity or an inflated radius. A third system also appears to defy the current hypothesis. KOI-126 (Carter et al. 2011) is a hierarchical triple system with two low-mass, fully convective stars in orbit around a 1.35 M⊙ primary. The two low-mass stars are orbiting each other with a period of 1.77 days. Therefore, they should show signs of inflated radii due to enhanced magnetic activity. However, it has been shown that the two low-mass, fully convective stars were in agreement with model predictions when considering their super-solar metallicity and the age of the higher mass primary (Chapter 2; Feiden et al. 2011). This agreement was further confirmed by Spada & Demarque (2012). Metallicity has been proposed previously as a solution to the observed MR discrepancies, but for the case of single field stars (Berger et al. 2006). This was contradicted shortly thereafter by López-Morales (2007), most notably for DEBs. Although, we must consider that the radius discrepancy-metallicity correlation is severely complicated by the fact that metallicities of M dwarfs are notoriously difficult to determine observationally. Finally, developments in light curve modeling of spotted stars has generated interesting results. The presence of large polar spots may alter the light curve analysis of DEBs by modifying the eclipse profile. These modifications lead to 2 – 4% uncertainties in the derived stellar radii (Morales et al. 2010; Windmiller et al. 2010; Kraus et al. 2011). Thus far, only two DEBs (GU Boo and CM Dra) have been thoroughly tested for their sensitivity to spots. Systematic uncertainties may therefore dominate the error budget, casting a shadow of doubt on the observed radius discrepancies, which are often made apparent due to the minuscule random uncertainties. The uncertainties and developments outlined above have motivated us to reevaluate the current state of the low-mass MR relationship. In what is to follow, we use a large grid of 73 theoretical stellar evolution isochrones to compare the low-mass models of the Dartmouth Stellar Evolution Program (DSEP) with DEB systems that have well constrained masses and radii. We then explore how potentially unaccounted for systematic uncertainties have the ability to create the appearance of discrepancies when neglected and mask real ones when considered. Section 4.2 will present the DEB sample followed by a description of the isochrone grid and fitting procedures in Section 4.3. Results will be presented in Section 4.4 and a discussion of the implications of our findings is in Section 4.5. 4.2 Data Our selection criteria mimic those of Torres et al. (2010, hereafter TAG10) in so far as we require the random uncertainties in the mass and radius measurements to be less than 3%. No data was disqualified due to perceived data quality issues or the original author’s attempts to constrain possible systematic uncertainties. We also applied the criterion that the DEB system must include at least one component with a mass less than 0.8 M⊙ . This cut-off in mass is used to designate “low-mass” stars. The mass cut-off was selected for two main reasons. First, the effects of age and metallicity on the structure of low-mass main sequence stars are suppressed compared to stars with masses of approximately 1.0 M⊙ or above. Overall, this allows for less flexibility in fitting models to the observations, providing a more critical analysis of the stellar evolution models. Second, some of the largest discrepancies between observations and models are seen in the low-mass regime. This is likely a consequence of the former reason: true discrepancies become more apparent as the models become less sensitive to the input parameters. Stars used in this study are listed in Table 4.1 along with their observationally determined properties and original references. Our final sample consisted of eighteen DEB systems for a total of thirty-six stars. Six of these systems are taken from TAG10 who reanalyzed the 74 available data using a common set of reduction and parameter extraction routines in an effort to standardize the process. While the original references are cited, the parameters listed are those derived by TAG10 whose results were similar to the original values. A majority of the systems, ten in total, were published after the release of the TAG10 review. Six are drawn from the study performed by Kraus et al. (2011), three are products of recent results from the Kepler Space Telescope mission (Carter et al. 2011; Doyle et al. 2011; Bass et al. 2012), and the final post-TAG10 system was discovered by the MEarth survey (Irwin et al. 2011). The remaining two systems from our sample were announced before TAG10 (López-Morales et al. 2006; López-Morales & Shaw 2007), however, they were not included in the review for reasons related to either data availability or data quality. One final note: the Kepler systems KOI-126 (Carter et al. 2011) and Kepler-16 (Doyle et al. 2011) were not analyzed in a similar manner to the rest of the double-lined DEB population. They are DEB systems whose parameters were derived from Kepler photometry using a dynamical-photometric model (see Supporting Online Material from Carter et al. 2011). However, they still satisfy the criteria for comparison with stellar evolution models and have been included for this reason. Table 4.1 DEBs with at least one low-mass component with precise masses and radii. Star P orb Mass Radius P rot Name (day) (M ⊙ ) (R ⊙ ) (day) UV Psc A 0.86 0.9829 ± 0.0077 1.110 ± 0.023 ··· 0.76440 ± 0.00450 0.8350 ± 0.0180 0.80 0.981 ± 0.012 1.061 ± 0.016 ··· 0.6644 ± 0.0048 0.6810 ± 0.013 1.31 0.924 ± 0.008 0.8807 ± 0.0017 ··· 0.683 ± 0.005 0.6392 ± 0.0013 ··· UV Psc B IM Vir A 1.309 IM Vir B KID 6131659 A KID 6131659 B 17.528 continued on next page 75 Source† 1 2 3 Table 4.1 – continued Star P orb Mass Radius P rot Name (day) (M ⊙ ) (R ⊙ ) (day) RX J0239.1-1028 A 2.072 0.7300 ± 0.0090 0.7410 ± 0.0040 ··· 0.6930 ± 0.0060 0.7030 ± 0.0020 ··· 0.6897 ± 0.0034 0.6489 ± 0.0013 ··· 0.20255 ± 0.0007 0.22623± 0.0005 ··· 0.61010 ± 0.00640 0.6270 ± 0.0160 0.49 0.59950 ± 0.00640 0.6240 ± 0.0160 0.54 0.59920 ± 0.00470 0.6194 ± 0.0057 0.87 0.59920 ± 0.00470 0.6194 ± 0.0057 0.82 0.584 ± 0.002 0.560 ± 0.005 ··· 0.544 ± 0.002 0.513 ± 0.009 ··· 0.567 ± 0.002 0.552 ± 0.004 ··· 0.532 ± 0.002 0.532 ± 0.004 ··· 0.557 ± 0.001 0.569 ± 0.002 ··· 0.535 ± 0.001 0.500 ± 0.003 ··· 0.5428 ± 0.0027 0.5260 ± 0.0028 ··· 0.4982 ± 0.0025 0.5088 ± 0.0030 ··· 0.527 ± 0.002 0.505 ± 0.008 ··· 0.491 ± 0.001 0.471 ± 0.009 ··· 0.499 ± 0.002 0.457 ± 0.010 ··· 0.443 ± 0.002 0.427 ± 0.008 ··· 0.469 ± 0.002 0.441 ± 0.004 ··· 0.382 ± 0.001 0.374 ± 0.004 ··· 0.43490 ± 0.00120 0.4323 ± 0.0055 ··· 0.39922 ± 0.00089 0.3916 ± 0.0094 ··· 0.3946 ± 0.0023 0.3860 ± 0.0052 ··· RX J0239.1-1028 B Kepler-16 A 41.08 Kepler-16 B GU Boo A 0.49 GU Boo B YY Gem A 0.81 YY Gem B MG1-506664 A 1.55 MG1-506664 B MG1-116309 A 0.827 MG1-116309 B MG1-1819499 A 0.630 MG1-1819499 B NSVS 01031772 A 0.368 NSVS 01031772 B MG1-78457 A 1.586 MG1-78457 B MG1-646680 A 1.64 MG1-646680 B MG1-2056316 A 1.72 MG1-2056316 B CU Cnc A 2.77 CU Cnc B LSPM J1112+7626 A 41.03 continued on next page 76 Source† 4 5 6 7 8 8 8 9 8 8 8 10 11 Table 4.1 – continued Star P orb Mass Radius P rot Name (day) (M ⊙ ) (R ⊙ ) (day) 0.2745 ± 0.0012 0.2978 ± 0.0046 ··· 0.2413 ± 0.0030 0.2543 ± 0.0014 ··· 0.2127 ± 0.0026 0.2318 ± 0.0013 ··· 0.23102 ± 0.00089 0.2534 ± 0.0019 ··· 0.21409 ± 0.00083 0.2398 ± 0.0018 ··· LSPM J1112+7626 B KOI-126 B 1.77 KOI-126 C CM Dra A 1.27 CM Dra B † References: (1) Popper (1997); (2) Morales et al. (2009b); Source† 12 13 (3) Bass et al. (2012); (4) López-Morales & Shaw (2007); (5) Doyle et al. (2011); (6) López-Morales & Ribas (2005); (7) Torres & Ribas (2002); (8) Kraus et al. (2011); (9) López-Morales et al. (2006); (10) Ribas (2003); (11) Irwin et al. (2011); (12) Carter et al. (2011); (13) Morales et al. (2009a) 4.3 Isochrone Fitting 4.3.1 Isochrone Grid Isochrones were computed for a wide range of age and metallicity values. The parameter space was defined to encompass a vast majority of stars typical of the local galactic neighborhood. Seven ages and seven scaled-solar metallicities were adopted for a total of forty-nine individual isochrones. The sets of values used in this study were: [Fe/H] = {−1.0, −0.5, −0.3, −0.1, 0.0, +0.1, +0.2} dex and age = {0.3, 1.0, 3.0, 5.0, 6.0, 7.0, 8.0} Gyr. For completeness, we also generated a set of isochrones that employed a smoothly varying convective mixing length. A second set of isochrones was generated in order to address the idea that low-mass stars may possess inflated radii due to inefficient convective energy transport. It has been posited that convection within low-mass stars may either be inherently inefficient or that other physical processes (i.e., magnetic fields) act to reduce the ability of convection to effectively transport energy. We do not attempt to prescribe a physical mechanism associated with this 77 conjecture. Instead, we attempt to parametrize convection in such a way so as to reduce the efficiency of convection in the lower mass regimes while still maintaining our solar calibration, necessary to properly model the Sun. The second grid of isochrones was computed with a mass-dependent mixing length, henceforth referred to as “variable αMLT ” models. Convection for these models was parametrized with a smooth quadratic function of the form αMLT µ ¶ µ ¶ M 2 M =a + b. M⊙ M⊙ (4.1) Selection of a quadratic was arbitrary and carries no physical justification, except to produce low-mass stars with relatively inefficient convection compared to those in the standard model case. Coefficients were determined by matching the convective mixing length to predetermined values at two different masses. Since the overall structure of very low-mass stars is rather insensitive to the precise value of the mixing length, we anchored αMLT = 1.00 at M = 0.1M ⊙ . The other end of the mass spectrum is constrained by our need to satisfy our solar calibration. Thus, at M = 1.0M⊙ , the convective mixing length was fixed to αMLT = 1.94. Subsequently, αMLT µ ¶ µ ¶ M 2 M = 0.949 + 0.991. M⊙ M⊙ (4.2) A relative comparison between models computed with a solar calibrated mixing length and those generated with a parametrized mixing length is presented in Figure 4.1. Our isochrone grid for these models covered a fraction of the parameter space compared to the standard model set. There was a total of 12 isochrones for the variable αMLT models: [Fe/H] = {−0.5, 0.0, +0.2} dex with ages = {1.0, 3.0, 5.0, 8.0} Gyr. Below the fully convective boundary, the models are rather insensitive to the mixing length, as stated above. For higher masses, the reduced mixing length increases the model radii by up to 3%. As designed, a 78 Sun-like star is unaffected by this mixing length prescription. We also see that above 1M⊙ the model radii begin to decrease due to an increased mixing length as a result of our parametrization. 4.3.2 Model Rotation Rotation is not included in our models, as was discussed in Section 1.5.1. Recalling that discussion, we find that the stars in the present sample all have χ < 10−4 , except for the two stars in NSVS 01031772. They have χ ∼ 10−3 as a consequence of their short orbital period. Regardless, all of the stars satisfy the slowly rotating approximation. For all of the systems in question we found F < 0.001, except for NSVS 01031772. This system is deviating slightly from sphericity with F ≈ 0.011. We feel justified in neglecting the physics of rotation in our stellar models, with the caveat that attempting to probe model precisions below 1% for NSVS 01031772 will likely require a more detailed treatment of rotational deformation. 4.3.3 Fitting Procedure Judging agreement between observations and models was performed on a system-by-system basis, as opposed to fitting individual stars. Critical to the process was ensuring that both components of a given system were consistent with isochrones of a common age and metallicity. For each object, corresponding model radii were derived by linearly interpolating in each isochrone using the observationally determined mass. A linear interpolation scheme was sufficient since the mass resolution along the isochrones was small (∆M = 0.02M⊙ ). Relative errors between the model radii and the observationally determined radius were 79 1.20 Standard Model Variable αMLT Radius (R⊙) 1.00 8 Gyr 0.80 1 Gyr 0.60 0.40 0.20 δR/R 0.05 [Fe/H] = 0.0 1 Gyr 8 Gyr 0.00 0.20 0.30 0.40 0.50 0.60 0.70 0.80 0.90 1.00 Mass (M⊙) Figure 4.1: (Top) A mass-radius comparison of isochrones computed using a solar-calibrated mixing length (solid line) and the mass dependent mixing length, described in Equation (4.2), (dashed line) at 1 Gyr (light-blue) and 8 Gyr (maroon). (Bottom) Radius variations induced by our prescription of a mass-dependent αMLT . Shown are 1 Gyr (light-blue, solid line) and 8 Gyr (maroon, dashed line) isochrones computed with a solar heavy element composition (Grevesse & Sauval 1998). Positive values indicate that the variable αMLT models are inflated compared to the solar-calibrated models. The “blip” observed near the fully convective boundary is likely due to a 3 He instability described by van Saders & Pinsonneault (2012). 80 then calculated. The relative error was defined as R obs − R model δR = R obs R obs (4.3) and will be presented as such throughout the rest of the paper. Formal agreement was determined by analyzing the number of standard deviations outside of the accepted range our model radii were located. Explicitly, # σR = R obs − R model σR (4.4) where σR is the random uncertainty of the observational radius. Different levels of compatibility were assigned to each system based on the individual agreement of each component in the system. For instance, if an isochrone matched both components at a common age within the 1σ limits set by the quoted random uncertainties, the system as a whole was designated as being “fit” by the models. Only within the 1σ level was a system considered to be accurately represented by the models. The analysis was applied to two sets of data, which differ in the adopted radius uncertainties (formal quoted uncertainties or fixed 3% uncertainties) that will be justified later. The process detailed above generated a list indicating the level of agreement between each isochrone and each DEB system. Narrowing the list to a single “best fit” isochrone involved minimizing the root mean square deviation (RMSD) v u 2 µ u 1 X δR i ¶2 RMSD =t . 2 i =1 σR,i where the sum is performed over the components of the DEB system. 81 (4.5) Table 4.2 Age and [Fe/H] priors for low-mass DEBs. DEB System Age (Gyr) [Fe/H] Status UV Psc 7.9 Rejected IM Vir ··· ··· -0.3±0.3 ≤ 1.0 ··· Kepler-16 GU Boo YY Gem CU Cnc KOI-126 CM Dra ··· 0.4 ± 0.1 0.4 ± 0.1 ≤ 0.0 Accepted +0.1±0.2 Accepted +0.15±0.8 Accepted +0.1±0.2 4.0 ± 1.0 4.0 ± 1.0 Rejected ≤ 0.0 Rejected Rejected Accepted 4.3.4 Age & Metallicity Priors DEBs that have been well studied have additional constraints that allow us to restrict the set of isochrones used in the fitting procedure. Specifically, our sample contains eight systems that have the added constraint of either an estimated age or metallicity, and in a couple cases, both. Before accepting the quoted age and metallicity priors, however, we performed a qualitative analysis to ensure that the estimates we adopted were reliable. We judged four of the eight systems with quoted age or metallicity priors to have been determined reliably. A summary of the adopted priors is given in Table 4.2. UV Psc The age of the UV Psc system is quoted by the TAG10 review to be 7.9 Gyr. Popper’s original paper describing the characteristics of the UV Psc system does not provide any evidence to support an age estimate (Popper 1997). An age of approximately 8 Gyr is typically assigned to the system due to the fact that stellar evolution models predict the physical properties of the primary star, UV Psc A, at that age. However, there has yet to be any set of models that can place both components on a consistent, coeval isochrone. Since the estimated age 82 of UV Psc A appears to be derived from stellar evolution models, we allowed our models to independently determine the most acceptable age. IM Vir Morales et al. (2009b) attempt to determine the metallicty of IM Vir by applying various photometric metallicity relations. The result of their efforts was that they found all of the various empirical methods quote different values with large uncertainties. Values for [Fe/H] vary from the metal-poor end with [Fe/H] = -0.8 up to a super-solar value of [Fe/H] = +0.15. This range also fits nicely within the set of model metallicity values selected for this study. Since the cited metallicity range would not provide any additional constraint on our analysis, we rejected the metallicity prior. Kepler-16 Kepler-16, the first confirmed binary system with a circumbinary planet, was provided with a metallicity estimate in its discovery paper (Doyle et al. 2011). A metallicity of [Fe/H] = −0.3±0.2 was determined spectroscopically. The authors indicated the spectroscopic anal- ysis was performed on the K-dwarf primary and that the general method was similar to that applied to KOI-126 using Spectroscopy Made Easy (SME; Valenti & Piskunov 1996; Carter et al. 2011). Reliability is lent to the method, in general, due to its success at deriving the metallicity of KOI-126 A. We therefore accepted the quoted metallicity prior. GU Boo The age estimate provided by López-Morales & Ribas (2005) in their characterization of GU Boo was primarily based on kinematics. Specifically, they conclude that GU Boo is an isolated system and the vertical component of its motion provides a hint that it has undergone 83 perturbations due to disk heating processes. Assuming that the system has been subjected to disk heating, one can only infer that the system has an age greater then 108 yr, the typical timescale for dynamical perturbations associated with an objects orbit around the galactic center (Soderblom 2010). Unfortunately, no further constraints were able to be placed on the age of the system. However, it is not possible to rule out the scenario that GU Boo was dynamically ejected from its stellar nursery. It would then appear that no reliable age estimation exists, prompting us to reject the age prior for this system. YY Gem YY Gem is thought to be physically associated with the Castor AB (α Gem) quadruple system. All three systems were found to be gravitationally bound based on a statistical analysis performed using a three-body interaction code (Anosova et al. 1989). While the Anosova et al. (1989) analysis was performed with pre-Hipparcos proper motions as their initial conditions, it is unlikely that the results will be effected at the level necessary to unbind the systems. Castor A and B are themselves both binaries. The primary in both systems is an A star and both are thought to have an M dwarf companion. Therefore, physical properties derived from spectroscopic and photometric observations of the two A stars will be essentially unaffected by the presence of their companions. Placing the two primaries on an MV -(log(Teff ) H-R diagram, we used DSEP to derive an age of about 400 Myr. This age is consistent with the average age of 370 Myr derived by Torres & Ribas (2002), who modeled the A-star primaries using multiple stellar evolution codes. The A stars also lend themselves well to a spectroscopic determination of the metallicity. Unfortunately, there appears to be only one result reported. Smith (1974) estimates an average metallicity for the two A stars to be about [Fe/H] = +0.7 relative to Vega. As described in Torres & Ribas (2002), this implies a rather uncertain metallicity relative to the Sun of 84 [Fe/H] = +0.1 ± 0.2. Despite this, we adopt this metallicity constraint due to its rather large uncertainty, which should presumably encompass the true value. CU Cnc When CU Cnc was originally investigated by Ribas (2003), it was found to have a space motion very similar to that of the Castor sextuple system (α Gem, YY Gem) and, subsequently, the proposed “Castor moving group.” For this reason, CU Cnc was deemed to be associated with the Castor moving group implying an age and metallicity similar to the Castor sextuple. However, determining membership of a moving group is complicated and has often lead to ambiguous results concerning the coeval nature of the group. It is not uncommon for members of the same kinematic moving group to have different metallicities, implying that members of a defined moving group may not have been born in the same galactic environment and, as such, are not necessarily coeval. The lack of an age estimation beyond its kinematic similarity to the Castor system led us to reject the age prior of CU Cnc. KOI-126 KOI-126 is a hierarchical triple eclipsing binary recently whose discovery was recently announced (Carter et al. 2011). The quoted metallicity prior of [Fe/H] = +0.15 ± 0.08 was determined using SME, as was mentioned above in the discussion of Kepler 16. Assuming this metallicity allowed for a relatively precise age constraint (∼ 4 ± 1 Gyr) to be placed on the primary, a 1.35M⊙ subgiant. Combining both the age and metallicity information led to two low-mass companions to also be fit by stellar models (Chapter 2; Feiden et al. 2011). Thus, there appears to be little question about the validity of the age and metallicity estimations, leading us to adopt the given priors for our study. 85 CM Dra Finally, we consider the very well studied CM Dra system. Spectroscopic observations of the system have produced varying results for the metallicity of the system, but all appear to be consistent with −1 ≤ [Fe/H] ≤ 0 (Viti et al. 1997, 2002; Morales et al. 2009a; Kuznetsov et al. 2012). Due to the difficulties in modeling, and thus fitting, the entire SED of an M dwarf, we determined there was no particular reason to strongly favor one metallicity result over another. An age was determined for the system by assuming that the common proper motion white dwarf (WD) companion, WD 1633+572, is bound to the CM Dra system. It is the possible to derive an age from WD cooling tracks combined with stellar evolution models of the approximate WD progenitor. The WD cooling age was found to be 2.38 ± 0.37 Gyr. Morales et al. (2009a) predicted a mass for the progenitor star of M = 2.1±0.4M⊙ and derived a total (stellar model + WD cooling) age of 4.1 ± 0.8 Gyr. For consistency, we calculated the lifetime of the progenitor star using DSEP with the full available suite of physics. Accounting for the metallicity constraint defined above, we derived an age of 3.7 ± 1.2 Gyr and, in doing so, found no reason to reject, or modify, the age prior. 4.4 Results 4.4.1 Standard Stellar Models Direct Comparison Figure 4.2 demonstrates that, in general, our models reduce the observed radius discrepancies to below 4% for 92% of the stars in our sample. Only a few outliers are seen to be largely discrepant. Across the entire sample, we find a mean absolute error of 2.3%. This broadly 86 δR/Robs (obs - model) 0.12 Quoted Uncertainties A) Fixed 3% Uncertainties B) 0.08 0.04 0.00 -0.04 δR/Robs (obs - model) -0.08 0.12 0.08 0.04 0.00 -0.04 -0.08 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Mass (M⊙) Figure 4.2: Relative errors between stellar evolution models and observationally determined radii. Reliable age and metallicity priors were accounted for in the statistical analysis. (Top) The adopted measurement uncertainties are the cited observational random uncertainties derived from the light curve fitting procedure. (Bottom) A fixed 3% uncertainty is adopted to represent possible systematic uncertainties (i.e., starspots). represents a factor of two reduction in the previously cited radius discrepancies. The age and metallicity of the best fit isochrone for each DEB system is listed in Table 4.3 along with the level of agreement between the best fit isochrone and the individual stars comprising each DEB system. Contrasting with previous studies, we do not observe an overwhelming systematic trend of the models grossly under predicting stellar radii. Instead, systems discrepant by more than 5% represent an exception to the broad agreement between the models and observations. While the agreement is admittedly not perfect, it is apparent that most observed discrepancies must now be judged according to the precision with which their radii were measured. 87 As we will see in the next subsection, systematic uncertainties have the potential to blur our interpretation of the agreement between models and observations, simply due to the factor of two improvement described above. Ignoring systematic uncertainties, for now, we are still confronted with the need to explain both the slightly discrepant as well as the largely discrepant radii. Discussion and comments pertaining to plausible explanations of the inflated radii will be addressed later in Section 4.5. The presence of stars that appear undersized compared to our models is likely the result of our statistical fitting method and not a physical phenomenon. Fixed Uncertainties Recent work in modeling eclipse profiles of DEB systems has led to the realization that large, polar starspots can potentially affect observationally derived radii and subsequently dominate over the quoted random uncertainties. Windmiller et al. (GU Boo; 2010) and Morales et al. (CM Dra; 2010) have found that systematic uncertainties may be present in DEB radii on the order of 2 – 4%. This is to be compared with quoted random uncertainties that are typically on the order of 1% or less. Similarly, Kraus et al. (2011) included an estimate of systematic uncertainties for their radius measurements and found typical uncertainties on the order of 2 – 3%. With the radius uncertainties potentially dominated by often unquoted systematics, we were curious to see what effect such uncertainties would have on comparisons with theoretical models. Quite obviously, observed discrepancies would be reduced by the level of systematic uncertainty. However, it has not been clear whether those systematics have the ability to completely mask the observed radius residuals, particularly when combined with isochrones that cover a wide age-metallicity parameter space. Since the radius uncertainty cut-off for our DEB sample was set at 3%, we elected to adopt a fixed 3% radius uncertainty in our analysis to mimic potential systematics. 88 Demonstrated in Figure 4.2 B is the effect of including a fixed 3% radius uncertainty. The overall distribution of points appears to be similar to the previous case, except that the larger uncertainty enables more systems (eleven in total) to be considered fit by our analysis (see Table 4.3). This result is clearly expected when introducing larger error bars, as mentioned above. Introducing a fixed systematic uncertainty also relieves the problem that some stars appear smaller than the models in Figure 4.2 A. What is important to realize is that we are now presented with a scenario where the systematic uncertainties mask our ability to draw firm conclusions about whether the observed radius discrepancies are inherently real or merely a consequence of neglected uncertainties. Such ambiguity was previously not present since the observed radius residuals were substantially larger than any potential systematic uncertainties. 89 Table 4.3 Best fit isochrone using a solar calibrated mixing-length. Quoted Star Name Age UV Psc A 8.0 [Fe/H] δR/R obs -0.10 UV Psc B IM Vir A 7.0 -0.10 IM Vir B KID 6131659 A 3.0 -0.50 90 KID 6131659 B RX J0239.1-1028 A 8.0 -0.10 RX J0239.1-1028 B Kepler-16 A 1.0 -0.10 Kepler-16 B GU Boo A 8.0 -0.10 GU Boo B YY Gem A YY Gem B 0.3 -0.10 Fixed 3% # σR Fit Age -0.0285 -1.375 No 8.0 0.1031 4.785 -0.0037 -0.248 0.0408 2.136 -0.0014 -0.750 0.0004 0.200 0.0297 5.510 0.0285 10.007 0.0004 0.210 0.0299 5.510 0.0292 1.144 0.0414 1.616 0.0813 8.834 0.0813 8.834 No Yes No No No No continued on next page 7.0 3.0 8.0 0.3 8.0 0.3 [Fe/H] δR/R obs -0.10 -0.10 -0.50 -0.10 -0.10 -0.10 -0.10 # σR Fit -0.0285 -0.950 No 0.1032 3.439 -0.0038 -0.125 0.0408 1.359 -0.0014 -0.047 0.0004 0.013 0.0297 0.991 0.0285 0.949 0.0099 0.330 0.0225 0.749 0.0292 0.973 0.0414 1.381 0.0813 2.710 0.0813 2.710 No Yes Yes Yes No No Table 4.3 – continued Quoted Star Name Age MG1-506664 A 1.0 [Fe/H] δR/R obs -0.10 MG1-506664 B MG1-116309 A 7.0 -0.50 MG1-116309 B MG1-1819499 A 8.0 -0.50 MG1-1819499 B 91 NSVS 01031772 A 8.0 -1.00 NSVS 01031772 B MG1-78457 A 5.0 +0.20 MG1-78457 B MG1-646680 A 1.0 +0.20 MG1-646680 B MG1-2056316 A 3.0 +0.20 MG1-2056316 B CU Cnc A 8.0 +0.20 Fixed 3% # σR Fit Age 0.0035 1.951 No 1.0 -0.0031 -1.612 -0.0084 -1.162 0.0150 1.993 0.0296 8.425 -0.0528 -8.792 -0.0021 -0.396 0.0412 6.991 -0.0008 -0.049 -0.0001 -0.006 -0.0226 -1.725 0.0184 1.309 -0.0074 -1.632 0.0067 1.248 0.0233 1.832 No No No Yes No No No continued on next page 7.0 7.0 8.0 5.0 1.0 3.0 8.0 [Fe/H] δR/R obs -0.10 -0.50 -0.10 -0.50 +0.20 +0.20 +0.20 +0.20 # σR Fit 0.0035 0.116 Yes -0.0031 -0.105 -0.0084 -0.281 0.0150 0.500 0.0404 1.348 -0.0408 -1.360 -0.0134 -0.447 0.0395 1.317 -0.0008 -0.026 -0.0001 -0.004 -0.0226 -0.755 0.0184 0.613 -0.0074 -0.247 0.0067 0.222 0.0233 0.777 Yes No No Yes Yes Yes Yes Table 4.3 – continued Quoted Star Name Age [Fe/H] δR/R obs # σR 0.0020 0.084 -0.0005 -0.035 0.0290 1.875 -0.0011 -0.205 -0.0003 -0.053 0.0316 4.221 0.0360 4.798 CU Cnc B LSPM J1112+7626 A 8.0 +0.20 LSPM J1112+7626 B KOI-126 B 1.0 +0.10 KOI-126 C CM Dra A 92 CM Dra B 5.0 Fixed 3% 0.00 Fit No Yes No Age 8.0 3.0 5.0 [Fe/H] δR/R obs # σR 0.0020 0.067 -0.0005 -0.016 0.0290 0.966 -0.0036 -0.120 -0.0004 -0.013 0.0316 1.055 0.0360 1.200 +0.20 +0.20 0.00 Fit Yes Yes No 4.4.2 Variable Mixing Length Models Non-standard, variable mixing-length models lead to only slightly better results over the standard models when considering the quoted random uncertainties (see Table 4.4 and Figure 4.3). We find that radius deviations are largely reduced to below 4%, seen clearly in Figure 4.3. The residuals do appear to be more tightly clustered around the zero point than they do in Figure 4.2. Although, only slightly. This is suggestive of an overall better agreement between models and observations when we vary the mixing length, which is reinforced by a slightly lower MAE of 2.1% across the entire sample. There was evidence in Figure 4.3 of a possible tendency for the best fit theoretical isochrone to over predict the observed radius at higher masses and under predict the radius for lower mass stars. A least squares regression performed on the data finds a linear slope of −0.031 ± 0.013, significant at the 2.5σ level (not shown). Half of the DEB systems that were not formally fit are characterized by an isochrone that over-predicts the higher mass primary but under-predicts the lower mass secondary. This is only an artifact of our functional parametrization of αMLT , suggesting convection was too heavily suppressed at higher masses and under-suppressed in the low-mass regime. Artificially fixing the uncertainties at 3% adds ambiguity as to whether there are any real discrepancies for a majority of the systems. This is illustrated in Figure 4.3. As was suggested by our study of the standard models, systematic uncertainties must be minimized to below 2% before an accurate comparison between models and observations may occur. 93 δR/Robs (obs - model) 0.12 Quoted Uncertainties A) Fixed 3% Uncertainties B) 0.08 0.04 0.00 -0.04 δR/Robs (obs - model) -0.08 0.12 0.08 0.04 0.00 -0.04 -0.08 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Mass (M⊙) Figure 4.3: Relative errors between stellar evolution models with a mass-dependent convective mixing length and observationally determined radii. Reliable age and metallicity priors were accounted for in the statistical analysis. (Top) The adopted measurement uncertainties are the cited observational random uncertainties derived from the light curve fitting procedure. (Bottom) A fixed 3% uncertainty is adopted to represent possible systematic uncertainties (i.e., starspots). 94 Table 4.4 Best fit isochrone using a mass dependent convective mixing length. Quoted Star Name Age UV Psc A 8.0 [Fe/H] δR/R obs 0.00 UV Psc B IM Vir A 8.0 +0.20 IM Vir B KID 6131659 A 3.0 -0.50 95 KID 6131659 B RX J0239.1-1028 A 8.0 -0.50 RX J0239.1-1028 B Kepler-16 A 1.0 0.00 Kepler-16 B GU Boo A 8.0 -0.50 GU Boo B YY Gem A YY Gem B 0.4 0.00 Fixed 3% # σR Fit Age -0.0072 -0.349 No 8.0 0.0866 4.019 -0.0032 -0.209 0.0238 1.245 -0.0113 -5.876 -0.0190 -9.352 -0.0075 -1.394 0.0039 1.356 -0.0085 -4.227 0.0244 11.061 0.0085 0.333 0.0190 0.740 0.0757 8.226 0.0757 8.226 No No No No Yes No continued on next page 8.0 3.0 8.0 1.0 8.0 0.4 [Fe/H] δR/R obs 0.00 +0.20 -0.50 -0.50 0.00 -0.50 0.00 # σR Fit -0.0072 -0.241 No 0.0866 2.888 -0.0032 -0.105 0.0238 0.792 0.0113 -0.378 -0.0190 -0.634 -0.0075 -0.251 0.0039 0.129 -0.0085 -0.282 0.0245 0.815 0.0085 0.283 0.0190 0.632 0.0757 2.523 0.0757 2.523 Yes Yes Yes Yes Yes No Table 4.4 – continued Quoted Star Name Age MG1-506664 A 1.0 [Fe/H] δR/R obs 0.00 MG1-506664 B MG1-116309 A 8.0 +0.20 MG1-116309 B MG1-1819499 A 5.0 -0.50 MG1-1819499 B 96 NSVS 01031772 A 8.0 +0.20 NSVS 01031772 B MG1-78457 A 3.0 0.00 MG1-78457 B MG1-646680 A 1.0 0.00 MG1-646680 B MG1-2056316 A 1.0 +0.20 MG1-2056316 B CU Cnc A 8.0 +0.20 Fixed 3% # σR Fit Age -0.0017 -0.964 No 1.0 -0.0099 -5.065 -0.0126 -1.743 0.0115 1.534 0.0240 6.839 -0.0575 -9.587 -0.0121 -2.276 0.0352 5.965 -0.0023 -0.143 0.0005 0.025 -0.0319 -2.428 0.0159 1.129 -0.0083 -1.829 0.0112 2.095 0.0127 0.998 No No No Yes No No Yes continued on next page 8.0 5.0 8.0 3.0 1.0 1.0 8.0 [Fe/H] δR/R obs 0.00 +0.20 0.00 +0.20 0.00 0.00 +0.20 +0.20 # σR Fit -0.0017 -0.057 Yes -0.0099 -0.329 -0.0126 -0.421 0.0115 0.384 0.0413 1.376 -0.0396 -1.320 -0.0121 -0.404 0.0352 1.172 -0.0023 -0.075 0.0005 0.016 -0.0319 -1.063 0.0159 0.529 -0.0083 -0.277 0.0112 0.373 0.0127 0.423 Yes No No Yes No Yes Yes Table 4.4 – continued Quoted Star Name Age [Fe/H] δR/R obs # σR -0.0028 -0.116 -0.0038 -0.278 0.0255 1.652 0.0020 0.365 -0.0017 -0.307 0.0282 3.757 0.0333 4.441 CU Cnc B LSPM J1112+7626 A 8.0 +0.20 LSPM J1112+7626 B KOI-126 B 3.0 +0.20 KOI-126 C CM Dra A 97 CM Dra B 5.0 Fixed 3% 0.00 Fit No Yes No Age 8.0 5.0 5.0 [Fe/H] δR/R obs # σR -0.0028 -0.092 -0.0038 -0.125 0.0255 0.850 0.0010 0.033 -0.0014 -0.047 0.0282 0.939 0.0333 1.111 +0.20 0.00 0.00 Fit Yes Yes No 4.4.3 Peculiar Systems Before continuing with a further interpretation of our results, we wanted to briefly comment on several individual DEB systems. These systems stood out in our mind as worth remarking upon separate from the ensemble. Our models never fit the UV Psc system to a coeval age. While we were able to fit UV Psc A, the secondary component was always found to have an 8 – 10% larger radius than would be expected from stellar evolution theory, consistent with Popper (1997). Both stars may realistically be discrepant with the models, however, if the age is truly younger than 8 Gyr. Constraining the actual magnitude of the discrepancies is difficult without observational age and metallicity estimates. Fittingly, the system is known to be very active based on spectroscopic analysis of Hα cores, multi-epoch photometric monitoring (Kjurkchieva et al. 2005), and the derived X-ray luminosity (this work, Section 4.5.2). Further observations of UV Psc to constrain its metallicity and to provide data for a more rigorous starspot analysis would be a worthwhile endeavor. MG1-189499, characterized by Kraus et al. (2011), was also never found to be in total agreement. The primary and secondary were found to deviate by 3.0% and -5.3%, respectively. This appeared to be a cause for concern, but Kraus et al. were one of the few groups to provide an estimate of potential systematic uncertainties. Uncertainties in the primary star’s radius were elevated to 4.6% with a comparatively large uncertainty of 3.4% in the secondary’s radius. Applying these uncertainties and rerunning the isochrone fitting procedure allows the predictions of the models to nearly fall inline with the observations. An age of 8 Gyr was again derived but with a new metallicity of [Fe/H] = -0.3. This isochrone yielded relative radius errors of 4.8% and -2.9% for the primary and secondary, respectively. Hence, the secondary is now considered fit by the models and the primary is only 0.2% outside the bounds of uncertainty. The last two systems we would like to discuss are two that have long posed problems for 98 modelers: YY Gem and CM Dra. YY Gem (Torres & Ribas 2002) is an equal mass, equal radius DEB system that is effectively represented by a single point at (0.6, 0.08) in Figure 4.2. Beginning with the low-mass models of Hoxie (1970), YY Gem has never been adequately reproduced by stellar models. The difficulty with YY Gem seems to be in the estimated age of 400 Myr, determined by its association with the Castor AB quadruple system. An older age is logically preferred by the models considering the notably inflated radius. Interestingly, this would have led previous studies to underestimate the observed radius discrepancy. Assuming either a 1 or 5 Gyr isochrone naturally results in models with larger radii than those computed with an age of 400 Myr. Further study of this system, particularly its distance (Section 4.5.2) and association with Castor AB (Section 4.3.4), will be beneficial toward fully understanding the nature of its inflated radii. Finally, CM Draconis is another constant thorn in theoreticians sides. Here, the primary and secondary deviate by 3.2% and 3.6%, respectively. CM Dra is one of the most studied DEB systems and has very well constrained physical properties measured from data spanning multiple decades (Morales et al. 2009a). Our models indicate that a solar composition is preferred1 . However, the composition of CM Dra, while not known positively, is very likely subsolar (Viti et al. 1997, 2002; Morales et al. 2009a; Kuznetsov et al. 2012), meaning our models are probably more discrepant than indicated by this study. Morales et al. (2010) found that large polar spots produce systematic uncertainties in the radius measurements of around 3%, potentially bringing CM Dra more in line or more out of line with model predictions. For the time-being, this system will continue to test our knowledge of stellar evolution. 1 A super-solar metallicity is actually favored, as was also noted by Spada & Demarque (2012), but the application of the metallicity prior in our analysis restricted the models to only solar or subsolar compositions. 99 4.5 Discussion Data presented in this study hint at two competing explanations for the occurrence of the differing model and observational MR relations. It is entirely plausible that stellar evolution models are not incorporating key physical processes that can account for the observed discrepancies. This view is not new and has always accompanied discussions of low-mass models (Hoxie 1970, 1973; Chabrier & Baraffe 1997; Baraffe et al. 1997; Baraffe et al. 1998). Typically cited is our incomplete knowledge dealing with the complex array of molecules present in M-dwarf atmospheres as well as the lack of structural changes induced by a large-scale magnetic field. Included in the latter are both the effects on convective energy transport and the emergence of spots on the stellar photosphere. The other scenario is one in which the neglect of systematic uncertainties is driving the apparent discrepancies. We have shown that by allowing for realistic variation in age and metallicity of the models, the radius residuals are of the same magnitude as the potential systematic uncertainties. Note, this is without any modification to the solar calibrated mixing length or the inclusion of any non-standard physics. Previous studies have indicated that systematics may help alleviate the size of the radius residuals, but have required additional modifications to the models in order to fully reconcile models with observations. It is now clear that we must work to constrain and minimize systematic uncertainties in observations of DEBs to allow for them to provide an accurate test of stellar evolution models. 4.5.1 Radius Deviations The MAE between our models and the observations was 2.3%, a factor of two improvement over the canonical minimum of 5%. We found deviations of no more than 4%, with the exception of a few stars, instead of ubiquitous 5 – 15% errors, as is often quoted. Despite improving the situation, panel A of Figure 4.2 illustrates that the models are still unable to 100 fully reproduce the observed stellar radii. Accepting the factor of two improvement presented in Section 4.4.1, the paradigm of broad disagreement between models and observations is shifted to one where agreement is broad, and large discrepancies are an exception. With the radius deviations typically less than 4%, an evaluation of the systematic errors becomes imperative. Formerly, systematic uncertainties of about 4% were incapable of relieving radius deviations greater than 5%. Stellar evolution models still appeared to disagree with DEB observations even after the inclusion of systematic errors. The reason for the factor of two improvement is twofold. First, we have calculated models with a finer grid of metallicities. DSEP utilizes an EOS that enables models to be more reliably calculated for super-solar metallicities, allowing for a greater range of stellar compositions to be considered. Low-mass stellar models with super-solar metallicity have previously been unavailable for comparison with low-mass DEB data. Thus, the ability to extend our model set to super-solar metallicities allows for more flexibility in attempting to match the observed properties of DEB components. Second, a far larger number of low-mass DEBs with precisely measured radii were available to us as opposed to previous studies. Before the publication of TAG10, there were only eight systems that met the criteria necessary to accurately constrain stellar evolution models. Following the TAG10 review, the total number of systems that met the necessary criteria more than doubled with the addition of ten newly characterized DEBs. These additional systems appear to be more in line with the results of standard stellar evolution theory. However, well-known discrepant systems remain noticeably discrepant and still require further explanation. We must now ask, “what belies the current discrepancies between models and observations?” Figure 4.2 favors the hypothesis that non-standard physics are absent from current stellar evolution models. Our larger data set allows us to notice that stars of similar masses 101 from different DEB systems appear to be discrepant at varying levels, an effect a single set of standard models cannot correct. However, this is contingent upon the accuracy our age and metallicity predictions. Until we have better empirical age and metallicity estimates for the various systems, it is too difficult to ascertain the true level of discrepancy for any individual star. The efficiency of convective energy transport is of greatest interest. It is possible that convection is naturally inefficient. Although, we gather from Figures 4.2 and 4.3 that suppression of convective energy transport must be tied to a stellar property that is largely independent of mass. Simple parametrization of the suppression of convection is too uniform over a given mass regime to fully account for the observed differences in stellar radii for stars with similar masses (see Section 4.4.2). Any effort, either theoretical or observational, to constrain the physics of convection in low-mass stars will lend crucial insight. The most favored option, is that convection is not intrinsically inefficient, but that a largescale magnetic field acts to suppress convective motions (Mullan & MacDonald 2001; Chabrier et al. 2007; MacDonald & Mullan 2012). Stellar evolution models self-consistently incorporating the effects of large-scale magnetic fields will help on this front. Observations of cool-star magnetic field strengths and topologies will then provide a means of judging the validity of any new models. One final hypothesis is that starspots may affect the structure of stars and generate the inflated radii we observe. Chabrier et al. (2007) investigated such a possibility by artificially reducing the total stellar bolometric luminosity in an effort to mimic the effects of spots. They found that radius discrepancies were relieved with their parametrization. However, it is still not apparent whether spots reduce the total bolometric flux or if they locally shift flux to longer wavelengths (Jackson et al. 2009), preserving the total luminosity. Ultimately, if starspots are required in stellar evolution models, their inclusion is necessitated in the analysis of DEB light curves. 102 Quantifying the effect starspots may have on observed DEB light curves is extremely difficult. Obtaining accurate knowledge of the total surface coverage of spots, the total number of spots, their individual sizes, temperature contrasts, and their overall distribution on the stellar surface is nearly an impossible task given only a light curve. From a theoretical perspective, finding a proper parametrization to mimic the effect of spots on a three dimensional volume within the framework of a one dimensional model provides its own complications. Currently, the only feasible method to include spots, is to include their potential effects on the radius measurement uncertainties. Unfortunately, while the inclusion of fixed 3% radius uncertainties in our analysis was able to alleviate many of the noted radius discrepancies, it also created a situation where the measurement uncertainty was on the order of the typical radius deviation. We are presented with a case where the observations are no longer effective at testing the models and the manifestation of most radius discrepancies can be attributed to under estimated error bars. At this point, we require observations that have been rigorously vetted for potential systematic uncertainties and are able to still provide mass and radius measurements to better than 2%. 4.5.2 Radius-Rotation-Activity Correlations Direct measurements of low-mass magnetic field strengths are rare, especially among fast rotating stars (Reiners 2012). Without a direct measure of the magnetic field strength, we are forced to rely on indirect measures to probe correlations between stellar magnetism and the appearance of inflated stellar radii. Ideally, these indirect measures are intimately connected with the dynamo mechanism, thought to generate and maintain stellar magnetic fields, or are the product of magnetic processes in the stellar atmosphere. The preferred indirect measures are typically stellar rotation or the observation of magnetically driven emission (Hα, Ca ii H and K, X-ray). 103 A) δR/Robs (obs - model) 0.12 B) 0.08 0.04 0.00 -0.04 -0.08 0 10 20 30 Porb (day) 40 1 2 Porb (day) 3 4 Figure 4.4: Observed correlation between radius deviations and DEB orbital periods. Maroon asterisks are stars with known X-ray flux measurements. Dashed vertical lines represent the various period cuts used in our statistical analysis to divide the sample into two subsamples. (Left) The full range of periods present in the current sample. (Right) Highlighting the short period regime as it encompasses a majority of the DEB systems. Rotation Typically, low-mass DEBs are found in tight, short period orbits (< 3d; see Figure 4.4). Tidal interactions spin-up the individual components and allow them to remain rapidly rotating throughout their life cycle. Stellar dynamo theory dictates that the large-scale magnetic field strength is tied to the rotational properties of a star (i.e., a rapidly rotating star should be more magnetically active than a comparable star that is slowly rotating; Parker 1979; Reiners et al. 2012), providing a natural starting point for our investigation. We performed two independent statistical tests on the distribution of radius residuals as a function of the orbital period (P orb ). Our primary objective was to determine whether rapidly rotating systems produce, on average, larger radius deviations than systems perceived to be slow rotators. Rotational periods were assumed to be synchronous with the orbital period unless a separate value for the rotational period was cited in the literature. The statistical tests performed were a Kirmogorov-Smirnov (K-S) test and another whereby we tested the probability of obtaining a given distribution of residuals via a Monte Carlo method. Both tests were performed on the observed difference in mean absolute error 104 δR/Robs (obs - model) 0.12 0.08 0.04 0.00 -0.04 -0.08 0.01 0.1 1 Rossby Number, Ro Figure 4.5: The theoretical Rossby number, Ro = P rot /τconv , versus the relative radius error. Ro is tied directly to the theoretical stellar dynamo mechanism and is empirically related to the ratio of a star’s X-ray to bolometric luminosity (a magnetic activity indicator). Asterisks in maroon are stars with known X-ray flux measurements. (MAE)2 between two data bins. The data bins were divided at preselected values of P orb , identified visually as vertical dashed lines in Figure 4.4. Comparing the radius deviations with the rotational periods, we find no evidence of any dominant correlation. Figure 4.4 displays the residual data as a function of the orbital period, with frame A showing the full range of observed periods and frame B highlighting the “short period” regime. The correlation of radius deviations with orbital period has been studied previously (e.g., Kraus et al. 2011) where a significant difference between the two bins was observed around P orb = 1.5 days. We performed the statistical tests using three values for the orbital period (1.0d, 1.5d, and 2.0d) that defined the two period bins. We confirm the results of Kraus et al. (2011) and find a 3.1σ difference in the distribution of radius deviations around 1.5d. Systems with P orb < 1.5d had a MAE of 3.4% while the longer 2 We selected the MAE over the RMSD as a measure of the mean radius deviation of a given ensemble in order to reduce the weight of any individual outlier in the final mean. 105 period systems had a MAE of 1.0%. While this is tantalizing, we can not necessarily attribute any physical significance to this particular division. We should expect this difference to be present for any two subsamples. However, we fail to find any evidence for a statistically meaningful difference when we divide the subsamples at 1.0d and 2.0d. Therefore, the significant difference noted at 1.5d is likely a spurious statistical result3 . There does not appear to be a physically meaningful explanation for why the divide should be made at 1.5d, but then also not hold for a division at 1.0d. The rotational (or orbital) period is not necessarily the most appropriate proxy for the magnetic field strength or potential magnetic activity level. For instance, two stars may have identical rotation periods, but they may have different rotational velocities depending on their radius. The latter variable may play a more important role in governing the stellar magnetic field. It is therefore ideal for our chosen rotational variable to have some connection to intrinsic stellar properties. Optimally, the rotational variable in question would also provide a direct link to either observable magnetic activity or a theoretical description of stellar magnetism. Accordingly, we advocate the use of the Rossby number (hereafter Ro). Ro is defined as the ratio of the rotational period of the star to its convective overturn time, Ro = P rot /τconv , and measures the strength of the Coriolis force acting on the vertical motion of convection cells. The dimensionless quantity Ro appears directly in standard mean-field dynamo theory (α-ω dynamo; Parker 1979)4 and is intimately related to the ratio of the stellar X-ray luminosity to bolometric luminosity (Wright et al. 2011). The latter quantity has been shown to be a strong indicator of a stellar corona heated to over 106 K by magnetic activity (Vaiana et al. 1981; Pallavicini et al. 1981; Noyes et al. 1984). 3 Kraus et al. (2011) posit the difference may actually be a by-product of the DEB light curve analysis methods. Providing a further examination is outside the scope of this study. 4 For fully convective stars, an α-ω dynamo cannot operate due to the lack of a tachocline. However, it is thought that an α2 dynamo can efficiently generate a magnetic field. Since it is the α mechanism that is related to the strength of the Coriolis force, the Rossby number should be just as applicable to fully convective stars (Chabrier & Küker 2006; Browning 2008). 106 We find no significant difference in the distribution of data points as a function of Ro. One cut in Ro was performed at Ro = 0.1, illustrated in Figure 4.5. The selection of Ro = 0.1 approximately corresponds to Rosat , or the value of Ro associated with an observed saturation in the ratio of the stellar X-ray luminosity to the bolometric luminosity (Wright et al. 2011). This suggests that all points with Ro values less than 0.1 should presumably be sufficiently magnetically active so as to display inflated radii. We find this is not the case. Unfortunately, the selection of Ro = 0.1 creates one bin with a small sample size which could affect the statistics. Caution must further be taken with the above results as secondary stars in the longer period (> 15d) systems do not have independently measured rotation periods, meaning they may actually have lower Ro values. A visual inspection of Figure 4.5 leads us to the same conclusion, if we ignore the two most discrepant points. Since we were comparing MAE values, the outliers do not significantly affect the results of the statistical analysis. Arguably, the MAE is not an effective measure of the degree of inflation of each sample as it treats deflated radii the same as inflated radii. Instead, the actual direct mean may provide a more compelling statistical measure. Thus, we ran our statistical analysis on the direct mean error. Overall, the typical degree of inflation among the radii of low-mass stars was found to be about 1.6%. No statistically significant correlations with either P orb or Ro were uncovered. The most significant result was found for the period cut at 1.5d, where the K-S test indicated a significant difference. However, the MC method produced a difference in the mean error of the two populations at 2.2σ, below our significance threshold of 3.0σ. All other bin divisions yielded results significant at only about the 1σ level. Curiously, if we consider only data points with measured X-ray fluxes (see below), there is evidence that systems with lower Ro values may have larger radius discrepancies. However, we are then prompted to explain why the trend does not continue to higher Ro values, a question we are currently not equipped to answer. There is still some ambiguity with the presence of the points near Ro = 0.01, which appear to contradict the presence of any 107 δR/Robs (obs - model) 0.12 0.08 0.04 0.00 -0.04 Variable αMLT Variable αMLT -0.08 0.0 10.0 20.0 30.0 Porb (day) 40.0 1.0 2.0 Porb (day) 3.0 4.0 Figure 4.6: Observed correlation between radius deviations and DEB orbital periods. Dashed vertical lines represent the various period cuts used in our statistical analysis to divide the sample into two subsamples. (Left) The full range of periods present in the current sample. (Right) Highlighting the short period regime as it encompasses a majority of the DEB systems. definite correlation. Deeper X-ray observations of X-ray faint low-mass DEBs will clarify this ambiguity. We subjected the data produced from fits to the variable αMLT models to the same statistical analyses described above. Again, we compared the radius deviations to the rotational period and Rossby number. The resulting distribution are shown in Figures 4.6 and 4.7. The strongest hint of a correlation was found when applying a cut at a period of 1.5d, although it was only significant at the 2.7σ level, below our 3σ significance threshold. Performing the period cut at 1.0d and 2.0d yielded results that were even less significant (2.4σ and 1.9σ, respectively), providing the same qualitative result as standard model scenario. Similarly, no strong correlation is observed between the radius deviations and the Rossby number. As with the standard models, we conclude that there are no strong correlations between radius deviations and rotational properties. X-ray Activity An interesting extension of our discussion in the previous subsection, is to compare our derived radius deviations with the observed X-ray to bolometric luminosity ratio (hereafter 108 Variable αMLT δR/Robs (obs - model) 0.12 0.08 0.04 0.00 -0.04 -0.08 0.01 0.1 1 Rossby Number, Ro Figure 4.7: The theoretical Rossby number, Ro = P rot /τconv , versus the relative radius error for models computed with a mass-dependent mixing length. Ro is tied directly to the theoretical stellar dynamo mechanism and is empirically related to the ratio of a star’s X-ray to bolometric luminosity (a magnetic activity indicator). R x = L x /L bol ). Since no correlation was noted with Ro, we expect that no correlation will be observed with R x , as it has been shown to be intimately connected with Ro (Wright et al. 2011). López-Morales (2007) previously performed such a comparison and found a clear correlation between R x and radius deviations, in contradiction with our predictions. Her comparison was performed under different modeling assumptions, which may have influenced the results. Specifically, she compared all observations to a 1 Gyr, solar metallicity isochrone from Baraffe et al. (1998). Without variation in age and metallicity, the observed radius discrepancies may have been over estimated (or under estimated in the case of YY Gem). We therefore perform the same comparison as López-Morales (2007) to probe whether age and metallicity variations affect the outcome. A subsample of sixteen DEBs from this study have identified X-ray counterparts in the ROSAT All-Sky Survey Bright and Faint Source Catalogues (Voges et al. 1999; Voges et al. 109 2000). Our analysis follows that of López-Morales (2007) whose previous analysis contained a fraction of our current X-ray detected sample. X-ray count rates were converted to X-ray fluxes according to the formula derived by Schmitt et al. (1995), F x = (5.30HR + 8.31) × 10−12 X cr (4.6) where HR is the X-ray hardness ratio5 , X cr is the X-ray count rate, and F x is the X-ray flux. Luminosities in the X-ray spectrum were computed with distances determined from either Hipparcos parallaxes (preferred when available; van Leeuwen 2007) or near-infrared photometry. Photometric distances were estimated using the luminosity calculated from the StefanBoltzmann relation assuming the observationally determined radius and effective temperature. Absolute magnitudes were then derived using the phoenix ames-cond model atmospheres, adopting the best fit isochrone metallicities. In an effort to reduce errors introduced by the theoretical atmospheres, distances were computed using the average distance modulus derived from Two Micron All Sky Survey (2MASS) J - and K -band photometry (Cutri et al. 2003). We derived R x values for all sixteen stars in our X-ray sample to ensure internal consistency. Slight discrepancies between values presented here and those of López-Morales (2007) are due entirely to differences in the adopted distances. Attributing the X-ray flux contribution from each star in an DEB system is a difficult task. As such, López-Morales performed her analysis using three reported empirical scaling relations (Pallavicini et al. 1981; Fleming et al. 1989): • Case 1 – each component contributes equal weight. • Case 2 – proportional to the respective rotational velocity, v sin i , of each component. 5 There are typically two hardness ratios listed in the ROSAT catalog, HR1 and HR2. The Schmitt et al. (1995) formula requires the use of HR1. 110 ROSAT Detections δR/Robs (obs - model) Linear Regression 0.12 Case 1 0.08 0.04 0.00 δR/Robs (obs - model) -0.04 0.12 Case 2 Case 3 0.08 0.04 0.00 -0.04 0.0000 0.0005 0.0010 Rx = Lx/Lbol 0.0015 0.0000 0.0005 0.0010 Rx = Lx/Lbol 0.0015 Figure 4.8: Relative error between observational and model radii for stars with detected X-ray emission. X-ray data is drawn from the ROSAT All-Sky Survey and combined with the radius residuals derived from this study. Data are shown as maroon filled circles. Illustrated are two least-square regressions performed on the data. The light-blue, dashed line demonstrates a non-negligible slope of ∼ 25 ± 17 across all of the data, while the indigo, dash-dotted line excludes the two most discrepant points, UV Psc B and YY Gem. • Case 3 – proportional to the square of the rotational velocity, (v sin i )2 , of each component. We present results from all three cases in Figure 4.8. Immediately we notice that the size of the stellar radius deviations appears to correlate with R x . A linear least-square regression performed on each data set (light-blue dashed line in Figure 4.8) suggests the slope for each case (1 – 3) is 26, 24, and 22 (± 17), respectively, all with reduced-χ2 values of ∼8. Pearson r coefficients were 0.72, 0.69, and 0.63, respectively. The statistics suggest that the likelihood of uncorrelated sets of data producing these particular correlations are 0.2%, 0.3%, and 0.9%, for case 1, 2, and 3, respectively. We can therefore rule out the null hypothesis with greater than 99% confidence. 111 Visually, however, we notice that the correlation is largely driven by the presence of YY Gem and UV Psc B, located in the upper-right region of each panel in Figure 4.8. If we were to remove those three points (YY Gem appears as a single point), the correlation vanishes and we only observe an offset from the zero point (Indigo dash-dotted line in Figure 4.8). Furthermore, the distance to YY Gem is highly uncertain. Assuming it is associated with Castor AB provides a distance of about 15 pc (discussed in Section 4.3.4), but our photometric analysis, as described above, places YY Gem at a distance of approximately 11 pc, 4 pc closer to the Sun than the Castor AB system. Instead of selecting a single distance estimate, we averaged the two estimates and adopted d = 13±2 pc, which also happens to be the distance assumed by Delfosse et al. (2000). The fact that our statistical correlation critically hinges upon three points, two of which are strongly distance dependent, is a cause for concern and implies that the statistics should be interpreted with care. If we believe the strong statistical correlation, then we are presented with a situation where the rotational data and the X-ray data disagree. This may be due to the physics underlying the stellar dynamo or those underlying X-ray saturation. However, there are two further interpretations that are contingent upon the role systematics play. First, we have that the correlation is entirely real and the presence of non-inflated stars with R x ∼ 0.0007 are outliers in the relation. Second, systematics play a large role, as proposed in Section 4.4.1. Here, the truly deviant stars exhibit very strong X-ray emission (R x > 0.001), while non-inflated stars show lower, varying levels of X-ray emission. For this view to hold, the relation between the level of radius inflation and magnetic field strength can not be linear. Strong magnetic fields would induce significant radius inflation while moderate and weak fields would produce little or no inflation. Accepting, on the other hand, that the statistical correlation is spurious, we are left with a picture that is coherent with our rotation analysis. Namely, magnetic activity may not be the leading cause for all of the observed inflated radii. Here, systematics may still play a 112 role in producing stars that appear inflated, but that are consistent with the models, leaving a few discrepant stars. Naturally inefficient convection, potentially dependent on particular stellar properties, may be operating. However, magnetic activity cannot be fully ruled out, as we do not yet have a fully self-consistent description of the interaction of magnetic fields with the stellar interior and atmosphere. It may be that magnetic fields acting within YY Gem and UV Psc have a more noticeable affect on stellar structure, as higher mass stars are affected more by changes to the properties of convection (see Figure 4.1). We tend to favor a hybrid interpretation. Here, most of the observed “inflation” is an artifact of unaccounted-for systematics, but significantly discrepant stars (YY Gem, UV Psc) are associated with very strong magnetic activity. We believe that CM Dra probably fits into the latter category due to a push in the literature towards subsolar metallicity. Effects of a large-scale magnetic field are presumably mass dependent, with higher mass stars showing a greater propensity to become inflated owing to their sensitivity to changes in convective properties. In other words, higher mass stars are more sensitive to weaker magnetic fields than lower mass, fully convective stars. Whether there is a characteristic magnetic field strength that induces substantial radius deviations is unclear. Dynamo saturation and the saturation of magnetic activity, as evidenced by the flattening of the Ro-R x relation in Wright et al. (2011), is not yet fully understood, but may provide further insight into the apparent disagreement between our rotation and X-ray analyses. Finally, clarity will be obtained with a better distance measurement to YY Gem, either from ground-based parallax programs or with eventual results from Gaia. Of all the points in Figure 4.8, the existence of a positive correlation is most dependent upon the distance to YY Gem. An accurate distance, as well as a reanalysis of its association with the Castor system, has the ability to not only relieve the ambiguity present in the X-ray data, but also to provide insight into the necessary constraints for the system (i.e., is YY Gem really about 400 Myr old?). As further low-mass DEB systems are discovered, X-ray observations 113 are strongly encouraged in order to develop a coherent picture of how radius deviations correlate with magnetic activity. 4.6 Summary This chapter focused on reevaluating the current state of agreement between the theoretical and observational low-mass, main sequence MR relationship. The DEB systems used in the analysis were required to have quoted random uncertainties in the mass and radius below 3% in order to provide an effective test of stellar evolution models. A large grid of DSEP models were computed with variation in age and metallicity characteristic of the local galactic neighborhood. Best fit isochrones were derived by allowing the age and metallicity to be optimized while maintaining the constraint that the system be coeval with a single composition. DEBs with reliably determined ages or metallicities were compared with a restricted set of isochrones in the range allowed by the observational priors. Overall, we find 92% of stars in our sample are less than 4% discrepant with the models, largely representing a factor of two improvement over the canonical 5% – 15% deviations. Our results suggest that low-mass stars with radii that deviate significantly from model predictions are exceptions to general agreement. Discrepancies may also be the result of unaccounted for systematic uncertainties (i.e., starspots) that may as large as 4%. With uncertainties as large as the typical radius deviations, we find it difficult to draw the firm conclusion suggesting that models are in broad disagreement with observations. Instead, we are left with a situation where the observational uncertainties may be too large to provide an adequate test of stellar models. The combination of random and systematic uncertainties for the sample of low-mass DEBs must be constrained and minimized below the 2% level before accurate model comparisons may be made. Radius correlations with orbital (rotational) period, Ro, and R x were also considered. No dis- 114 tinct trends were identified with either orbital period or Ro. However, we find evidence for a strong correlation between radius deviations and R x (previously noted by López-Morales 2007) in contradiction with our Ro analysis. The trend is not as tight as that derived by López-Morales (2007), owing to the age and metallicity variations allowed by our analysis. This correlation is also largely contingent upon the veracity of the distance estimate to YY Gem. Accurately determining the distance to YY Gem and evaluating its association with the Castor AB quadruple would alleviate much of the uncertainty. Finally, we must not leave the theoreticians out of the spotlight. The degree to which a magnetic field can alter the interior structure of low-mass stars is still only partially known and further investigations are required. Development of models with self-consistent magnetic field perturbations will begin to shed light on this unknown. Comparisons between predicted magnetic field strengths from self-consistent models and observational data (either direct or indirect) will provide a measure of the validity of the ability of magnetic fields to inflate stellar radii. Whatever the final solution may be to this long-standing problem, it is now apparent that the level of inflation required by theory is not a ubiquitous 5% – 15%, but only so in extreme cases. 115 Chapter 5 Magnetic Perturbation within the Framework of DSEP Material presented in the chapter to follow provides mathematical derivations of the magnetic perturbations now contained within DSEP. The mathematical formalism and the general approach to the problem were originally presented by Lydon & Sofia (1995) and within the context of DSEP in a publication in The Astrophysical Journal under the title “SelfConsistent Magnetic Stellar Evolution Models of the Detached, Solar-Type Eclipsing Binary EF Aquarii.” 5.1 Introduction Despite the identification of magnetic field as a potential culprit responsible for inflating the radii of DEB components, only ad hoc procedures for treating the effects of magnetic fields have been introduced (Mullan & MacDonald 2001; Chabrier et al. 2007). The method examined by Chabrier et al. (2007) included artificially decreasing the convective mixing length parameter, so as to mimic the effect of a global magnetic field within the star, as 116 well as artificially reducing the star’s bolometric flux in an effort to reproduce the effects of photospheric spots. A second method, proposed by Mullan & MacDonald (2001), altered the Schwarzschild criterion by perturbing the adiabatic gradient in a manner consistent with the work of Gough & Tayler (1966). Investigations by both groups appear to be at odds with one another. Chabrier et al. (2007) and Morales et al. (2010) claim that starspots appear to be the dominate mechanism inflating stellar radii, and that modifications to convection require unrealistic magnetic field strengths (i.e., reductions in the mixing length in their formulation). On the other hand, Mullan & MacDonald (2001) and MacDonald & Mullan (2012) conclude the opposite that reduction in convective efficiency is ultimately the dominant mechanism. Regardless of which is really the dominant mechanism, both approaches are inherently ad hoc, yet both are capable of reproducing the observed inflated stellar radii. In this chapter, we introduce a self-consistent treatment of a globally pervasive magnetic field embedded in the framework of the Dartmouth stellar evolution code (Dotter et al. 2007, 2008). Our approach follows the outline provided by Lydon & Sofia (1995), though we deviate from their method in a number of ways that are described below. All of the stellar structure equations, including those in the equation of state, are self-consistently modified, as opposed to arbitrarily altering a single quantity. In this way, modifications to the efficiency of thermal convection are accounted for in a more complete fashion, owing to the full thermodynamic treatment of the magnetic field. Overall, the approach used to model magnetic effects can be considered analogous to the parameterized mixing length treatment of convection.1 The viability of the models is tested against results from a recent study that characterized the DEB EF Aquarii (Vos et al. 2012). EF Aquarii (HD 217512; henceforth EF Aqr) is a solar-type DEB found to contain two components displaying drastically inflated radii (Vos et al. 2012). Fundamental properties of the 1 In so far as reducing an inherently nonlinear, three-dimensional process into terms suitable for a onedimensional model. 117 Table 5.1 Fundamental stellar properties of EF Aquarii. Property EF Aqr A EF Aqr B M (M ⊙ ) 1.244±0.008 0.946±0.006 R (R ⊙ ) Teff (K) 1.338±0.012 [Fe/H] 6150±65 0.956±0.012 0.00±0.10 5185±110 system are quoted in Table 5.1. What is most striking is the similarity of the secondary to the Sun and the entire system to α Centauri A and B, in terms of the stellar masses and composition. Although the secondary appears similar to both α Cen B and the Sun, its radius appears to be about 10% larger than one would expect based on standard stellar evolutionary calculations. The effective temperature of the primary is nearly 300 K cooler than standard models predict, further revealing that both components suffer from substantial radius inflation. In an effort to reconcile the observations with predictions from theoretical models, Vos et al. (2012) reduced the value of the convective mixing length. They found αMLT of 1.30 and 1.05 were required for the primary and secondary, respectively, compared to their solarcalibrated value of 1.68. They concluded that fine-tuning the models allows for an accurate description of the observed properties. Reduction of the required convective mixing length may be physically motivated in two ways: (1) naturally inefficient convection and (2) magnetically suppressed convection. While we must be careful to not read too much into the reality of mixing length theory, in stellar evolution models the mixing length is an intrinsic “property” of convection. Thus, reducing the mixing length is akin to saying convection is not very efficient at transporting excess energy. Bonaca et al. (2012) calibrated the convective mixing length for solar-like stars using as- 118 teroseismic results provided by the Kepler Space Telescope. They found that the value of αMLT in stellar models is tied to stellar properties (i.e., log g , log Teff , and [M /H]). Applying the Bonaca et al. (2012) empirical calibration to the stars in EF Aqr, we find that the primary and secondary component require αMLT = 1.68 and 1.44, respectively. Again, compared with their solar calibrated value of αMLT = 1.68. The asteroseismically adjusted mixing lengths are significantly larger than the fine-tuned values determined by Vos et al. (2012). Therefore, it appears that naturally inefficient convection is insufficient to explain the inflated radii of EF Aqr. We are left with the option that magnetic fields may possibly be to blame. In what is to follow, we describe a self-consistent approach to modeling the effects of a globally pervasive magnetic field with application to the EF Aqr system. A complete description of the magnetic perturbation is provided in Sections 5.2 – 5.4. Section 5.5 demonstrates the ability of the invoked perturbation to reconcile the models with observations through a case study of the EF Aqr system. We conclude with a further discussion of the results and their implications in Section 5.6. 5.2 Magnetic Perturbation 5.2.1 Magnetic Field Characterization Investigating the effects of a global magnetic field on the interior structure of a star over long time baselines, requires formulating a purely three-dimensional (3D) phenomenon in terms suitable for a one-dimensional (1D) numerical model. Unfortunately, full 3D magnetohydrodynamic (MHD) models are not yet capable of modeling stellar magnetic fields over the long time baselines required for stellar evolutionary calculations. This is in part due to the rapid2 diffusion of the magnetic field and the immense computational time required. Therefore, in order to probe the effects of a magnetic field, we seek to avoid directly solving 2 Relative to a typical stellar lifetime. 119 the induction equation ∂B = ∇ × (u × B) + η∇2 B. ∂t (5.1) While not actively seeking a solution to the full suite of 3D MHD equations, it is possible to use the theoretical framework of MHD to provide a reasonably accurate 1D description of a magnetic field and its associated properties. Ultimately, we are able to describe a magnetic field in terms of the MHD equations and then project out the radial component, the component necessary for stellar evolutionary model computations. The spatial and temporal evolution of a given magnetic field are governed, quite naturally, by Maxwell’s equations, presented here in Gaussian-cgs, ∇·E = 0 ∇×E = − (5.2) 1 ∂B c ∂t ∇·B = 0 ∇×B = 4π J c (5.3) (5.4) (5.5) where within the stellar plasma, we assume any regions of excess charge inducing an electric potential will rapidly neutralize owing to the mobility of other charges (Debye shielding). Thus, we can safely assume that the plasma is electrically neutral, ρ e = 0. For simplicity, we here made another assumption, that temporal variations of the large-scale field are small, suggesting that the conduction current dominates the displacement current. Now, let us consider the interactions between the electric and magnetic fields within a dense, ionized fluid moving with arbitrary velocity, u. For slow temporal evolution, nonrelativistic dynamics may be described by a single conducting fluid that obeys the classical equations of hydrodynamics coupled with the equations of electromagnetism; the MHD equations (Jackson 1999). Considering a perfectly conducting, non-viscous, non-rotating, compressible fluid in the presence of a gravitational field, the MHD equations governing 120 the system are Ohm’s law for a moving fluid, ¶ µ u×B , J = σ E+ c (5.6) ∂ρ m + ∇ · (ρ m u) = 0, ∂t (5.7) the equation of mass continuity, where ρ m is the mass density, and the fluid equation of motion, ρm du J × B ← → = − ∇ · P + ρm g dt c (5.8) ← → with g being the gravitational field vector and P representing the gas pressure tensor. The electromagnetic term in the fluid equation of motion is associated with the assumption that a magnetic field permeates the plasma. However, we have neglected forces associated with any electric fields, for reasons detailed above. Since, a priori, we have no knowledge of the current density within a given fluid, we replace the current density within Equation (5.8) using Equation (5.5). The equation of motion may now be written as ρm 1 du ← → = (∇ × B) × B − ∇ · P + ρ m g. d t 4π (5.9) The magnetic force term may be rewritten with the help of the vector identity ∇(a · b) = (a · ∇)b + (b · ∇)a + a × (∇ × b) + a × (∇ × b), which for a = b yields 1 − a × (∇ × a) = (a · ∇)a − ∇a 2 2 (5.10) (5.11) Knowing that magnetic fields are divergenceless by Equation (5.4), the momentum equation 121 becomes ρm 1 1 du ← → = (B · ∇)B − ∇B 2 − ∇ · P + ρ m g d t 4π 8π (5.12) as a result. The equation of motion may be further simplified by recognizing that the magnetic force terms may be rewritten as the divergence of a tensor (Gurnett & Bhattacharjee 2005). Let us define the components of this tensor as Tr,s = − Br B s B2 + δr,s 4π 8π (5.13) where δr,s is the Kronecker delta. Then, in vector notation, BB ← →B2 ← → + I T =− 4π 8π (5.14) ← → with I representing the identity tensor. Taking the s component of the divergence of this newly formed tensor, ¸ · X ∂ B2 Br B s ← → − (∇ · T )s = − + δr,s , − 4π 8π r ∂x r (5.15) ¸ µ ¶ · ∂ B2 ∂B r ∂B s 1 X ← → − + Bs Br . − (∇ · T )s = 4π r ∂x r ∂x r ∂x s 8π (5.16) which becomes Equation (5.4) imposes the condition that ∇·B = X ∂B r r ∂x r = 0. (5.17) reducing Equation (5.16) to ¶ µ ¶ µ ∂ B2 1 X ∂B s ← → − − (∇ · T )s = Br . 4π r ∂x r ∂x s 8π 122 (5.18) In vector notation, the above equation is precisely the same as the Lorentz force term in the equation of motion, 1 1 ← → (B · ∇)B − ∇B 2 . − (∇ · T )s = 4π 8π (5.19) The final form of Equation (5.8) can then be written in terms of the divergence of two tensors, ρm ³← du → ← →´ = −∇ · T + P + ρ m g. dt (5.20) The magnetic tensor introduced above may be considered an anisotropic pressure tensor, where the pressures it describes are intrinsic properties of the magnetic field. Further justification of this interpretation is provided below. Stars, however, are considered to be in hydrostatic equilibrium, suggesting that the macroscopic variables of the system change only on timescales longer than the free-fall time (Weiss et al. 2004). Therefore, the lefthand side of the equation of motion vanishes, ³← → ← →´ ∇ · T + P = ρ m g, (5.21) which is a statement of magnetohydrostatic equilibrium. Let us now analyze the two components of the electromagnetic contribution to the equation of motion. There are two terms associated with the Lorentz force. The first term, −∇ B2 8π (5.22) is the force generated by a gradient in the square of the magnetic field strength. Drawing an analogy with the gas pressure of the system (∇P gas ), we recognize the above term represents the force associated with the gradient of a “magnetic pressure.” This force is directed isotropically from regions of strong magnetic field to regions of weak magnetic field. 123 The second term in the Lorentz force, 1 (B · ∇)B 4π (5.23) is directed opposite along the radius of curvature of the magnetic field line. This is analagous to tension in a string and may be interpretted as the magnetic tension force. Together, the magnetic pressure and tension exert a force on the plasma in which the magnetic field is threaded. Since the magnetic field exerts a force on the plasma, there is an energy associated with the magnetic field, Umag = B2 . 8π (5.24) It is now possible to consider how the introduction of a magnetic field would affect the canonical stellar structure equations. 5.2.2 Stellar Structure Perturbations At the most fundamental level, one-dimensional stellar evolution codes simultaneously solve a set of four coupled, first-order differential equations.3 They are dr dm dP dm dT dm dL dm 1 4πr 2 ρ Gm = − 4πr 4 T dP = ∇s P dm dU P d ρ = ǫ− + 2 dt ρ dt = (5.25) (5.26) (5.27) (5.28) the equation of mass conservation, hydrostatic equilibrium, energy transport, and energy conservation, respectively. Qualitatively, we can easily predict how these equations will be 3 There are additional equations often included to account for atomic diffusion. While included in DSEP, we do not seek perturbations to these equations at the present time. See Mathis & Zahn (2005) for a rigorous treatment of mixing associated with magnetic fields. 124 altered by the presence of a magnetic field which may then be translated into a quantitative description. The equation of mass conservation should be unaltered by any magnetic perturbation. Of course, this is assuming that mass removed by stellar winds is negligible and that transient events that may remove mass (i.e., flares, coronal mass ejections) are neglected. The stated conditions hold for our approach, thus, 1 dr = . d m 4πr 2 ρ (5.29) Note that we have dropped the subscript m on the density and assume all references to density are specifically to the mass density, unless otherwise noted. Hydrostatic equilibrium, as we saw earlier in Equation (5.21), is modified through the inclusion of the magnetic pressure and tension. Projecting out the radial component of the magnetic pressures, we are able to adapt the three-dimensional concept for one-dimensional models. Therefore, we have µ 2 ¶¸ · Gm 1 B (B · ∇)B dP =− −∇ + · r̂. 4 2 dm 4πr 4πr ρ 4π 8π (5.30) The precise handling of the vector magnetic field within the code will be discussed later. The final form of the energy transport equation is the same as if there were no perturbation. Namely, T dP dT = ∇s dm P dm (5.31) where ∇s is the local temperature gradient of the ambient plasma. Magnetic perturbations to the stellar structure equations will self-consistently alter the temperature gradient through various thermodynamic considerations. Of greatest importance will be the affects on the treatment of convection. The full treatment will be discussed in the next section. 125 Finally, there are changes to the quantities present in the canonical equation of energy conservation in stellar evolution. Modifications to these quantities arise from the treatment of the specific thermodynamic equations (discussed in the next section). Additionally, there are terms that are purely electromagnetic in origin. The final form of the energy conservation equation is dU P d ρ Q ohm F Poynt dL = ǫ− + 2 + + . dm dt ρ dt ρ ρ (5.32) Aside from the first three standard terms on the right-hand side, there are two additional electromagnetic terms. First, there is a Poynting flux associated with the field, F Poynt = c E × B, 4π (5.33) although as discussed above, we assume the E -field is zero everywhere. Next, energy is also associated with the Ohmic dissipation of electric currents brought about by the resistive nature of the plasma. Q ohm ∝ I 2 R, (5.34) where I is the electric current and R is the resistance of the medium. Here, electrical currents are converted to heat that then is transmitted to the surrounding plasma. Since we have assumed an infinitely conducting plasma, this energy term goes immediately to zero. 5.2.3 Thermodynamic Structure The effects of a global magnetic field are introduced into the thermodynamic framework supplied by DSEP following the approach outlined by Lydon & Sofia (1995, hereafter LS95). Our aim in this section is to provide a detailed mathematical description of the magnetic perturbation introduced by LS95. Places where we diverge from their original prescription 126 will be noted. At the core of the LS95 method is the specification of a new thermodynamic state variable, χ, such that χ = χ(r, ρ) = Umag ρ = B (r )2 . 8πρ (5.35) The state variable χ is the magnetic energy per unit mass and B (r ) is the magnetic field strength at radius r . Unlike LS95, our definition of χ depends on the radial distribution of the magnetic field strength and also on the density of the stellar plasma. Originally, LS95 favored a mass-depth-dependent function, χ(Mr ). However, we moved away from this prescription when we realized several thermodynamic derivatives became divergent. Once the magnetic field strength is specified throughout the star, it is straightforward to calculate χ at each point within the model. The energy associated with the magnetic field arises due to forces exerted by the magnetic field on the plasma. These forces are represented by the anisotropic pressure tensor presented in Equation (5.12). To handle the stress tensor in 1D, we convert the pressure tensor to a scalar pressure, by taking the trace of the pressure tensor to yield the mean magnetic pressure, 1 ³← →´ 〈P mag 〉 ∼ Tr T . 3 (5.36) Substituting the expression for the magnetic stress tensor from Equation (5.14), µ ¶ 1 BB ← →B2 〈P mag 〉 ∼ Tr − + I . 3 4π 8π (5.37) Since we are not solving the full set of MHD equations, we look, instead, to set approximate upper and lower limits on the scalar pressure. Assuming a Cartesian coordinate system, if we imagine the magnetic field is parallel to the z -axis, or for a star, the rotational axis, then 127 we may expand the pressure tensor to read 0 0 B 2 /8π ← → T = 0 B 2 /8π 0 0 0 −B 2 /4π + B 2 /8π . (5.38) Note that there is an isotropic magnetic pressure associated with each diagonal element along with the additional magnetic tension term in the final element. Since tension is directed along the field line, the tension exists in the z -direction only, in this instance. Taking the trace, we find µ ¶ 1 B 2 3B 2 〈P mag 〉 ∼ − + 3 4π 8π (5.39) since the negative tension term only acts along a single direction at any point along a given field line while the pressure acts isotropically. Therefore, µ ¶ 1 B2 1 〈P mag 〉 ∼ = χρ. 3 8π 3 The above equation is satisified for a field which has a strong magnetic tension component. However, if we assume that there is no tension at all, then, following the same procedure, µ ¶ 1 3B 2 = χρ. 〈P mag 〉 ∼ 3 8π These two approaches limit the strength of the scalar magnetic pressure within the onedimensional framework. Explicitly, 1 χρ ≤ 〈P mag 〉 ≤ χρ. 3 (5.40) Defining a “geometry parameter,” akin to LS95, allows us to emulate the effects of having 128 a strongly curved field or a field with no curvature, and varying degrees between the two extremes. This geometry parameter is defined such that we recover the average magnetic pressure for each case above, ¡ ¢ 〈P mag 〉 = γ − 1 χρ ≡ P χ (5.41) where the geometry parameter γ= 2 4/3 tension-free . (5.42) maximum tension In both cases, the appropriate expression for the scalar magnetic pressure is returned. With the magnetic energy density and pressure formulated as scalars, we have successfully converted the inherently three-dimensional magnetic field into a one-dimensional magnetic perturbation. In the process, we have also reproduced the scalar parameters originally presented by LS95. As we have seen, magnetic fields carry along an associated pressure, tension, and energy density. The introduction of magnetic terms into the equations of stellar structure then necessitates the inclusion of these quantities in the EOS of the system. Using the definitions introduced above for the magnetic energy density (χ) and the magnetic pressure (Pχ ) it is possible to develop a magnetic version of the thermodynamics involved in standard stellar evolution. Beginning with a statement of the first law of thermodynamics, dQ = T d S = dU + P dV, (5.43) we recognize that each term contains both the standard gas and radiation terms as well as 129 a new magnetic contribution, dQ = T (d S 0 + d S χ ) = (dU0 + dUχ ) + P 0 dV, (5.44) where no work is performed by the magnetic perturbation. Hereafter it is assumed that any unsubscripted variables refers to the total quantity while gas and radiation are lumped together under the subscript 0 (zero) convention and magnetic variables carry a subscript χ. To write the first law in terms of the total pressure, we substract off the magnetic contribution, £ ¤ T d S = dU + P − (γ − 1)χρ dV. (5.45) Stellar evolution calculations are performed using specific quantities, meaning that ρ = V −1 . So, T d S = dU + P dV − (γ − 1) χ dV V (5.46) Equation (5.46) is the new, “non-standard” first law of thermodynamics. It suggests that ¡ ¢ V or ρ = f P, T, χ (5.47) ¡ ¢ U or S = f ρ, T, χ . (5.48) and Following the derivation by LS95, explicitly writing out the other state variables illustrates the effects of adding the magnetic perturbation. The total pressure is the sum of the individual contributions, 1 P = RρT + aT 4 + χρ(γ − 1), 3 (5.49) where R is the specific gas constant and a = 4σ/c is the radiation constant. Rearranging 130 and solving for the density gives, ρ= P − aT 4 /3 . RT + (γ − 1)χ (5.50) Additionally, the total specific internal energy of the system is the sum of the individual energies, 3 aT 4 U = RT + + χ. 2 ρ (5.51) These thermodynamic quantities are subject to the EOS, d ln ρ = αd ln P − δd ln T − νd ln χ. (5.52) The coefficients in the EOS are thermodynamic derivatives that complete the total derivative d ln ρ , ∂ ln ρ α= ∂ ln P µ ¶ (5.53) , T,χ µ ∂ ln ρ δ=− ∂ ln T ¶ , (5.54) ∂ ln ρ ν=− ∂ ln χ ¶ . (5.55) P,χ and µ P,T The coefficients δ and ν carry a negative sign in anticipation that without it, they would be negative quantities. Note that α carries no subscript and should not be confused with the convective mixing length parameter, αMLT . An immediate consequence of altering the thermodynamic variables are the effects on the specific heats. The specific heat at constant pressure, µ dQ cP = dT ¶ µ dU = dT P,χ ¶ µ ¶ £ ¤ dV + P − (γ − 1)χρ , d T P,χ P,χ 131 (5.56) and at constant volume, cV = µ dQ dT ¶ V,χ = µ dU dT ¶ (5.57) , V,χ The specific heats may be rewritten by deriving expressions for each of the terms in the above equations. Assuming a small infinitesimal change in the total specific energy, µ ∂U dU = ∂V ¶ µ ∂U dV + ∂T T,χ ¶ µ ¶ ∂U dT + ∂χ V,χ (5.58) d χ, V,T we can divide take the derivative with respect to temperature at constant pressure and magnetic energy (d P = d χ = 0), µ dU dT ¶ µ ∂U = ∂V P,χ ¶ T,χ µ dV dT ¶ µ ∂U + ∂T P,χ ¶ (5.59) . V,χ The difference between the two specific heats is then c P − cV = (µ ∂U ∂V ¶ T,χ £ ¤ + P − (γ − 1)χρ )µ dV dT ¶ (5.60) . P,χ Simplifying the above equation is possible using the reciprocity relation (Weiss et al. 2004). Starting with the first law, Equation (5.46), T d S = dU + P dV − (γ − 1) χ dV V (5.61) we may expand the internal energy term producing the equation, µ ∂U T dS = ∂V ¶ µ ∂U dV + ∂T T,χ ¶ µ ∂U dT + ∂χ V,χ ¶ V,T d χ + P dV − (γ − 1) χ dV. V (5.62) In Equations (5.62) and (5.58), d S and dU are total differentials. Therefore, we must have ∂2 S/(∂V ∂T ) = ∂2 S/(∂T ∂V ) and ∂2U /(∂V ∂T ) = ∂2U /(∂T ∂V ). Utilizing this fact, we may 132 substitute Equation (5.62) into Equation (5.58) to find µ ∂U ∂V ¶ µ ∂P =T ∂T T,χ ¶ − P + χV −1 (γ − 1). V,χ (5.63) This is the reciprocity relation mentioned earlier. With the reciprocity relation, the difference in specific heats may be simplified. Substituting Equation (5.63) into Equation (5.60), µ ∂P c P − cV = T ∂T ¶ V,χ µ ∂V ∂T ¶ . (5.64) P,χ Writing the EOS in Equation (5.52) in terms of the volume reveals that we may further simplify Equation (5.64), µ ¶ µ ¶ µ ¶ dP d T ∂ lnV dχ ∂ lnV ∂ lnV dV = + . V ∂ ln P T,χ P ∂ ln T P,χ T ∂ ln χ P,T χ (5.65) At constant volume and magnetic energy, µ dP dT ¶ µ ∂V =− ∂T V,χ ¶ P,χ µ ∂V ∂P ¶−1 T,χ = Pδ , Tα (5.66) which yields our final equation for the difference in specific heats c P − cV = P δ2 . ρT α (5.67) The difference in specific heats is written just as it would be without any magnetic perturbation. However, the individual quantities have been self-consistently modified to include the contribution from the magnetic perturbation. Knowledge of the specific heats enables the computation of the change in energy through a 133 change in heat within a given stellar layer. Recall, dQ = "µ ∂U ∂V ¶ T,χ + P − (γ − 1)χV −1 # µ ∂U dV + ∂T ¶ V,χ d T + d χ. (5.68) The change in heat may be simplified to read µ ∂P dQ = cV d T − T ∂T ¶ V,χ dV + d χ, (5.69) which is further reducable using Equation (5.66) and by using the relation dV = ρ −2 d ρ , giving, dQ = cV d T − Pδ dρ + d χ. ρα ρ (5.70) Rewriting the preceding equation using the EOS from Equation (5.52), · ¸ dT dχ dP Pδ −δ −ν α + d χ, dQ = cV d T − ρα P T χ (5.71) which can be rearranged to be µ µ ¶ ¶ P δ2 Pδ dP P δν dQ = cV + + + 1 d χ. dT − ρT α ρ P ραχ (5.72) Finally, recognize that the quantity multiplying d T is c P according to Equation (5.67), meaning the change in heat becomes µ ¶ δ P δν dQ = c P d T − d P + + 1 d χ. ρ ραχ (5.73) An alternative form of this equation is used by the stellar evolution calculation, where the infinitesimal changes are logarithmic quantites. To be explicit, µ ¶ P δν Pδ d ln P + + χ d ln χ. dQ = c P T d ln T − ρ ρα 134 (5.74) We now have an equation for the change in heat within a stellar layer. However, the presence of a purely magnetic term must be considered. The introduction of the pure magnetic term suggests that energy is removed from the internal energy of the system and converted into magnetic energy. If we assume that magnetic phenomena are generated through a rotationally driven dynamo action, then we instead infer that rotational energy is the source of pure magnetic term. Since our models do not specifically treat rotation, we must discard the pure magnetic energy term when computing the change in heat in the stellar layer. Thus, the change in heat to be considered in the stellar luminosity equation is dQ = c P T d ln T − Pδ P δν d ln P + d ln χ, ρ ρα (5.75) or for the purposes of the derivations to follow, P δν δ d χ. dQ = c P d T − d P + ρ ραχ (5.76) This concludes the changes to the general thermodynamic structure of a stellar evolution model. The rest of this section contains the effects a magnetic perturbation would have on the equations for convective motion. The changes made are a direct consequence of the changes made to the thermodynamic structure and are, thus, closely related. We preface with the adiabatic temperature gradient. Adiabaticity requires constant entropy (no heat exchange in the system) and constant magnetic energy, d ln T ∇ad = d ln P µ ¶ . (5.77) S,χ Therefore, dQ = 0 = c P T d ln T − 135 Pδ d ln P. ρ (5.78) This can be easily solved to give the definition of the adiabatic gradient, d ln T ∇ad = d ln P µ ¶ S,χ = Pδ , ρT c P (5.79) which has precisely the same form as the non-magnetic derivation. 5.2.4 Magnetic Mixing Length Theory Convection is determined to occur in regions where a given fluid parcel is unstable to a small displacement in the radial direction. The primary method of determining convective stability is to analyze the density of a generalized fluid parcel. Parcels that are less dense than their surroundings will travel radially outward until they reach a height within the star at which the surrounding fluid has the same density as the parcel. Upon reaching this point, the fluid parcel is assumed to fully mix with its surroundings becoming indistinguishable from the rest of the fluid. Conversely, if a fluid element is more dense than surrounding fluid, it will sink down to a greater depth in the star, following the same trend as a rising convective element. In either case, gravity is the restoring force. So long as the velocity of the unstable convective parcel is slower than the local sound speed, the parcel will remain distinguishable from its surroundings. Namely, information about pressure changes outside the parcel can be transmitted throughout the parcel sufficiently fast to keep the parcel in constant pressure equilibrium. The distance over which a fluid parcel travels before mixing is the well-known “mixing length.” Mixing length theory (MLT) has been well established as a local means of prescribing convection for a one-dimensional stellar evolution code (Vitense 1953; Böhm-Vitense 1958). At locations where various differences in prescriptions of MLT occur, we will specify our assumptions. Stability of a fluid parcel against convection is determined by comparing the density of a 136 fluid parcel against its surroundings. For the rest of the document, let us use the definition that (5.80) D A = Ae − A s where e and s represent values for the fluid element and surroundings, respectively. Then, Dρ = ·µ dρ dr ¶ µ dρ − dr e ¶¸ ∆r. (5.81) s To ensure stability, we require Dρ > 0, meaning the element is stable against small radial displacements when µ dρ dr ¶ µ dρ − dr e ¶ s > 0. (5.82) Calling upon the EOS from Equation (5.52) allows us to expand the above equation. The change in density of the element is µ ¶ µ ¶ µ ¶ µ ¶ 1 dρ d ln P d ln T d ln χ =α −δ −ν ρ dr e dr e dr e dr e (5.83) while the change in density of the surrounding material is µ ¶ ¶ ¶ ¶ µ µ µ 1 dρ d ln P d ln T d ln χ =α −δ −ν ρ dr s dr s dr s dr s (5.84) Included in the previous two equations is the radial gradient of the magnetic energy. We can calculate how the background magnetic energy changes with radius, but it is difficult to specify how the magnetic energy changes for the fluid parcel. This is because the parcel may exchange magnetic energy with its surroundings. If the parcel carries its original magnetic energy as it moves, then µ d ln χ dr ¶ e = 0. (5.85) On the other hand, if the magnetic energy in the parcel is always equal to that of its sur- 137 roundings, there must be a flux of χ as the parcel moves. Thus, µ d ln χ dr d ln χ = dr e ¶ µ ¶ (5.86) . s It is therefore advantageous to relate the spatial gradient of the magnetic energy of the parcel to that of the surroundings by introducing a free parameter, f , such that µ d ln χ dr d ln χ =f dr e ¶ µ ¶ (5.87) s where f varies between 0 and 1. Later, we shall attempt to physically motivate a fixed value for f . Substituting Equations (5.83), (5.84), and (5.87) into Equation (5.82) gives d ln T −δ dr d ln χ + (1 − f )ν dr e µ ¶ µ ¶ d ln T +δ dr s µ ¶ s >0 (5.88) The pressure derivatives cancelled due to our assumption that the parcel is in pressure equilibrium with its surroundings. If we multiply the above equation through by the local pressure scale height,4 HP = − dr d ln P (5.89) we end up with an expression for stability against convection, d ln T δ d ln P µ d ln χ − (1 − f )ν d ln P e ¶ µ ¶ d ln T −δ d ln P s µ ¶ s >0 (5.90) Before we go any further, let us now define three new gradient variables. The temperature gradient of the ambient plasma, d ln T ∇s ≡ d ln P µ 4 ¶ , (5.91) s Other variations of MLT use the temperature or density scale height, but our assumption of pressure equilibrium makes the pressure scale height a natural choice. 138 which is the same gradient that appears in Equation (5.31), the temperature gradient of the displaced fluid parcel, d ln T ∇e ≡ d ln P µ ¶ (5.92) , e and the gradient of the magnetic energy per mass unit, d ln χ ∇χ ≡ d ln P µ ¶ (5.93) . s This final gradient we can expand to also read d ln χ ∇χ = dr µ ¶ µ s dr d ln P ¶ (5.94) s This allows us to write the stability criterion as ν ∇e − (1 − f ) ∇χ − ∇s > 0. δ (5.95) The stability criterion above can be simplified further by considering the temperature gradient of the element, ∇e . Our selection of the free parameter f is of importance. By assuming f = 0, we are at liberty to assume complete adiabaticity because no energy is exchanged between the fluid parcel and its surroundings. Therefore, ∇e → ∇ad . When f 6= 0, then any heat transferred away from the parcel will be in the form magnetic energy (dQ = d χ). Equation (5.73) then tells us that P δν δ d χe . 0 = c P d Te − d P e + ρ ραχ (5.96) Multiplying through by the pressure scale height gives µ d ln T d ln P ¶ µ P δν d ln χ Pδ −f = , T ρc P T ραc P d ln P s e ¶ 139 (5.97) which can be reduced using the gradient definitions, ∇e = ∇ad − f ν ∇ad ∇χ . α (5.98) Substituting this back into the stability criterion given by Equation (5.95) yields ∇ad − f ν ν ∇ad ∇χ − (1 − f ) ∇χ − ∇s > 0. α δ (5.99) This is the final form of the stability criterion. For a fluid parcel to be stable against convection, this criterion must be satisfied. Under the assumption that convection will occur whenever a fluid parcel is convectively unstable, convection will occur when ³ ν ´ ν ∇s − ∇ad 1 − f ∇χ + (1 − f ) ∇χ > 0. α δ (5.100) Expressed compactly with the definition of ∇e , ν ∇s − ∇e + (1 − f ) ∇χ > 0 δ (5.101) This completes our discussion of convective stability. However, it requires that we know the value of ∇s , the temperature gradient of the surrounding material. In principle, ∇s is not known a priori. We may not know ∇s at any given point in a stellar model, but we do know the total flux of energy that must be transported across any given layer. Stars are finely balanced. The energy generated by gravitational contraction or nuclear fusion in the core must be transferred to the surface so that it may be radiated away. We can use this to our advantage and solve for the ambient temperature gradient. Imagine a fictitious gradient, ∇rad , that represents what the background temperature gradient would have to be to transport all of the energy by radiation. The total energy flux is 140 then F tot = 4acG T 4 M r ∇rad . 3 κP r 2 (5.102) Here, a is the radiation constant, G is Newton’s gravitational constant, c is the speed of light in a vacuum, and κ is the opacity of the plasma. Additionally, Mr is the mass contained within the radius r . However, if convection is present F tot = 4acG T 4 M r ∇s + F conv . 3 κP r 2 (5.103) The convective flux may be considered a flux of enthalpy D H = DU + P DV + V DP (5.104) because it is not only transporting excess internal energy, but is also doing work on the surroundings. It has to “push” material out of the way to move. The last term in Equation (5.104) is zero due to the assumed pressure equilibrium. Therefore, D H = DU + P DV = DQ. (5.105) We may then rationalize that the convective flux is F conv = ρu conv DQ, (5.106) where uconv is the velocity of the convecting element. From Equation (5.73), we may rewrite DQ as µ ¶ P δν DQ = c P DT + + 1 Dχ, ραχ (5.107) · µ ¶ ¸ P δν F conv = ρu conv c P DT + + 1 Dχ ραχ (5.108) which yields, 141 In the original formulation by LS95, they separated out the pure magnetic flux term, (5.109) F conv, χ = ρu conv Dχ, which represents the magnetic energy transported by a convecting element when f 6= 1. Since it is not presently clear what the value of f should be, we leave the pure magnetic flux in the equation. We now introduce the concept of a mixing length to aid our evaluation of the temperature excess, DT , and magnetic energy excess Dχ of the convecting element. After the parcel travels a distance ℓm (= αMLT HP ), we assume the fluid parcel completely mixes with the surrounding material and deposits its excess energy. The parcels crossing any given layer within a convection zone will not have necessarily started their ascent or descent at the same location. Therefore, we assume that a typical parcel started at a distance ℓm /2 from the point under consideration (Böhm-Vitense 1958). The results of MLT are relatively insensitive to this choice (Henyey et al. 1965). The average temperature excess taking into account all of the various parcels that may be dissolving at a given location may be written, D ln T = 1 ∂(DT ) 1 ∂(DT ) ℓm ∆r = , T ∂r T ∂r 2 (5.110) where, ∂ ∂(D ln T ) = ∂r ∂r ·½µ d ln T dr ¶ e − µ d ln T dr ¶¾ s ¸ ∆r . (5.111) Multiplying the terms inside the curly braces by 1 = −(d r /d ln P )HP−1 , · ¸ ∂(D ln T ) ∆r ∂ = , (∇s − ∇e ) ∂r ∂r HP 142 (5.112) which implies that 1 ∂(D ln T ) = (∇s − ∇e ) . ∂r HP (5.113) The final expression for the temperature excess D ln T is then D ln T = (∇s − ∇e ) ℓm . 2HP (5.114) Evaluating the magnetic energy excess in a similar manner gives D ln χ = (1 − f )∇χ ℓm 2HP (5.115) The above equation nicely illustrates that if f = 1, the magnetic energy excess between an element and its surroundings is zero because they are continually transferring energy to one another. The rate at which the excess energy is deposited by a convective parcel is also governed by the convective velocity uconv . An expression for uconv may be derived knowing that buoyancy is the restoring force acting on the parcel. The radial buoyancy force per unit mass is k r = −g D ln ρ. (5.116) Once again, Equation (5.52) can be used to substitute for D ln ρ , again accounting for pressure equilibrium. Thus, ¡ ¢ k r = g δD ln T + νD ln χ (5.117) It is now possible to calculate the work done on the fluid parcel. Over half a mixing length, the work done is µ ¶ £ ¤ ℓm 1 ℓm kr . = g δD ln T + νD ln χ 2 2 4 143 (5.118) Using the definition of the energy excesses derived above, µ ¶ h i ℓ2 ν ℓm 1 m kr . = g δ (∇s − ∇e ) + (1 − f )∇χ 2 2 δ 8HP (5.119) Now make the crude assumption that half of the energy derived from the buoyancy force is converted into kinetic energy of the fluid parcel. The other half is transferred to the surroundings because the parcel has to move the surrounding material out of the way. The convective velocity may then be written, h i ℓ2 ν m 2 u conv = g δ (∇s − ∇e ) + (1 − f )∇χ δ 8HP (5.120) The addition of the convective velocity does not yet close the system of equations presented thus far. We need yet another equation, which we can develop by account for the fact that as the parcel travels, it does so non-adiabatically. There are three ways for the parcel to change temperature: adiabatic expansion (or contraction), radiative losses, and the flux of magnetic energy. We start Equation (5.73), µ dQ dr ¶ e = cP µ dT dr ¶µ µ ¶ µ ¶ dχ δ dP P δν +1 − + . ρ dr e ραχ dr e e ¶ (5.121) where the change in heat on the left-hand-side is now the sum of the radiative losses and the change of magnetic energy. In presenting the terms associated with radiative losses, we will not provide a full derivation, but only cite the final form. The radiative losses for a semi-opaque bubble is taken to be the average between the flux for a transparent bubble and that of a completely opaque bubble (Henyey et al. 1965; Mihalas 1978). The resulting expression is, ¸ · 2acT 4 ω (∇s − ∇e ), (radiative losses) = − ρu conv HP 1 + ÿω2 (5.122) where ÿ is a constant describing the shape and temperature distribution of the convective 144 parcel (Henyey et al. 1965) and ω = κρℓm , related to the mean free path of a photon. The change of magnetic energy is given by Equation (5.87). Combining the radiative loses and magnetic energy transfer between the parcel and the surrounding medium, ¸ µ ¶µ ¶ ¶ µ ¶ µ ¶ · µ dχ dχ 2acT 4 δ dP P δν ω dT (∇s − ∇e ) + +1 − = cP − + . ρu conv HP 1 + ÿω2 dr e dr e ρ dr e ραχ dr e (5.123) The preceding equation is easily reducible by recognizing the appearance of the pure magnetic terms on each side cancel out, ¸ µ ¶ µ ¶ µ ¶µ ¶ · dT δ dP 2acT 4 P δν d χ ω (∇s − ∇e ) = c P − + . − ρu conv HP 1 + ÿω2 dr e ρ dr e ραχ d r e (5.124) To yield a better form for the equation of heat exchange between the parcel and the surroundings, we multiply by HP T −1 = −T −1 (d r /d ln P ) to get ¸ µ ¶ · d ln T P δν ω Pδ 2acT 3 (∇s − ∇e ) = c P +f ∇χ − 2 ρu conv 1 + ÿω d ln P e T ρ ραT (5.125) If we now divide by c P ¸ · 2acT 3 Pδ P δν ω (∇s − ∇e ) = ∇e − +f ∇χ 2 ρu conv c P 1 + ÿω T ρc P ραT c P (5.126) ¸ · ν ω 2acT 3 (∇s − ∇e ) = (∇e − ∇ad ) + f ∇ad ∇χ . 2 ρu conv c P 1 + ÿω α (5.127) which becomes We now have a final set of five equations which we can use to solve for the temperature gradient of the system. The total flux of energy transported by a fictitious radiative energy gradient (Equation 5.102), F tot = 4acG T 4 M r ∇ad , 3 κP r 2 145 the total flux transported by a combination of radiation and convection (Equation 5.103), F tot = 4acG T 4 M r ∇s + F conv , 3 κP r 2 the convective flux (Equation 5.108), ¶ ¸ · µ P δν + 1 Dχ , F conv = ρu conv c P DT + ραχ the convective velocity (Equation 5.120), h i ℓ2 ν m 2 u conv = g δ (∇s − ∇e ) + (1 − f )∇χ , δ 8HP and the exchange of heat between the parcel and the surrounding as the parcel moves (Equation 5.127), ¸ · ν ω 2acT 3 (∇s − ∇e ) = (∇e − ∇ad ) + f ∇ad ∇χ . 2 ρu conv c P 1 + ÿω α Using these five equations, we are in a position to solve for the temperature gradient of the ambient medium. Solution of the Magnetic Mixing Length Theory Equations Solving the system of five equations presented above is quite tedious. If we first define several new variables, the process algebra becomes slightly more clean. First, we set the value of the parcel shape parameter, ÿ = 1/3, as is done in the standard version of MLT employed by our code (Böhm-Vitense 1958; Henyey et al. 1965). The new variables are defined in anticipation of the fun to come, ∂ ln ρ Q =δ=− ∂ ln T µ 146 ¶ , P,χ (5.128) γ0 = c P ρ 1 + ω2 /3 , 2acT 3 ω C= gQℓ2m 8HP , (5.129) (5.130) ¸ · ³ ν ´1/2 −1 ν 1/2 V = γ0C , ∇rad − ∇ad + f ∇ad ∇χ + (1 − f ) ∇χ α δ (5.131) 9 ω2 , A= 8 3 + ω2 (5.132) y = u convV γ0 (5.133) and With these new variables defined, we will work towards a solution for the equations in a series of seven steps. Each step will be labeled in bold lettering. Step 1: Rewrite the convective velocity (Equation 5.120) in terms of the new variables. µ y V γ0 ¶2 ¸ · ν = C ∇s − ∇e + (1 − f )∇χ Q (5.134) Step 2: Equation (5.127), the heat exchange equation, may be redefined with the new variables, ν 1 (∇s − ∇e ) = (∇e − ∇ad ) + f ∇ad ∇χ . γ0 u conv α (5.135) This enables us to derive an expression for ∇e . Replace (γ0 uconv )−1 with (V /y) using Equation (5.133) and distribute the first term, V ν V ∇s − ∇e = (∇e − ∇ad ) + f ∇ad ∇χ , y y α (5.136) and then combine like-terms for the gradients, ¶ µ ³ V V ν ´ ∇s + ∇ad 1 − f ∇χ = ∇e 1 + , y α y 147 (5.137) and divide by (V /y), ∇s + ³ ³ ν ´ y´ y ∇ad 1 − f ∇χ = ∇e 1 + . V α V (5.138) This now defines the temperature gradient of the convecting parcel as ∇e = ¡ ¢ ∇s + ∇ad 1 − f ν∇χ /α y/V (5.139) 1 + y/V Step 3: Find an expression for (∇s − ∇ad ) by substituting the Equation (5.139) into Equation (5.134). We start by expanding Equation (5.134), µ y γ0V ¶2 = C ∇s −C ∇e +C ν (1 − f )∇χ , Q (5.140) which becomes, upon substitution, µ y γ0V ¶2 = C ∇s −C à ! ¡ ¢ ∇s + ∇ad 1 − f ν∇χ /α y/V 1 + y/V +C ν (1 − f )∇χ . Q (5.141) Add the first two terms on the right hand side, µ y γ0V ¶2 = " # ¡ ¢ C ∇s +C ∇s (y/V ) −C ∇s −C ∇ad 1 − f ν∇χ /α y/V 1 + y/V +C ν (1 − f )∇χ . (5.142) Q Cancel equal terms and divide by C , # " ¡ ¢ µ ¶ ∇s (y/V ) − ∇ad 1 − f ν∇χ /α y/V ν y 2 1 + (1 − f )∇χ . = C γ0V 1 + y/V Q (5.143) Rearrange the terms such that we have (∇s − ∇ad ) in the numerator and bring the magnetic contribution to the left hand side, · µ ¶ ¸ (∇s − ∇ad )(y/V ) + ( f ν∇ad ∇χ /α)(y/V ) 1 y 2 ν . − (1 − f )∇χ = C γ0V Q 1 + y/V 148 (5.144) Divide both sides of the equation by (y/V ) to get, µ ¶ · ¸ (∇s − ∇ad ) + ( f ν∇ad ∇χ /α) ν V − (1 − f )∇χ = . y 1 + y/V C γ20V Q y (5.145) Finally, solve for (∇s − ∇ad ), " µ ¶# ³ ν V ν y´ (∇s − ∇ad ) = − (1 − f )∇ − f ∇ad ∇χ . 1 + χ 2 y V α C γ0V Q y (5.146) Step 4: Rewrite Equation (5.108) in terms of the convective velocity. Begin with Equation (5.120) and solve for (∇s − ∇e ), 2 u conv ¸ · ν = C (∇s − ∇e ) + (1 − f )∇χ . Q (5.147) Therefore, (∇s − ∇e ) = 2 u conv C − ν (1 − f )∇χ . Q (5.148) Now, using Equations (5.114) and (5.115) rewrite the convective flux (Equation 5.108), · µ ¶ ¸ P δν ℓm F conv = ρu conv c P T (∇s − ∇e ) + + 1 (1 − f )χ∇χ . ραχ 2HP (5.149) Expanding the above equation and recalling the definition of ∇ad from Equation(5.79), i ℓ h ν m . F conv = ρu conv c P T (∇s − ∇e ) + (1 − f )∇ad ∇χ + (1 − f )χ∇χ α 2HP (5.150) Replacing (∇s − ∇e ) with Equation (5.148), F conv = ρu conv c P T · 2 u conv C ¸ ν ν ℓm − (1 − f )∇χ + (1 − f )∇ad ∇χ + (1 − f )χ∇χ , Q α 2HP 149 (5.151) which rearranges to read F conv = ρu conv c P T · 2 u conv C + ν(1 − f )∇χ µ ∇ad 1 χ − + α Q ν ¶¸ ℓm , 2HP (5.152) Step 5: Substitute Equation (5.152) into Equation (5.103) to derive an expression for the total flux, · 2 µ ¶¸ u conv 4acG T 4 M r ∇ad 1 χ ℓm F tot = ∇s + ρu conv c P T + ν(1 − f )∇χ − + . 2 3 κP r C α Q ν 2HP (5.153) Distribute the terms associated with the convective flux to produce, ¶ µ 3 u conv ℓm ∇ad 1 χ ℓm 4acG T 4 M r ∇s + ρc P T − + + ρu conv c P T ν(1 − f )∇χ . F tot = 3 κP r 2 C 2HP α Q ν 2HP (5.154) Since it will be advantageous going forward, let us define ¶ ∇ad 1 χ − + ∇χ . Ξ ≡ ν(1 − f ) α Q ν (5.155) 3 u conv 4acG T 4 M r ℓm ℓm F tot = ∇s + ρc P T + ρu conv c P T Ξ . 2 3 κP r C 2HP 2HP (5.156) µ Therefore, The step involves equation Equations (5.102) and (5.156) and rearranging the first term on the right, 3 u conv ℓm 4acG T 4 M r ℓm (∇ − ∇ ) = ρc T + ρu c T Ξ . s P conv P rad 3 κP r 2 C 2HP 2HP (5.157) Next, we know that g = G Mr /r 2 , HP = P /(ρg ), and ω = κρℓm . This allows us to write, u3 4acT 4 ℓm ℓm ℓm (∇rad − ∇s ) = ρc P T conv + ρu conv c P T Ξ . 3HP C 2HP 2HP 150 (5.158) Dividing by T ℓm /HP gives, ρc P 3 ρc P 4acT 3 (∇rad − ∇s ) = u conv + Ξu conv . 3 2C 2 (5.159) Isolating the gradient difference on the left hand side, (∇rad − ∇s ) = 3 ρc P 3 ρc P 3 u conv + Ξu conv , 3 4C 2acT 4 2acT 3 (5.160) which allows us to rewrite the equation in terms of γ0 , (∇rad − ∇s ) = ³ ´ ´ 3 ω ω 3 ³ 3 γ0 γ u + Ξu conv . 0 4C 1 + ω2 /3 conv 4 1 + ω2 /3 (5.161) Pulling the factor of 1/3 out of the term in parenthesis, (∇rad − ∇s ) = ³ ω ´ 9 ³ ω ´ 9 3 γ0 γ0 u Ξu conv , + 4C 3 + ω2 conv 4 3 + ω2 (5.162) which is ripe in the making for using one of our predefined variables, A , (∇rad − ∇s ) = 2A 3 γ0 u conv + 2Aγ0 Ξu conv . C (5.163) Finally, substitute for uconv using Equation (5.133) and simplify to find (∇rad − ∇s ) = 2Ay 3 C γ20V 3 + 2Ay Ξ. V (5.164) Before we proceed to the next step in the solution, it is worth noting that our result for Equation (5.164) is not the same as that of LS95. Whereas there are three terms incorporated in our variable Ξ, LS95 have only a single term. One of the extra terms we picked up was due to LS95 neglecting the pure magnetic term in the convective flux, as we mentioned earlier. The second term, however, was spuriously dropped by LS95. 151 Step 6: This step is rather unintuitive, but we want to rewrite Equation (5.164) in terms of quantities that we have previously derived or defined. Thus, we essentially add zero to the left hand side by adding ³ ν ν ´ ³ ν ´ ν 0 = ∇ad − f ∇ad ∇χ − (1 − f ) ∇χ − ∇ad − f ∇ad ∇χ − (1 − f ) ∇χ . α δ α δ (5.165) We then have that ¢ ¡ ¢¤ £¡ C γ20V 3 ∇rad − ∇ad + αν f ∇ad ∇χ + (1 − f ) νδ ∇χ − ∇s − ∇ad + αν f ∇ad ∇χ + (1 − f ) νδ ∇χ = 2Ay 3 + 2AC γ20V 2 yΞ . (5.166) Utilizing the definition for our variable V , ³ ν ν ´ V −C γ20V 3 ∇s − ∇ad + f ∇ad ∇χ + (1 − f ) ∇χ = 2Ay 3 + 2AC γ20V 2 yΞ α δ (5.167) Step 7: The final step in our solution involves rewriting Equation (5.146) and substituting it into Equation (5.167). First we write µ ¶ ν ν ν V y2 − (1 − f )∇ − (1 − f )∇ − f ∇ad ∇χ + (∇s − ∇ad ) = χ χ y α C γ20V δ C γ20V 2 δ y (5.168) and then substitute into Equation (5.167) to get V −C γ20V 3 " # µ ¶ ν y2 V − (1 − f )∇χ + = 2Ay 3 + 2AC γ20V 2 yΞ. y C γ20V δ C γ20V 2 y (5.169) Distribute the terms on the left hand side so we have that C γ2V 4 ν − V y 2 = 2Ay 3 + 2AC γ20V 2 yΞ. V − V 2 y + (1 − f )∇χ 0 δ y 152 (5.170) Then multiply by y , ν V y − V 2 y 2 + (1 − f )∇χC γ20V 4 − V y 3 = 2Ay 4 + 2AC γ20V 2 y 2 Ξ. δ (5.171) If we perform a final rearrangement of the terms to set the equation equal to zero, ν 2Ay 4 + V y 3 + (2AC γ20 Ξ + 1)V 2 y 2 − V y − (1 − f )∇χC γ20V 4 = 0. δ (5.172) This is the final quadratic equation for which we need a solution. Standard MLT, in contrast to the equation above, requires finding the root of a cubic equation and lacks the first quadratic and the zeroth order coefficients (Böhm-Vitense 1958; Henyey et al. 1965; Weiss et al. 2004). If we are able to obtain a solution for y , then we know the rest of the variables in the system, including the convective velocity, uconv , and the temperature gradient, ∇s . To find the root of Equation (5.172), we make an educated guess as to the solution for y . A series of Newton-Raphson corrections are used to converge to the proper solution with a convergence tolerance of 10−10 selected to reduce the propagation of large numerical errors. There are several intelligent first guesses for y , as outlined by LS95. We have found that ¤−1 £ y = 2AC γ20 Ξ + 1 (5.173) provides the quickest convergence in our code, typically converging to a solution in approximately 10 iterations. One might notice that our definition of f can greatly reduce the complexity of Equation (5.172). The selection of f , as you might recall, defines the transfer of magnetic energy from the parcel to the surroundings or vice versa. We may interpret this parameter as determining whether magnetic flux tubes are frozen into the fluid or whether they will, instead, efficiently diffuse through the medium. 153 5.2.5 The Parameter f and the Frozen Flux Condition In Section 5.2.1 we specified that the plasma under consideration was perfectly conducting and, thus, had zero resistivity. One consequence of assuming an ideal MHD plasma is that magnetic field lines become physical objects that are transported by the plasma, the so-called frozen flux condition (FFC; Alfvén 1942). The magnetic flux is Φ(t ) = Z S(t ) B(r, t ) · d Â. (5.174) For an ideal plasma, the evolution of Φ in time is dependent upon not only the time rate of change of the magnetic field, B(r, t), but also on any distortion occurring to the bounding surface, S(t ), as the plasma moves. The net effect is that the time rate of change of the magnetic flux is equal to zero. Therefore, we must have ∂B = 0, ∂t (5.175) which may be rewritten using Equation (5.1), the induction equation, with η = 0. This results in the well-known FFC condition, ∇ × (u × B) = 0. (5.176) The FFC enforces the restriction that, for a spherically symmetric bubble of plasma undergoing isotropic expansion or contraction (Kulsrud 2004), B ρ 2/3 = constant. (5.177) Applying this constraint to the magnetic energy gradient definition of a convecting fluid element (Equation 5.87), we are able to physically motivate the definition of the parameter f which governs the flux of energy between the fluid element and the surrounding material. 154 Imagine a region in a star where a small bubble begins to grow convectively unstable. Initially, the bubble has the same properties as the surrounding fluid. It is only because of the change in density that other properties begin changing as well. The FFC allows us to write the magnetic energy contained within a fluid parcel as a function of the magnetic energy of the surrounding material. Since a convecting fluid parcel has a slight under- or overdensity compared to its surroundings, we perturb the element’s density ρe = ρ s + ξ (5.178) where it is understood that ξ ≪ ρ s . We also drop the subscript s hereafter. Assuming, for simplicity, that the bubble expands isotropically, the magnetic field strength within a convectively unstable bubble is µ ¶ ¢2/3 ξ 2/3 B ¡ = B 1+ B e = 2/3 ρ + ξ ρ ρ (5.179) meaning the magnetic energy per mass may be written as µ ¶ B2 ξ 1/3 χe = = 1+ . 8π(ρ + ξ) 8πρ ρ B e2 (5.180) We now have a direct relation between the magnetic energy density of the convecting fluid element and the surrounding material, ¶ µ ξ 1/3 . χe = χs 1 + ρ (5.181) Taking the radial, logarithmic derivative, µ d ln χ dr d ln χ = dr e ¶ µ ¶ µ ¶· ¸ 1 ξ d ln ξ d ln ρ − + . 3 ρ +ξ dr dr s (5.182) The first of the bracketed terms goes to zero, as the density perturbation is independent of 155 radial location. Using the definition of χs (Equation 5.35) to expand the derivative, and after rearranging the resulting terms, the equation becomes µ d ln χ dr µ 2¶ µ ¶· µ ¶¸ B d d ln ρ 1 ξ ln = − 1+ . dr 8π dr 3 ρ +ξ e ¶ (5.183) By definition, we know that ξ/ρ ≪ 1, meaning the perturbation term in the square brackets is negligible. As such, µ d ln χ dr d ln χ ≈ dr e ¶ µ ¶ (5.184) s allowing us to conclude that the FFC implies that f ≈ 1. 5.3 Implementation in the Dartmouth Code In the previous sections, we developed the mathematical framework necessary to include the effects of a globally pervasive magnetic field within a 1D stellar evolution code. Here, we briefly outline some of the technical details of the actual implementation. 5.3.1 Magnetic Field Strength Distribution The strength of the perturbations described in the preceding sections are determined by the magnitude and spatial gradient of χ. Mentioned in Section 5.2.3, was that we deviate from the prescription of χ proposed by LS95. Instead of defining · µ ¶ ¸ 1 M D − M Dc 2 χ = χmax exp − , 2 σ (5.185) ¶¸ · µ Mr , M D = log10 1 − M∗ (5.186) where 156 1 14 Magnetic Field Strength Tachocline Location 12 10 0.6 8 0.4 |B| (kG) Radius Fraction 0.8 6 4 0.2 2 0 0 0.2 0.4 0.6 Radius Fraction 0.8 1 0 Figure 5.1: Magnetic field strength profile for a 1.0 M⊙ star with a 5.0 kG photospheric magnetic field strength (maroon, solid). The green dash-dotted line indicates the location of the stellar tachocline, the interface between the radiative and convective regions. The plot is meant only to illustrate the field strength profile. A small gap in the field strength profile is barely perceptible near the surface of the star. This artifact is due to the separation of the stellar interior and envelope integration regimes in the code. we opt to directly prescribe the radial magnetic field profile. Approaching the problem in this manner introduces the difficulty of selecting a particular radial profile. Without any real confidence of the radial profile inside stars, we are left to our own devices. One of the simplest profiles to select is that of a dipole configuration, where the field strength drops off as r −3 from the magnetic field source location. This is illustrated in Figure 5.1. The radial profile may then be prescribed as B (R) = B (R tach ) · (R tach /R)3 R > R tach (R/R tach )3 R < R tach (5.187) with the peak magnetic field strength defined to occur at the radius Rtach . The radial location described by R tach is the location of the stellar tachocline, an interface between the 157 convective envelope and radiative core. This interface region is thought to be characterized as a strong shear layer where the differentially rotating convection zone meets the radiative core rotating as a solid body. Theory suggests that the tachocline is the source location for the standard mean-field stellar dynamo (i.e., the α–ω dynamo; Parker 1975). Since DSEP monitors the shell location of the boundary to the convection zone, the tachocline appeared to be a reasonable location, both theoretically and numerically, to base the scaling of the magnetic field strength. However, defining the magnetic field strength at the tachocline (B (Rtach )) is required. In an effort to allow for direct comparisons between field strengths input into the code and observed magnetic field strengths, the field strength at the tachocline is anchored to the photospheric (surface) magnetic field strength, B (R tach ) = B surf µ R∗ R tach ¶3 . (5.188) where B surf = B (R ∗ ) is introduced as a free parameter. The advantage of B surf as a free parameter is that it has potential to be constrained observationally. Fully convective stars do not possess a tachocline. However, a dynamo mechanism still has the potential to drive strong magnetic fields through an α2 mechanism (Chabrier & Küker 2006). Full 3D MHD modeling suggests that the magnetic field strength peaks at about 0.15 R ∗ (Browning 2008). Unfortunately, the adopted micro-physics were solar-like and may not be entirely suitable for fully convective M-dwarfs. Regardless of these shortcomings, Browning’s investigation provides the only estimate, to date, for the location of the peak magnetic field strength in fully convective stars. As such, we adopt Rtach = 0.15R ∗ as the dynamo source location in our models of fully convective stars. 158 5.3.2 Numerical Implementation Although we have laid out the mathematical construction of the magnetic perturbation, we have yet to illuminate precisely how the perturbation is treated numerically. When a magnetic model is first executed, the user provides a surface magnetic field strength, the geometry parameter γ, and the age at which the magnetic perturbation will occur. The model proceeds to evolve the same as a standard model until the initial perturbation age is reached. Once the perturbation age is reached, the magnetic field strength profile is prescribed based on the assumed photospheric field strength and the location of the tachocline, as in Figure 5.1. The magnetic energy and magnetic pressure are then computed for each of the model’s mass shells. Here, the total pressure associated with each mass shell is also perturbed. Following the introduction of the perturbation, the code must recompute the structure of the stellar envelope, which is separate from the stellar interior integration. The envelope comprises the outer 2%–3%, by radius, of the stellar model. Surface boundary conditions are determined prior to the envelope integration by interpolating within the phoenix model atmosphere tables using log g and Teff . This interpolation returns P gas at the surface of the star, and defines the start of the inward integration. The magnetic perturbation is then explicitly included in the calculation of the analytic EOS. This leads into the convection routines, where the non-standard stability criterion in Equation (5.100) is evaluated and judged. Either the equations of magnetic MLT are solved, or the radiative gradient is selected. The envelope integration scheme proceeds until it reaches a pressure commensurate with the pressure for the interior regime. From the newly calculated envelope, the interior integration begins using a Henyey integration scheme (Henyey et al. 1964) with the magnetic perturbation implemented. The EOS 159 log(R/R⊙) 0.08 Non-mag 2.0 kG 0.5 kG 3.0 kG 3.0 kG Non-mag 1.0 kG 0.04 0.00 -0.04 -0.08 (a) ZAMS Pre-ZAMS 0.08 log(R/R⊙) (b) Post-ZAMS Non-mag 500 Models 150 Models Non-mag 0.04 0.00 -0.04 -0.08 (c) 0.01 (d) 0.1 1 0.01 Age (Gyr) 0.1 1 Age (Gyr) Figure 5.2: Tests of numerical stability for a 1 M⊙ magnetic model with γ = 2. (a) Influence of various magnetic field strengths. (b) Evidence of a smooth perturbation at a large magnetic field strength with a relatively large radius change. (c) Consistency among models with the perturbation turned on at various evolutionary stages. (d) Demonstrating the insensitivity of the perturbation to the size of the time steps after the perturbation is enabled. Here, more models means smaller evolutionary time steps. and convection routines are evaluated as in the envelope. Once a final solution is converged upon, the code iterates in time and the process is repeated. For each temporal iteration, the magnetic field profile is adjusted to adapt to changes in the location of the tachocline and changes in the number of mass shells. 5.4 Testing and Validation 5.4.1 Numerical Stability In Sections 5.2 – 5.3, we outlined the formulation and implementation of a magnetic perturbation within the framework of the Dartmouth stellar evolution code. With the perturbation 160 implemented, it was crucial to perform a series of numerical tests and common-sense checks to validate that the code was operating properly. The four key numerical tests were to ensure that: 1. The relative change in radius between magnetic models of monotonically increasing photospheric magnetic field strength must also be monotonically increasing with respect to a non-magnetic model. 2. All model adjustments after the initial perturbation must be continuous and smooth. 3. The final perturbed model properties must be independent of the evolutionary stage at which the perturbation is made. 4. The resulting perturbed model must be consistent, regardless of the number of time steps taken after the perturbation. All of these tests were performed to confirm that the code was producing consistent results and that it was doing so in a numerically stable manner (i.e., no wild fluctuations). Figure 5.2 demonstrates that all model adjustments to a magnetic perturbation satisfy each of the four criteria we required for numerical validation. Panel (a) demonstrates that the radius monotonically increases as the surface field strength applied monotonically increases. Changes are observed to be smooth, as seen in panel (b) and are independent of the number (or size) of the evolutionary time steps taken (panel (d)). Finally, the plot in panel (c) indicates that the relative change to the model asymptotes to the same value, regardless of the evolutionary phase at which the perturbation is applied. 5.4.2 Solar Model Beyond testing for numerical stability, we must be assured that the code produces results consistent with reality. Typically, a comparison with previous studies would be utilized. 161 However, the only such examples computed for evolutionary timescales are for CM Dra (Chabrier et al. 2007; MacDonald & Mullan 2012). The stellar mass regime occupied by CM Dra would require the implementation of FreeEOS, a task reserved for a future investigation. With the analytical EOS, it would appear there are no models generated with which to compare. Even LS95 operate over timescales on the order of a solar-cycle and not evolutionary times. In the absence of previous results with which to compare, we opted to impose a weak magnetic field (5 G) on our solar-calibrated model. Seeing as the properties of the Sun do not require a magnetic field in order to produce an adequate solar model (Bahcall et al. 1997), we would expect the solar properties to remain almost entirely unaffected. As expected, the magnetically perturbed model still meets our requirements for it to be considered solarcalibrated (see Section 1.5.2). The model radius increases by 3 parts in 105 while effective temperature decreases by 4 parts in 106 given the presence of the weak field. 5.4.3 Ramped Versus Single Perturbation The nature of the procedure developed in this chapter requires that the magnetic perturbation be introduced in the model during runtime. Therefore, we must explore how the method of introducing the magnetic perturbation affects the structural evolution of a model. We consider whether the magnetic perturbation is slowly ramped from an initially very weak field up to full strength or whether the magnetic perturbation is applied at a single moment, regardless of the pertrubation strength. Numerical stability of the magnetic perturbations was discussed previously in Section 5.4.1 and it was shown that the code produces smooth, consistent results independent of the age of perturbation and the size of model time steps adopted. However, the perturbation was enacted in a single step, regardless of the surface magnetic field strength. Stars do not operate in this manner. We were curious to probe the effects of ramping the magnetic field 162 3.60 Non-Magnetic Single Step Ramped Field 0.4 3.58 0.2 3.56 log(Teff/K) log(R/R⊙) Non-Magnetic Single Step Ramped Field 0.0 3.54 3.52 -0.2 3.50 -0.4 (a) (b) 3.48 0.1 1 10 Age (Myr) 100 1000 0.1 1 10 Age (Myr) 100 1000 Figure 5.3: Standard DSEP (black, solid) and magnetic mass tracks for a 0.4 M⊙ star are plotted to illustrate the consequences of using a ramped magnetic perturbation (light blue, dash-dotted) as opposed to a single-step perturbation (maroon, dashed). The effects are considerably different depending on if we view (a) the age–radius plane or (b) the Teff radius plane. strength from an initial weak field up to full strength. Ramping was performed by assuming that the intial magnetic perturbation occurs along the pre-main sequence (pre-MS) at an age of approximately 50 kyr. The initial perturbation could be no larger than 100 G at the model photosphere. If the prescribed field strength was larger than 100 G, the initial 100 G field was increased by 0.005 G per year. This rate corresponds to about 50 G per model time step, thereby allowing the field to reach maximum strength within 0.6 Myr. A comparison between the morphology of a stellar mass track along the pre-MS with no magnetic field, a single-step peturbation, and field ramping is illustrated in Figure 5.3. The particular procedure for introducing the magnetic perturbation is irrelavant for the final stellar parameters along the main sequence, once the star has settled into its final configuration. In Figure 5.3(a) the magnetic tracks converge smoothly and rapidly, regardless of the treatment of the perturbation. This is true despite the appearance of a discontinuity that develops in the effective temperature track for strong, single perturbation models (Fig- 163 ure 5.3(b)). We hesitate to label this spike an instability, because while the temperature does fluctuate immediately after the perturbation, it settles quite rapidly into a stable configuration. Interestingly, there is no such discontinuity in the model radius. Ramped models do not display this discontinuity but still produce results consistent with the single perturbation models along the main sequence. Therefore, all stellar parameters from single perturbation magnetic models are trustworthy beyond the zero-age main sequence (ZAMS). Along the pre-MS and on the approach to the ZAMS, ramping tends to change to the overall characteristics of the mass track. Changes to the track morphology are gradual and lack the abrupt spike displayed by single perturbation models, as one would expect. Even after the single perturbation model has relaxed, there are significant differences in the track morphology leading up to the equilibration of 3 He burning (the bump that appears around 200 Myr). Such differences are accentuated depending on the location of the single perturbation (not shown, but the earlier the perturbation, the worse the disagreement between the ramped and single step models). Ramping should clearly be the preferred method to analyze how the structure of pre-MS stars are affected. This is not surprising since pre-MS stars are still contracting and are presumably still building up strong magnetic fields. It is not clear why the single perturbation models do such a poor job along the pre-MS, though, as mentioned above, the situation improves when the perturbation age is increased and we find better agreement between the ramped and single step models. The dependence of the pre-MS stellar properties on the initial age of the perturbation and the rate of ramping is unclear and is outside the scope of this chapter. What we are primarily concerned with is that the MS properties are unaffected, in line with our discussion of numerical stability in Section 5.4.1. However, Figure 5.4 indicates that the precise details have an ability to play a significant role in what properties are derived for pre-MS systems. Only the effect on the effective temperature is shown as it demonstrates the differences moreso than the effect on the radius, which appear to be negligible. 164 3.60 Non-Magnetic τpert = 0.10 Myr τpert = 0.04 Myr 3.58 log(Teff/K) 3.56 3.54 3.52 3.50 3.48 0.1 1 10 Age (Myr) 100 1000 Figure 5.4: The effect of altering the magnetic perturbation age for the same 0.4 M⊙ model in Figure 5.3. Significant differences along the pre-main sequence are observed when comparing the two magnetic mass tracks that were perturbed at 0.10 Myr (light blue, dashed) and 0.04 Myr (maroon, dash-dotted). The standard model mass track (black, solid) is provided for reference. We note that it may be possible to use ramping techniques for modeling single stars over their entire lifetime. Essentially, increase the field strength along the pre-MS, steadily decrease it along the MS, thereby mimicing the magnetic breaking process, and the assuming constant magnetic flux along the red giant branch (Reiners 2012). Tying the field strength to rotation would be ideal, however, rotation is currently not implemented in the Dartmouth code. 5.5 Case Study: EF Aquarii 5.5.1 Standard Models Standard, non-magnetic mass tracks with solar metallicity were computed for both EF Aqr A and B with masses of 1.24 M⊙ and 0.95 M⊙ , respectively. Additional mass tracks were 165 also generated for a scaled-solar metallicity of +0.1 dex. The two components were unable to be fit with a coeval age, regardless of the adopted metallicity. This is consistent with the conclusions of Vos et al. (2012). Two fitting methods were performed to this end. We first attempted to fit both components on an age–radius plane, which is equivalent to fitting on the mass–radius (M –R ) plane. This is illustrated by the solid lines in Figure 5.5. The primary star evolves much more rapidly than secondary, owing to the rather large mass difference. Thus, the model radius of the primary inflates to the observed radius at an age of 2.0 Gyr. The radius of the primary then quickly exceeds the observational bounds within about 0.1 Gyr. The observational bounds on the primary radius places tight constraints on the allowed age our models predict for the system. However, our model of the secondary does not reach the observed radius until an age of 6.3 Gyr, an age difference of 4.3 Gyr between the two components. This 4.3 Gyr difference is consistently present in our models, including when mass and composition are allowed to vary within the observed limits (not shown). Assuming the age of the system was predicted accurately using models of EF Aqr A in the M –R diagram, we find that the model radius of EF Aqr B underpredicts the observed radius by 11%. Again, consistent with the findings of Vos et al. (2012), who found a radius discrepancy of 9%. Such a disagreement is also broadly consistent with results from other studies of active EBs (Ribas 2006; Torres et al. 2010). A second approach was to fit the system on an HR diagram using the individual mass tracks. Figure 5.6 demonstrates that the standard model tracks do not match the observed Teff –luminosity of either star, despite fitting the stars individually in the M –R plane. Our models predict temperatures that are 250 K (4%) and 430 K (8%) hotter than the observations for the primary and secondary, respectively. The noticeable temperature disagreements in both stars may be the result of two possible effects. On the one hand, we might assume that the age predicted from the M –R plane 166 1.6 (a) γ=2 γ = 4/3 Non-Mag. Radius (R⊙) 1.4 1.2 EF Aqr A 1.0 EF Aqr B 0.8 10 100 1000 Age (Myr) Figure 5.5: Individual stellar mass tracks representing EF Aqr A and B (labeled) in the age–radius plane. The observational constraints on the stellar radii are represented by the gray swaths. Non-magnetic mass tracks are shown as a red, solid line while γ = 2 and γ = 4/3 magnetic tracks are indicated by the blue, long-dash and light-blue, short-dash lines, respectively. The corresponding photospheric magnetic field strengths for EF Aqr A are 1.6 kG (γ = 2) and 2.6 kG (γ = 4/3). For EF Aqr B they are 3.2 kG (γ = 2) and 5.5 kG (γ = 4/3). Vertical dotted lines define the age constraint imposed by EF Aqr A, suggesting the system has an age of 1.3 ± 0.2 Gyr. is correct and that there only exists a discrepancy with the effective temperatures. This implies that either the effective temperature from the models or the observations is incorrect. However, since the stars are quite similar to the Sun, it is more likely the case that our standard models underpredict the radius of the primary as well, driving up the modelderived effective temperature. This particular scenario is supported by Vos et al. (2012) who found both components display obvious Ca ii emission that is likely the result of each star having a magnetically heated chromosphere. The age inferred from the M –R diagram would then be older than the true age of the system. Unfortunately, this scenario further complicates the situation regarding EF Aqr B. If the age of the system is younger than inferred from standard models of the primary, then the radius discrepancy for the secondary becomes larger than originally quoted. 167 1.2 (b) log(L/L⊙) 0.8 0.4 EF Aqr A 0.0 -0.4 γ=2 γ = 4/3 Non-mag. 6500 EF Aqr B 6000 5500 5000 Teff (K) 4500 4000 Figure 5.6: Individual stellar mass tracks representing EF Aqr A and B (labeled) in the Teff – radius plane (HR diagram). The observational constraints are indicated by the black points. Non-magnetic mass tracks are shown as a red, solid line while γ = 2 and γ = 4/3 magnetic tracks are indicated by the blue, long-dash and light-blue, short-dash lines, respectively. The corresponding photospheric magnetic field strengths for EF Aqr A are 1.6 kG (γ = 2) and 2.6 kG (γ = 4/3). For EF Aqr B they are 3.2 kG (γ = 2) and 5.5 kG (γ = 4/3). 5.5.2 Magnetic Models Following the results discussed above, magnetic models were computed for both EF Aqr components using a scaled-solar heavy element composition. Several models with a mass of 1.24 M⊙ were generated with various surface magnetic field strengths. We then found the model with the weakest field strength required to produce the observed radius and Teff of EF Aqr A. With γ = 2, we had to prescribe a photospheric magnetic field strength of 1.6 kG, while with γ = 4/3, a more intense 2.6 kG field was necessary. The magnetic model tracks are displayed in Figures 5.5 and 5.6 as blue dashed lines. The magnetic models of the primary suggest a younger age of 1.3 ± 0.2 Gyr for the EF Aqr system, as opposed to the 2.0 Gyr age determined from standard models. This younger age is consistent with the 1.5 ± 0.2 Gyr age derived for the primary by Vos et al. (2012) after fine-tuning the mixing length. 168 We next had to select a magnetic field strength that would allow a 0.95 M⊙ model to have a radius and Teff compatible with EF Aqr B at 1.3 Gyr, if finding that unique combination was possible. Surface magnetic field strengths of 3.2 kG and 5.5 kG were able to produce the required stellar parameters, with γ = 2 and 4/3, respectively, at an age of 1.35 Gyr. In both cases, the models were able to reproduce the stellar radius and Teff within the quoted 1σ uncertainties. Figures 5.5 and 5.6 demonstrate that the magnetic models do indeed match both component radii and Teff s at a common age. Structurally, the addition of a magnetic perturbation within the models reduces the radial extent of the surface convection zone. For both stars in EF Aqr, we find the magnetic models that are sufficient to correct the observed discrepancies have surface convection zones that are 4% smaller than those in the standard models, at the same age. The reduction of the surface convection zone can be attributed to the modified stability criterion as well as modified convective velocities. While only speculation, we attribute the equality of the percent reduction of convection zone sizes to coincidence. 5.6 Discussion 5.6.1 Field Strengths The implementation of a magnetic perturbation within stellar evolution models is quite capable of reconciling predicted model fundamental stellar properties with those determined observationally, at least for EF Aqr. While it seems plausible that magnetic fields may suppress thermal convection inside solar-type stars, how are we to be sure that magnetic fields may be reasonably invoked for this particular system? Even if invoking magnetic fields is rational, are the field strengths required by the models realistic? Addressing the first question, we showed in Section 5.1 that naturally inefficient convection, 169 as described by the Bonaca et al. (2012) calibration, was unable to account for the small values of αMLT required to mitigate the observed model-observation disagreements. But, the inability of naturally inefficient convection to provide a solution does not positively identify magnetic fields as the root cause. However, there is additional evidence that invoking magnetic fields is reasonable. High-resolution spectroscopy of the Ca ii H and K lines for both stars in EF Aqr reveals incredibly strong emission cores superposed on the absorption troughs (Vos et al. 2012). A search of the ROSAT Bright Source Catalogue (Voges et al. 1999) also shows that EF Aqr is a strong X-ray emitter. Coupling these observations with high projected rotational velocities extracted from line broadening measurements, suggests that there is the potential for a strong dynamo mechanism to be operating. This evidence is only circumstantial, but does provide tantalizing clues. For the sake of argument, let us assume that the stars are significantly magnetically active. It would then be worthwhile to compare the strength of the magnetic field for each EF Aqr component to those values required by the models. Unfortunately, no direct magnetic field strength estimates of EF Aqr are available, forcing us to base our analysis on indirect magnetic field strength estimates. A natural first step would be to compare the EF Aqr components to other known solar-type stars, such as the Sun and α Cen A and B (see Table 5.2). The Sun’s mean photospheric magnetic field strength is between 0.1 G and 1 G (Babcock & Babcock 1955; Babcock 1959; Demidov et al. 2002) with local patches of very intense fields (i.e., sunspots) with strengths on the order of 2–3 kG (Hale 1908). Similarly, the average longitudinal field strength of α Cen A was determined to be less than 0.2 G, after a null detection of a Stokes V polarization signature (Kochukhov et al. 2011). The field strengths required by the models of the EF Aqr components therefore suggest that the stars are pervaded by magnetic fields that typically characterize the intense regions of 170 sunspots. This at first appears detrimental to the validity of the models. However, studies of the active and quiet Sun, particularly sunspot regions, has led to multiple scaling relations allowing for an indirect determination of stellar photospheric magnetic field strengths. One of these scaling relations was observed to exist between the X-ray luminosity, L x of an active region and its total unsigned magnetic flux, Φ, (Fisher et al. 1998; Pevtsov et al. 2003). The two observables were found to exhibit a power-law relation, L x ∝ Φp , (5.189) where the power-law index, p , was determined to be 1.19±0.04 by Fisher et al. (1998). The index was later revised by Pevtsov et al. (2003) using a more diverse data set, including both solar and extrasolar sources.5 Their revised analysis decreased the index to p = 1.15. The magnetic flux is defined in the usual manner (Equation 5.174), Φ= Z S B · d  = Z S |B z | d A. (5.190) with |B z | represents the vertical magnetic field strength. Therefore, if we are able to determine the X-ray luminosity of the EF Aqr components, it is possible to place a lower limit on the magnetic field strength at the surface of the two components. The system has a confirmed X-ray counter-part in the ROSAT All-Sky Survey Bright Source Catalogue (Voges et al. 1999). The X-ray count rate was converted to an X-ray flux according to the formula derived by Schmitt et al. (1995), F x = (5.30HR + 8.31) × 10−12 X cr , (5.191) where HR is the X-ray hardness ratio,6 X cr is the X-ray count rate, and F x is the X-ray flux. 5 The total unsigned magnetic flux of the stellar sources was obtained using direct observational techniques (Saar 1996). 6 There are typically two hardness ratios listed in the ROSAT catalog, HR1 and HR2. The Schmitt et al. 171 Finally, the X-ray flux was converted to a luminosity using the 172 pc distance quoted by Vos et al. (2012). The count rate measured by ROSAT was 0.0655±0.0154 counts s−1 with a hardness ratio of 0.32±0.22. This yields an X-ray flux of 6.55×10−13 erg cm−2 . Weighting the contribution of each star to the total X-ray flux is reliably performed in one of two ways: by assuming both stars contribute equally (valid if for binaries if stars are similar in radius; Fleming et al. 1989) or weighting proportional to v r ot sin i (Pallavicini et al. 1981; Fleming et al. 1989). Given the similarity of the two stars in EF Aqr, the precise weighting does not affect the results. If both stars contribute equally to the total X-ray flux, then at a distance of 172 pc, the Xray luminosity of each component is L x = 1.16 × 1030 erg s−1 . Alternatively, weighting the two stars based on their projected rotational velocity gives L x, A = 1.25 × 1030 erg s−1 and L x, B = 1.07 × 1030 erg s−1 . To provide a comparison, the X-ray luminosity of a typical solar active region is on the order of 1027 erg s−1 (Fisher et al. 1998; Pevtsov et al. 2003). The Sun, on average, has a total X-ray luminosity of 1027 erg s−1 up to nearly 1028 erg s−1 , depending on where in its activity cycle it is located (Ayres 2009). Similarly, Ayres (2009) monitored the X-ray luminosity of α Cen and found the primary had an X-ray luminosity around half that of the Sun (approximately 1027 erg s−1 ) and the secondary had about twice the X-ray luminosity of the Sun (about 1028 erg s−1 ). Further estimates for the X-ray luminosity of α Cen A and B comes from ROSAT, which yields luminosities between 1027 erg s−1 and 1028 erg s−1 for each component, consistent with Ayres’ analysis. Table 5.2 provides a comparison of how these quantities translate to magnetic field strengths. Comparisons with the Sun and α Cen show that each component in EF Aqr has an Xray luminosity 2–3 orders of magnitude greater than “typical” G and early-K stars. Again, while not indicative of causation, the correlation between high levels of X-ray emission and (1995) formula requires the use of HR1. 172 Table 5.2 Comparison of estimated magnetic field strengths (in G). Star Direct X-Ray Ca ii DSEP Sun 0.1 - 1.0 5 - 20 < 0.2 ∼3 ··· ··· α Cen A α Cen B EF Aqr A EF Aqr B ··· ··· ··· ∼49 ··· ··· ··· ··· 1300 830 1600 - 2600 2500 3300 3200 - 5500 magnetic activity strongly suggests that EF Aqr is incredibly active. Given this information, our initial assumption that the stars are active seems valid. Therefore, the implementation of a magnetic perturbation in stellar models of this system appears warranted. The amount of vertical magnetic flux near the surface of each star may then be found using the Pevtsov et al. (2003) scaling relation. This suggests Φ = 1.39 × 1026 Mx associated with each component (given equal flux contribution). Converting to a magnetic field strength involves dividing the unsigned magnetic flux by the total area through which the field is penetrating. For our purposes, the area is the entire surface area of the star. Therefore, we find the vertical magnetic field strength for the primary and secondary of EF Aqr to be 1.3 kG and 2.5 kG, regardless of the adopted flux weighting, respectively. Note, this is the vertical magnetic field strength and sets a lower limit to the total magnetic field strength. It should also be mentioned that the X-ray luminosities calculated for EF Aqr A and B are near the edge of the data sample utilized by Pevtsov et al., although no extrapolation of the relation was required. Further estimates of the magnetic field strengths may be found by applying a scaling relation using the Ca ii K line core emission (Schrijver et al. 1989). The scaling relation was developed by correlating Ca ii K line core emission and the magnetic flux density of solar 173 active regions and their surroundings. The relation between the two was found to be I c − 0.13 = 0.008〈B f 〉0.6 Iw (5.192) where I c is the intensity of the emission line core, I w is the intensity of the line wing. While the relation was derived for small, local active regions, Schrijver et al. (1989) suggest that there is no reason to believe that the relation would not hold for hemispherical averages of solar-type stars. Using the spectra provided by Vos et al. (2012) for the Ca ii K core emission lines from each EF Aqr component, we were able to estimate the magnetic field strength of each component. Spectra for the primary indicate that the average magnetic field strength, 〈B f 〉, is equal to 830 G. Similarly, for EF Aqr B, 〈B f 〉 = 3.3 kG. The values quoted above are derived from rough approximations of the core and wing intensities. However, we do not foresee the values of 〈B f 〉 changing radically with more precise line intensity measurements. We do caution that the results for EF Aqr B require an extrapolation of the Ca ii relation and the data for EF Aqr A place it near the edge of the derived relation where only a few data points exist. Based on the scaling relations for X-ray emission and Ca ii K line core emission, the magnetic field strength for the primary and secondary is seen to be approximately 1 kG and 3 kG, respectively. The magnetic field strengths required by the models are therefore within a factor of two of the predicted field strengths, regardless of the adopted γ value. Since we expect the X-ray emission prediction to be a lower limit to the full magnetic field strength, this is extremely encouraging. The models do not require abnormally large field strengths to reconcile the model properties with those from observations, particularly when a γ value of 2 is adopted. 174 5.6.2 Implications The introduction of self-consistent magnetic stellar evolution models has multiple applications, ranging from studies of exoplanet host stars (Torres 2007; Charbonneau et al. 2009; Muirhead et al. 2012) to investigations of cataclysmic variable (CV) donor stars (Knigge et al. 2011), as well as to studies attempting to probe the stellar initial mass function of young clusters, where stars are typically very magnetically active (Johns-Krull 2007; Jackson et al. 2009; Yang & Johns-Krull 2011). Although, most obvious, are the implications for studies of low-mass eclipsing binary systems (see, e.g., Torres et al. 2010; Parsons et al. 2012, and references therein). Low-mass stellar evolution models have been highly criticized for being unable to predict the radii and effective temperatures of DEB stars. Models incorporating magnetic effects open the door to probing the underlying cause of the model-observation disagreements and providing semi-empirical corrections to models. Magnetic fields have long been theorized as the culprit, but previous generations of models have only treated magnetic fields in an ad hoc manner. Comparing the results of these methods with the one presented in this work, both in terms of surface parameters and the underlying interior structure, will provide an interesting test of their validity. Ultimately, the ad hoc models disagree on the dominant physical mechanism underlying the observed discrepancies. The availability of self-consistent magnetic models should help to settle the debate as to which mechanism (suppressed convection or starspots) is most at work. For EF Aqr, the models suggest that magnetic suppression of thermal convection is sufficient to reconcile the models with the observations. Since stars with small convective envelopes, such as those discussed in this work, are more sensitive to adjustments of the convective properties, it is not wholly surprising that suppressing convection is sufficient to explain the observations. Whether this mechanism will be adequate for stars near the fully convective boundary has yet to be seen. Future work modeling the lowest-mass DEB systems will clarify this ambiguity. Regardless, this may suggest why the largest radius deviations 175 are predominantly observed at higher masses (Feiden & Chaboyer 2012a), with the notable exception of CM Dra (Terrien et al. 2012). The nature of our models allows for independent verification of the magnetic field strengths required as input. While the indirect estimates provided by X-ray emission and Ca ii K emission are encouraging, confirmation of these results using high resolution Zeeman spectroscopy, spectropolarimetry, or Zeeman–Doppler imaging (ZDI) is preferred. Unfortunately, these observations are difficult for fast rotators, such as those that comprise most DEB systems. They are also difficult for distant systems, where the short integration time required by ZDI inhibits the ability of acquiring measurements with sufficient signal-tonoise (see the reviews by Donati & Landstreet 2009; Reiners 2012). Once a magnetic field is detected, there exists the question of whether the observed strength is indicative of the total magnetic field strength. Field strengths derived for stars with spectral-type K and M using Stokes V observations appear to yield only around 10% of the total magnetic field strength compared to observations in Stokes I (Reiners & Basri 2009). This is a consequence of the fact that regions of opposite polarity tend to cancel out in Stokes V , making it most sensitive to the large-scale component, not the small-scale fields thought to pervade low-mass stars. How to accurately account for this when testing the models is not fully clear and will require investigation. As more stars across all spectral types are observed in both Stokes V and I , a more coherent picture is sure to develop. Magnetic models may also be useful for transiting exoplanet surveys, particularly those focused on M-dwarfs (e.g., MEarth transit survey; Nutzman & Charbonneau 2008; Irwin et al. 2009). One of the largest uncertainties in deriving the properties of a transiting planet is the radius of the host star. The lack of reliability involved in predicting low-mass stellar radii from evolution models has deterred the use of models as predictors of exoplanet host-star radii (Torres 2007; Charbonneau et al. 2009). Shoring up these deficiencies may lead to more accurate predictions of host star radii from stellar models, circumventing, for a time, the 176 need for lengthy and costly observations. This would be most useful in identifying interesting follow-up targets by providing a better estimate of the habitable zone (Muirhead et al. 2012). There are certainly caveats with models, as other large uncertainties exist in predicting the properties of a single star from stellar evolution mass tracks (Basu et al. 2012). However, work is being performed to alleviate some of these uncertainties by calibrating models to asteroseismic data (Bonaca et al. 2012). Low-mass stars are also less sensitive to the input parameters of stellar models than their solar-type counterparts, reducing the associated uncertainties. Stellar evolution models may therefore provide a fast and reliable estimate of the host star properties, depending on the required level of precision. Since most M-dwarfs being surveyed are nearby, there is a good chance that they may have an X-ray counterpart in either the Bright Source or Faint Source Catalogue from the ROSAT All-Sky Survey (Voges et al. 1999; Voges et al. 2000). As was demonstrated in Section 5.6.1, magnetic field strengths required by the models are within about a factor of two (or better, if γ = 2 is adopted) of those predicted by the X-ray scaling relation of Pevtsov et al. (2003). This will allow an intelligent choice of the magnetic field strength used as an input for the models, thus producing more reliable results from stellar models. All told, the introduction of a self-consistent set of magnetic stellar evolution models provides the potential for models to be used with greater reliability in a wide range of applications. There still exist several challenges that require attention (Boyajian et al. 2012), but this is a first step in addressing key issues that have been raised in the past two decades. 177 Chapter 6 Magnetic Models of Low-Mass Stars with a Radiative Core 6.1 Introduction The previous chapter introduced the mathematical formulation and numerical implementation of a self-consistent magnetic field perturbation within the Dartmouth stellar evolution code. Magnetic perturbations were added using a method formally based on the one outlined by Lydon & Sofia (1995). After the internals of the code were verified, a first test of the magnetic models was carried out by modeling a solar-type DEB, EF Aquarii (EF Aqr; Vos et al. 2012). Results of the test case were positive. They indicated that the presence of a magnetic field in a solar-type star could lead to inflated radii at 10% level observed in EF Aqr (see Chapter 5; Feiden & Chaboyer 2012b). The magnetic field strengths required to produce agreement between the model properties and the observational properties of EF Aqr appeared consistent with estimates from activity data (Feiden & Chaboyer 2012b). Comparisons between the Sun, α Centauri B, and the stars in EF Aqr displayed no obvious inconsistencies regard178 ing levels of magnetic activity and the predicted (or measured) magnetic field strengths. However, the lack of direct magnetic field strength measurements among solar-type stars (Donati & Landstreet 2009; Reiners 2012), especially among rapid rotators, means that we are only able to speculate about the permitted magnetic field strengths of the surfaces of solar-type stars. The situation is different for low-mass stars. Considerable observational effort has been directed towards directly measuring the magnetic field characteristics of low-mass stars (Saar 1996; Reiners & Basri 2007; Johns-Krull 2007; Reiners & Basri 2009; Morin et al. 2010; Shulyak et al. 2011; Reiners 2012). K- and M-dwarfs have been the focus of magnetic field studies because around mid-M spectral type, about 0.35M⊙ , M-dwarfs are thought to become fully convective (Limber 1958a). The standard description of the stellar dynamo mechanism posits that magnetic fields are generated near the base of the outer convection zone (Parker 1979). A strong shear layer, called the tachocline, forms between the differentially rotating convection zone and the radiation zone, which rotates as a solid body. Fully convective stars, quite obviously, would not possess a tachocline. According to the standard dynamo theory put forth by Parker, this would leave fully convective stars unable to generate or sustain a strong magnetic field through the dynamo mechanism. Despite the lack of a tachocline, low-mass stars are observed to possess strong magnetic fields with surface strengths on the order of a few kilogauss (Saar 1996; Reiners & Basri 2007; Reiners et al. 2009; Reiners & Basri 2010). Instead of a dynamo primarily powered by rotation, it was suggested that turbulent convection may be driving the stellar dynamo (Chabrier & Küker 2006). Observational evidence of the dynamo transitioning from a rotationally driven dynamo in solar-type stars to a purely turbulent dynamo in fully convective stars has been tentatively identified by studying the topology of low-mass stars. There appears to be a shift in the large-scale field topology from non-axisymmetric in solar-type stars to an axisymmetric field topology among low-mass stars (Morin et al. 2008; 179 Donati & Landstreet 2009; Phan-Bao et al. 2009; Morin et al. 2010). Whether this apparent topology change is the hallmark of a transitioning dynamo or whether the shift in field topology actually exists is still a matter of debate (Donati & Landstreet 2009; Reiners & Mohanty 2012) An additional curiosity was identified by Reiners & Basri (2009), who found that Stokes V polarization observations tend to only yield about 10% of the total magnetic field strength for low-mass stars. Instead of using Stokes V , Reiners & Basri performed their observations using a Stokes I polarization filter. Since Stokes V only measures the net, signed magnetic flux, it is thought that Stokes I , which measures the total unsigned magnetic flux, is more sensitive to small-scale magnetic fields of opposite polarity on the stellar surface. The interest in low-mass stellar magnetic fields also benefits investigations of the low-mass stellar mass-radius problem. Attempts to resolve the observed discrepancies between lowmass stellar radii and stellar evolution model predictions using magnetic fields can be confronted with empirical data. In particular, non-standard stellar evolution models that incorporate magnetic perturbations (Mullan & MacDonald 2001; Feiden & Chaboyer 2012b) should not require unrealistically large surface magnetic field strengths to reconcile models with observations. With the dynamo dichotomy in mind, we have elected to divide our analysis of the lowmass stellar mass-radius problem into two parts. The first part presented in this chapter concerns low-mass stars in DEBs that should possess a radiative core and convective outer envelope. The second part, pertaining to fully convective low-mass stars will be presented in the next chapter. Our reason for splitting the analysis is that the models described in Chapter 5 assume that the energy for the magnetic field—and thus the dynamo mechanism— is sourced from rotation. This was explicitly stated in the discussion of Equation (5.75). With the possible transition from a rotationally driven dynamo to a turbulent dynamo at the fully convective boundary (Reiners & Basri 2010), we risk invoking a theory that may 180 Table 6.1 Sample of DEBs whose stars possess a radiative core. DEB Star System UV Psc A UV Psc B YY Gem A YY Gem B CU Cnc A CU Cnc B P orb Mass Radius Teff [Fe/H] (day) (M ⊙ ) (R ⊙ ) (K) (dex) 0.86 0.9829 ± 0.0077 1.110 ± 0.023 5780 ± 100 ··· 0.6194 ± 0.0057 3820 ± 100 +0.1 ± 0.2 0.81 2.77 0.76440 ± 0.00450 0.59920 ± 0.00470 0.59920 ± 0.00470 0.43490 ± 0.00120 0.39922 ± 0.00089 0.8350 ± 0.0180 0.6194 ± 0.0057 0.4323 ± 0.0055 0.3916 ± 0.0094 4750 ± 80 ··· 3820 ± 100 +0.1 ± 0.2 3125 ± 150 ··· 3160 ± 150 ··· not be entirely suitable for models fully convective stars. Here, we will describe and present results of detailed modeling of three DEB systems with radiative cores and convective envelopes. The sample of DEB systems selected for analysis in this chapter represent only a subset of the total systems available to us. Three representative systems were chosen to avoid muddling the results. The three systems we chose were UV Piscium (Carr 1967; Popper 1997), YY Geminorum (Adams & Joy 1920; Torres & Ribas 2002), and CU Cancri (Delfosse et al. 1999; Ribas 2003). We recall the properties of these three systems in Table 6.1. Note that the systems cover three different mass regimes of low-mass stars. This will provide us with information regarding the applicability of our magnetic models as we begin to model masses closer to the fully convective boundary. The chapter will proceed as follows. In Section 6.2 we present a detailed study for each of the three DEB systems. We discuss how realistic model magnetic field strengths are in Section 6.3. Based on this discussion, we discuss means for reducing the model surface magnetic field strengths, including introducing a way to include the effects of a turbulent dynamo in Section 6.4. Section 6.5 then discusses the broad conclusions of our study and also explores possible applications of magnetic models beyond attacking the mass-radius problem. Finally, we provide a summary of the study and of the main conclusions in Section 6.6. 181 6.2 Analysis of Individual DEB Systems 6.2.1 UV Piscium UV Piscium (HD 7700; hereafter UV Psc) is a DEB that contains a solar-type primary with a K-dwarf companion. It was likely discovered to be a variable star by Huth in 1959 (Kjurkchieva et al. 2005), but we were unable to locate the reference. However, this is reinforced by the fact that the first citation in NASA’s Astrophysics Data System is by Carr (1967) who reported the observation of the eclipse ephemerides without surprise. Numerous attempts at determining the fundamental stellar properties have been performed since its discovery, with the most precise measurements produced by Popper (1997). These measurements were later confirmed and slightly revised by Torres et al. (2010), who standardized the reduction and parameter extraction routines for a host of DEB systems. The mass and radius for each component of UV Psc recommended by Torres et al. (2010) is summarized in Table 6.1. Interestingly, no metallicity estimate exists despite the system being relatively bright (V = 9.01). One of the most notable features of UV Psc is that the secondary component is unable to be properly fit by standard stellar evolution models at an age commensurate with the primary (see e.g., Popper 1997; Lastennet et al. 2003; Torres et al. 2010; Feiden & Chaboyer 2012a). The radius of the secondary appears to be nearly 10% larger than models predict and the effective temperature is around 6% cooler than predicted. Metallicity and age are known to affect the stellar parameters predicted by models, typically allowing for better agreement with observations. However, the best fit age and metallicity of UV Psc still display the large disagreements mentioned above (Chapter 4; Feiden & Chaboyer 2012a). An investigation by Lastennet et al. (2003) found that it was possible to fit the components on the same theoretical isochrone. Their method involved independently adjusting the helium mass fraction Y , the metal abundance Z , and the convective mixing-length αMLT . The 182 authors were able to constrain a range of Y , Z , and αMLT values that produced stellar models compatible with the fundamental parameters of each component while enforcing that the stars be coeval. Lastennet et al. (2003) found that a sub-solar metal abundance ( Z = 0.012)1 combined with an enhanced helium abundance (Y = 0.31) and drastically reduced mixinglengths for each star (αMLT = 0.58αMLT, ⊙ and 0.40αMLT, ⊙ , respectively) produced the highest quality fit to the data at an age of 1.9 Gyr. The age inferred from their models is a factor of 4 lower than the 8 Gyr age commonly cited for the system. Despite properly fitting the two components, the investigation by Lastennet et al. (2003) did not provide any physical justification for the reduction in mixing-length. Furthermore, they required an abnormally high helium abundance given the required sub-solar heavy element abundance. Assuming that Y varies linearly with Z according to the formula µ ¶ ¢ ∆Y ¡ Y = Yp + Z − Zp , ∆Z (6.1) where Y p is the primordial helium mass fraction and Z p = 0, implies that ∆Y /∆Z > 5 for the Lastennet et al. (2003) study. Empirically determined values all converge around 2 ± 1 (Casagrande et al. 2007). The empirical relations are by no means certain and there is no guarantee that all stars conform to this prescription. However, a single data point suggesting ∆Y /∆Z > 5 is a significant outlier, at 3σ above the empirical relation, introducing some doubt as to whether that particular Y and Z combination is achievable. We must then look elsewhere to reconcile stellar models with observations of the secondary. The stars in UV Psc are known exhibit strong magnetic activity, showcasing a wide variety of phenomena associated with stars having magnetically heated atmospheres. Soft X-ray emission (Agrawal et al. 1980), Ca ii H & K emission (Popper 1976; Montes et al. 1995a), and Hα emission (Barden 1985; Montes et al. 1995b) have all been observed and 1 We calculate this implies [Fe/H] = −0.14 considering the required Y and the fact that they were using the GN93 heavy element abundances 183 associated with UV Psc. Flares have been recorded in Hα (Liu et al. 1996) and at X-ray wavelengths (Caillault 1982), further suggesting the components are strongly active. Star spots betray their presence in the modulation and asymmetries of a several light curves (Kjurkchieva et al. 2005), although some of these modulations have also been attributed to intrinsic variability in one of the components (Antonopoulou 1987). There does not appear to be any further evidence supporting this claim (Ibanoglu 1987; Popper 1997), leading us to believe any observed light curve variations are the result of spot modulation or flare activity. The aforementioned evidence provides clues that magnetic fields may be the source of the observed radius discrepancies. Lastennet et al. (2003)’s finding that a reduced convective mixing-length was required could then be explained by the need for a mechanism to inhibit the transport of thermal energy by convection (Chabrier et al. 2007). All studies of UV Psc have found that standard stellar evolution models are entirely capable of reproducing the fundamental stellar properties of the primary star (Popper 1997; Lastennet et al. 2003; Torres et al. 2010; Feiden & Chaboyer 2012a). Therefore, we begin by assuming that UV Psc A conforms to the predictions of stellar evolution theory, but that magnetic effects must be invoked to reconcile models with UV Psc B. Given this assumption, UV Psc A may be used to constrain the age and metallicity of the system. Using a large grid of stellar evolution isochrones, Feiden & Chaboyer (2012a) found UV Psc A was best fit by a 7 Gyr isochrone with a slightly metal-poor composition of −0.1 dex. The metallicity estimate is commensurate with Lastennet et al. (2003), though two independent methods were utilized to achieve the result. We adopt this sub-solar value as the initial target age and metallicity for the system. Standard model mass tracks are illustrated in Figures 6.1(a) and 6.1(b) for two metallicities. The age of the system is anchored to the narrow region in Figure 6.1(a) where the models agree with the observed radius of the primary. Figure 6.1(b) indicates that the [Fe/H] = −0.1 model yields satisfactory agreement with the observed effective temperature, as well as the 184 1.20 [Fe/H] = -0.1 1.20 [Fe/H] = -0.3 1.10 1.10 UV Psc A Radius (R⊙) Radius (R⊙) UV Psc A 1.00 0.90 0.90 0.80 0.80 UV Psc B 0.70 0.01 1.00 0.1 1 UV Psc B 0.70 6000 10 Age (Gyr) 5600 5200 Teff (K) 4800 4400 Figure 6.1: Standard DSEP mass tracks of UV Psc A (maroon) and UV Psc B (light blue) computed with [Fe/H] = −0.1 (solid line) and [Fe/H] = −0.3 (dashed line). (a) The ageradius plane. Horizontal swaths delineate the observed radii with associated 1σ uncertainty. The vertical region indicates the age predicted by the primary. (b) The Teff -radius plane. Shaded regions denote the observational constraints. radius. We infer an age of 7.2 Gyr, more precise than Feiden & Chaboyer (2012a) as we are not constrained to a discretized set of ages. The standard model for the secondary is shown to reach the observed radius at an age of 18 Gyr, according to Figure 6.1(a), an 11 Gyr difference between the two components. We also see that the model effective temperature of the secondary is too hot compared to observations. Whereas the empirical effective temperature (see Table 6.1) was measured to be 4 750 ± 80 K, standard models are hotter by approximately 250 K. Magnetic models of the secondary component were computed using a dipole profile, singlestep perturbation at 10 Myr (see Section 5.3 for details) for several values of the surface magnetic field strength. A surface magnetic field strength of 4.0 kG (corresponding to a tachocline field strength of 11 kG) produced a model radius that was commensurate with the observed radius at an age consistent with the primary. This fact is depicted in Figure 6.2(a) by the dash line representing the magnetic model of the secondary. The vertical lines bisecting the entire figure provide the age range at which the model must match the 185 1.20 |B|surf = 4.0 kG Non-magnetic 1.20 [Fe/H] = -0.1 1.10 1.10 UV Psc A Radius (R⊙) Radius (R⊙) UV Psc A 1.00 0.90 0.90 UV Psc B 0.80 0.80 |B|surf = 4.0 kG UV Psc B 0.70 0.01 1.00 0.1 1 0.70 Non-magnetic 6000 10 Age (Gyr) [Fe/H] = -0.1 5000 Teff (K) 4000 Figure 6.2: Similar to Figure 6.1 but with a single metallicity of [Fe/H] = −0.1 dex. Magnetic mass track for UV Psc B with a 4.0 kG surface magnetic field strength (light blue, dashed line). Standard DSEP mass tracks are plotted for comparison. (a) Age-radius plane. (b) Teff -radius plane. observed radius. Next, we checked that the effective temperature predicted by the magnetic model agreed with the temperature inferred from observations (Figure 6.2(b)). The same 4.0 kG magnetic mass track required to fit the secondary in the age-radius plane over-suppresses the effective temperature, forcing the model to be too cool compared to the empirical value. Intuitively, one might suggest lowering the surface magnetic field strength so as to maintain agreement in the age-radius plane while allowing for a hotter effective temperature. However, all field strengths that provide agreement in the age-radius plane are unable to predict an effective temperature sufficiently hot enough, they are all too cool. How might we interpret the remaining temperature disagreement? One possible solution is that the effective temperature measurement is incorrect. We feel this scenario is unlikely considering the temperatures are hot enough where large uncertainties associated with complex molecular bands are not present. The uncertainties quoted in Table 6.1 seem to us large enough to encompass the actual value. Another possibility is that we have not treated convection properly. Convection within the component stars may not have the same inherent 186 1.20 |B|surf = 3.0 kG Non-magnetic Bonaca et al. (2012) αMLT 1.20 [Fe/H] = -0.1 1.10 1.10 UV Psc A 1.00 Radius (R⊙) Radius (R⊙) UV Psc A 0.90 0.80 1.00 UV Psc B 0.90 0.80 UV Psc B |B|surf = 3.0 kG 0.70 0.01 0.70 Bonaca et al. (2012) αMLT 0.1 1 Non-magnetic 6000 10 Age (Gyr) [Fe/H] = -0.1 5000 Teff (K) 4000 Figure 6.3: Similar to Figure 6.2 except that all of the mass tracks have a αMLT reduced according to the Bonaca et al. (2012) empirical relation. The surface magnetic field strength used in modeling the secondary is 3.0 kG. (a) Age-radius plane. (b) Teff -radius plane. properties as convection within the Sun. This idea was what led Lastennet et al. (2003) to allow for an arbitrary mixing-length in each component. However, while mixing-length theory is not entirely realistic and allows for such an arbitrary choice of the mixing-length, arbitrary reduction without a necessary physical motivation (other than providing better empirical agreement) is not wholly satisfying and does not fully illuminate the reasons for the appearance of the discrepancies. Instead of applying an arbitrary adjustment to the convective mixing-length, we may attempt to modify the convective mixing-length parameter according to the relation developed by Bonaca et al. (2012) who calibrated the mixing-length parameter in stellar models to asteroseismic data. The physical motivation is that convection is inherently different among various stars due to their physical properties (i.e., log g , Teff , [M/H]). Modifications to the convective mixing-length are, therefore, no longer arbitrary and may not take on any random value that happens to allow the models to fit a particular case. The Bonaca et al. (2012) relation ascribes a mixing-length of αMLT = 1.71 for the primary and αMLT = 1.49 for the secondary of UV Psc, given a metallicity of −0.1 dex and our solar calibrated αMLT, ⊙ = 1.94. The resulting mass tracks in the Teff -radius plane are plotted in Figure 6.3. 187 Directly altering convection in this manner does not provide an adequate solution. The Bonaca et al. (2012) formulation indicates that the convective mixing-length is proportional to metallicity and Teff while inversely proportional to log g . Essentially, convection is less efficient in low-mass, metal-poor stars as compared to the solar case. Reducing the mixinglength in this fashion inflates both of the stellar radii and forces the temperature at the photosphere to decrease. The mixing-length primarily affects the outer layers of the star, where energy is transported by convection. A lower mixing-length implies that there is less energy flux across a given surface within the convection zone. Since the star must remain in equilibrium, the outer layers puff up to increase the energy flux, thereby reducing the effective temperature. Inflating the primary component means the models of the secondary must now agree with the observational properties at an age younger than 7.2 Gyr. This is well illustrated in Figure 6.3(a), where the vertical dashed lines anchoring the age of the system to UV Psc A are shifted to the left by 0.5 Gyr of where they were in Figure 6.1(a). It is true that a weaker magnetic field is required to alleviate the radius disagreements with the secondary due to inflation caused by the reduced mixing-length. Shown in Figure 6.3(a) is a magnetic model with a surface field strength of 3.0 kG. We do not find agreement between the model and empirical radius, but more importantly, Figure 6.3(b) demonstrates that the effective temperature of the secondary is already too cool. Increasing the magnetic field strength to yield agreement in the age-radius plane would only worsen the lack of agreement in the Teff -radius plane. Reducing the mixing-length is unable to provide relief to the oversuppression of the effective temperature induced by the magnetic field in Figure 6.2(b). We must seek another method to rectify the effective temperature of the magnetic model. The metallicity is an unconstrained input parameter for models of UV Psc. Recall, our selection of [Fe/H] = −0.1 was based solely on the agreement of standard stellar evolution models with the primary. Updating the adopted metallicity (and consequently, the helium 188 abundance) significantly affects the structure and evolution of the UV Psc components. Stars with masses above approximately 0.45M⊙ are similarly affected by altering the chemical composition. Both the stellar radiative opacity and the efficiency of the proton-proton (pp) chain are influenced. Helium plays an important role in governing the p-p chain by increasing the mean molecular weight. The abundance of metals is relatively unimportant in the nuclear energy generation as the CNO cycle does not contribute significantly to the overall energy budget at the given masses and the impact of metals on the mean molecular weight is small in comparison to the impact of helium. Radiative opacities are influenced by both helium and heavy element abundances. Decreasing the metallicity reduces the helium abundance in our models, as we assume a linear relation between Y and Z according to equation (6.1) with d Y /d Z = 1.6. The result is that the efficiency of energy generation via the p-p chain is reduced. However, a critical energy generation rate must be achieved for the star to maintain hydrostatic equilibrium. The star contracts, heating up the stellar material, causing the star to become hotter. The decrease in metals also reduces the stellar radiative opacity, causing the layers below the convection zone to contract in order to maintain a constant energy flux across a given surface. Ultimately, the star is left with a smaller radius and higher effective temperature. Adopting a lower metallicity of [Fe/H] = −0.3 for UV Psc increases the effective temperature of both components in the standard models while shrinking their radii at younger ages2 . Doing so also destroys the agreement between models effective temperatures and UV Psc A. These effects are demonstrated for standard models in Figures 6.1(a) and 6.1(b). Accurately reproducing the observed stellar properties now requires the use of magnetic models for both components. One should have expected this from the outset considering both stars are synchronously rotating with a period of 0.86 days. Magnetic models with a dipole profile and single-step perturbation were constructed for 2 At older ages, evolutionary effects begin to play a role as the stellar lifetimes are decreased at lower metallicity owing to the higher temperatures within the stellar interior. 189 1.20 Magnetic 1.20 Non-magnetic UV Psc A 1.10 1.10 [Fe/H] = -0.3 1.00 Radius (R⊙) Radius (R⊙) UV Psc A 0.90 0.80 0.90 UV Psc B 0.80 UV Psc B 0.70 0.01 1.00 0.1 1 0.70 [Fe/H] = -0.3 6000 10 Age (Gyr) Magnetic Non-magnetic 5000 Teff (K) 4000 Figure 6.4: The UV Psc system assuming a lower heavy element abundance of [Fe/H] = −0.3. Standard DSEP mass tracks are drawn as maroon and light blue solid lines for UV Psc A and B, respectively. Magnetic tracks are represented by dashed lines with the same color coding as the standard tracks. Surface magnetic field strengths are 2.0 kG and 4.6 kG for UV Psc A and B, respectively. (a) Age-radius plane. (b) Teff -radius plane. both components. We find that it is possible to wholly reconcile the models with the observations if the primary has a 2.0 kG surface magnetic field and the secondary has a 4.6 kG surface magnetic field. Model radii and temperatures match the empirical values within the age range anchored to the primary. These facts are shown in Figures 6.4(a) and 6.4(b). The revised age of the UV Psc system found from Figure 6.4(a) (bracketed by the two vertical lines) is between 4.4 Gyr and 5.0 Gyr. Averaging the two implies the age is 4.7±0.3 Gyr. This age is considerably younger than the 7 Gyr or 8 Gyr age commonly prescribed. While feedback from the models was necessary to adjust and improve the initial metallicity and to determine the required magnetic field strengths, we believe that the final result is entirely consistent with the available observational data. In fact, due to the lack of constraint on the stellar composition, our reliance on such a feedback cycle was inevitable. The metallicity range allowing for complete agreement is not limited to −0.3 dex. Further reducing the metallicity would also likely produce acceptable results, deduced from the fact that the models of UV Psc B just barely skirt the boundaries of the empirical values. Our final 190 recommendation is that the metallicity of UV Psc is approximately [Fe/H] = −0.3 ± 0.1 dex with surface magnetic fields of 2.0 kG and 4.6 kG for the primary and secondary, respectively. Verification of the metallicity, and possibly the magnetic field strengths, should be obtainable using spectroscopic methods. 6.2.2 YY Geminorum YY Geminorum (also Castor C and GJ 278 CD; hereafter YY Gem) has been the subject of extensive investigation after hints of its binary nature were uncovered spectroscopically (Adams & Joy 1920). The first definitive reports of the orbit were published simultaneously using spectroscopic (Joy & Sanford 1926) and photographic methods (van Gent 1926), which revealed the system to have an incredibly short period of 0.814 days. The photographic study of van Gent (1926) also showed that the two components eclipsed one another with the primary and secondary eclipse depths appearing nearly equal, suggesting the components were very similar. Rough estimations of the component masses and radii were performed using the available data, but the data was not of sufficient quality to extract reliable values (Joy & Sanford 1926). The system has since been confirmed to consist of two equal mass, early M-dwarfs with masses and radii established with a precision under 1% (Torres & Ribas 2002), presented in Table 6.1. An age and metallicity have also been attributed to YY Gem, which has a common proper motion with the Castor AB quadruple. Considered gravitationally bound, these three systems have been used to define the Castor moving group (Anosova et al. 1989). Spectroscopy of Castor Aa and Ba, both spectral-type A stars, yielded a metallicity of [Fe/H] = +0.1 ± 0.2, where the uncertainty is dominated by the fact that the original metallicity estimate was relative to Vega and not the Sun (Smith 1974; Torres & Ribas 2002). Stellar evolution models of Castor Aa and Castor Ba provide an age estimate of 359 ± 34 Myr, obtained by combining estimates from DSEP (400 Myr; Feiden & Chaboyer 2012a) and the values quoted 191 Dipole Profile 0.90 |B|surf = 2.5 kG |B|surf = 3.0 kG |B|surf = 4.0 kG |B|surf = 5.0 kG Radius (R⊙) 0.80 Non-magnetic 0.70 YY Gem A/B 0.60 [Fe/H] = -0.1 10 100 1000 Age (Myr) Figure 6.5: Standard DSEP (solid, black line) and magnetic (colored, broken lines) stellar evolution models of YY Gem in the age-radius plane. The horizontal swath represents the observational radius with associated 1σ uncertainty. The vertical shaded region marks the age limits set by Castor A and B. in the Appendix of Torres & Ribas (2002). Over half a century after its binarity was discovered, advanced (at the time) low-mass stellar evolution models suggested that the observed radii of YY Gem may not agree with predictions (Hoxie 1970, 1973). Although a subsequent generation of models appeared to find full agreement with the observations (Chabrier & Baraffe 1995), confirmation of true discrepancies remained veiled as the models and observations were fraught with uncertainties. Modern observational determinations of the stellar properties for the components of YY Gem (Ségransan et al. 2000; Torres & Ribas 2002), compared against the latest set of sophisticated stellar evolution models (Baraffe et al. 1998; Dotter et al. 2008), have now solidified the fact that the components of YY Gem appear inflated compared to model predictions by approximately 8% (Torres & Ribas 2002; Feiden & Chaboyer 2012a). Figure 6.5 shows a single standard stellar evolution mass track for the components of YY 192 0.90 Radius (R⊙) 0.80 0.70 0.60 4400 4200 4000 3800 3600 3400 3200 3000 Teff (K) Figure 6.6: Standard DSEP (solid, black line) and magnetic (colored, broken lines) stellar evolution models of YY Gem in the Teff -radius plane. The shaded region denotes the observational constraints. Gem as a black, solid line. The vertical dash lines border the adopted age thought to represent YY Gem. It is clear from Figure 6.5 that the standard stellar model significantly under predicts the radius measured by Torres & Ribas (2002) (shown as the light blue horizontal swath) and in Figure 6.6 the over predicted effective temperature is evident. The adopted metallicity for this particular set of standard models is [Fe/H] = -0.1, found to provide the best agreement between models and observations (Feiden & Chaboyer 2012a). As an aside, it may be noted from Figure 6.5 that our models are consistent with the properties of YY Gem at a young age. Previous studies have considered the possibility that YY Gem is still undergoing its pre-main sequence contraction (Chabrier & Baraffe 1995; Torres & Ribas 2002), as would be inferred from our models if no age prior were assigned. The standard model in Figure 6.5 predicts an age of about 60 Myr, around 50 Myr before our models indicate a 0.6 M⊙ star settles onto the main sequence. Torres & Ribas (2002) provided a detailed analysis of this consideration and conclude that it is erroneous to assume 193 YY Gem is a pre-main sequence system, primarily due to YY Gem’s association with the Castor quadruple. Without evidence to the contrary, this fact places YY Gem firmly on the main sequence and insures that the system remains discrepant with stellar models. YY Gem exhibits numerous features indicative of intense magnetic fields and activity. Light curve modulation has been continually observed (Kron 1952; Leung & Schneider 1978; Torres & Ribas 2002), suggestive of the presence of star spots. Debates linger about the precise latitudinal location and distribution (e.g., Güdel et al. 2001) of star spots, but spots contained below midlatitude (between 45◦ and 50◦ ) appear to be favored (Güdel et al. 2001; Hussain et al. 2012). The components display strong Balmer emission (Young et al. 1989; Montes et al. 1995b) and X-ray emission (Güdel et al. 2001; Stelzer et al. 2002; López-Morales 2007; Hussain et al. 2012) during quiescence and have been observed to undergo frequent flaring events (Doyle et al. 1990; Doyle & Mathioudakis 1990; Hussain et al. 2012). Furthermore, YY Gem has been identified as a source of radio emission, attributed to partially relativistic electron gyrosynchrotron radiation (Güdel et al. 1993; McLean et al. 2012). Given this evidence, it is widely appreciated that the components possess strong magnetic fields. Therefore, it would not be entirely surprising if the interplay between convection and magnetic fields contributed, at least in part, to the disagreements mentioned previously. Magnetic stellar evolution mass tracks were computed for various values of the surface magnetic field strength using a single-step perturbation with a dipole field configuration. These tracks are plotted in Figure 6.5 and 6.6 as dashed, colored lines. The adopted metallicity is, once again, -0.1 dex. The degree of radius inflation and temperature suppression increases as subsequently stronger values of the surface magnetic field strength are applied with the strongest field strength of 5.0 kG over predicting the observed stellar radius. Figure 6.5 also demonstrates that a 4.0 kG field strength matches the lower-bound of the observed radius. We, therefore, may confine the allowable magnetic field strength to be within the range of 4.0 kG to 5.0 kG. 194 Testing the agreement with the effective temperature, Figure 6.6 reveals that the models are just able to match the observed effective temperature with a 4.0 kG magnetic field. Any stronger of a field over-suppresses the effective temperature, forcing the model to be cooler than the actual observations. Recall, we encountered this same issue when attempting to model UV Psc in the previous section. Lowering the metallicity provided a solution for UV Psc, but doing so for YY Gem would amount to jeopardize the metallicity prior established by the association with the other Castor components (Smith 1974; Torres & Ribas 2002). Before ruling out the option of a lower metallicity, we decided to recompute the approximate metallicity of YY Gem using the Smith (1974) data. First, we needed to determine the metallicity of Vega, the reference for the (Smith 1974) study. Vega has 21 listed metallicity measurements in SIMBAD, of which, the 8 most recent appear to be converging towards a common value. Using the entire list of 21 measurements, Vega has a metallicity [Fe/H] = -0.4 ± 0.4 dex. If, instead, we adopt only those measurements performed since 1980, we find [Fe/H] = -0.6 ± 0.1 dex. Due to the convergence of values in more recent years, we elect to adopt the latter value for the metallicity of Vega. The metallicities measured by Smith (1974) for Castor A and Castor B were +0.98 dex and +0.45 dex, respectively. Taking the average of these two quantities as the metallicity for the Castor AB system, we have that [Fe/H] = +0.7 ± 0.3 dex. The difference in metallicity of Castor A and B might be explained by diffusion processes (e.g., Richer et al. 2000) and is not necessarily a concern. However, the fact that we are not observing the initial abundances for Castor A and B is a concern when it comes to prescribing a metallicity for YY Gem. Therefore, a conservative estimate for the metallicity of YY Gem relative to the Sun would be [Fe/H] = +0.1 ± 0.4 dex, providing more freedom in our model assessment of YY Gem. We note that this reassessment neglects internal errors associated with the abundance determination performed by Smith (1974). However, given the radical difference in metallicity measured for Castor A and B, the real metallicity is presumed to lie within the statistical 195 Dipole Profile 0.90 |B|surf = 4.3 kG Non-magnetic Radius (R⊙) 0.80 0.70 YY Gem A/B 0.60 [Fe/H] = -0.2 10 100 1000 Age (Myr) Figure 6.7: Same as Figure 6.5 except with a metallicity of −0.2 dex. error. The Castor sextuple system would certainly benefit from revised abundances using modern methods. New abundance determinations would not only enhance our understanding of YY Gem, but also provide confirmation that the three binaries comprising the Castor system have a common origin. Presented with greater freedom in modeling YY Gem, we computed additional standard and magnetic mass tracks with [Fe/H] = -0.2 dex. The magnetic tracks were computed in the same fashion as the previous set to provide a direct comparison on the effect of metallicity. Figures 6.7 and 6.8 illustrates the results of these models. While difficult to discern without the previous [Fe/H] = -0.1 standard mass track for comparison, the standard model mass track with [Fe/H] = -0.2 is smaller in radius by approximately 1% along the main sequence. The standard model with the revised metallicity also shows a hotter effective temperature, as anticipated. The magnetic mass track with a surface magnetic field of 4.3 kG was found to best fit the observations. At 360 Myr (indicated by the dashed vertical line in Figure 6.7), it is 196 0.90 Radius (R⊙) 0.80 0.70 0.60 4400 4200 4000 3800 3600 3400 3200 3000 Teff (K) Figure 6.8: Same as Figure 6.6 except with a metallicity of −0.2 dex. apparent that the magnetic models of YY Gem satisfy the radius restrictions enforced by the observations. The precise model radius inferred from the mass track is 0.620 R ⊙ , compared to the observed radius of 0.6194 R⊙ , a difference of 0.1%. Figure 6.8 further demonstrates that while the model is consistent with the observed radius, the effective temperature of the mass track is also in agreement with the observations. The model effective temperature at 360 Myr is 3773 K, well within the 1σ observational uncertainty (see Table 6.1). There is one additional constraint that we have yet to mention. Lithium has been detected in the stars of YY Gem (Barrado y Navascués et al. 1997). They find logN(7 Li) = 0.11, where logN(7 Li) = 12 + log(X Li /A Li X H ). Standard stellar models, however, predict that lithium should be completely depleted from the surface after about 15 Myr—well before the stars reach the main sequence. Since magnetic fields can shrink the surface convection zone, it is possible for the fields to extend the lithium depletion timescale (MacDonald & Mullan 2010). This is precisely what our magnetic models predict. With a metallicity of [Fe/H] = −0.2 at 375 Myr and with a 4.3 kG magnetic field our models predict logN(7 Li) ∼ 0.9 while with 197 [Fe/H] = −0.1 at 360 Myr and with a 4.0 kG we find logN(7 Li) = 0.1. The latter value is consistent with the lithium abundance determination of Barrado y Navascués et al. (1997). In summary, we find that the best fit magnetic models have a magnetic field strength between 4.0 and 4.5 kG. A sub-solar metallicity of [Fe/H] = −0.2 provides the most robust fit in the age-radius and Teff -radius planes, but a metallicity as high as [Fe/H] = −0.1 is allowed. The latter result also provides a lithium abundance estimate consistent with observations, whereas the lower metallicity model predicts too much lithium leftover at 360 Myr. All of these results should be verifiable using spectroscopic methods. 6.2.3 CU Cancri The variable M-dwarf CU Cancri (GJ 2069A, hereafter CU Cnc; Haro et al. 1975) was recently discovered to be a double-lined spectroscopic binary system by Delfosse et al. (1998). Follow up observations provided evidence that CU Cnc underwent periodic eclipses, making it the third known M-dwarf DEB at the time (Delfosse et al. 1999). Shortly thereafter, Ribas (2003) published a detailed re-analysis of the CU Cnc system using more detailed light curves in combination with the Delfosse et al. (1998). Ribas was able to derive precise masses for the two components, similar to Delfosse et al.. However, the detailed light curves obtained by Ribas (2003) allowed him to estimate the stellar radii with high precision. These values were presented in Table 6.1. Initial comparisons with the Baraffe et al. (1998) models indicated that the components of CU Cnc were peculiar. They were found to be 1 magnitude under luminous in the V band compared to solar metallicity models. Additionally, the prescribed spectral type was two subclasses later than expected for two 0.4M⊙ stars (M4 instead of M2; Delfosse et al. 1999). These oddities were used as evidence that CU Cnc has a super-solar metallicity. The increased metallicity would increase the TiO opacity at optical wavelengths thereby producing stronger TiO absorption features used for spectral classification (see Section 1.1). The 198 absolute V band magnitude would also be lowered since TiO bands primarily affect the opacity at optical wavelengths, shifting flux from the optical to the near-infrared. Using the BCAH98 models with metallicity 0.0 and −0.5, Delfosse et al. (1999) performed a linear extrapolation to estimate the metallicity of CU Cnc. They found [Fe/H] ∼ +0.5. The super-solar metallicity quoted by Delfosse et al. (1999) was supported by the space velocity of CU Cnc. It has galactic velocities U ≈ −9.99 km s−1 , V ≈ −4.66 km s−1 , and W ≈ −10.1 km s−1 . CU Cnc is therefore posited to be kinematically young and a member of the thin-disk, which is characterized by young, more metal-rich stars. However, the space velocities were used by Ribas (2003) to refute the metallicity estimate provided by Delfosse et al. (1999). Instead of supporting the notion that CU Cnc had a super-solar metallicity, Ribas conjectured that CU Cnc was a member of the Castor moving group, which is defined by U = −10.6 ± 3.7 km s−1 , V = −6.8 ± 2.3 km s−1 , and W = −9.4 ± 2.1 km s−1 ). Therefore, Ribas prescribed the metallicity of the Castor system to CU Cnc (see Section 6.2.2). He then assumed that CU Cnc had a near-solar metallicity With a metallicity (and age) estimate in hand, Ribas (2003) performed a detailed comparison between stellar models and the observed properties of CU Cnc. Models of the CU Cnc components were found to predict radii that were 10 – 14% smaller than observed. Furthermore, the effective temperature of the models were 10 – 15% hotter than the effective temperature estimated by Ribas (2003). CU Cnc was found to under luminous in the V band and the K band by 1.4 magnitudes and 0.4 magnitudes, respectively. Ribas proceeds to then lay out detailed arguments that neither stellar activity or metallicity provide a satisfactory explanation for the observed radius, Teff , and luminosity discrepancies. Instead, he proposes that CU Cnc may possess a circumstellar disk. The disk would then disproportionately affect the observed V band flux and not the K band. This would also force the effective temperatures to be reconsidered, leading to a change in the observed luminosities. The Ribas (2003) results rely heavily on the estimated effective temperature of the individual 199 components. Determining the effective temperature of M-dwarfs is fraught with difficulty. There is a strong degeneracy between metallicity and effective temperature for M-dwarfs when considering photometric color indicies. We will therefore return to a detailed discussion of the luminosity discrepancies later and focus on the radius deviations, for the moment. Radius estimates will be less affected by the presence of a circumstellar disk since radius determinations rely on differential photometry. In Chapter 4, our models preferred a super-solar metallicity when attempting to fit CU Cnc. The maximum metallicity permitted in that analysis was only [Fe/H] = +0.2 dex. We begin with a standard model analysis of CU Cnc assuming a super-solar metallicity with [Fe/H] ≥ +0.2 dex. This is valid even if CU Cnc is a member of the Castor moving group, as suggested by Ribas (2003). Recall that in Section 6.2.2 we placed constraints on the metallicity of the stars in the Castor sextuple of [Fe/H] = +0.1 ± 0.4. The radius evolution of both components is presented in Figure 6.9 for multiple metallicities. The results are nearly independent of the adopted metallicity. All of the mass tracks show that the models do not match the observed stellar radii at the same age along the main sequence. Models of the primary appear to deviate from the observations more than models of the secondary. This may just be a consequence of the larger radius uncertainty quoted for the secondary creating the illusion of better agreement. Agreement between the models and observations for both components is found at 120 Myr. At this age, the stars are both undergoing gravitational contraction along the pre-mainsequence (pre-MS). Can we rule out the possibility that the stars of CU Cnc are still in the pre-MS phase? At this point we can not. Ribas (2003) tentatively detect lithium in the spectrum of CU Cnc, which strongly suggests that the stars would be in the pre-MS phase. However, our models predict that lithium disappears entirely around 20 Myr for the stars of CU Cnc. We do note that the agreement between the models and observations occurs right near the boundary of the gray vertical box (centered on 120 Myr) in Figure 6.9. The 200 0.44 0.42 Radius (R⊙) CU Cnc A 0.40 0.38 CU Cnc B [Fe/H] = +0.2 [Fe/H] = +0.3 [Fe/H] = +0.4 0.36 10 100 1000 10000 Age (Myr) Figure 6.9: Standard DSEP mass tracks for CU Cnc A (maroon) and CU Cnc B (lightblue) at three different metallicities: +0.2 (solid), +0.3 (dashed), +0.4 (dotted). Horizontal bands identify the observed radius with 1σ uncertainty while the vertical band identifies the region in age-radius space where the models match the primary stars observed radius. evidence for agreement is certainly not definitive. Let us assume that the stars have reached the main sequence. There is evidence that the stars are magnetically active. The stars show strong X-ray emission based on ROSAT observations3 (López-Morales 2007; Feiden & Chaboyer 2012a), indicative of a magnetically heated corona. CU Cnc is also considered to be an optical flare star. It was first identified as such by (Haro et al. 1975) and optical flaring has been continually noted since then (e.g., Qian et al. 2012). Furthermore, the stars show strong chromospheric Balmer and Ca ii K emission during quiescence (Reid et al. 1995; Walkowicz & Hawley 2009). These tracers point toward the presence of at least moderate magnetic activity on the stellar surface. Magnetic models were computed using the dipole field profile. Two values for the surface magnetic field strength were chosen, 2.6 kG and 3.5 kG. Mass tracks that include a magnetic 3 ROSAT observations actually contain both CU Cnc and its companion CV Cnc and so are difficult to interpret. 201 CU Cnc A 0.44 Radius (R⊙) 0.42 0.40 0.38 CU Cnc B 0.36 Non-magnetic |B|surf = 2.6 kG |B|surf = 3.5 kG [Fe/H] = +0.3 10 100 1000 10000 Age (Myr) Figure 6.10: Magnetic stellar evolution mass tracks of CU Cnc A (maroon) and CU Cnc B (light-blue) with surface field strengths of 2.6 kG (dotted) and 3.5 kG (dashed). A nonmagnetic mass track for each star is shown as a solid line. All models have [Fe/H] = +0.3 following the discussion in the text. The horizontal swaths signify the observed radius with associated 1σ uncertainty. field are shown in Figure 6.10. We fixed the metallicity to [Fe/H] = +0.3 since it makes a relatively small difference to the standard model mass tracks in the age-radius plane. Note that the perturbation time is different between the 2.6 kG and 3.5 kG tracks. The perturbation age was pushed to 100 Myr in the 3.5 kG model to ensure convergence of the model immediately following the perturbation. Results of the model along the main sequence are insensitive to this change of perturbation age, as was shown in Sections 5.4.1 and 5.4.3. Figure 6.10 shows that our model of the secondary star with a surface magnetic field strength of 3.5 kG matches the observed radius of the secondary from 300 Myr (ignoring the pre-MS) until about 6 Gyr. The lower field strength model is nearly identical, except that it shows slight disagreement near the zero-age main sequence (ZAMS) around 300 Myr. Unlike the models of the secondary, the magnetic models of the primary reproduce the observations 202 7.0 M = 0.44 M⊙ 7 6.5 Li Fuses 7 log10(T) (K) Li Stable 6.0 5.5 Radiation Convection 5.0 0 0.2 0.4 0.6 Radius Fraction 0.8 1 Figure 6.11: Temperature profile within a model of CU Cnc A showing that the base of the convection zone exists at a higher temperature than the 7 Li fusion temperature. Note that the temperature profile from the stellar envelope calculation is not included. between about 900 Myr and 6 Gyr. Finding agreement between the model and observed radius near the ZAMS would require a stronger surface magnetic field. The need for a stronger magnetic field in the primary depends on the real age of the system. Ribas (2003) invoked the possible Castor moving group membership to estimate the age. The Castor moving group is thought to be approximately 350 – 400 Myr. According to Figure 6.10, this would place CU Cnc near the ZAMS if we account for magnetic fields. It also means that a stronger magnetic field would be needed in modeling the primary star. A surface magnetic field strength of 4.0 kG is required to produce agreement for the primary if it is coeval with the Castor system. However, as we discussed in Section 4.3.4, there is no compelling argument that leads us to believe that CU Cnc is coeval with the Castor sextuple. Kinematic membership is not sufficient for assigning an age or metallicity. The destruction of lithium is unaffected by the presence of the magnetic field. Unlike YY Gem, where lithium was preserved to a significantly older age, the stars in CU Cnc destroy 203 lithium in about 20 Myr independent of the magnetic field. This can be explained by the depth of the convective envelope in the stars of CU Cnc. At ∼ 0.4M⊙ , the stars are expected to have deep convection zones that extend from the stellar surface down to about 50% of the stellar radius. The depth at which lithium is destroyed (T = 2.5 × 106 K) is located at about 70% of the stellar radius. This is illustrated in Figure 6.11. Lithium will be mixed down to the base of the convection zone and be completely destroyed. Introducing a strong magnetic field (3.5 kG) reduces the size of the convection zone in CU Cnc A by 7% and by 9% in CU Cnc B. However, to preserve lithium the size of the convection zone would have to be reduced by nearly 30%, corresponding to the base of the convection zone moving from 50% to 65% of the total stellar radius. If lithium does exist on the surface of CU Cnc, then there is another process keeping lithium from being destroyed. We conclude this section on CU Cnc by returning to the photometric issues raised by both Delfosse et al. (1999) and Ribas (2003). Since the publication of Ribas (2003), the Hipparcos parallax data has been revised and updated to provide more accurate solutions (van Leeuwen 2007). The parallax for CU Cnc underwent a revision from π = 78.05 ± 5.69 mas to π = 90.37±8.22 mas, changing the distance estimate from 12.81±0.92 pc to 11.07± 1.01 pc. The absolute magnitudes must be adjusted for this revised distance. Using the V - and K -band magnitudes listed in Weis (1991) and the 2MASS survey (Cutri et al. 2003), we find MV, A = 12.27 mag, MV, B = 12.63 mag, and an integrated MK = 6.382 mag. From SIMBAD we can also obtain integrated color: (J − K ) = 0.906 and (H − K ) = 0.291,4 drawn from the 2MASS survey. Using a theoretical color-Teff transformation (Chapter 2; Dotter et al. 2008) we converted model surface properties to photometric magnitudes and colors. We were not able to reproduce the set of integrated colors and magnitudes or the individual V -band magnitudes using super-solar metallicity models alone. However, combining a super-solar metallicity 4 The integrated V -band magnitude listed on SIMBAD is 0.2 mag fainter than is quoted by Weis (1991). This is likely because the photometry listed in SIMBAD was taken during eclipse where the V -band flux drops by ∼ 0.2 mag (Ribas 2003). 204 with a magnetic field, we were able to produce models showing the appropriate trends: total V - and K - band magnitudes were reduced due to the decrease in luminosity associated with the magnetic field with a steeper decrease in the V -band due to increased metallicity (greater TiO opacity) and decreased Teff due to the metallicity and the magnetic field. The final photometric properties of our models did not exactly match the properties of CU Cnc. However, we note there is considerable uncertainty in the color-Teff transformation using the phoenix ames-cond theoretical models (Baraffe et al. 1998; Delfosse et al. 1998), particularly in the V -band. A larger exploration of the model parameter space and upgrading to the latest phoenix bt-settl models would help to determine if metallicity and magnetic fields are able to resolve the CU Cnc photometric anomalies. 6.3 Magnetic Field Strengths 6.3.1 Surface Field Strengths Section 6.2 demonstrates that the magnetic models can reconcile low-mass stellar model predictions with observations of stars in low-mass DEBs. The real predictive power of the models relies on the ability of the models to do so with realistic magnetic field strengths tied to physical observables. As with the case study of EF Aqr presented in Chapter 5, we may estimate magnetic field strengths using X-ray emission as an auxiliary measure. For the stars studied in this chapter, estimates based on Ca ii emission cores (Schrijver et al. 1989) will not be possible because Ca ii line profiles are not available. Estimates of magnetic field strength can be obtained using the scaling relation derived between magnetic flux and the total X-ray luminosity (Fisher et al. 1998; Pevtsov et al. 2003). The relation appears to extend over several orders of magnitude, from individual solar active regions up to hemispherical averages of stars. The basic method was presented in Section 5.6.1 using Equations (5.189), (5.190), and (5.191). 205 Table 6.2 X-ray properties for the three DEB systems. DEB X cr HR System (cts s−1 ) UV Psc 0.92 ± 0.07 −0.10 ± 0.07 0.73 ± 0.05 −0.14 ± 0.06 YY Gem CU Cnc 3.70 ± 0.09 π Nstars (erg s−1 ) (mas) 14.64 ± 1.03 d = 13 ± 2 pc −0.15 ± 0.02 90.37 ± 8.22 Lx 2 6 4 (2.0 ± 0.3) × 1030 (9.4 ± 0.2) × 1028 (2.0 ± 0.2) × 1028 Note: Nstars is the total number of stars thought to be contributing to the total X-ray counts detected by ROSAT. The value L x is quoted as the X-ray luminosity per star in the system. X-ray properties of the three DEB systems studied in this chapter are determined using X-ray data from the ROSAT All-Sky Survey Bright Source Catalogue (Voges et al. 1999), consistent with our previous X-ray analyses in Chapters 4 and 5. ROSAT count rates ( X cr ) and hardness ratios (HR) are given in Table 6.2. We convert to X-ray fluxes using the calibration of (Schmitt et al. 1995). The conversion to X-ray fluxes is complicated by the fact that ROSAT has relatively poor spatial resolution, meaning any nearby companions to these DEBs may also be contributing to X cr . The X-ray flux quoted in Table 6.2 is therefore calculated as the estimated flux per star, determined by dividing the total flux by the number of stars thought to be contributing to X cr . This is not a problem for UV Psc, which appears isolated. YY Gem and CU Cnc, on the other hand, have known, nearby companions. A search of the ROSAT Bright Source Catalogue for Castor A and B yield the same data as is found when searching for YY Gem, indicating that ROSAT cannot spatially resolve any of three systems. Castor A and B are both binaries, thought to have M-dwarfs companions, meaning that ROSAT is detecting X-ray emission from up to six sources. Similar to YY Gem, CU Cnc has a 12" companion, CV Cnc, another M-dwarf binary system. Both systems are likely contributing to the X cr listed in the ROSAT Bright Source Catalogue. Distances to the systems are calculated using updated Hipparcos parallaxes (van Leeuwen 2007), except for YY Gem, for which no parallax is available. Instead, we adopt a distance of 206 13 pc, which was used earlier in Section 4.5.2. The distances then allowed for the calculation of the total X-ray luminosity per star, quoted in Table 6.2. Converting X-ray luminosity measures to an estimate of the magnetic field strength requires first transforming the X-ray luminosity into an estimate of the surface magnetic flux, Φ. We discussed this relation in Section 5.6.1, where we adopted the Pevtsov et al. (2003) relation based on the work of (Fisher et al. 1998). Pevtsov et al. (2003) found that L x ∝ Φ1.15 , (6.2) as was mentioned in Section 5.6.1. The adopted power-law index depends on the particular data set used in the analysis. For instance, Pevtsov et al. (2003) found a power-law index p = 0.98 when they only considered the dwarf stars supplied by Saar (1996). When the took the entire solar (quiet regions, active regions, sun spots) and stellar data sets together, they found p = 1.15. The latter extends over twelve orders of magnitude in Φ and L x . The question of which power law index to adopt is important and can strongly influence the magnetic flux derived from X-ray luminosities. Seeing as we are focused on deriving the approximate magnetic flux for dwarf stars, is seems a natural choice to use the relation specifically derived from dwarf star data. However, the relation derived by Pevtsov et al. (2003) was based on data from Saar (1996). Since then, many more stars have had their magnetic field strengths measured, allowing us to expand the data set and derive a more reliable scaling relation. We computed the approximate surface magnetic flux for the dwarf stars presented in the Reiners (2012) review. Reiners (2012) attempted to collect all of the reliably known magnetic field measurements for cool dwarfs stars. The stars all had their magnetic field strengths measured directly using either Stokes I or Stokes V observations. We selected only those stars that had their fields measured using Stokes I , which is thought to yield a more accurate estimate of the average surface magnetic field strength 〈B f 〉 (Reiners & Basri 2009), which 207 26.5 26.0 log10(Φ) (Mx) 25.5 25.0 24.5 24.0 23.5 Pevtsov et al. (2003) 23.0 22.5 26.5 This work 27.0 27.5 28.0 28.5 29.0 -1 log10(Lx) (erg s ) 29.5 30.0 30.5 Figure 6.12: Stellar surface magnetic flux (Φ) as a function of total stellar X-ray luminosity for a collection of G-, K-, and M-dwarfs. Gray points are taken from Pevtsov et al. (2003) while the maroon points were calcuated for this work. Three separate linear regressions are presented: P = Pevtsov et al. (2003) for dwarf stars only (light-gray, dotted), T = Pevtsov et al. (2003) for (black, solid) the total comibnation of solar and stellar data, and R = data compiled for this work drawn from the Reiners (2012) sample (maroon, dashed). is the quantity predicted by our models. Results of extending the sample with are presented in Figure 6.12 with the full data set given at the end of the chapter in Table 6.5. The list of objects from the Reiners (2012) review were cross-correlated with the ROSAT Bright and Faint Source Catalogues (Voges et al. 1999; Voges et al. 2000) to extract X-ray count rates and hardness ratios. If objects did not have an X-ray counterpart they were excluded from the data set. We then identified all of the stars in the remaining subset that had parallax estimates from Hipparcos (van Leeuwen 2007). The final subset consisted of 25 objects with both X-ray counterparts and parallax measurements. Two additional data points for G-dwarfs not found in Reiners (2012) review were also included (Anderson et al. 2010) bringing the total number of objects to 27. We 208 also see in Figure 6.12 that the data points derived for this study follow the same general trend as those in Pevtsov et al. (2003). We caution that the data presented in Figure 6.12 suffer from uncertainty in the adopted stellar radius, which is largely based on spectral type, and the limited spatial resolution of ROSAT. As an example, CV Cnc is an mid-M-dwarf binary included in our final data set. Estimates of it’s X-ray luminosity include contributions from CU Cnc, as discussed above. Only for CV Cnc did we correct for a known binary companion. The uncertainty in the stellar radius only affects the magnetic flux and the lack of spatial resolution only affects the X-ray luminosity. Therefore, we believe that the estimates are robust, particularly given that Figure 6.12 is plotted using logarithmic units, where a factor of two does not contribute to a significant shift in the data points. The regression line and associated uncertainty limits shown in Figure 6.12 were obtained by performing an ordinary least squares (OLS) regression on the data. A standard OLS was selected because we are not attempting to establish a causal relationship between Φ and L x (Isobe et al. 1990). Our primary concern is developing a relationship that permits the prediction of Φ given a value for L x . Following the recommendations of Isobe et al. (1990) ¡ ¢ and Feigelson & Babu (1992), we perform an OLS log10 Φ| log10 L x , where log10 L x is the predictor variable and log10 Φ is the variable to be predicted. Notice that we transformed the variables to a logarithmic scale because the data extends over several orders of magnitude in Φ and L x . The result of the OLS analysis yielded the regression line ¢ ¡ log10 Φ = (24.873 ± 0.004) + (0.459 ± 0.018) log10 L x − µx , (6.3) where µx is the mean value of log10 L x taken over all of the data points in the sample. The shift of the dependent variable was performed before the regression analysis. By doing this, we were able to minimize the error associated with the y -intercept. However, the standard 209 Table 6.3 Surface magnetic field properties of the stars in UV Psc, YY Gem, & CU Cnc DEB log Φ Star (Mx) UV Psc A UV Psc B YY Gem A 〈B f 〉 〈B f 〉model 25.77 ± 0.45 0.79+1.43 −0.51 2.0 25.16 ± 0.45 0.62+1.13 −0.40 4.3 (kG) .. . YY Gem B .. . CU Cnc A 24.85 ± 0.45 CU Cnc B .. . (kG) 1.39+2.53 −0.90 4.6 .. . .. . 0.62+1.13 −0.40 4.0 0.76+1.38 −0.49 3.6 deviation of the mean, σx , becomes the largest source of error in the regression analysis. We found µx = 28.34 ± 0.97 where we took v u u σx = t N 1 X (x i − µx )2 . N − 1 i =1 (6.4) We opted for computing the standard deviation with N − 1 because of our small sample (27 data points) and to provide conservative estimates of the associated errors. The error associated with the standard deviation of the mean and the errors associated with the OLS coefficients in Equation (6.3) produce an uncertainty region about the regression line. We show this with the outlined shaded region in Figure 6.12. Finally, we also note that Equation (6.3) is strictly valid for 26.5 < log10 L x < 30.0, corresponding to the range of the data. Our estimates for the DEB surface fluxes, surface magnetic field strengths, and their errors are given in Table 6.3. With a non-causal relationship established between Φ and L x , we convert the X-ray luminosities for our DEB stars presented in Table 6.2 to estimates of surface magnetic flux. Care was taken to properly propagate the errors so that we have an estimate of the uncertainty of our predictions. Again, the largest source of error was the standard deviation of the mean in the shift of the dependent variable. This translates into 210 26.5 26.0 log10(Φ) (Mx) 25.5 25.0 24.5 24.0 23.5 Pevtsov et al. (2003) This work Model Predictions 23.0 22.5 26.5 27.0 27.5 28.0 28.5 29.0 -1 log10(Lx) (erg s ) 29.5 30.0 30.5 Figure 6.13: Same as Figure 6.12 except that the results of our theoretical models are included (light-blue stars). Magnetic fluxes are derived from the results in Sections 6.2.1, 6.2.2, and 6.2.3 while X-ray luminosities are computed from ROSAT data. large uncertainties when we calculate the surface magnetic field strengths from the surface fluxes. The X-ray data reveal that the magnetic field strengths required by our models are probably too large. We demonstrate the this visually in Figure 6.13, where we use the model surface magnetic field strengths to compute a surface magnetic flux. The data suggest that the only realistic magnetic fields strengths are those of UV Psc. However, the data points for UV Psc lie outside the strict range of the data set, meaning we must assume the extrapolation is valid. Even if we assume the extrapolation is valid, UV Psc B lies just above the error bounds of the linear correlation. Conversely, UV Psc A lies just inside the bounds. We must therefore be cautious with our interpretation of the accuracy of our models for UV Psc. This is further reinforced by nothing that only two G-dwarfs (the spectral of UV Psc A) were included in the data set and are therefore under represented. We do mention that estimates 211 of G-dwarf magnetic field strengths by Saar (1990) show that magnetic field strengths may be on the order 500 G, consistent with the values derived for UV Psc A from the X-ray luminosity relation. What is clear from Figure 6.13 and Table 6.3 is that the magnetic field strengths required by the models are too large for the stars in YY Gem and CU Cnc. The models of YY Gem need a magnetic field strength that is around 6 times too strong. Similarly large magnetic fields are needed by the models for the stars of CU Cnc. A more qualitative approach to evaluating YY Gem and CU Cnc would lead us to believe that our magnetic field strengths are entirely reasonable. Numerous studies of M-dwarfs find that surface magnetic field strengths of a few kilogauss are quite common (Saar 1996, 2001; Reiners & Basri 2007, 2009). However, these DEBs would then have noticeably stronger magnetic fields for their observed X-ray luminosity than the rest of M-dwarf population. It appears that our approach to modeling magnetic fields in low-mass stars may be incomplete. 6.3.2 Interior Field Strengths Assessing the validity of our predicted interior magnetic field strengths is inherently more difficult. Without any real measure currently available for the magnetic field strengths present inside stars, we elect to compare the theoretical magnetic field strengths required by our models with those predicted by 3D MHD models (see, e.g., Brandenburg & Subramanian 2005). Table 6.4 presents the peak interior magnetic field strengths for the models presented above for both the dipole and the Gaussian field profiles (see below). The peak magnetic field strengths in Table 6.4 are pre-defined to be at the base of the stellar convection zone. They are all on the order of a few times 103 to a few times 104 gauss. By comparison, 3D MHD models routinely find peak magnetic field strengths of a few times 103 gauss (consistent with equipartition estimates) to 105 gauss for solar-like stars (see review by 212 Table 6.4 Peak interior magnetic field strengths. DEB |B |dipole |B |Gaussian UV Psc A 4 40 UV Psc B 12 400 YY Gem A YY Gem B 13 .. . 500 .. . CU Cnc A 21 1500 CU Cnc B 21 2000 Star (kG) (kG) Brandenburg & Subramanian 2005). In the immediate context, “solar-like” is loosely taken to mean stars with a radiative core and a solar-like rotation profile. The peak magnetic field strengths of our models then appear consistent with those predicted from 3D MHD models. Furthermore, while helioseismological investigations have yet to reveal the interior magnetic field profile for our Sun, initial indications suggest the peak magnetic field strength is below 300 kG (Antia et al. 2000). Estimates place the strengths in the vicinity of several tens of kG (Antia et al. 2003). 6.4 Reducing the Magnetic Field Strengths Given the results that the model surface field strengths are likely too large, we seek to reformulate our magnetic perturbation. Let us first consider some of the assumptions used to formulate our magnetic models in Chapter 5. We have identified four key assumptions: (1) our prescribed magnetic field radial profile, (2) the plasma obeys the equations of ideal magnetohydrodynamics (MHD), (3) our neglect of electric fields, and (4) the dynamo is driven purely by rotation. 213 6.4.1 Magnetic Field Radial Profile All of the models presented up to this point, we have prescribed what we called a “dipole radial profile.” The interior magnetic field is set to a maximum at the boundary of the convection zone and falls off as r 3 towards the core and surface of the star. What if we instead used a much steeper profile so that the peak magnetic field strength is stronger at the base of the convection zone for a given magnetic field strength? To this end, we define a Gaussian radial profile. Again, the peak field strength is prescribed in the manner described above. However, instead of the field falling off as r 3 , the source location represents the peak of a Gaussian distribution. Functionally, this may be expressed as µ ¶ ¸ · 1 R src − R 2 B (R) = B (R src ) exp − 2 σg (6.5) where the width of the Gaussian, σg , is an additional free parameter which we will discuss in a moment, Rsrc is the radius of the peak magnetic field strength, as before. The advantage of introducing a Gaussian profile is that the field strength increases much more rapidly towards the source location. This rapid increase allows the field strength at the model photosphere to be weaker than in the dipole profile for a given value of the peak magnetic field strength. Determining the width of the Gaussian is arbitrary. In most applications, it would seem reasonable to set the width of the Gaussian to a constant value. A value of σg = 0.2 was favored by Lydon & Sofia (1995), who parameterized their Gaussian as a function of mass depth, instead of radius. However, since we are to be considering stars with a wide variety of convection zone depths and not just the Sun, a single parameter would not be appropriate. Our approach is to correlate the width of the Gaussian with the depth of the convection zone. Parameterizing the width in this way allows the magnetic field to be localized in the convection zone. Or, as for fully-convective stars, spread throughout the entire convective 214 1.2 100 15 (a) 80 4 40 2 |B|surf = 2.0 kG 0.8 10 0.4 5 Gaussian Scaled |B| (MG) 60 (b) 0.4 M⊙ Gaussian Scaled |B| (kG) Dipole Scaled |B| (kG) 1.0 M⊙ |B|surf = 2.0 kG Dipole Scaled |B| (kG) 6 20 0 0.0 0 0.0 0 0.2 0.4 0.6 0.8 1.0 Radius Fraction 0.2 0.4 0.6 Radius Fraction 0.8 0.0 1.0 Figure 6.14: Comparison of the dipole (light blue, dashed) and Gaussian (maroon, solid) radial profiles for the interior magnetic field strength. (a) a 1.0M⊙ model. (b) a 0.4M⊙ model. Note that the two field profiles are plotted on different scales with the dipole on the left y -axis and the Gaussian on the right y -axis. region. Fixing the width of the Gaussian to σg = 0.2 in fully-convective objects and σg = 0.1 in the Sun, we find σg = 0.2264 − 0.1776ζ (6.6) where ζ = R src /R ∗ is the dynamo source location in fractional units. A direct comparison of the shape of the magnetic field profiles used in this study is given in Figure 6.14. Two masses are shown to make clear the variable width of the Gaussian. Figure 6.15 illustrates the influence of using the Gaussian radial profile. We plot the relative difference in the radius evolution between the Gaussian and dipole profiles for a series of stellar masses. All of the models have an equivalent surface magnetic field strength of 2.0 kG and solar metallicity. Despite the increased magnetic field strength at the tachocline (see Table 6.4), using a Gaussian magnetic field profile instead of a dipole profile has a sub-1.5% effect on model radius predictions. Therefore, we find no compelling reason to alter our default field strength profile. Two additional comments on the results presented in Figure 6.15. The first comment concerns the fact that the different radial profiles produce similar results. We see this despite 215 (RGaussian - Rdipole)/Rdipole 0.030 1.1 M⊙ 0.025 1.0 M⊙ 0.9 M⊙ 0.020 0.8 M⊙ 0.7 M⊙ 0.015 0.6 M⊙ 0.5 M⊙ 0.4 M⊙ 0.010 0.005 0.000 -0.005 7.0 7.5 8.0 8.5 9.0 log10(Age/yr) 9.5 Figure 6.15: The effect on stellar radius predictions caused by using a Gaussian radial profile instead of a dipole profile. Shown is the relative radius difference as a function of age for a series of stellar models. All of the models have equivalent surface magnetic field strenghts (2.0 kG) but different magnetic field radial profiles. the peak magnetic field strength in the Gaussian formalism being an order of magnitude— at least—larger than in the dipole formalism. It would then seem that the deep interior field strength is relatively unimportant in stars with radiative cores. Instead, the surface magnetic field strength appears to be responsible for driving the radius inflation. To understand why, we plot the difference between the temperature gradient, ∇s , and the adiabatic gradient, ∇ad , as a function of density in Figure 6.16. Consider the non-magnetic model. When the line dips below the zero point, radiation carries all of the excess energy. Near the stellar surface, where ρ is small, ∇s is noticeably super-adiabatic, indicative of inefficient convective energy transport. Deeper in the star, where −5 < log10 ρ < 0, the temperature gradient is super-adiabatic, but only slightly (∇s − ∇ad < 10−8 ). This suggests convective energy transport is highly efficient. Our approach modifies ∇ad by a factor proportional to (ν∇χ ). Deep in the stellar interior, ∇χ ∼ 0.1, and ν ∼ 10−8 − 10−9 . This is not sufficient to inhibit convection deep within the star. Near the surface, however, any 216 Non-magnetic Magnetic 0.2 super-adiabatic convection (∇s - ∇ad) 0.1 near-adiabatic convection 0.0 0.04 radiative zone 0.00 -0.1 -0.04 -0.08 -6.5 -6.0 -5.5 -0.2 -7 -6 -5 -4 -3 -2 log10(ρ) (g cm-3) -1 0 1 2 Figure 6.16: The difference between the plasma temperature graident, ∇s , and the adiabatic temperature gradient, ∇ad , as a function of the logarithmic plasma density for a M = 0.6M⊙ star. We show this for two models: a non-magnetic model (maroon, solid line) and a magnetic model (light-blue, solid). The zero point is marked by a gray dashed line, dividing locations where convection (positive) or radiation (negative) is the dominant flux transport mechanism. The inset zooms in on the near-surface region where radiation becomes the dominant flux transport mechanism. inhibition of convection causes a steepening of ∇s , forcing radiation to carry more energy. We would then expect to see the region near the stellar surface where radiation carries all of the flux grow. This is precisely what we see in Figure 6.16. The structural changes caused by the steep temperature gradient near the surface are enough to reconcile the models with the observations before the deep interior magnetic field strength becomes appreciable in magnitude. Therefore, the outward movement of the convection zone boundary occurs largely as a response to changes near the stellar surface. Second, we see in Figure 6.15 that the 1.1M⊙ and 1.0M⊙ lines display an upturn. Occurring near 1 Gyr, the models are too young to be undergoing rapid evolutionary changes. Instead, we believe this is related to the physical properties of the stars near the boundary of the 217 convection zone. As we just discussed, the magnetic field strengths for both profiles are typically too weak near the tachocline to affect stellar structure. However, the models indicate that the opposite is true for the 1.1M⊙ and 1.0M⊙ stars. The convection zones are thin and convection is generally super-adiabatic throughout much of the convection zone. This results in the Gaussian radial profile having a stronger effect on the convection properties. It also explains why the Gaussian radial profile has a weaker influence on the stellar radius as we decrease the stellar mass. Convection in the deep interior becomes more adiabatic (re: efficient) at lower masses. 6.4.2 Finite Electrical Conductivity Assuming that the stellar plasma has an infinite conductivity, and thus obeys the equations of ideal MHD, has several implications. Therefore, it is advantageous to probe what happens when we assume a finite electrical conductivity. Generally speaking, the introduction of a finite electrical conductivity inhibits the generation of favorable current networks, making it more difficult for a dynamo mechanism to generate a magnetic field compared to the case of ideal MHD (Mohanty et al. 2002). While magnetic field strengths would be decreased due to the plasma’s inability to set up strong current networks, finite electrical conductivity affects the magnetic inhibition of convection. This has been treated previously by MacDonald and Mullan in a series of papers (MacDonald & Mullan 2009; Mullan & MacDonald 2010; MacDonald & Mullan 2010, 2013) where they applied an analytical correction factor to their magnetic suppression term discussed in Section 1.4.1. The correction factor acted to reduce the effect of their magnetic inhibition parameter when the plasma resistivity was sufficiently high. Although important, this would ultimately force us to increase our magnetic field strengths to offset the reduction in magnetic inhibition. The most significant implication for our purposes is that assuming a perfectly conducting 218 plasma lead to the definition that f ≈ 1 (see Section 5.2.5). Recall that f parametrizes the flux of magnetic energy between a convecting bubble and the surrounding plasma. If we assume a finite electrical conductivity, then f = 1 is no longer strictly enforced. This has the effect of introducing the Lorentz force into the magnetic buoyancy equation. Results of changing the parameter f from f = 1 to f = 0 are shown in Figure 6.17. As with Figure 6.15, we plot the relative difference in the radius evolution for a series of models. All of the models were calculated with a 2.0 kG magnetic field and a dipole field profile (see the previous section). The largest mass model shown is 0.7M⊙ because models with larger masses failed to converge for a 2.0 kG magnetic field. It is evident from Figure 6.17 that lowering the value of f can have a significant impact on the model radius evolution. Differences between 2% and 9% were observed, depending on the model mass. Therefore it is worthwhile to study the effect of setting f = 0 on the individual DEB stars discussed above. We tested what happened when we set f = 0 for each of the stars discussed in Section 6.2. In each case, the strength of the surface magnetic field is reduced by nearly a factor of two. YY Gem, for instance, now only requires a magnetic field strength of ∼ 2 500 G at the surface instead of the 4 300 G required when f = 1. However, when we compare with the X-ray predicted surface magnetic field strengths, the f = 0 model values are still too large by a factor of 5, save UV Psc, but we have already stated reason to be skeptical of the X-ray predicted values for UV Psc. Whether or not we have reason to believe that f should be zero is a more difficult question. The stellar plasma should be highly conducting throughout most of the star because hydrogen is completely ionized at temperatures above ∼ 104 K. In the near surface regions, hydrogen is no longer ionized. Since finite electrical conductivity is driven by the presence of neutral species, this would seem to favor a lower f value. But, there are other species which contribute to the total free electron population (e.g., Ca and Na), meaning the plasma 219 0.10 0.7 M⊙ 0.6 M⊙ (Rf = 0 - Rf = 1)/Rf = 1 0.08 0.5 M⊙ 0.4 M⊙ 0.06 0.04 0.02 0.00 7.0 7.5 8.0 8.5 9.0 log10(Age/yr) 9.5 Figure 6.17: The influence of the parameter f on model radius predictions. Plotted is the relative difference in radius evolution between magnetic models computed with f = 1 and f = 0. All models have a prescribed surface magnetic field strength of 2.0 kG. is still at least partially conducting. Therefore, it is likely that near the surface f 6= 1, but the fact that the plasma is conducting suggests that 0 < f < 1. Deriving a more realistic value for f would be a worthwhile endeavor. We plan to investigate this free parameter in a future study. Ultimately, reducing f aides our modeling work, but it does not provide enough leverage to bring our models into agreement with the observations. We then seek another method through which we can reduce our predicted surface magnetic field strengths. 6.4.3 Presence of E-fields It was discussed in Feiden & Chaboyer (2012b) that the presence of electric fields were neglected due to rapid neutralization of excess charge in stars. We still believe that this is a valid assumption. Small-scale electric fields may be permitted, but it is still true that the 220 plasma is electrically neutral on scales larger than the Debye length. The Debye length is λD = s kTe 4πn e q e2 ≈ s kTe m H 4πρq e2 , (6.7) where kTe is the temperature of the electron population, ne is the number density of electrons, and q e is the electron charge. For a plasma with a temperature of 1 eV at a density of about 1 g cm−3 , λD ∼ 10−8 cm. We therefore feel justified in neglecting the presence of large-scale electric fields. If we also then assume that the stellar plasma is highly conducting, there will be little heat loss to ohmic dissipation. 6.4.4 Dynamo Energy Source The final assumption that we identified earlier in this section was that the magnetic field is driven purely by rotation. By doing so, we have allowed for the unintended consequence that the magnetic field strength—both the surface and the interior—can grow without limit.5 There is no “natural” limit imposed upon the field strengths as the mechanism from which the field is drawing energy (i.e., rotation) is neglected entirely. Limiting the magnetic field strengths, however, does not modify in any way the results of the models. It can only validate or invalidate the field strengths the models require (see Section 6.3.2). Instead, we reconsider the physical source of the stellar dynamo. There are questions as to whether the solar magnetic field is generated at the tachocline by the strong shear induced by rotation (i.e., the standard Parker model; Parker 1955) or if it is primarily generated within the solar convection zone by turbulent convection without explicit need for a tachocline (Brandenburg & Subramanian 2005; Brown et al. 2010). Consequently, the generation of a magnetic field from turbulent convection would suppress convective velocity fluctuations thereby reducing the total heat flux transported by convec5 We note that there is a computational limit whereby too strong of a magnetic field will prevent the model from converging to a solution. 221 tion. Early magneto-convection simulations in three-dimensions support this assessment (Stein et al. 1992). Given that thick convective envelopes are a ubiquitous feature of lowmass stars, it is not unreasonable to posit that suppressing the total heat flux transported by convection would strongly impact stellar structure. Assuming that the magnetic field sources its energy directly from the kinetic energy of turbulent convection has important consequences for a magnetic theory of convection.6 Consider a single convecting bubble. Beginning with the conservation of energy, we must have that 1 2 B2 1 2 ρu conv, 0 = ρu conv + , 2 2 8π (6.8) where ρ is the mass density, uconv, 0 is the convective velocity prior to the introduction of a magnetic field, uconv is the convective velocity after generation of a magnetic field, and the final term is the magnetic energy. We have implicitly assumed that a characteristic convecting bubble is responsible for generating the magnetic field in its vicinity. This local treatment is an extreme oversimplification, but is a zeroth order approximation given the framework of mixing length theory. The result is that the characteristic convective bubble will have a lower velocity in the presence of magnetic field, 2 2 u conv = u conv, 0− B2 , 4πρ (6.9) by a value equal to the square of the local Alfvén velocity, u 2A = B2 = 2χ. 4πρ (6.10) 3 The total convective flux is proportional to uconv , meaning that the magnetic field can sig- nificantly alter the total flux transported by convection. To make up for the decreased convective flux, the radiative temperature gradient will grow steeper by an amount propor6 The theory advanced here grew out of an evening of intense discussions and derivations with Axel Brandenburg during a visit to NORDITA amidst the author’s stay in Uppsala, Sweden. 222 tional to u 2A . There are a two different approaches that we can take to include the energy conversion from turbulent kinetic energy to magnetic energy. The first involves using the new definition for the convective velocity give by Equation (6.9) and re-deriving the equations for mixing length theory given in Section 5.2.4. After modifying the convective velocity and the convective flux (to remove a term introduced by the new definition of the convective velocity) we find that the final quartic becomes h i ν 2Ay 4 +V y 3 +(1+2AC γ20 Ξ)V 2 y 2 +(γ20V 2 u 2A −1)V y + u 2A − (1 − f )∇χC γ20V 4 = 0. (6.11) δ The additional terms introduced are those proportional to u 2A . Unfortunately, when we then attempt to find the root of Equation (6.11) we are unable to obtain convergence. Closer inspection of the coefficients in the equation reveals that there is no real root in the outer envelope. The low density present in the envelope drives up u 2A , which causes the range of Equation (6.11) to lie above zero for all real values of y . We also attempted to modify the “non-adiabaticity” equation, Equation (5.127), so as to model the transformation of turbulent kinetic energy into magnetic energy as a heat loss. However, we were again unable to converge upon a real root for the quartic equation. Further investigation into the precise cause of non-convergence is required. The second method we have identified is less consistent, but provides a first-order approximation. If we first solve the MLT equations as normal, we can modify the results to mimic the conversion of turbulent kinetic energy to magnetic energy. The convective velocity is reduced by the Alfvén velocity, as in Equation (6.9). The reduction in the convective velocity therefore reduces the amount of flux carried by convection. While energy is transferred to the magnetic field, the energy is still confined to the local region under consideration, reducing the total flux of energy across the region. Assuming that flux is conserved, flux must then be carried by radiation, which leads to an increase in the local temperature gradient of 223 (Rconv - Rrot)/Rrot 0.20 0.18 0.16 0.14 0.12 0.10 0.08 0.06 0.04 0.02 0.00 7.0 1.0 M⊙ 0.9 M⊙ 0.8 M⊙ 0.7 M⊙ 0.6 M⊙ 0.5 M⊙ 0.4 M⊙ 7.5 8.0 8.5 9.0 log10(Age/yr) 9.5 10.0 Figure 6.18: The effect on model radius predictions when the dynamo source is turbulent convection instead of purely rotation. Shown is the relative radius difference as a function of age for a series of stellar models. All of the models have an equivalent surface magnetic field strenghts of 0.5 kG. the ambient plasma, further increasing the temperature excess over the convecting element. The increase in temperature excess is proportional to the total energy removed from the convecting element in the form of magnetic energy, ∆(∇s − ∇e ) ∝ u 2A C , (6.12) where C was defined in Equation (5.130). Increasing the temperature excess will create feedback onto the convective element, causing it to increase its velocity. However, since we are treating the reduction in convective flux after solving the MLT equations, we do not concern ourselves with the feedback on the convective velocity. This would have presumably been accounted for in the first approach we outlined. We again computed a series of models using this new modified MLT to compare the relative differences in radius evolution. These are shown in Figure 6.18. We assumed a 500 G 224 surface magnetic field strength with a dipole radial profile and f = 1. The 0.8M⊙ , 0.7M⊙ , and 0.9M⊙ tracks in Figure 6.18 show slightly different characteristics. Models assuming a turbulent dynamo readjusted to the presence of the magnetic field at an older age than did the models with a rotational dynamo; all perturbations were introduced at the same age, τage = 0.1 Myr. This new formulation of the magnetic models, whereby we source the magnetic field energy from the convective flux, has the potential to significantly reduce the magnetic field strengths required to inflate the model radii. We recomputed models for UV Psc, YY Gem, and CU Cnc using the best fit metallicities found in Section 5.2. We were only able to achieve model convergence for the stars in YY Gem and CU Cnc. Models of UV Psc with surface magnetic fields up to 650 G did converge, but field strengths greater than that were needed to reconcile the models. The resulting magnetic fields strengths required for YY Gem and CU Cnc A and B were 0.7 kG, 0.7 kG, and 0.8 kG, respectively. We also attempted to recompute models for EF Aquarii (EF Aqr) presented in (Feiden & Chaboyer 2012b), but ran into the same convergence issues as with UV Psc. The surface magnetic field strengths derived from sourcing the magnetic field energy from convection are nearly identical to the X-ray estimated values. This agreement lends credence to our latest approach, at least for lower mass models. We found it difficult to attain model convergence using this approach for models with masses greater than about 0.75M⊙ . However, below that approximate mass limit the turbulent dynamo approach can inflate stellar radii with relative ease compared to our original formulation. Above 0.75M⊙ , more investigation examining the structure of the convection zones in these stars is required to diagnose the convergence issues. The solution could be that the radial profile produces magnetic field strengths that are simply too large compared to the predicted convective velocities. 225 6.5 Further Discussion 6.5.1 Interior Structure We saw in the previous section that there are two approaches to modeling the effects of magnetic fields: stabilizing the plasma against convection and reducing convective efficiency. These approaches hint at different physical mechanisms by which stars produce magnetic fields. They also hint at the dominant mechanism responsible for inflating stellar radii (see Section 6.5.2). It is therefore worth exploring whether or not the different approaches affect stellar interior structure—and as a consequence observable properties—in any discernible manner. Figure 6.19 shows the run of density and the development of the parameter η2 (R) through the model interior. Recall from Equations (3.7) and (3.9) that η2 is related to the stellar deviation from sphericity and the interior structure constant, k 2 . All three models were computed with a mass of 0.599M⊙ , a metallicity of −0.20 dex, and were halted at an age of 370 Myr. The non-magnetic model we know has a radius smaller than the actual radius of YY Gem (here considered a single star). The two magnetic models, on the other hand, were calculated with the surface magnetic field strength required to reconcile the model radius with the observational radius (4.3 kG and 0.7 kG for the LS95 prescription and turbulent dynamo, respectively). The presence of a magnetic field causes the density profile to be steeper deep within the star, although the central density remains nearly constant. However, there is no noticeable difference between the two magnetic models, despite having two vastly different magnetic field strengths throughout the star. Given that the density profile is steeper, the density is considered more centrally condensed, meaning the interior structure constants should be smaller in the magnetic case than in the non-magnetic case. This is confirmed by the curves of η(R) in Figure 6.19, which are larger throughout the magnetic models. This leads to a 226 80 Non-magnetic LS95 prescription 2 60 η2(R) Density, ρ(R) (g cm-3) Turbulent dynamo 40 1 M = 0.599 M⊙ 20 [Fe/H] = -0.20 Age = 370 Myr 0 0.0 0.2 0.4 0.6 0.8 0 1.0 Fractional Radius (R/R*) Figure 6.19: Interior density profile for a M = 0.599M⊙ model with and without the presence of a magnetic field. Also shown is the η2 parameter, related to the stellar deviation from sphericity and the interior structure constant, k 2 . The magnetic models were computed with a surface magnetic field strength strong enough to reconcile the model radius with the observed radius of YY Gem (4.3 kG and 0.7 kG for the LS95 prescription and turbulent dynamo approach, respectively). Note that the two lines for the magnetic models directly on top of one another. lower value of k 2 because k j is inversely proportional to η j at constant j , ¯ ∂k j ¯ j + 2η j ¯ = −¡ ¢2 , ¯ ∂η j j j +ηj (6.13) where we have differentiated Equation (3.9). We confirmed that the value of k 2 in the nonmagnetic case was k 2 = 0.060 while the magnetic models had k 2 = 0.045 and k 2 = 0.046 for the LS95 and turbulent dynamo models, respectively. The radius to the boundary of the convection zone was also equivalent in both magnetic models. In the non-magnetic model, the convection zone boundary was located at R/R ∗ = 0.671, but it receded to R/R ∗ = 0.691 and R/R ∗ = 0.690 for the magnetic models. Thus, 227 the only way of differentiating between the two proposed methods using stellar surface properties is to evaluate the strength of the surface magnetic fields. 6.5.2 Comparison to Previous Work The magnetic perturbation presented in Section 6.4.4 is qualitatively different from earlier approaches (Chapter 5; Mullan & MacDonald 2001; MacDonald & Mullan 2012; Feiden & Chaboyer 2012b). These earlier approaches assess the ability of a magnetic field to stabilize a fluid against convection. Qualitatively our approach in Chapter 5 is very similar to the method favored by Mullan & MacDonald (2001). In fact, the variable ν appearing in our modified MLT equations is essentially the Mullan & MacDonald (2001) convective inhibition parameter, δMM . We have ∂ ln ρ ν≡− ∂ ln χ µ ¶ P,T = P mag P mag + P gas (6.14) whereas δMM was defined in Section 1.4.1 as δMM = B2 . B 2 + 4πγP gas (6.15) One significant difference is that our approach also introduces the magnetic energy gradient, ∇χ , defined in Equation (5.94). This gradient reduces the effect of ν on the adiabatic gradient by an order of magnitude. We therefore expect that magnetic fields used by Mullan & MacDonald (2001) to be about an order of magnitude smaller than ours. The tactic used in Section 6.4.4 has attempted to quantify a reduction in convective efficiency. Chabrier et al. (2007) had previously explored this concept, although they did so by reducing the convective mixing length via a reduction in αMLT . Note that this is conceptually different than the methods discussed above where we were concerned with the stabilization of convection. Convective efficiency can be defined through the framework of MLT by considering heat 228 losses that are “horizontal” to the radial motion of a bubble (Böhm-Vitense 1958; Weiss et al. 2004). Quantitatively, the efficiency can be written as Γ= κg δ1/2 ρ 5/2 ℓ2m cP (∇s − ∇e )1/2 p 1/2 T 3 P 12 2ac (6.16) where the individual variables were defined in Chapter 5. We see that the temperature excess, (∇s − ∇e ), factors into the efficiency, as well as the convective mixing length, ℓm . Rewriting the above equation in terms of the convective velocity and mixing length parameter αMLT , we have Γ= c P κρ 2 u conv αMLT HP . 6ac T3 (6.17) Using this expression, we can attempt to compare reductions in uconv with the reductions in αMLT used by Chabrier et al. (2007). Consider a reduction in uconv caused by a magnetic field that has a strength that is some fraction, Λ, of the equipartition field strength. Then the ratio of the convective efficiency in the presence of a magnetic field to the non-magnetic efficiency is Γmag /Γ0 = (1 − Λ2 )1/2 . Therefore, to achieve the same results by reducing αMLT , we need only multiply the solar value by (1 − Λ2 )1/2 . We can now relate αMLT to Λ, Λ= s µ αMLT 1− αMLT, ⊙ ¶2 . (6.18) The drastically reduced αMLT values (αMLT = 0.1, 0.5) used by Chabrier et al. (2007) correspond to Λ ∼ 0.999 and Λ ∼ 0.966, respectively. We find that our models require such strong magnetic fields, at least in the outer layers. The model of YY Gem in Section 6.4.4 has Λ ∼ 1 throughout a large portion of the envelope calculation. Λ decreases from 1 at the photosphere (where T = Teff ) to 0.3 at the point where we match the envelope to the interior. The comparison is by no means clear given that we have prescribed a radial profile that is independent of Λ. 229 80 Non-magnetic 2 60 η2(R) Density, ρ(R) (g cm-3) Turbulent dynamo Reduced αMLT 40 1 M = 0.599 M⊙ 20 [Fe/H] = -0.20 Age = 370 Myr 0 0.0 0.2 0.4 0.6 0.8 0 1.0 Fractional Radius (R/R*) Figure 6.20: Interior density profile for a M = 0.599M⊙ model with and without the presence of a magnetic field. Also shown is the η2 parameter, related to the stellar deviation from sphericity and the interior structure constant, k 2 . The magnetic field was prescribed using the turbulent dynamo formuation and the reduced αMLT method (Chabrier et al. 2007). In the turbulent dynamo model, a 0.7 kG field was used and αMLT = 0.6 was used in the reduced αMLT model. Note that the magnetic model and reduced αMLT model profiles lie on top of one another. To make the comparison more direct, we compare the interior structure of the magnetic model with that from a reduced αMLT model. Our models required αMLT = 0.60 to reproduce the observed radius of YY Gem, corresponding to Λ = 0.951 for reference. Based on Figure 1(a) of Chabrier et al. (2007) it appears that they require αMLT ∼ 0.4 to reproduce the properties of YY Gem. When we look at the density distribution associated with our reduced αMLT model and our magnetic model (using the turbulent dynamo formulation) we find there is no significant difference. This is illustrated in Figure 6.20. From this it is apparent that our turbulent dynamo magnetic models produce results consistent with reduced αMLT models. The advantage of the magnetic models is that they can be tested observationally. Our models 230 predict a specific surface magnetic field strength whereas reduced αMLT models are difficult to constrain. We have introduced the possibility of constraining reduced αMLT models using the equipartition magnetic field strength, Λ. Still, the use of a magnetic field prescription provides a more realistic solution. 6.5.3 Starspots Up to this point we have avoided any mention of specifically incorporating effects due to starspots. Previously magnetic investigations attempted to account for spots by reducing the flux at the model photosphere (Chabrier et al. 2007; Morales et al. 2010; MacDonald & Mullan 2012, 2013). The details of this approach were outlined in Chapter 1. Reductions of photospheric flux were combined with the aforementioned techniques to magnetically suppress convection to reconcile model radii with observations of DEBs. The explicit introduction of spots into stellar models can significantly affect stellar structure (Chabrier et al. 2007). By reducing the flux at the model photosphere the model is forced to inflate. The prescription for incorporating spots, however, has several uncertainties including the spot filling factor, the spot temperature contrast, and the spot distribution. These are not very well constrained for low-mass stars, particularly for M-dwarfs (Jackson & Jeffries 2013). It would therefore be advantageous to eliminate having to specify spot properties in stellar models. In this Chapter, we have seen that our models do not explicitly require starspots. This is because our models effectively use the average surface magnetic field strength, 〈B f 〉, as input. Therefore, strong localization of magnetic flux is averaged together with the typical “quiet” regions on the stellar surface. Since measurements of stellar surface magnetic fields generally cannot distinguish between the sources of magnetic flux (with the exception of Zeeman Doppler Imaging), we only receive information regarding the average surface field strength when attempting to validate our magnetic models (Reiners 2012). 231 Conceptually, starspots reduce the photospheric flux by suppressing convection. Therefore, there is little reason to manually reduce the photospheric flux in the stellar models. A correct treatment of magnetic suppression of convection should implicitly account for this effect. We see that this is the case with our models. Investigations looking at the effects on stellar colors using our models in comparison with those that explicitly include spots (Jackson et al. 2009; Jackson & Jeffries 2013; MacDonald & Mullan 2013) are ongoing. 6.5.4 Implications for Asteroseismology Figure 6.19 showed that magnetic models have a lower density throughout their interior than do non-magnetic models. One result is that the magnetic models have a lower k 2 , indicative of the mass being more centrally condensed. This is not unexpected if we consider that the magnetic field acts to inflate the radius of the model. Related to the reduction in mass density throughout the stellar interior is the effect on the sound speed profile. We plot the sound speed profile for two M = 0.6M⊙ models in Figure 6.21. The models have a scaled-solar composition of [Fe/H] = −0.2 dex. One model has a magnetic field with a surface field strength of 0.7 kG using the turbulent dynamo formulation. Note that the density profile for these two models are shown in Figure 6.19. Throughout the model interior the magnetic model has a lower sound speed. Comparing the sound speed between the two models, we see that the magnetic model has a sound speed that is slower by about 5%. The bottom panel in Figure 6.21 shows the relative sound speed difference between the two models with respect to the non-magnetic models. We define δc s c s, mag − c s, 0 = , cs c s, 0 (6.19) where c s, 0 is the sound speed in the non-magnetic model and c s, mag is that for the magnetic 232 0.5 cs (Mm s-1) 0.4 0.3 0.2 0.1 0.0 Non-magnetic Bsurf = 0.7 kG δcs/cs 0.00 -0.04 -0.08 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 Fractional Radius, (R/R*) 0.8 0.9 1.0 Figure 6.21: (Top) Sound speed profile for a M = 0.599M⊙ model with and without the presence of a magnetic field. (Bottom) Relative difference in the sound speed profile between the two models with respect to the non-magnetic model. The magnetic model was computed with a surface magnetic field strength strong enough to reconcile the model radius with the observed radius of YY Gem (0.7 kG). model. We defined the sound speed such that c s2 µ ¶ P P ∂ ln P = Γ1 = ρ ρ ∂ ln ρ ad (6.20) where P is the total pressure and ρ is the mass density. Using the total pressure is valid everywhere except in the outermost layers (∼ 1% by radius) because the magnetic pressure is significantly lower (over 4 order of magnitude lower) than the gas pressure. The change in sound speed between the two models affects the p-mode frequencies and frequency spacing, which consequently alters the interpretation of seismic data. We now explore these effects. Imagine that observational data is obtained for a M = 0.6M⊙ star that has a 0.7 kG magnetic field. These two facts about the star are unknown to the observer. Instead, the observer has 233 taken great care to obtain high resolution spectra with a high signal-to-noise of the star. From this data they are able to ascertain the effective temperature, Teff , and the approximate composition. They also have photometric data from Kepler that allows for a seismic analysis, where to first order, they are able to obtain the stellar radius from the p-mode frequencies. The properties they derive for the star are R = 0.619 ± 0.006R ⊙ , Teff = 3 820 ± 100 K, and [Fe/H] = −0.2 ± 0.1.7 Taking these properties in combination with the p-mode frequencies and frequency spacings, what inferences would the observer make about the star? Using the grid of nonmagnetic models presented in Chapter 4 the observer would estimate the mass of the star to be between 0.65M⊙ and 0.69M⊙ , or M = 0.67 ± 0.02M⊙ . The age of the star would be difficult in this particular case due to the lack of significant radius evolution of low-mass stars on the main sequence. One concession, is that the models would not match the observed Teff within the 1σ limit quoted above. However, one can easily find a match within 2σ. Moreover, the sound speed profiles between the higher mass non-magnetic models and the lower mass magnetic models are quite similar With magnetic models, a slight degeneracy in mass and magnetic field strength would be introduced, as varying the magnetic field at a given mass can alter the radius and Teff . However, the true solution would be encompassed by the formal uncertainties. Compare this with the case where non-magnetic models are used, where the mass estimate is too large by ∼10% and the true solution lies 3.5σ below the derived value. Future investigations exploring the use of magnetic models for asteroseismic analyses are warranted. While the case above was exploratory and does not represent the typical stars selected for asteroseismic analysis, the conclusions should be valid for all stars with an outer convective envelope. Magnetically active stars are also not favored for asteroseismology, due to issue with removing possible spots variations in the light curve. However, the 7 Conveniently, these are the properties of YY Gem. 234 lack of photometric variability due to spots is not necessarily indicative of an inactive star (Jackson & Jeffries 2013). 6.5.5 Exoplanets & the Circumstellar Habitable Zone The primary effect of introducing a magnetic perturbation into a model is that the model radius is increased and the Teff is decreased. What has not been discussed is that the luminosity of the model shows a marginal decrease of about 10%. This means the radius increase does not precisely balance the Teff decrease, in contrast with what has been assumed previously (see, e.g., Morales et al. 2008). This reduction in luminosity occurs as the model readjusts its internal structure to compensate for the reduction of convective heat flux. These changes in the stellar surface properties can have a significant impact on studies of extra-solar planets (exoplanets) around low-mass stars. Transiting exoplanets have radii that are directly proportional to the radii of their host stars (e.g., Seager & Mallén-Ornelas 2003). Determining the radius of a low-mass host star is difficult. Interferometry allows for the direct measurement of the stellar radius, but is only available for the brightest targets (von Braun et al. 2012; Boyajian et al. 2012). One could attempt to perform an asteroseismic analysis, but it would suffer due to the faintness of low-mass stars and also from the considerable variability that is characteristic of these stars. In lieu of other methods, stellar evolution models are used to aid in the interpretation of transiting exoplanetary systems (e.g., Muirhead et al. 2012; Gaidos 2013; Dressing & Charbonneau 2013). Studies of transiting exoplanets have relied on standard stellar evolution models (Muirhead et al. 2012; Gaidos 2013; Dressing & Charbonneau 2013). Using magnetic models can have a significant impact on these studies. For instance, Muirhead et al. (2012) measured the metallicity and Teff for the low-mass stars tagged as planet candidates in the Kepler mission (Batalha et al. 2013). They then interpolated within DSEP isochrones to determine stellar 235 radii. Had they used magnetic stellar models, for a given Teff , the stellar radius would be larger, with the precise factor dependent on the magnetic field strength assumed. This would result in larger, less dense (and therefore less Earth-like) planets. Estimating the surface magnetic field is the largest hindrance to using magnetic models. Stars in the Kepler field are typically distant, meaning X-ray measurements will not necessarily be available. Deep observations of the Kepler field with Chandra could provide X-ray magnetic activity information. Alternatively, it might be feasible to obtain spectra and use the Hα equivalent width to estimate magnetic activity. Hα is less revealing, as we are not aware of a relation linking Hα equivalent widths to magnetic field strengths. However, the presence of Hα in emission would be indicative of an active star. At that point, several magnetic model could be used to constrain the stellar radius. Aside from affecting the determination of host star radii, magnetic models may affect the edges of the circumstellar habitable zone (HZ). This can happen over two timescales: (1) evolutionary timescales, and (2) timescales on the order of a stellar cycle. The first timescale refers to how the luminosity of a star changes over its lifetime. We said that magnetic fields can reduce the total luminosity of a given stellar model. Therefore, the luminosity evolution of a particular star will be altered by the evolution of the magnetic field. The second timescale refers to stellar activity cycle, much like the 11 yr solar cycle. Changes in the magnetic field strength over the cycle may affect the host star’s internal structure in such a way so as to alter the boundaries of the HZ. Regular fluctuations in the stellar radius, effective temperature, and luminosity could cause corresponding fluctuations in the boundaries of the HZ. Here, we explore the second possibility. Investigations into the first affect requires more sophisticated magnetic models that have evolving magnetic fields. We compute three models for a M = 0.6M⊙ star. One model has no magnetic field, one has a 0.5 kG magnetic field, and the final model has a 0.7 kG magnetic field. Both of the magnetic models were calculated 236 Figure 6.22: Habitable zone boundaries for M = 0.6M⊙ models with no magnetic field, a 0.5 kG magnetic field, and a 0.7 kG magnetic field. The Teff and luminosity used to compute the HZ were calculated at 1 Gyr. The order of the inner HZ edges correspond to the 0.7 kG, 0.5 kG, and non-magnetic model in order of increasing distance from the star. The outer HZ edges are in the same order. Transparency was used in the coloring, meaning the green shell is a combination of the colors used for each HZ. assuming a turbulent dynamo. We used the Habitable Zone Calculator8 based on the most recent HZ models by Kopparapu et al. (2013) to estimate the HZ boundaries for each of our models. The Kopparapu et al. (2013) HZ models improve on the Kasting one-dimensional radiative-convective climate models (Kasting 1988, 1991; Kasting et al. 1993) by including updated H2 O and CO2 opacities. We find that the HZ boundaries are robust. This is demonstrated in Figure 6.22 where we show the estimated habitable zone boundaries for each of our models. We adopted the conservative limits defined by Kopparapu et al. (2013), which place the inner edge of the HZ at the location where their climate models predict a fully saturated troposphere and 8 http://depts.washington.edu/naivpl/sites/default/files/HZ_Calc.html 237 the outer edge of the HZ is where the maximum greenhouse warming can occur for a CO2 dominated atmosphere. In Figure 6.22, the large green region is considered habitable for all three models. Fluctuations of the HZ associated with possible stellar activity cycle are under 0.1 AU. Even for the extreme case of a fluctuation between no magnetic field and 0.7 kG the HZ outer edge changes by 0.14 AU. The reason for the stability is that, while the stellar effective temperature is lower by nearly 300 K for the 0.7 kG model compared to the non-magnetic model, the luminosity decreases by ∼ 0.06L ⊙ (or about 10%). Therefore, only planets very near the to the HZ boundaries will be affected by the modest variation in luminosity. Note that our approach considered very dramatic changes in the stellar properties throughout the magnetic activity cycle. The sizable changes in the model properties between the different magnetic field strength occur over thermal timescales instead of short timescales associated with stellar cycles. Fluctuations in the global stellar properties due to magnetic fields on the timescales of a stellar cycle are expected to be less than those observed between two models with different magnetic field strengths. In fact, helioseismic studies have indicated that the Sun does not appear to undergo significant structural changes throughout its activity cycle (Libbrecht & Woodard 1990; Foukal et al. 2006). Acoustic frequencies do change with the solar cycle, but have their origin in the outermost layers of the Sun (Libbrecht & Woodard 1990). There is evidence that the total solar irradiance (TSI) does change over the course of the solar cycle. TSI is the amount of solar radiation received by the Earth. We discussed this previously in Section 1.3.4, that the TSI increases up to 0.1% with increasing sun spot number (Foukal et al. 2006). Whether planets around other stars would observe similar fluctuations in total irradiance is unclear. Addressing this question requires a better understanding of the development of spots and faculae on active stars. 238 6.6 Conclusions In this Chapter we addressed the question of how the presence of a magnetic field affects the structure of a low-mass star with a radiative core. We approached this problem by taking a careful look at three DEB systems that show significant radius and Teff deviations from standard stellar evolution models. Using the magnetic stellar evolution models introduced in Chapter 5 (Feiden & Chaboyer 2012b), we attempted to reproduce the observational properties of UV Psc, YY Gem, and CU Cnc. The models probe the ability of a magnetic field to stabilize plasma against convection. After finding that the magnetic models were able to reconcile the stellar models with the observed radii and Teff s (consistent with the models of Mullan & MacDonald 2001), it was show in Section 6.3.1 that the surface magnetic field strength were likely too strong. This was done by taking the coronal X-ray luminosity as an indirect diagnostic for the surface magnetic field strength (Fisher et al. 1998; Pevtsov et al. 2003). In contrast to the surface magnetic fields, the interior field strengths are of a plausible magnitude, consistent with the range of field strengths predicted within the Sun by 3D MHD models (Brandenburg & Subramanian 2005). We then attempted to reduce the surface magnetic field strengths while still maintaining the newfound agreement between stellar models and DEB observations. The most plausible explanation we uncovered is that kinetic energy in turbulent convective flows is actively converted into magnetic energy. Introducing changes to mixing length theory that mimic this physical process yielded accurate models with surface magnetic field strengths nearly equal to those predicted by X-ray emission. We were unable, however, to implement this prescription for the stars of UV Psc. This is likely a consequence of the chosen magnetic field profile, which caused the models to fail to converge. We note that by invoking the “turbulent dynamo” approach, we are able to generate substantial radius inflation with relatively modest (sub-kG) magnetic fields. This could be a clue to solving the cases of extreme 239 radius inflation, as observed with V405 Andromeda, which appears between 70% and 80% larger than standard stellar evolution models (Vida et al. 2009; Ribeiro et al. 2011). Beyond direct comparisons to observational data, we explored additional implications of the present study. We found that different theoretical descriptions leading to similarly inflated stars produce nearly identical stellar interiors. The radial density profile within models invoking the methods of Chapter 5, Section 6.4.4, and Chabrier et al. (2007) are all nearly identical. The latter uses a reduced mixing length theory approach to simulate the introduction of a turbulent dynamo. We provide a variable Λ to translate a reduction in mixing length into a magnetic field strength (normalized to the equipartition field strength, which depends on the convective velocity). We also argue that asteroseismic studies and planetary habitability studies should be relatively unaffected by structural changes in the stellar models. First, asteroseismic studies typically known the stellar composition and effective temperature. Anomalous values of the Teff due to magnetic fields should provide a clue that magnetic fields are required. For a given density profile, it is difficult to reproduce the Teff for a star of a given mass without a magnetic field. If Teff is ignored, then masses would typically be overestimated by about 10%. Second, studies concerning the size of planetary habitable zones are rather insensitive to the effects of magnetic fields on stellar structure. Stellar effective temperature are decreased, but the stellar radius is increased, leading to only a modest decrease in the stellar luminosity in the presence of a magnetic field. We used the habitable zone models of (Kopparapu et al. 2013) to illustrate this fact. Even if a star were to have a magnetic cycle that fluctuates between no magnetic field and a sizable field of several hundred Gauss, only the innermost and outermost edges of the habitable zone are affected. Increased magnetic field strength would, however, increase the level of surface activity, which presents separate problems for planetary habitability around low-mass stars. Since the habitable zones are relatively near 240 to low-mass stars, sizable flares and coronal mass ejections could devastate a nearby planet. While planetary habitable zones may be relatively stable to changes in stellar structure due to a magnetic field, the radii of transiting exoplanets are not. The radii of transiting exoplanetary radii scale with the stellar radius. This is but one reason that further exploration of the role of magnetic field in low-mass stars is required. Interpretation of low-mass star observations depend critically on the properties of model predictions. 241 Table 6.5 Low-mass stars from Reiners (2012) with direct magnetic field measurements used to derive our Φ − L x relation. Star Other Name Name HD 115383 59 Vir HD 115617 SpT 242 〈B f 〉 R⋆ log(Φ) π X cr (kG) (R ⊙ ) (Mx) (mas) (cnts s−1 ) G0 0.5 0.74 25.22 56.95 1.12 61 Vir G6 0.1 0.95 24.74 116.89 σ Dra - K0 0.1 0.80 24.59 40 Eri - K1 0.1 0.75 ǫ Eri - K2 0.1 LQ Hya - K2 GJ 566 B ξ Boo B Gl 171.2 A HR Fx log(L x ) (erg s−1 cm−2 ) (erg s−1 ) -0.14 8.48E-12 29.50 0.01 -0.96 4.58E-14 26.60 173.77 0.26 -0.80 1.06E-12 27.62 24.53 200.62 0.80 -0.28 5.46E-12 28.21 0.70 24.59 310.94 2.82 -0.44 1.69E-11 28.32 2.5 0.70 25.86 53.70 2.73 -0.04 2.21E-11 29.96 K4 0.5 0.70 25.14 149.26 2.44 -0.31 1.63E-11 28.94 - K5 1.4 0.65 25.56 55.66 2.69 -0.04 2.18E-11 29.93 Gl 182 - M0.0 2.5 0.60 25.74 38.64 0.65 -0.19 4.75E-12 29.58 Gl 803 AU Mic M1.0 2.3 0.50 25.54 100.91 5.95 -0.07 4.72E-11 29.74 Gl 569 A - M2.0 1.8 0.40 25.24 - 0.49 -0.40 3.03E-12 - Gl 494 DT Vir M2.0 1.5 0.40 25.16 85.54 1.57 -0.01 1.30E-11 29.33 Gl 70 - M2.0 0.2 0.33 24.12 87.62 0.04 -0.67 2.08E-13 27.51 Gl 873 EV Lac M3.5 3.8 0.31 25.35 195.22 5.83 -0.16 4.35E-11 29.14 continued on next page Table 6.5 – continued Star Other Name Name Gl 729 V1216 Sgr Gl 87 SpT 243 〈B f 〉 R⋆ log(Φ) π X cr (kG) (R ⊙ ) (Mx) (mas) (cnts s−1 ) M3.5 2.1 0.20 24.71 336.72 0.94 - M3.5 3.9 0.30 25.33 96.02 Gl 388 AD Leo M3.5 3.0 0.39 25.44 GJ 3379 - M3.5 2.3 0.25 GJ 2069 B CV Cnc M4.0 2.7 Gl 876 IL Aqr M4.0 GJ 1005 A - Gl 490 B HR Fx log(L x ) (erg s−1 cm−2 ) (erg s−1 ) -0.43 5.67E-12 27.78 - - - - 213.00 3.70 -0.27 2.55E-11 28.83 24.94 190.93 0.40 -0.20 2.90E-12 27.98 0.25 25.01 78.10 0.24 -0.14 1.82E-12 27.95 0.2 0.31 24.07 213.28 - - - - M4.0 0.2 0.23 23.81 191.86 - - - - G 164-31 M4.0 3.2 0.20 24.89 50.00 0.84 -0.22 6.00E-12 29.46 Gl 493.1 FN Vir M4.5 2.1 0.20 24.71 123.10 0.14 -0.16 1.04E-12 27.92 GJ 4053 LHS 3376 M4.5 2.0 0.17 24.55 137.30 0.06 -0.49 3.43E-13 27.34 GJ 299 - M4.5 0.5 0.18 23.99 148.00 - - - - GJ 1227 - M4.5 0.2 0.19 23.64 120.00 - - - - GJ 1224 - M4.5 2.7 0.18 24.73 132.60 0.21 -0.45 1.24E-12 27.93 Gl 285 YZ Cmi M4.5 4.5 0.30 25.39 167.88 1.47 -0.21 1.06E-11 28.65 GJ 1154 A - M5.0 2.1 0.20 24.71 - 0.10 -0.23 7.09E-13 - continued on next page Table 6.5 – continued Star Other Name Name GJ 1156 GL Vir Gl 905 SpT 244 〈B f 〉 R⋆ log(Φ) π X cr (kG) (R ⊙ ) (Mx) (mas) (cnts s−1 ) M5.0 2.1 0.16 24.51 152.90 0.13 HH And M5.5 0.1 0.17 23.24 316.70 GJ 1057 CD Cet M5.5 0.1 0.18 23.29 GJ 1245 B - M5.5 1.7 0.14 GJ 1286 - M5.5 0.4 GJ 1002 - M5.5 Gl 406 CN Leo Gl 412 B HR Fx log(L x ) (erg s−1 cm−2 ) (erg s−1 ) -0.25 9.08E-13 27.67 0.18 0.15 1.64E-12 27.29 117.10 - - - - 24.31 220.00 0.20 -0.37 1.27E-12 27.50 0.14 23.68 138.30 - - - - 0.2 0.13 23.31 213.00 - - - - M5.5 2.4 0.13 24.39 418.30 0.23 -0.22 1.64E-12 27.05 WX Uma M6.0 3.9 0.13 24.60 206.94 0.18 -0.64 8.85E-13 27.39 GJ 1111 DX Cnc M6.0 1.7 0.12 24.17 275.80 - - - - Gl 644C VB 8 M7.0 2.3 0.10 24.15 153.96 - - - - GJ 3877 LHS 3003 M7.0 1.5 0.10 23.96 157.80 - - - - GJ 3622 - M7.0 0.6 0.10 23.56 221.00 - - - - LHS 2645 - M7.5 2.1 0.08 23.91 - - - - - LP 412-31 - M8.0 3.9 0.08 24.18 - - - - - VB 10 V1298 Aql M8.0 1.3 0.08 23.70 164.30 - - - - continued on next page Table 6.5 – continued Star Other Name Name LHS 2924 - LHS 2065 - SpT 〈B f 〉 R⋆ log(Φ) π X cr (kG) (R ⊙ ) (Mx) (mas) (cnts s−1 ) M9.0 1.6 0.08 23.79 90.00 - M9.0 3.9 0.08 24.18 116.80 - HR Fx log(L x ) (erg s−1 cm−2 ) (erg s−1 ) - - - - - - 245 Chapter 7 Magnetic Models of Low-Mass, Fully Convective Stars 7.1 Introduction The radial extent of the outer convection zone in low-mass stars increases as stellar mass decreases. Around 0.35M⊙ , the interior of low-mass stars transitions from being partially convective to fully convective (Limber 1958a; Baraffe et al. 1998). Fully convective stars should be the simplest stars to describe from a theoretical perspective. The interior structure predicted by stellar models is largely independent of many of the input variables (e.g., αMLT ) and input physics (e.g., diffusion physics) (Baraffe et al. 1998; Dotter et al. 2007). We saw in Chapters 2, 3, and 4 that metallicity, αMLT , and age can alter the predictions of stellar models, but only at the few percent level. Observations of significant radius inflation in fully convective stars is therefore puzzling. Multiple DEBs are known to host fully convective stars. Those with precise mass and radius determinations are Kepler-38 (Orosz et al. 2012), Kepler-16 (Doyle et al. 2011; Winn et al. 2011; Bender et al. 2012), LSPM J1112+7626 (Irwin et al. 2011), KOI-126 (Carter et al. 2011), 246 and CM Draconis (CM Dra; Lacy 1977; Metcalfe et al. 1996; Morales et al. 2009a). Of these systems, the only fully convective stars to be accurately reproduced by stellar evolution models are KOI-126 B and C (Chapter 2; Feiden et al. 2011; Spada & Demarque 2012). The rest display varying degrees of radius inflation compared to model predictions. Most significant are the inflated radii of the stars in CM Dra. Historically, the stars of CM Dra are the fully convective stars against which to benchmark stellar models. The discovery of KOI-126 in 2011 introduced the first pair of well-characterized fully convective stars in a DEB since Lacy (1977). Over the years, the discrepancy between the model radii of CM Dra and those determined from observations has grown. Initial modeling efforts were optimistic (Chabrier & Baraffe 1995), but this quickly changed with the introduction of more sophisticated models (Baraffe et al. 1998) and more precise observations (Metcalfe et al. 1996; Morales et al. 2009a). Recent metallicity estimates have widened the gap, yet again (Rojas-Ayala et al. 2012; Terrien et al. 2012). This chapter is concerned with the second half of our magnetic analysis of the low-mass stellar mass-radius problem: fully convective stars. Chapter 6 dealt with using the magnetic models introduced in Chapter 5 to reconcile stellar evolution model predictions with observations of DEB stars that have a radiative core. We showed that the magnetic models presented in Chapter 5 were likely incomplete and required the addition of terms in the mixing length equations to account for the conversion of turbulent kinetic energy into magnetic energy. A brief overview of the magnetic field formulations is presented in Section 7.2. Two wellcharacterized DEBs, Kepler-16 and CM Draconis, are then used as case studies for investigating the effects of the magnetic models in fully convective stars (Section 7.3). We have elected to focus on Kepler-16 in addition to CM Dra because it has a more massive K-dwarf companion. If there is any significant difference between a system with two fully convective stars and a system with only one, we hope that it will become apparent. The validity of 247 our model results are discussed and we provide comparisons to previous investigations in Section 7.4. After reasoning that magnetic models are not sufficient to describe the radius inflation observed in fully convective stars, we re-advance an old hypothesis: that metallicity is the origin of the radius discrepancies. 7.2 Magnetic Models The magnetic models to be used in this chapter have been described in the two previous chapters. However, we want to consolodate the terminology and remind the reader about the various formulations after the development of several branches in Chapter 6. Below we describe the three model formulations that will be referenced throughout this chapter. 7.2.1 Dipole Radial Profile Models of the “dipole radial field” variety are the standard sort first introduced in Chapter 5. Given a surface magnetic field strength, the radial profile of the magnetic field is determine by calculating the peak magnetic field strength. Stars with radiative cores have this point defined by the tachocline (Feiden & Chaboyer 2012b), but fully convective stars have the point placed at 15% of the stellar radius. This is loosely based on results from 3D MHD models of fully convective stars, which find the strongest magnetic field at 0.15R ⋆ (Browning 2008). The rest of the field is then automatically defined because we assume the magnetic field strength falls off as r 3 towards the core and surface of the star. 7.2.2 Gaussian Radial Profile Looking to increase the peak magnetic field strength for a given surface magnetic field strength, we used a Gaussian profile in Section 6.4.1. The peak magnetic field strength 248 was still defined at the tachocline in partially convective stars and at R = 0.15R ⋆ in fully convective stars. However, instead of a power-law decline of the interior magnetic field strength from the peak, the peak was set as the peak of a Gaussian distribution. The width of the Gaussian was made variable, based on the width of the convection zone. Larger convection zones were given broarder Gaussians compared to those with thin convective envelopes (see Section 6.4.1). These models will be referred to as “Gaussian radial profile” models throughout. 7.2.3 Constant Λ Turbulent Dynamo Section 6.4.4 introduced a third method of including the effects of a magnetic field, the so-called “turbulent dynamo” formulation. Conceptually, the turbulent dynamo formalism seeks to source the energy for the magnetic field from turbulent convection. This is different from our approach used in the dipole and Gaussian magnetic field models, which investigate the stabilization of convection more so than suppressing convective efficiency. For completeness, we wanted to compute models of fully convective stars with this turbulent dynamo formulation. However, combining this formulation with a dipole or Gaussian field strength profile causes the models to not converge. The interior magnetic field strengths are too large. The Alfvén velocity, u 2A , quickly dominates the convective velocity, uconv, 0 , term in Equation (6.9), meaning there is more magnetic energy than available kinetic energy from convection. To mitigate this issue, we created a new radial profile based on the equipartition field strength parameter, Λ, introduced in Section 6.5.2. We defined B = ΛB eq , (7.1) where B is the magnetic field strength at a given point in the model and B eq is the equipar- 249 tition magnetic field strength. The latter is defined such that B eq = q 2 4πρu conv , (7.2) where ρ is the plasma mass density and uconv is the convective velocity. With these definitions, we reformulated the magnetic mixing length theory such that u conv = u conv, 0 p 1 − Λ2 . (7.3) Additionally, the term associated with the increase of the plasma temperature gradient, ∇s , became ∆∇s = 2 (Λu conv 0 )2 u A = , C C (7.4) as before, where C is defined in Chapter 5. The radial profile was then prescribed by assigning a constant value to Λ, hereafter the “constant Λ” formulation. This has the added benefit of mimicking an α2 dynamo, where the magnetic field is generated throughout the convection zone (Chabrier & Küker 2006; Browning 2008). 7.3 Analysis of Individual DEB Systems 7.3.1 Kepler-16 The report of the first circumbinary exoplanet was announced by Doyle et al. (2011) with their discovery of Kepler-16b. While the planet is interesting in its own right, what made the discovery even more remarkable was the fact that the two host stars comprise a lowmass detached eclipsing binary with a period of just over 41 days. This fact allowed for a very precise determination of the component masses and radii using the same photometricdynamical analysis that led to the characterization of the hierarchical triple KOI-126 (Chap- 250 Table 7.1 Fundamental properties of Kepler-16. Property Kepler-16 A Kepler-16 B D11 Mass (M⊙ ) 0.6897 ± 0.0034 0.20255 ± 0.00066 Radius (R⊙ ) 0.6489 ± 0.0013 0.2262 ± 0.0005 B12 Mass (M⊙ ) Teff (K) [Fe/H] 0.654 ± 0.017 4337 ± 80 Age (Gyr) 0.1959 ± 0.0031 −0.04 ± 0.08 ··· 3±1 ter 2; Carter et al. 2011). Kepler-16 was found to contain a K-dwarf primary with a fullyconvective M-dwarf secondary. Individual stellar parameters are presented in Table 7.1. Following the discovery paper, Winn et al. (2011) provided a spectroscopic determination of the primary star’s composition and a preliminary estimate of the star’s age. The age estimation was obtained using gyrochronology and a Ca ii age-activity relation. Winn et al. (2011) estimate that the system is 3±1 Gyr old with a near-solar metallicity of [Fe/H] = −0.04 ± 0.08. Their study was also the first to directly compare the Kepler-16 components to stellar evolution models in the context of the mass-radius problem. The authors found the Baraffe et al. (1998) model predictions to be in agreement with the properties of primary. However, the fully-convective secondary was seen to be inflated by a few percent, relative to the models. This assessment was echoed in an independent study by Feiden & Chaboyer (2012a, Chapter 4). Less than a year after the initial discovery, radial velocity confirmation of the component masses—within 2σ—was obtained (Bender et al. 2012). The results from the Bender et al. (2012) study display slight disagreements with the original masses (Doyle et al. 2011). Of particular note, is that the precise mass ratio is different between the two studies. Bender et al. (2012) attempted to pin-point the origin of the discrepancy, but were unable to do so with complete confidence. 251 0.72 [Fe/H] = -0.12 0.72 Kepler-16 A [Fe/H] = -0.12 [Fe/H] = -0.04 [Fe/H] = -0.04 [Fe/H] = 0.04 0.70 Radius (R⊙) Radius (R⊙) 0.70 0.68 0.66 0.64 Doyle et al. (2011) 0.62 Bender et al. (2012) 1 [Fe/H] = 0.04 0.68 Bender et al. (2012) 0.66 0.64 Doyle et al. (2011) 0.62 Kepler-16 A 4800 10 Age (Gyr) 4600 4400 4200 Teff (K) 4000 3800 Figure 7.1: Standard DSEP models computed at the exact masses measured by Doyle et al. (2011) (maroon) and Bender et al. (2012) (light blue) for Kepler-16 A. Mass tracks for the adopted metallicity of Winn et al. (2011) and the two boundaries of the associated 1σ uncertainty are given by the solid, dash, and dash-dotted lines, respectively. (a) Age-radius diagram with the observed radius indicated by the gray horizontal swath. (b) Teff -radius plane where the gray box indicates observational constraints for Kepler-16 A. The agreement displayed between the original photometric-dynamical masses (Doyle et al. 2011) and the radial velocity masses (Bender et al. 2012) at first seems comforting, they largely agree. However, the difference in mass for each component drastically alters the comparison with stellar evolution models, assuming that the derived masses do not heavily influence the derivation of the component radii and effective temperatures. The two mass estimates will come to play a key role in our analysis. Hereafter, the masses originally quoted by Doyle et al. (2011) will be referred to as the D11 masses, whereas the revised values of Bender et al. (2012) will be referred to as the B12 masses. Table 7.1 provides a comparison of the different quantities. We focus our attention on Kepler-16 A first, as the analysis will directly influence our interpretation of the secondary. Plotted in Figure 7.1 are standard DSEP models computed at the exact masses of the primary provided by D11 and B12. Three different tracks are illustrated for each mass, corresponding to [Fe/H] = −0.12, −0.04, and +0.04 (Winn et al. 2011). Similar to the figures in Chapter 6, the horizontal swath designates the measured 252 radius constraints. Given the primary mass from D11, we find that standard stellar evolution models match the stellar radius and temperature with high accuracy at an age of 2.8 Gyr. When the model radius equals the precise observed radius (0.6489R⊙ ), the associated model effective temperature is 4337 K, the precise measured temperature (Winn et al. 2011). Additionally, the age predicted by the mass track is consistent with the estimated ages (Winn et al. 2011). The agreement between the models and observations does not guarantee the validity of the D11 masses over the B12 masses. Agreement may be expected. The effective temperature and metallicity for the primary were measured using Spectroscopy Made Easy (Valenti & Piskunov 1996, hereafter SME), which relies on theoretical stellar atmospheres. DSEP also relies on theoretical atmospheres, although the phoenix model atmospheres used by DSEP use a different line list database (see Hauschildt et al. 1999a, and references therein) than the theoretical atmospheres used by SME (VALD: Vienna Atomic Line Database Piskunov et al. 1995; Stempels et al. 2001). In a sense, the fact that we find the precise effective temperature for a given log g and metallicity (i.e., those of Kepler-16 A) may be a better test of the agreement between different stellar atmosphere models rather than a test of the interior evolution models. What is encouraging, is that we derive the appropriate stellar properties at an age consistent with the Winn et al. (2011) study. The age consistency is not ensured by the agreement between the stellar atmosphere models. The previous discussion may be erroneous if we are adopting a mass that is too large compared to the actual mass of the primary. B12 suggest this is the case. The effect of adopting the lower B12 masses is displayed in Figure 7.1. Since the radius measurement has remained constant, we derive an age of 12 Gyr for the primary star. The effective temperature associated with the lower mass model also becomes too cool compared to the observations at the measured metallicity (−0.04 dex). Relief is found by lowering the metallicity by 0.1 dex, which increases the temperature by 30 K. This is enough to bring the model temperature to 253 within 1σ of the measured value Winn et al. (2011). There is on caveat with the above discussion: the spectroscopic analysis by Winn et al. (2011) relied on fixing the stellar log g as input into SME. Thus, the temperature and metallicity are intimately tied to the adopted log g . Reducing the mass of the primary by 5% (B12) but leaving the radius fixed to the D12 value leads to a decrease in log g of 0.02 dex. Such a change in the fixed value of log g has the potential to decrease the derived effective temperature by as much as 100 K (Stempels, priv. comm.). This implies the temperature could be as cool as 4 230 K. This would bring the observed temperature into agreement with the model derived temperature. Considering the discussion above, we believe it is safe to assume that changing the mass of the DSEP models will not produce any large effective temperature offset at a given radius, predicated on the fact that the temperature is always safely above ∼ 4 000 K. Below this value, theoretical atmosphere predictions begin to degrade. More simply stated, we have no reason to doubt the DSEP predicted temperature for Kepler-16 A, regardless of the adopted mass. The model age of 12 Gyr for the B12 primary mass appears old given the multiple age estimates provided by Winn et al. (2011). Is it possible that the system is actually 12 Gyr old, but appears from rotation and activity to be considerably younger? After all, 12 Gyr is not unreasonably old given the current estimate for the age of the Universe (Planck Collaboration et al. 2013). Consider that the 35.1±1.0 day rotation period of the primary was found by Winn et al. (2011) to be nearly equal to the pseudo-synchronization rotation period, predicted to be 35.6 days (Hut 1981). Winn et al. (2011) rightly point out that intuition suggests tidal effects are unimportant in a binary with a 41 day orbital period. However, subsequent tidal interactions briefly endured when the components are near periastron have the ability to drive the components towards pseudo-synchronous rotation. The potential pseudo-synchronization of the components acts to keep the stars rotating at a faster rate than they would if they were completely isolated from one another. The time scale for this to occur is approximately 3 254 [Fe/H] = -0.12 0.24 Kepler-16 B [Fe/H] = -0.04 Radius (R⊙) [Fe/H] = 0.04 0.23 Doyle et al. (2011) 0.22 0.21 Bender et al. (2012) 1 10 Age (Gyr) Figure 7.2: Identical to Figure 7.1(a), except the mass tracks are computed at the masses measured for Kepler-16 B. Gyr (Zahn 1977; Winn et al. 2011), meaning that the rotation period is not necessarily indicative of the system’s age. The primary will have approximately the same rotation period at 12 Gyr as it will at 3 Gyr. Furthermore, the timescale for orbital circularization is safely estimated to be ∼ 104 Gyr (Zahn 1977; Winn et al. 2011). Tidal evolution calculations are subject to large uncertainties and should approached as an order of magnitude estimate. We are therefore unable to immediately rule out the possibility that Kepler-16 is 12 Gyr. We have so far neglected any remark on the agreement between standard models and Kepler-16 B. This comparison is carried out in Figure 7.2. Unfortunately, no effective temperature estimates have been reported, explaining our neglect of the Teff -radius plane in the figure. As with Figure 7.1, mass tracks are shown for multiple metallicities. The standard models mass tracks for Kepler-16 B are unable to correctly predict the observed radius at a realistic age. The disagreement is relatively independent of the adopted metallicity, which introduces slight (i.e., ∼0.5%) variations in the stellar radius at a given age. Only the D11 mass track with a super-solar metallicity is able to match the observed radius at an age less 255 than 20 Gyr. No evidence is available to support the idea that Kepler-16 B is magnetically active. Still, we look to magnetic fields to reconcile the model predictions with the observations. All possible scenarios relating to the various stellar mass estimates are considered. Explicitly, we compute models for both the D11 and B12 masses and then attempt to fit the observations using the magnetic Dartmouth stellar evolution models. D11 Masses The D11 primary mass implies that the age of the Kepler-16 system is approximately 3 Gyr, as shown in Figure 7.1(a). Recall, this age is reinforced by Winn et al. (2011) who used age-activity and age-rotation relations to estimate an age of 3±1 Gyr. Due to the consistency between the model derived age and the empirically inferred age, we see no reason to introduce a magnetic perturbation into the model of Kepler-16 A. Winn et al. (2011) observe moderately weak chromospheric activity coming from the primary, further supporting our decision. Thus, we seek to reconcile models of Kepler-16 B with D11 masses at 3 Gyr. Magnetic models of Kepler-16 B were computed for a range of surface magnetic field strengths. The perturbation was of the form presented in Chapter 5. A dipole radial profile was used and the perturbation was applied in a single time step at an age of 10 Myr. Mass tracks with a 4.0 kG and 5.0 kG surface magnetic field strength are shown in Figure 7.3. We are unable to produce radius inflation larger than 1%, even for the strongest magnetic field strength of 5.0 kG. The peak field strength for the 5.0 kG model (located at R = 0.15R ∗ ) is approxi- mately 1.5 MG. Discussion about how real such a magnetic field might be is deferred until Section 7.4. For the moment, we are interested in knowing what magnetic field strength is required to reconcile the models with the observations. We next constructed magnetic models using a Gaussian radial profile. The magnetic field 256 was again introduced as a single perturbation. However, the model was perturbed at an age of 1 Gyr to avoid convergence issues that arise with the Gaussian perturbation along the pre-main sequence (pre-MS). It would also have been viable to ramp the magnetic field strength from zero to full strength over a series of time steps. Both options lead to equivalent results along the main sequence (Chapter 5). Figure 7.4 presents mass tracks that include a magnetic perturbation with a Gaussian radial profile. Two values of the surface magnetic field strength were used, 4.0 kG and 5.0 kG, and are shown in the figure. The model with a 5.0 kG surface magnetic field strength reconciles the model radius with the observed radius at an age of 3 Gyr. In contrast to the results from the previous chapter, the dipole and Gaussian radial profiles produce radically different results for a given surface magnetic field strength. This is due to the difference in the peak magnetic field strengths. We will return to this issue in Section 7.4. Lastly, figure 7.5 shows how the influence of the constant Λ profile on a model of Kepler-16 B. We started with a rather larger value of Λ = 0.9999 to gauge the reaction of the stellar models to this formulation. The impact on the radius evolution of a M = 0.202M⊙ star is negligible. Further increasing Λ had no effect on the resulting radius evolution. 257 Radius (R⊙) 0.24 [Fe/H] = -0.04 Kepler-16 B Dipole Profile Doyle et al. (2011) 0.23 0.22 |B|surf = 4.0 kG |B|surf = 5.0 kG 0.21 Non-magnetic 1 10 Age (Gyr) Figure 7.3: Standard DSEP (solid line) and magnetic models (broken lines) of Kepler-16 B with D11 masses. The models were computed with [Fe/H] = −0.04 and a solar calibrated αMLT . The magnetic model was calculated using a dipole radial profile with a 4.0 kG (dotted line) and a 5.0 kG (dash-dotted line) surface magnetic field. The observed radius constraints are shown as a shaded horizontal region and the age constraint is a vertical shaded region. [Fe/H] = -0.04 0.24 Kepler-16 B Doyle et al. (2011) Radius (R⊙) Gaussian Profile 0.23 0.22 |B|surf = 4.0 kG |B|surf = 5.0 kG 0.21 Non-magnetic 1 10 Age (Gyr) Figure 7.4: Standard DSEP (solid line) and magnetic models (broken lines) of Kepler-16 B with D11 masses. The models were computed with [Fe/H] = −0.04 and a solar calibrated αMLT . The magnetic model was calculated using a Gaussian radial profile with a 4.0 kG (dotted line) and a 5.0 kG (dash-dotted line) surface magnetic field. The observed radius constraints are shown as a shaded horizontal region and the age constraint is a vertical shaded region. 258 [Fe/H] = -0.04 Radius (R⊙) 0.24 Kepler-16 B Constant Λ Doyle et al. (2011) 0.23 0.22 Λ = 0.9999 0.21 Non-Magnetic 1 10 Age (Gyr) Figure 7.5: Standard DSEP (solid line) and magnetic model (dotted line) of Kepler-16 B with D11 masses. The models were computed with [Fe/H] = −0.04 and a solar calibrated αMLT . The magnetic model was calculated using a constant Λ = 0.9999 profile. The observed radius constraints are shown as a shaded horizontal region and the age constraint is a vertical shaded region. 0.24 |B|surf = 5.0 kG Kepler-16 B |B|surf = 6.0 kG [Fe/H] = -0.04 Gaussian Profile Non-magnetic Radius (R⊙) Bender et al. (2012) 0.23 0.22 0.21 1 10 Age (Gyr) Figure 7.6: Standard DSEP (solid line) and magnetic models (broken lines) of Kepler-16 B with B12 masses. The models were computed with [Fe/H] = −0.04 and a solar calibrated αMLT . The magnetic model was calculated using a Gaussian radial profile with a 5.0 kG (dotted line) and a 6.0 kG (dash-dotted line) surface magnetic field. The observed radius constraints are shown as a shaded horizontal region and the age constraint is a vertical shaded region. 259 B12 Masses Adopting the B12 masses primarily alters the age derived from stellar models. Instead of 3 Gyr, we infer an age of 12 Gyr from models of Kepler-16 A, as was shown in Figure 7.1(a). The relative radius discrepancy noted between models and Kepler-16 B is increased by approximately 1% over the D11 case when using the B12 masses. This can be attributed to the larger influence of age for a M = 0.65M⊙ star than for a M = 0.20M⊙ star. Magnetic models were computed for Kepler-16 B with the B12 mass estimate using a Gaussian radial profile introduced at an age of 1 Gyr. These models are shown in Figure 7.6. We did not generate models with a dipole radial profile models or with a constant Λ given the lack of radius inflation observed for these magnetic field profiles. We find that a 6.0 kG surface magnetic field model was required, which translates to a nearly 50 MG peak magnetic field strength. The stronger field strength required when using the B12 mass instead of the D11 mass is a consequence of the 1% increase in the radius discrepancy mentioned above. 7.3.2 CM Draconis CM Draconis (GJ 630.1 AC; henceforth CM Dra) contains two low-mass stars below the fully-convective boundary and is arguably one of the most important systems for benchmarking stellar evolution models. Shortly after CM Dra was discovered by Luyten, Eggen & Sandage (1967) uncovered that the star was actually a DEB. The nature of the secondary was subject to debate. It was not clear from observations whether the secondary was a dark, very lowmass companion such that no secondary eclipse occurred or whether the two components were of nearly equal mass. Evidence was tentatively provided in favor CM Dra being a single dMe star with a dark companion (Martins 1975), although more observations were encouraged as the author found a possible hint of a secondary eclipse. Any speculation that the secondary companion to the dMe star of CM Dra was a dark, lower-mass object was 260 laid to rest by Lacy (1977) who obtained sufficient data to provide the first determination of stellar parameters for both stars. Following Lacy (1977)’s determination of the stellar properties, subsequent studies refined and improved the masses and radii of the CM Dra stars (Metcalfe et al. 1996; Morales et al. 2009a). The currently accepted values (Morales et al. 2009a; Torres et al. 2010) are presented in Table 7.2. Additional information about the CM Dra stars has been revealed in recent years. Morales et al. (2009a) provided a detailed analysis of a nearby white dwarf (WD) common proper motion companion and estimated the age of the WD to be 4.1 ± 0.8 Gyr. Assuming the WD and CM Dra share a common origin, we can infer that the age of CM Dra is comparable. Deriving the metallicity of CM Dra has proven more difficult than estimating its age. Nearinfrared (NIR) spectroscopic studies that fit theoretical model atmospheres to atomic and molecular features have consistently favored the interpretation that the system is metalpoor ([M/H] ≈ −0.6; Viti et al. 1997, 2002; Kuznetsov et al. 2012). Optical spectroscopy of molecular features (CaH & TiO; Gizis 1997) and NIR photometric colors (Leggett et al. 1998), on the other hand, have suggested that the system might have a near-solar metallicity. More recent techniques relying on empirically-calibrated narrow-band NIR (H & K band) spectral features have started to converge on a value of [M/H] = −0.3 ± 0.1 (Rojas-Ayala et al. 2012; Terrien et al. 2012). Further support is provided by the photometric color-magnitudeTable 7.2 Fundamental properties of CM Draconis. Property Mass (M⊙ ) Radius (R⊙ ) Teff (K) [Fe/H] CM Dra A CM Dra B 0.23102 ± 0.00089 0.21409 ± 0.00083 3130 ± 70 3120 ± 70 0.2534 ± 0.0019 Age (Gyr) 0.2398 ± 0.0018 −0.30 ± 0.12 4.1 ± 0.8 261 0.30 [Fe/H] = -0.18 [Fe/H] = -0.30 Radius (R⊙) 0.28 [Fe/H] = -0.42 0.26 CM Dra A 0.24 0.22 CM Dra B WD Age 0.20 0.1 1 10 Age (Gyr) Figure 7.7: Standard DSEP models of CM Dra A (maroon) and B (light-blue). The models were computed for Fe/H] = −0.18 (dashed line), −0.30 (solid line), and −0.42 (dotted line) with a solar calibrated αMLT . Observed radius constraints are shown as shaded horizontal regions and the age constraint is a vertical shaded region. metallicity relation of Johnson & Apps (2009). Their relation predicts [Fe/H] ≈ −0.4. The components of CM Dra are known to be inflated compared to standard stellar models (Ribas 2006; Morales et al. 2009a; Torres et al. 2010; Feiden & Chaboyer 2012a; Spada & Demarque 2012; Terrien et al. 2012). This fact has become steadily more apparent since the initial comparative study of Chabrier & Baraffe (1995), which found little disagreement. A precise estimate of the level of disagreement depends on the adopted metallicity (see, e.g., Feiden & Chaboyer 2012a; Terrien et al. 2012), but the problem is robust. Figure 7.7 demonstrates the level of disagreement compared to standard DSEP mass tracks. It also illustrates the effect of altering the metallicity in the stellar models. The level of disagreement observed in Figure 7.7 is between 5% – 6% for the primary and 7% – 8% for the secondary (Terrien et al. 2012). The latest study by Terrien et al. (2012) has essentially doubled the level of radius disagreement from 3% – 4% to 6% – 8%. This is purely a consequence of their metallicity estimate. 262 0.30 |B|surf = 5.0 kG Non-Magnetic Radius (R⊙) 0.28 0.26 CM Dra A 0.24 CM Dra B 0.22 Dipole Profile WD Age [Fe/H] = -0.18 0.20 0.1 1 10 Age (Gyr) Figure 7.8: Standard DSEP (solid line) and magnetic model (dashed line) of CM Dra A (maroon) and B (light-blue). The models were computed with [Fe/H] = −0.18 and a solar calibrated αMLT . The magnetic model was calculated using a dipole radial profile with a 5.0 kG surface magnetic field. Observed radius constraints are shown as shaded horizontal regions and the age constraint is a vertical shaded region. It is well appreciated that CM Dra is magnetically active. Balmer emission and light curve modulation due to spots have been recognized since very early investigations (Zwicky 1966; Martins 1975; Lacy 1977). Frequent optical flaring has also been continually noted (e.g., Eggen & Sandage 1967; Lacy 1977). Further details on the flare characteristics of CM Dra may be found in the work by MacDonald & Mullan (2012). The system is also a strong source of X-ray emission according to the ROSAT All-Sky Survey Bright Source Catalogue (Voges et al. 1999). The high level of magnetic activity and the short orbital period of CM Dra (1.27 days) have been used to justify the need for magnetic perturbations in low-mass stellar evolution models. Such studies were carried out by Chabrier et al. (2007) and MacDonald & Mullan (2012) using the methods described in Sections 1.4 and 6.5.2. CM Dra therefore provides a pivotal test of our models and of the magnetic field hypothesis advanced in Section 1.3.4. 263 Results from our magnetic models are displayed in Figures 7.8, 7.9, and 7.10. We computed magnetic models using all three formulations addressed thus far: dipole radial profile, Gaussian radial profile, and the constant Λ turbulent dynamo formulation. Perturbations were introduced over a single time step, but the precise age of the perturbations varied (dipole: 10 Myr, Gaussian: 1 Gyr, constant Λ: 100 Myr). The models all had [Fe/H] = −0.18, the maximum allowed metallicity (Terrien et al. 2012), and a solar-calibrated αMLT . The models performed similar to those for Kepler-16 B. A strong 5.0 kG surface magnetic field strength (1.5 MG peak field strength) with a dipole radial profile was insufficient. It relieved the radius disagreements but was unable to provide the necessary inflation. This is evidenced in Figure 7.8. The Gaussian radial profile model was able to largely reconcile the models with the observations. Surface field strengths of ∼ 6.0 kG were required (slightly stronger for CM Dra B), which translates to a 50 MG peak field strength. These models are displayed in Figure 7.9. Lastly, the constant Λ models were also unable to induce any significant radius inflation. We used Λ = 0.5, 0.9, 0.99, and 0.9999. Only the Λ = 0.9999 model is shown in Figure 7.10 to more clearly illustrate the ineffectiveness of this method. 264 0.30 |B|surf = 6.0 kG Non-Magnetic Radius (R⊙) 0.28 0.26 CM Dra A 0.24 CM Dra B 0.22 Gaussian Profile WD Age [Fe/H] = -0.18 0.20 0.1 1 10 Age (Gyr) Figure 7.9: Standard DSEP (solid line) and magnetic model (dashed line) of CM Dra A (maroon) and B (light-blue). The models were computed with [Fe/H] = −0.18 and a solar calibrated αMLT . The magnetic model was calculated using a Gaussian radial profile with a 6.0 kG surface magnetic field. Observed radius constraints are shown as shaded horizontal regions and the age constraint is a vertical shaded region. 0.30 Λ = 0.9999 Non-Magnetic Radius (R⊙) 0.28 0.26 CM Dra A 0.24 CM Dra B 0.22 Constant Λ WD Age [Fe/H] = -0.18 0.20 0.1 1 10 Age (Gyr) Figure 7.10: Standard DSEP (solid line) and magnetic model (dashed line) of CM Dra A (maroon) and B (light-blue). The models were computed with [Fe/H] = −0.18 and a solar calibrated αMLT . The magnetic model was calculated with a constant Λ = 0.9999 turbulent dynamo formulation. Observed radius constraints are shown as shaded horizontal regions and the age constraint is a vertical shaded region. 265 7.4 Discussion 7.4.1 Magnetic Field Radial Profiles The three magnetic field profiles we introduced in Section 7.2 produce different results in the stellar models. We can understand this phenomenon by following the same line of reasoning used in Section 6.4.1. There, we were able to diagnose why magnetic stellar models produced nearly identical results for the dipole and Gaussian radial profiles in stars with a radiative core. Recall that the super-adiabatic layer near the surface of those stars was more heavily influenced by the presence of the magnetic perturbation than the deeper near-adiabatic convective layers. Thus, the peak magnetic field strength had little effect compared to the surface magnetic field strength, meaning the dipole and Gaussian profiles produced similar results. For fully convective stars, the situation is reversed. We display the run of (∇s − ∇ad ) in two models of CM Dra A in Figure 7.11. One standard DSEP model and one magnetic model are shown. The magnetic model is the Gaussian radial profile model with a 6.0 kG surface magnetic field that was plotted in Figure 7.9. The surface radiative zone and the superadiabatic layer are significantly altered by the presence of the magnetic field. However, these layers have little influence over the structure of the star. We verified this by looking at the profile of a dipole radial profile model with a similar surface magnetic field strength. The (∇s − ∇ad ) profile exhibited in the surface layers by the dipole radial profile model is nearly identical to the Gaussian profile model. The difference between the dipole profile and the Gaussian profile models was the appearance of a radiative core in the Gaussian profile model. This is highlighted in the inset of Figure 7.11. The final radial profile is the constant Λ profile, which also relies on the turbulent dynamo formulation. This profile slightly perturbs the shape of the (∇s − ∇ad ) curve in the superadiabatic layer. It also produces a marginally larger surface radiative zone than the standard 266 Non-magnetic Magnetic 0.2 (∇s - ∇ad) 0.1 super-adiabatic convection near-adiabatic convection 0.0 0.001 0.000 -0.1 -0.001 -0.002 1.0 1.5 2.0 -3 -2 -1 -3 log10(ρ) (g cm ) 0 1 -0.2 -6 -5 -4 2 Figure 7.11: The difference between the plasma temperature graident, ∇s , and the adiabatic temperature gradient, ∇ad , as a function of the logarithmic plasma density for a M = 0.231M ⊙ star. We show this for two models: a non-magnetic model (maroon, solid line) and a magnetic model (light-blue, solid). The zero point is marked by a gray dashed line, dividing locations where convection (positive) or radiation (negative) is the dominant flux transport mechanism. The inset zooms in on the deep interior where the magnetic field creates a small radiative core. non-magnetic model. Convection in the deep interior is relatively unaffected, so a radiative core does not develop. We opted to not display these features in a figure because the overall profile is almost identical to the non-magnetic profile in Figure 7.11. Given the insensitivity of the overall stellar structure of fully convective stars to the size of the super-adiabatic layer, the constant Λ models have a negligible influence on the stellar radius. 7.4.2 Surface Magnetic Field Strengths We use the relationship between the surface magnetic flux, Φ, and X-ray luminosity, L x , developed in the previous chapter to estimate the surface magnetic field strengths for the 267 stars of CM Dra. Surface field strength estimates can not be calculated for Kepler-16 because it does not appear in any of the ROSAT catalogs. The ROSAT Bright Source Catalogue (Voges et al. 1999) indicates that CM Dra has X cr = 0.18 ± 0.02 cts s−1 with HR = −0.30 ± 0.07. This translates into an X-ray luminosity per star of L x = (1.57 ± 0.40) × 1028 erg s−1 , where we have used the parallax, π = 68 ± 4 mas, quoted by Harrington & Dahn (1980). Note that the X-ray luminosities are upper limits due to possible X-ray contamination in the ROSAT data. There are several stars nearby to CM Dra in the plane of the sky, but it is difficult to judge whether they contribute to the ROSAT count rate. From the X-ray luminosity derived above, we find log10 (Φ) = 24.81 ± 0.45 Mx. The errors associated with the surface magnetic field strength estimates are substantial due to the large error on the surface magnetic flux. Converting the magnetic fluxes to a surface magnetic +3.36 field strengths, we estimate that 〈B f 〉 A = 1.65+3.00 −1.07 kG and 〈B f 〉B = 1.85−1.19 kG for CM Dra A and B, respectively. These estimates imply that the 6.0 kG surface magnetic field strengths required by our Gaussian radial profile models are likely too strong. However, this does not invalidate the magnetic field models, only our choice of radial profile. The magnetic field strength in the deep interior is of greater consequence, so it may be possible to construct a radial profile to greater reflect this fact (see below). Models that use the constant Λ formalism predict surface magnetic fields strengths up to ∼ 3.0 kG. This upper limit is set by the magnetic field coming into equipartition with the kinetic energy of convective flows (Chabrier & Küker 2006; Browning 2008). Values of 3.0 kG are consistent with the estimated field strengths. Additionally, the equipartition magnetic field strengths are consistent with typical field strengths measured on the surface of M-dwarfs (Saar 1996; Reiners & Basri 2009; Shulyak et al. 2011; Reiners 2012). Although the magnetic field strengths are consistent with observations and 3D MHD modeling, our models were unable to produce radii consistent with the observations. Although there are no X-ray measurements for Kepler-16 B, we can infer from CM Dra 268 how realistic we believe our models to be. Kepler-16 B has a similar mass to the stars in CM Dra and may have a similar age, as well. Based on this information, it is our opinion that Kepler-16 B should have a magnetic field strength of a similar magnitude to the stars of CM Dra. We do not believe it would have a significantly larger field strength owing to its age and orbital period. Therefore, we believe that the 6.0 kG surface magnetic field required to bring the models into agreement with the observations is probably too strong. 7.4.3 Comparison to Previous Studies Previous attempts to reconcile model radii with the observed radii of CM Dra were seemingly successful (Chabrier et al. 2007; Morales et al. 2010; MacDonald & Mullan 2012). It is therefore prudent to compare the results of our modeling effort with these previous studies. The methods used by the two groups were in some respects similar. Morales et al. (2010) used the methods outlined by Chabrier et al. (2007) to include possible effects of magnetic fields. This approach relies on reducing the convective mixing length, αMLT , to mimic a reduced convective efficiency. On the other hand, MacDonald & Mullan (2012) implemented their convective inhibition parameter, δMM , to help stabilize regions in the model against convection (Mullan & MacDonald 2001). While these two groups used different methods to alter convection, they both treated star spots according to the formalism introduced by Chabrier et al. (2007). We will briefly summarize the results of these studies. Morales et al. (2010) found that they were able to reproduce the observed properties of CM Dra with a reduced αMLT and a star spot factor β = 0.17. This is after they applied a downward correction to the radii of CM Dra to compensate for possible 3% systematic errors in the light curve analysis due to star spots. Their value for β = 0.20 ± 0.04 combined a spot coverage fraction of 42%±8% with a spot temperature contrast of approximately 85% (spots are 15% cooler than the unblemished photosphere). The study by MacDonald & Mullan (2012) identified a region of δMM − β ( f 269 in their paper) parameter space that was able to reconcile their models with CM Dra (see their Figures 13 – 15). Their investigation reinforced the work by Chabrier et al. (2007) and Morales et al. (2010) that the spot parameter β has a greater influence on the stellar structure. The full range of β was 0.15 . β . 0.28 with 0.0 < δMM < 0.025. Using the revised radii from Morales et al. (2010) and the recommended β = 0.17, MacDonald & Mullan (2012) required a magnetic inhibition parameter between 0.020 < δMM < 0.025. These values of δMM correspond to vertical magnetic field strengths of approximately 500 G. Before we compare our models to the previous studies, we first evaluate their accuracy in light of recent findings. The metallicity of the models adopted in both studies was [Fe/H] = 0.0. 1 Recent measurements have measured the metallicity to be [Fe/H] = −0.30 ± 0.12 (Rojas-Ayala et al. 2012; Terrien et al. 2012). Consequently, the stellar models used in previous studies will be over-estimating the actual model-predicted radii by about 2.5%, based on metallicity variations observed in the DSEP models. This may seem like an insignificant amount, but consider that the Morales et al. (2010) study only corrected a 2% radius offset between the Baraffe et al. (1998) models and CM Dra (after correcting for systematic radius errors of about 3%). Now, Morales et al. (2010) used a 1 Gyr isochrone, which is 3 Gyr younger than the estimated age of CM Dra. Therefore, the radii of their models were about 1% smaller compared to what a 4 Gyr isochrone predicts. Including the 1% age effect, decreasing the metallicity of their models would decrease their radius predictions by about 1.5%. This nearly doubles the original radius offset. The resulting β needed to correct for an additional 1.5% radius offset would be ≥ 0.30 based on Figure 8 of Morales et al. (2010). This translates into a spot coverage of over 63% for a fixed temperature contrast of 85% or a temperature contrast of 73% for a fixed spot coverage of 42%. These latter values may be consistent with TiO modeling of spot properties (O’Neal et al. 2004) and modeling of spot characteristics of late-type stars in clus1 MacDonald & Mullan (2012) did explore varying combinations of Y , Z , αMLT , and δMM , but were unable to find a suitable combination to reproduce the stars in CM Dra. 270 ters (Jackson et al. 2009; Jackson & Jeffries 2013). However, Jackson & Jeffries (2013) assume randomly distributed spots whereas Morales et al. (2010) used large polar spots—spots preferentially located near the poles defined by the rotation axis. Their preference for large polar spots was adopted to justify the 3% systematic errors used to reduce the radii of CM Dra. Further investigations are needed to evaluate the consistency of these two studies, as the precise spot configuration could place the 3% systematic radius correction applied by Morales et al. (2010) in jeopardy. If the 3% radius error has to be discarded, then a β > 0.37 will be needed to bring the models into agreement with CM Dra. It was implied by Morales et al. (2010) that a β of this magnitude may be physically unrealistic and should be approached with caution. Keeping this in mind, we compare our models to those of Morales et al. (2010) and MacDonald & Mullan (2012). Our results are largely consistent with the results of Morales et al. (2010) and MacDonald & Mullan (2012). Comparing with MacDonald & Mullan (2012), we look at the dipole and Gaussian radial profile models (Figures 7.8 and 7.9). The magnetic perturbation applied in these models is qualitatively similar to the one applied by MacDonald & Mullan (2012), except we neglect star spots. In Section 6.5.2 we estimated that our magnetic field strengths would need to be an order of magnitude larger than those of Mullan & MacDonald (2001) as a consequence of our formulation. Our “inhibition parameter” largely boils down to the quantity (ν∇χ ). It was previously shown that ν ∼ δMM . Since ∇χ ∼ 0.1, our magnetic field strength should need to be about an order of magnitude larger. This is seen in our requirement of a 6.0 kG surface magnetic field with a 50 MG peak field strength to reproduce the properties of CM Dra. MacDonald & Mullan (2012) required a 500 G field with a 1 MG peak magnetic field strength. It is a little surprising that this played out considering they also added a factor to account for star spots, but is in line with our initial approximations. The pitfall of the dipole and Gaussian radial profiles is the sizable peak magnetic field. There is no evidence that a magnetic field on the order of a MG or greater could exist 271 within a fully convective star. There is certainly no physical mechanism to generate such a strong magnetic field from a turbulent dynamo (Chabrier & Küker 2006; Chabrier et al. 2007; Browning 2008). To avoid invoking a turbulent dynamo to justify a strong 1 MG magnetic field in the stars of CM Dra, MacDonald & Mullan (2012) suggest that a primordial µG seed magnetic field was present in the proto-stellar molecular cloud. This seed field was amplified due to flux conservation throughout the collapse of the individual proto-stars. We see two issues with this idea. First, the magnetic field must then survive for several billion years without diffusing through the plasma. Since the magnetic field is likely frozen into the highly conducting stellar plasma, the timescale for diffusion is proportional to the convective timescale. For a fully convective M-dwarf this is around 107 s, or ∼ 1 yr. If convection were largely suppressed by the magnetic field, this convective timescale would increase, but not by nine orders of magnitude. Even with slow convection, the molecular diffusion timescale for such a field is, in all likelihood, quite rapid compared to the lifetime of the star (Chabrier et al. 2007). Full 3D MHD models of fully convective stars with a large primordial magnetic field could lend insight into a more realistic diffusion timescale. The second problem we see with the interior MG field strength is KOI-126. Nearly identical in mass and rotational period to CM Dra, the fully convective stars of KOI-126 are well reproduced by stellar models (Chapter 2; Feiden et al. 2011; Spada & Demarque 2012). MacDonald & Mullan (2012) argue that model agreement based on KOI-126 should be expected given its orbital period, which is half a day longer than CM Dra’s. They calculate that the surface magnetic fields would be significantly lower than those CM Dra. However, if we assume the 1 MG interior field is of primordial origin, then we see no reason why KOI-126 would not be expected to have a similar interior field strength (assuming the field was stable for billions of years). Seeing as the interior magnetic field strength largely determines the induced radius inflation for fully convective stars, we would draw the opposite conclusion of MacDonald & Mullan (2012). KOI-126 should appear inflated compared to 272 standard stellar evolution models. We therefore doubt the existence of super-MG magnetic fields deep within fully convective stars. This leads us to believe the agreement between our Gaussian radial profile model and the observations is unphysical. Comparing our constant Λ models to the Morales et al. (2010) results, we find they are consistent. Chabrier et al. (2007) and Morales et al. (2010) find that the influence of reducing αMLT is negligible compared to the radius inflation created by their star spot formulation. Section 6.5.2 demonstrated that our turbulent dynamo approach was quite similar to the reduced αMLT approach. In fact, we formulated reductions in αMLT as a function of Λ. It is therefore not surprising that we find agreement between the two methods. So what are we to make of the previous discussions? On the one hand, the suppression of convection appears to be unable to reconcile the model predictions with the observations. But, on the other hand, reducing the total flux due to the presence of star spots is able to do so. We find these two pieces of evidence contradictory. Star spots have a physical origin and that origin is the suppression of convection by a vertical magnetic field (Chandrasekhar 1961; Gough & Tayler 1966). We have difficulty understanding why models that suppress convection in a manner consistent with the physical picture of star spots (Gough & Tayler 1966; Mullan & MacDonald 2001) would be insufficient at inflating radii, yet an independent method based on reducing the total flux—essentially the same physical effect as inhibiting convection to model spots—would suffice (i.e., Chabrier et al. 2007). It is true that any onedimensional treatment of magnetic fields and spots will be overly simplistic, and further work on this matter is warranted. However, even when we almost entirely suppress convection using a constant Λ profile with Λ = 0.9999 and a reduced αMLT = 0.5, there is a negligible impact on the model radius predictions. It may be argued that the two effects are easy to reconcile when considering that magnetic fields can have global and local effects. A sub-kG large-scale magnetic field may exist in combination with strong, localized magnetic field structures that form spots. The result is an 273 average magnetic field strength that appears to be only a few kG. We test this by introducing a strong suppression factor (Λ = 0.9999) and an independent stabilization factor using a dipole radial field with a 3.0 kG surface magnetic field in our models. This formulation barely inflates the radius of a M = 0.23M⊙ model. We mentioned before that the deep interior magnetic field strength is the most important factor when it comes to inflating fully-convective stars. Thus, having strong localizations of magnetic fields on the surface is insufficient to cause the necessary structural changes. Further testing of this hypothesis is necessary. The case for magnetic fields inflating the radii of fully convective stars appears to be tenuous. The more we learn about these systems, the harder it becomes to reconcile model predictions with the observations. When we weigh the evidence, we find it harder to believe that magnetic fields are actively inflating the radii of fully convective stars. Furthermore, although the reduced-flux star spot formulation appears to be able to provide the necessary radius inflation, this does not include variations due to metallicity. Our models use the largest possible value of the metallicity for CM Dra. If we consider the difficulty in obtaining small, 1% changes in the radii of these stars, moving to lower metallicities presents new challenges. Radius changes due to metallicity have the ability to push the star spot parameters into unphysical regimes. We are of the opinion that magnetic fields may not be the appropriate solution. While more work is needed to investigation the influence of star spots and improved methods for treating a magneto-convection in 1D models, we feel that it is not too early to seek other solutions. 7.4.4 Alternatives If magnetic inhibition and stabilization against convection are unable to reconcile fully convective model predictions with observations, what might the solution be? As an exercise, 274 we plotted the radius discrepancies of fully convective stars against several intrinsic properties: mass, metallicity, orbital period, and Rossby number. Data for all fully convective stars with a measured metallicity were included. This sample contains six stars: CM Dra (A, B), KOI-126 (B, C), Kepler-16 (B), and Kepler-38 (B) (Orosz et al. 2012). LSPM J1112+7626 (B) (Irwin et al. 2011) was not included because it lacks a proper metallicity estimate. Figure 7.12 shows the results of this exercise. We were surprised to find that metallicity provided the best correlation with radius deviations, as seen in panel (a). Radius deviations clearly do not correlate with stellar mass, evidenced by Figure 7.12(b). Orbital period and Rossby number show a hint of correlation, but the relationships are broken by the presence of KOI-126 (B, C), which are located along the δR/R obs = 0 line. We note that the rotation periods of the secondary stars in Kepler-38 and Kepler-16 are not known. Spin-orbit synchronization was assumed to create Figure 7.12, but is not necessarily a valid assumption with these stars. This would also affect the calculation of the Rossby number in Figure 7.12(d). The relationship between radius deviations and metallicity is surprising for two primary reasons. First, it contradicts the results presented by López-Morales (2007), which we described in Section 1.3.1. The second reason is that the two variables are anti-correlated. If a trend with metallicity had been expected, we would have predicted that the more metal rich objects would show the largest discrepancies. With only six data points, the sample size is admittedly small. However, that the data points follow such a tight trend is either a remarkable coincidence or is an important clue to help unravel the radius discrepancies. Evidence of this anti-correlation may also exist in the study by Boyajian et al. (2012). They measured the radii of over 20 K- and M-dwarfs using the CHARA interferometer. One key result from their study was that the empirical radius-Teff relation for low-mass stars was independent of metallicity (see their Figure 13). Plotting the relative difference between their measured radii and DSEP models, they found that smaller objects were typically under- 275 0.10 0.10 (a) 0.06 0.04 0.02 0.00 0.06 0.04 0.02 0.00 -0.02 -0.5 -0.4 -0.3 -0.2 -0.1 0 [Fe/H] -0.02 0.1 0.2 0.3 0.18 0.10 0.20 0.22 0.24 Mass (M⊙) 0.26 0.10 0.08 0.06 0.04 0.02 (d) 0.08 (c) δR/Robs δR/Robs (b) 0.08 δR/Robs δR/Robs 0.08 0.06 0.04 0.02 0.00 0.00 -0.02 0.01 -0.02 0 10 20 30 40 Porb (day) 0.1 Ro 1 Figure 7.12: Radius residuals between models and observations of fully convective stars as a function of [Fe/H], mass, orbital period (proxy for rotation period), and Rossby number. Only fully convective stars with precise masses, radii, and metallcities were included. The stars in the sample are CM Dra (Morales et al. 2009a), KOI-126 (B, C) (Carter et al. 2011), Kepler-16 B (Doyle et al. 2011), and Kepler-38 B (Orosz et al. 2012). predicted by the models. If the anti-correlation shown in Figure 7.12 is real, we would then expect the stars with the largest observed discrepancies in Boyajian et al. (2012) to be more metal poor. This appears to be the case (see Figure 14 in Boyajian et al. 2012). Now, they did not find any direct correlation with metallicity and radius deviations (see their Figure 16), but precise radius deviations are difficult to determine because their mass estimates are highly uncertain. Still, it would appear as though the data in Boyajian et al. (2012) lend some support to the existence of a metallicity-radius anti-correlation among fully convective stars. We would like to offer a compelling argument to justify the existence of this relationship. However, we do not have a definitive explanation at the present time, although we may 276 be able to rule out some scenarios. It is likely not a connection between metallicity and the stellar convective properties. We have shown that modifications to convection can not account for any significant radius inflation. Unless there are multi-dimensional affects of convection that can alter the structure of the star, convection seems to be an insufficient explanation. Opacity would also seem to be ruled out. One would expect that the significant sources of error from opacity calculations are related to the presence of metals. An anticorrelation between model errors and metallicity would suggest that hydrogen opacities are in error. One possibility is that the model [α/Fe] abundances are incorrect. It has been suggested that below solar metallicity, M-dwarfs show a total metallicity, [M/H], that is greater than the iron abundance, [Fe/H] (Rojas-Ayala et al. 2012; Muirhead et al. 2012; Mann et al. 2013a,b). For our models, we have been using a scaled-solar metallicity, which effectively equates [Fe/H] to [M/H]. The differences between [M/H] and [Fe/H] noted by Rojas-Ayala et al. (2012) are on the order of 0.1 dex. This translates, roughly, to an alpha-element enhancement of [α/Fe] ≈ +0.2 dex. Given a star with a mass similar to CM Dra A and a [Fe/H] = −0.30, increasing [α/Fe] by 0.2 dex increases the radius by about 2%. A modest increase compared to the 6% to 8% radius deviations we observe with CM Dra. Providing a complete solution for this anti-correlation would require a generous helping of alpha-element enhancement.2 There is also the question of the adopted helium abundance. We noted in Equation (6.1) that the initial helium abundance is scaled linearly with the heavy element mass fraction. Figure 7.13 shows how the metallicity-radius deviation correlation changes when we assume a fixed helium abundance, Y = 0.33. This value represents a significant enhancement compared to the linear-scaled value for all of the stars in our sample. It is also intermediate 2 There is also the larger issue of the overall solar composition. We use the Grevesse & Sauval (1998) solar abundance composition, but it seems likely that the solar heavy element composition is lower (Asplund et al. 2009). The global lowering of the heavy element composition may offset the benefits of an enhanced alpha abundance. 277 0.10 Z-scaled Yi Yi = 0.33 0.08 δR/Robs 0.06 0.04 0.02 0.00 -0.02 -0.5 -0.4 -0.3 -0.2 -0.1 [Fe/H] 0 0.1 0.2 0.3 Figure 7.13: Radius residuals between fixed-Y models (Y = 0.33) and observations of fully convective stars as a function of [Fe/H]. The fixed-Y points are plotted in maroon and those with Z -scaled Y values (from Figure 7.12) are plotted in gray, for reference. between estimated helium abundances in old star clusters (Piotto et al. 2005; Brogaard et al. 2012) and should therefore not be entirely unphysical. We see that arbitrarily increasing the helium abundance relieves some of the radius discrepancies, but does not fully explain the observed trend with metallicity. We must also consider that all of the stars will not necessarily have such an enhancement in helium abundance and that altering the helium abundance of the systems will effect other properties (most notably the age). All of the stars have ages estimated from companions, which must also be modeled with the helium enhancement to extract meaningful results. 7.5 Summary This chapter addressed the question of whether magnetic fields can realistically inflate the radii of fully convective stars. We approached the problem by modeling two individual 278 DEBs that contain fully convective stars, Kepler-16 and CM Draconis. Ultimately, we find that magnetic fields do not provide a physically plausible solution of the observed radius discrepancies. Multiple formulations of the magnetic perturbation in our models were used to no avail. There was a hint that strong MG magnetic field deep in the stellar interior could reconcile the models, but it is our opinion that this is a physically unrealistic scenario. Our results, while largely consistent with previous studies, are in disagreement that magnetic fields provide an adequate solution. Without invoking magnetic fields, we are at a loss to explain the large radius discrepancies observed with fully convective stars. However, an anti-correlation between these discrepancies and [Fe/H] was uncovered. No physical explanation for this occurrence was put forth. Alpha-element enhancement may provide part of the solution, but can only account for about 2% of the total radius deviations. Further investigation into this anti-correlation may reveal the true cause of the observed radius inflation. Studies to collect more data and generate a larger model grid, covering the available magnetic and non-magnetic parameter space, are on-going. 279 Chapter 8 Conclusions & Future Investigations 8.1 Thesis Summary The body of work contained in this dissertation was aimed at deepening our theoretical understanding of low-mass stars. From the outset, an end goal of the project was to have developed a set of new stellar evolution models with magnetic fields. Along the way, however, we found the need to evaluate the present capability of standard low-mass stellar evolution models. We began our journey with KOI-126, the hierarchical triply eclipsing system with two lowmass stars below the fully convective boundary (Carter et al. 2011). Chapter 2 presented stellar evolution models that accurately reproduced the properties of all three stars. This was the first successful modeling of fully convective stars in a DEB. Given their similarity to CM Dra A and B, the low-mass stars in KOI-126 provide a unique constraint on theoretical descriptions of the radius inflation observed with CM Dra. Any theory mitigating the problems with CM Dra must also not undermine the agreement with KOI-126. We also predicted theoretical interior structure constants for KOI-126 B and C. A weighted average of these quantities helps to define the rate of apsidal motion of the binary orbit. Observational 280 constraints on the apsidal motion rate at the 1% level predicted by Carter et al. (2011) would provide a stringent test of our stellar evolution models. Not only would the constraints test our models, but we discussed how the observed apsidal motion could validate various equations of state. KOI-126 B and C can therefore be used to probe the interior structure of low-mass stars. Chapter 3 delved deeper into applying our model predictions for the internal structure constants. We developed a novel method for estimating the ages of young DEB systems. Most age dating techniques involve comparing the predicted surface properties of stellar evolution models to observations of young stars. The problem with this approach is that the surface properties of stellar models and real stars may be heavily influenced by magnetic activity. By using the interior structure constants, we can achieve an age dating method that is more robust by avoiding the interpretation of surface features. We tested the sensitivity of this technique to specific model input properties (i.e., mass, metallicity, helium abundance) and possible magnetic activity effects (e.g., suppressed convection). Generally speaking, we were able to achieve an age estimate with a precision to better than 20% with favorable observational data. Returning to the mass-radius discrepancies in Chapter 4, we reanalyzed the severity of the problem. Numerous literature references to the mass-radius problem quoted a systematic 5% – 10% radius inflation.1 By coupling a large data set (nearly double the number of DEBs in previous studies) and modeling variations due to metallicity and age, we showed that the radius problem was not systematic. Average deviations of about 2.5% were the norm and only a couple systems exhibited significant inflation above 5%. Previous misidentification of large systematic errors was likely the result of studies using only one or two isochrones instead of properly fitting systems to a grid of isochrones. Furthermore, we provided a statistical analysis comparing the observed radius deviations to potential magnetic field strength proxies (i.e., rotation period, Rossby number, and X-ray luminosity). We found no 1 43 papers between 2007 and 2012... not that we were ever compelled to count. 281 significant difference in the distribution of radius deviations between slowly rotating and rapidly rotating populations. However, there was evidence for a correlation between radius deviations and X-ray luminosity. The original idea when this project began was to develop a new set of low-mass stellar evolution models that included magnetic field perturbations. This goal was achieved in Chapter 5. We lay out, in full detail, the mathematical formalism of our approach to including magnetic field perturbations. Following the seminal work of Lydon & Sofia (1995), we tweaked the formulation and developed our own set of magnetic stellar evolution models. A series of tests were performed to ensure the numerical stability and overall believability of the models. After verifying that the models were producing reliable results, we applied our models to a recently characterized solar-type DEB, EF Aquarii (Vos et al. 2012). Specifically, we found that models of EF Aqr A required a surface magnetic field strength between 1.6 – 2.6 kG. EF Aqr B, on the other hand, needed a stronger surface magnetic field strength of 3.2 – 5.5 kG. The magnetic field strengths were evaluated using scaling relations based on X-ray luminosity and Ca ii emission strengths. The derived surface magnetic fields were consistent with the model predicted field strengths. The final two chapters analyzed several important low-mass DEBs to test the hypothesis that magnetic fields are inflating low-mass stars. Here, we split the analysis into two parts: Chapter 6 dealt with stars that have a radiative core and convective envelope and Chapter 7 looked at fully convective stars. Stars with radiative cores proved to be susceptible to significant radius inflation from magnetic fields. Three systems, UV Psc, YY Gem, and CU Cnc, were fit by magnetic models, but only after we reformulated the mixing length theory of magneto-convection to account for the conversion of convective energy into magnetic energy. The resulting magnetic models were able to reproduce the radii of the stars in YY Gem and CU Cnc with realistic surface magnetic field strengths. Estimates of the actual surface magnetic field strengths were derived from an X-ray–magnetic-flux scaling relation. 282 Additional analysis was carried out to explore the impact of magnetic models on studies of exoplanet habitability and asteroseismology. Fully convective stars were not so easily influenced by the magnetic models. All but models with super-MG magnetic fields in the deep interior were ineffective at inflating stellar radii. This is in sharp contrast to other studies that find magnetic fields up to the task. Extensive discussion comparing and contrasting the individual studies was provided so that the reader could form their own opinion. We, however, are skeptical that magnetic fields are inflating fully convective stars. Instead, there appears to be an anti-correlation between radius deviations and metallicity of fully convective stars. It is not clear to us what the physical justification is for this anti-correlation. 8.2 Future Investigations Throughout the course of developing this dissertation, numerous avenues for continuing research were exposed. The most natural of these is the development of a large database of magnetic model tracks and relations. With the magnetic models now rigorously tested against individual DEBs, we are confident that the magnetic field perturbations are correct. An extensive grid of models will be required to allow for the type of individual or statistical tests performed throughout this thesis. Additionally, a functional web interface will be necessary. Progress on both pieces to the database will begin in the near future. Developing a database of magnetic mass tracks will enable additional studies using the magnetic models. From Chapter 2, we learned that the apsidal motion of KOI-126 (B,C) could be used to evaluate the validity of different equations of state. Following the publication of Feiden et al. (2011), I began work on testing the sensitivity of the theoretical interior structure constants to the available micro-physics. This includes several equations of state (FreeEOS; Saumon et al. 1995) and subtle variations of the FreeEOS. With the introduction 283 of reliable magnetic field models, it would be advantageous to include how variations in a stellar magnetic field might influence the interior structure constants. Eventually we hope to place constraints on individual components of the equation of state using observations KOI-126 thereby allowing for the computation of more accurate equations of state suitable for low-mass stars and giant planets. A comprehensive database will also allow for large statistical tests of the stellar evolution models. I am involved in a campaign to evaluate stellar evolution and stellar atmosphere models using the Boyajian et al. (2012) sample of K- and M-dwarfs with interferometric radii and effective temperatures. Observers are obtaining spectra for each star so that they may determine [Fe/H] and [α/Fe] (Mann et al. 2013a,b). A fine grid of over 3700 standard stellar evolution mass tracks is nearly complete and adding a grid of magnetic models would provide a means of testing magnetic stellar evolution models against a sample of wellcharacterized single field stars. Single field stars are thought to not show signs of magnetic inflation (Boyajian et al. 2012), but with uncertainties in mass and metallicity, it is difficult to constrain at the present time. I am also beginning an effort to directly test the validity of the magnetic stellar evolution models in the partially convective regime. Results in Chapter 6 suggest that the influence of a magnetic field is very particular. When we computed magnetic models for UV Psc and YY Gem we had to shift our models to a lower metallicity than initially anticipated to obtain the correct radii and Teff s. The predicted metallicity-Teff -magnetic field strength combinations are particular and fairly robust. Therefore, if we can measure the metallicities and Teff s, we can test the model predictions. To get accurate metallicities and Teff s, we will obtain high resolution optical spectra at multiple phases of the DEB orbits using the Fibre-fed Echelle Spectrograph (FIES) on the Nordic Optical Telescope (NOT). Using SME (Valenti & Piskunov 1996), it will be possible to decompose the spectra of the stars in each binary system (Stempels et al. 2008). Observations for YY Gem were completed for a sep- 284 arate observing program and a program to obtain spectra for UV Psc and CU Cnc will be carried out in late 2013 or early 2014. When it comes to the physics of the magnetic perturbation, there are other paths to explore. For instance, the turbulent dynamo models produce the most realistic results for DEB star with a radiative core. However, the balance between turbulent kinetic energy and magnetic energy is not necessarily dependent on previous time steps. The models calculate the convective properties using mixing length theory and then perturb uconv and ∇s afterward. Thus, only changes to the large scale structure affect the non-magnetic mixing length calculation. Instead of performing the perturbation in this way, it would be ideal to have feedback between convection and the magnetic field (Chabrier & Küker 2006; Browning 2008). This is akin to implementing an actual dynamo mechanism in the models. An extension of this would be to then allow the magnetic field ramp up with age along the pre-MS and ramp down along the main sequence, as single stars are thought to do (Skumanich 1972; Barnes 2010). Finally, the issues with fully convective stars still need to be addressed. This is true from both a magnetic and standard stellar model perspective. The contradictions between the results of our magnetic models (Chapter 7) and the star spot formulations (Chabrier et al. 2007; Morales et al. 2010; MacDonald & Mullan 2012) need to be addressed further. Additionally, the anti-correlation that we revealed between radius deviations and metallicity deserve more attention. Exploring parameter spaces in Y , [Fe/H], and [α/Fe] would be the first step. However, the results from a similar exploration of CM Dra in MacDonald & Mullan (2012) suggest that this may not provide any suitable explanation. In the meantime, we continue seek a plausible explanation. 285 Appendix A DMESTAR User’s Guide Dartmouth Magnetic Evolutionary Stellar Tracks And Relations Throughout the latter portion of this dissertation, the development of a magnetic version of the Dartmouth stellar evolution code was presented. Here, we will briefly summarize the different variations introduced above and then present a short guide for those wishing to use the code. Chapter 5 outlined the basic mathematical formulation of a magnetic perturbation assuming a rotationally driven dynamo and that equipartition only occurs between the magnetic field pressure and the surrounding gas pressure. The radial profile of the magnetic field strength within the model was prescribed using a “dipole radial profile.” In Chapter 6, a second radial profile was introduced, the “Gaussian radial profile.” Additionally, the mathematical formulation was modified to include effects associated with a turbulent convection driven dynamo. Finally, a third radial profile was added in Chapter 7. This radial profile assumed the magnetic field strength was equal to some fraction, Λ, of the turbulent equipartition magnetic field strength. It was coined the “constant Λ radial profile.” The code was written in Fortran 90, which is backwards compatible with older FORTRAN 77 that forms the foundation of the Dartmouth stellar evolution code. The magnetic field 286 routines are contained in a single module, mag_field.1 There are eight (8) subroutines within mag_field: b_initialize, b_field, find_tach_rad, b_dipole, b_gauss, b_equip, b_envelope, and b_rescale. These subroutines perform the following functions, • b_initialize – initializes the magnetic field properties using a combination of default values and user-supplied input. • b_field – the primary magnetic field routine that calls all of the necessary routines for determining the interior magnetic field strength profile. • find_tach_rad – finds the radius of the stellar tachocline, if it exists. • b_dipole – sets the run of the magnetic field strength using the dipole radial profile. • b_gauss – sets the run of the magnetic field strength using the Gaussian radial profile. • b_equip – sets the run of the magnetic field strength using the constant Λ radial profile. • b_envelope – calculates the run of the magnetic field strength in the model envelope for the various radial profiles. • b_rescale – rescales the magnetic field profile as the number of mass shells is readjusted. Running DMESTAR may be performed by setting a single logic variable, lzmag to .true. in the physics namelist file. By doing so, the user opts to run the magnetic code with the magnetic field variables set to the default values. However, the default values are designed to minimize the impact of the magnetic perturbations, such that it is equivalent to running the code without a magnetic field. Instead, the user may create a magnetic namelist file that specifies the desired magnetic field options. The entire list of variable options, their data types, descriptions, and their default values are listed in Table A.1. 1 Hereafter, Fortran modules, subroutines, and variables will be written using a monospace font. 287 Table A.1 Magnetic namelist variables. Variable Type Description b_pert_age real Age of first perturbation (in Gyr) 0.1 b_surf real Surface magnetic field strength (G) 0.1 gammag real Magnetic field “geometry” 2.0 chi_f real Magnetic flux parameter, f 1.0 fc_tach real Fully convective “tachocline” (R⋆ ) 0.15 eq_lambda real Equipartition fraction, Λ 0.0 b_rad_prof char Radial profile ‘dipole’ dynamo char Dynamo type ‘rot’ lbramp logical Magnetic field ramping toggle Default .false. Again, the user should set lzmag = .true. to turn the magnetic perturbation option on in DMESTAR. The rest of the magnetic model properties should be set in the the magnetic namelist file, which should then be copied to the Fortran input unit 75 by a run script. The exact purpose of lzmag is to inform the program that a magnetic model is being requested. Instead of immediately introducing the magnetic perturbation, there is a delay between the start of the program and the actual initiation of the magnetic perturbation. Segmentation faults may occur if the field is introduced too early on the Hayashi track. Currently, the default initialization age is 0.1 Gyr, but this may be altered through the use of the variable b_pert_age. This variable defines the age (in Gyr) of the first perturbation. Segmentation faults may also be mitigated by ramping the magnetic field from zero up to the requested surface magnetic field strength using the logic variable lbramp. The step size of the magnetic field ramping technique is hardwired to be 1.0×10−5 multiplied by the size of the time step (in Gyr). So note that the rate of ramping is governed by number of time steps taken during the model run. Given how crude this method is, it will be updated in the future. The strength of the perturbation is determined by the surface magnetic field strength and 288 the type of radial profile selected. These properties are controlled by the variables b_surf and b_rad_profile. The surface magnetic field strength must be declared in gauss (G) and is set to 0.1 G by default. Three options exist for the radial profile: “dipole,” “gauss,” and “equip,” which instruct the program to use either a dipole, Gaussian, or constant Λ radial profile. There is also the option to use either a rotationally driven dynamo or a turbulent dynamo, defined by setting dynamo to either “rot” or “turb.” If a constant Λ profile is selected, there will be no option for a rotational dynamo. The code will automatically confirm that dynamo = ‘turb’. The fraction of the equipartition magnetic field strength requested by the user is controlled by the variable eq_lambda. If the users selects a dynamo = ‘turb’, then any of the radial profile options are viable. Once a radial profile is selected, the peak magnetic field strength is prescribed to the tachocline (except for the case of constant Λ). While the tachocline is defined by the code for models with a radiative core, fully convective models must have one specified arbitrarily. Based on 3D MHD models by Browning (2008), the default value is set to 0.15R ⋆ . DMESTAR allows for this point to be changed by setting fc_tach to a value between 0.001R ⋆ and 0.999R⋆ . Lastly, there are options to control the variables f and γ, which were defined in Chapter 5. Recall, f controls the amount of flux exchanged between a convecting bubble and the surrounding plasma and γ mimics the geometry of the magnetic field. By default, these values are set to chi_f = 1.0 and gammag = 2.0, respectively. Cases for these values were made within the text. If the user is unsure, it is recommended to use the default values. A sample magnetic namelist file is provided at the end of this Appendix. Monitoring the magnetic perturbation is possible using data contained in multiple output files. The program tracks the magnetic field variables throughout the interior (unit 76) and the envelope (unit 80). Both of these output files are new to the code. Explicitly, the code tracks the magnetic pressure (p_mag) and the magnetic energy per unit mass (u_mag) as a function of radius and density. With these variables, the magnetic field strength radial 289 profile may be extracted. Additionally, the magnetic field strength at the tachocline is output to the model track file (unit 37) to provide a quick diagnostic. A complete list of the input and output unit numbers allocated in DMESTAR is provided below. The list is actually the new module that DMESTAR keeps track of allocated unit numbers. This is different from DSEP, which used to declare all unit numbers if the subroutine parmin. By using a module, former common blocks associated with file i/o are no longer required. The module, or individual variables, are imported where needed. Listing A.1: Sample Magnetic Namelist File ! Example of a magnetic namelist file for DMESTAR ! File: magnetic.nml ! Author: Gregory A. Feiden ! Date: 07 May 2013 !---------------------------------------------------------$magnetic lmag = .false. ! toggle magnetic field: .true. (on) or .false. (off) b_pert_age = 0.1 ! perturbation age (in Gyr) b_surf = 0.6d2 ! surface magnetic field (G) gammag = 2.0 ! magnetic geometry parameter chi_f = 1.0 ! magnetic flux parameter fc_tach = 0.15 ! fully convective tachocline in units of R_star eq_lambda = 0.9 ! equipartition parameter b_rad_prof = ’equip’ ! Gaussian radial profile; options: ’dipole’, ’gauss’, ’equip’ dynamo = ’rot’ ! turbulent dynamo; options: ’rot’, ’turb’ ramp_b_field = .false. ! ramp the magnetic field; options: .true. or .false. $end 290 Listing A.2: Allocated Unit Numbers for I/O module file_lib implicit none integer, protected :: iowr ! output: status file integer, protected :: igrad ! output: temperature gradients integer, protected :: ilast ! output: last model (text) integer, protected :: ifirst ! input: first model (text) integer, protected :: irun ! input: physics namelist integer, protected :: istand ! input: contral namelist integer, protected :: ifermi ! input: Fermi tables integer, protected :: idebug ! output: reserved for debugging integer, protected :: itrack ! output: track integer, protected :: ishort ! output: all diagnostic info integer, protected :: imilne ! output: Milne invariant variables integer, protected :: imodpt ! output: shell by shell info integer, protected :: istor ! output: saved models integer, protected :: iopmod ! output: pulsation, interior integer, protected :: iopenv ! output: pulsation, envelope integer, protected :: iopatm ! output: pulsation, atmosphere integer, protected :: ilstbn ! output: binary last model integer, protected :: istobn ! output: binary stored model integer, protected :: ifstbn ! input: binary starting model integer, protected :: idyn ! output: dynamo information integer, protected :: illdat ! input: OPAL92 opacity table integer, protected :: iliv95 ! input: OPAL95 opacity table integer, protected :: isnu ! output: SNU fluxes integer, protected :: icond ! input: conduction opacities integer, protected :: ikur ! input: Kurucz low T opacities integer, protected :: iiso ! output: isochrone information integer, protected :: ioatm ! input: Kurucz atmospheres integer, protected :: ialxo ! input: Alex Low T opacities integer, protected :: iopale ! input: OPAL EOS integer, protected :: io95co ! input: OPAL binary dump integer, protected :: iolaol ! input: LAOL opacities integer, protected :: iopurez ! input: LAOL opacities, pure CN integer, protected :: iolaol2 ! input: opacities for diffusion integer, protected :: ioopal2 ! input: opacities for diffusion 291 integer, protected :: ikur2 ! input: opacities for diffusion integer, protected :: ioatma ! input: Allard atmospheres integer, protected :: imonte1 ! input: monte carlo for SNUs integer, protected :: imonte2 ! input: monte carlo for SNUs integer, protected :: iscvh ! input: SCVH95 EOS integer, protected :: iscvhe ! input: SCVH95 EOS integer, protected :: iscvz ! input: SCVH95 EOS integer, protected :: imag ! input: magnetic field namelist integer, protected :: imagpro ! output: magnetic field profile integer, protected :: iapsid ! output: apsidal motion info integer, protected :: imenv ! output: magnetic envelope integer, protected :: iatmphx ! input: phoenix atmosphere logical, protected :: files_initialized contains subroutine file_init() ! set unit numbers for i/o files if (files_initialized) then write(*, *) ’Warning: unauthorized call to file setup.’ return end if iowr = 6 ! output: status file igrad = 8 ! output: temperature gradients ilast = 11 ! output: last model (text) ifirst = 12 ! input: first model (text) irun = 13 ! input: physics namelist istand = 14 ! input: contral namelist ifermi = 15 ! input: Fermi tables idebug = 18 ! output: reserved for debugging itrack = 19 ! output: track ishort = 20 ! output: all diagnostic info imilne = 21 ! output: Milne invariant variables imodpt = 22 ! output: shell by shell info istor ! output: saved models = 23 iopmod = 24 ! output: pulsation, interior 292 iopenv = 25 ! output: pulsation, envelope iopatm = 26 ! output: pulsation, atmosphere ilstbn = 27 ! output: binary last model istobn = 28 ! output: binary stored model ifstbn = 29 ! input: binary starting model idyn ! output: dynamo information = 30 illdat = 32 ! input: OPAL92 opacity table isnu = 33 ! output: SNU fluxes icond = 35 ! input: conduction opacities ikur = 36 ! input: Kurucz low T opacities iiso = 37 ! output: isochrone information ioatm = 38 ! input: Kurucz atmospheres ialxo = 39 ! input: Alex Low T opacities iliv95 = 48 ! input: OPAL95 opacity table iopale = 49 ! input: OPAL EOS io95co = 50 ! input: OPAL binary dump iolaol = 61 ! input: LAOL opacities iopurez = 62 ! input: LAOL opacities, pure CN iolaol2 = 63 ! input: opacities for diffusion ioopal2 = 64 ! input: opacities for diffusion ikur2 = 65 ioatma = 66 ! input: opacities for diffusion ! input: Allard atmospheres imonte1 = 70 ! input: monte carlo for SNUs imonte2 = 71 ! input: monte carlo for SNUs iscvh = 72 ! input: SCVH95 EOS iscvhe = 73 ! input: SCVH95 EOS iscvz = 74 ! input: SCVH95 EOS imag = 75 ! input: magnetic field namelist imagpro = 76 ! output: magnetic field radial profile iapsid = 77 ! output: apsidal motion info imenv ! output: magnetic envelope = 80 iatmphx = 95 ! input: phoenix atmosphere(s), 95 --> 99 files_initialized = .true. end subroutine file_init subroutine file_close(file_index) 293 integer, intent(in) :: file_index close(file_index) end subroutine end module file_lib 294 Appendix B Additional Model Physics While the majority of additions to the Dartmouth stellar evolution code surrounded the introduction of the magnetic perturbation, other physics were either added or updated along the way. This Appendix briefly summarizes these changes. Each section describes the physics and any control variables added to the code, which should be specified in the physics namelist file. Default values for these new control variables are shown in parentheses. B.1 Turbulent Mixing We have introduced turbulent diffusion as described by Richard et al. (2005). Turbulent diffusion modifies the atomic diffusion coefficient and acts to extend the mixing region below the convection zone. The magnitude of the turbulent diffusion coefficient is tied to an adjustable reference temperature, T0 , and varies with density via: · ρ D T = ωD He (T0 ) ρ(T0 ) 295 ¸−3 (B.1) where ω characterizes the relative strength of turbulent diffusion and D He (T0 ) and ρ(T0 ) are the helium diffusion coefficient and density at the prescribed reference temperature, respectively. Proffitt & Michaud (1991) motivate the ρ −3 dependence in order to reproduce the solar beryllium abundance, which appears to be unchanged over time. Thus, any nonstandard mixing in the Sun must be localized to a narrow region below the solar convection zone. We select ω = 400 and leave it fixed, in accordance with Richard et al. (2005). The reference temperature used primarily in our models is T0 = 106 , which was found to best reproduce the observed abundance trends of the globular cluster NGC 6397 (Korn et al. 2007). The turbulent diffusion coefficient is then introduced into the gravitational setting routines already present in DSEP. Bahcall & Loeb (1990) indicate the time rate of change of the hydrogen mass fraction satisfies the following partial differential equation, · ¸ 1 ∂ r 2 X T 5/2 ξH (r ) ∂X = 2 , ∂t ρr ∂r (ln Λ/2.2) (B.2) where the dimensionless function ξH (r ) is ¸ · 5(1 − X ) ∂ ln P ∂ ln T ∂ X (1 + X ) ξH (r ) = + ΦH (X ) + ln . 2 4 ∂r ∂r (3 + 5X ) ∂r (B.3) The first term on the right hand side of the above equation can be identified as the gravitational settling term, the second term is term associated with atomic diffusion and the third term accounts for thermal diffusion. We may compare the equation presented by BL with the diffusion equation presented by Richard et al. (2005) which adds a turbulent diffusion coefficient to the diffusion equation presented by Aller & Chapman (1960) · ¸ ¸ · ZH mp g A H mp ∂ lnC X ∂ ln T v D = (D H + D T ) − (g rad − g ) + + kT + DH . ∂r kT 2kT ∂r (B.4) Note, there is a difference in sign between the Richard et al. (2005) formulation and the 296 Bahcall & Loeb (1990) formulation which is only indicative of which direction refers to inward or outward diffusion. Now, we may directly compare the equations presented by both BL and RMR, but it must be noted that the diffusion equation from RMR defines atomic diffusion in terms of the natural logarithm of the hydrogen concentration whereas BL define atomic diffusion in terms of the hydrogren mass fraction gradient. If we simplify the second term of Bahcall & Loeb (1990)’s dimensionless function ξH (r ), we have ¸ · X (1 + X ) ∂ = ln ∂r (3 + 5X )2 = = = = ¸ · (3 + 5X )2 X 10X (1 + X ) ∂X 1+ X + − X (1 + X ) (3 + 5X )2 (3 + 5X )2 (3 + 5X )3 ∂r · ¸ 1 1 10 ∂X + − X 1 + X 3 + 5X ∂r · ¸ (1 + X )(3 + 5X ) + X (3 + 5X ) − 10X (1 + X ) ∂X X (1 + X )(3 + 5X ) ∂r ¸ · 3 + 8X + 5X 2 + 3X + 5X 2 − 10X − 10X 2 ∂X X (1 + X )(3 + 5X ) ∂r ¸ · ∂X 3+ X X (1 + X )(3 + 5X ) ∂r (B.5) To compare with Richard et al. (2005), we must determine the partial derivative of the concentration of hydrogen in terms of the hydrogen mass fraction gradient. Thoul et al. (1994) define the concentration gradient for a particular ion species as à !−1 µ ¶ X Zi X i X Z j X j ∂ ln X j ∂ lnC s ∂ ln X s = − . ∂r ∂r Ai Aj ∂r i j (B.6) If we assume a composition of only hydrogen and helium ions, then we may simplify the above equation and write it in terms of the hydrogen mass fraction gradient, ∂ lnC H ∂r = = = à !−1 µ ¶ X Zi X i X Z j X j ∂ ln X j ∂ ln X H − ∂r Ai Aj ∂r i j µ ¶ µ ¶ 2(1 − X ) −1 ∂ ln X 2(1 − X ) ∂ ln(1 − X ) ∂ ln X − X+ X + ∂r 4 ∂r 4 ∂r µ ¶ 2 1 ∂X ∂X 1 ∂X − − . X ∂r 1 + X ∂r 2 ∂r 297 Thus, we find the negative gradient of the logarithm of the concentration of hydrogen − 1 ∂X ∂l nC H =− . ∂r X (1 + X ) ∂r (B.7) Now, consider the concentration gradient of the metallicity, Z . Assume the gas is one of fully ionized iron atoms. This was assumed for the derivation of the gravitational settling of heavy elements currently employed in DSEP. By this assumption, ZF e = 26 and A F e = 56. Therefore, we can write !−1 µ à ¶ X Z j X j ∂ ln X j X Zi X i ∂ lnC Fe ∂ ln X Fe = − ∂r ∂r Ai Aj ∂r j i ¶ µ ∂ ln X Fe Z H X H ZHe X He ZFe X Fe −1 = · − + + ∂r AH A He A Fe µ ¶ Z H X H ∂ ln X H ZHe X He ∂ ln X He ZFe X Fe ∂ ln X Fe + + . AH ∂r A He ∂r A Fe ∂r Now, recall that X H = X , X He = Y and X Fe ∼ Z , where we may eliminate one of the mass fraction variables knowing that the total mass fraction of elements in the solar interior is equal to 1, Y = 1− X − Z. (B.8) Then, we have µ ¶ µ ¶ ∂ lnC Fe 1 ∂Z 2Y 26Z −1 ∂X 2 ∂Y 26 ∂Z = − X+ + + + ∂r Z ∂r 4 56 ∂r 4 ∂r 56 ∂r µ ¶ µ ¶ 2X + 1 − X − Z 13Z −1 ∂X 1 ∂X 1 ∂Z 13 ∂Z 1 ∂Z − + − − + = Z ∂r 2 28 ∂r 2 ∂r 2 ∂r 28 ∂r ¶ µ µ ¶ ∂Z 14 14 + 14X ∂X − . = 2 14Z + 14X Z − Z ∂r 14 + 14X − Z ∂r (B.9) We now make the approximation that the addition of Z in the denominator is negligible 298 compared to the value 14, in other words, Z ≪ 14. Therefore, 1 ∂Z 1 ∂X ∂ lnC F e = − ∂r Z ∂r 1 + X ∂r (B.10) We can now define the atomic diffusion coefficient, D H , in the diffusion equation from Richard et al. (2005) based on the equation presented by Bahcall & Loeb (1990). Since the gradient of the logarithm of the concentration is defined as above, comparing with Bahcall & Loeb (1990), we find · ¸ r 2 T 5/2 X X + 3 DH = − ln Λ/2.2 5X + 3 (B.11) As defined by Richard et al. (2005), the turbulent diffusion coefficient is · ρ D T = 400D He (T0 ) ρ(T0 ) ¸−3 (B.12) where T0 is selected in order to minimize Li depletion due to the diffusion of Li into the nuclear burning region, near T = 2.5 · 106 K . Since we are following the convention of Bahcall & Loeb (1990) in defining the diffusion coefficient for helium, we may replace the diffusion coefficient for helium with that of hydrogen evaluated at T0 , save a minus sign to indicate the two species are diffusing in opposite directions, D T = 400 · R 02 · T05/2 · X 0 ln Λ/2.2 · X0 + 3 5X 0 + 3 ¸· ρ ρ(T0 ) ¸−3 , (B.13) where R 0 and X 0 are the radius and the hydrogen mass fraction of the shell at the reference temperature T0 , respectively. We may now rewrite the atomic diffusion term (for hydrogen) of the diffusion equation with the new expressions for the diffusion coefficients, µ ¶ µ ¶µ ¶ #· ¸ R 02 · T05/2 · X 0 1 X0 + 3 ρ −3 ∂X r 2 T 5/2 X + 3 − , + 400 · vH = − ln Λ/2.2 5X + 3 ln Λ/2.2 X (5X 0 + 3) ρ(T0 ) 1 + X ∂r " (B.14) which we may further simplify following the sign convention of Richard et al. (2005) and 299 Thoul et al. (1994), · ¸" ¶ # µ ¶2 µ ¶5/2 µ ¶ µ X +3 T0 X 0 ρ(T0 ) 3 ∂X r 2 T 5/2 R0 . 1 − 400 · vH = ln Λ/2.2 (X + 1)(5X + 3) r T X ρ ∂r (B.15) To remain consistent with our initial assumption that hydrogen and helium diffuse at the same rate in opposite directions, we can say that for helium # · ¸" ¶ µ ¶5/2 µ r 2 T 5/2 ∂X X +3 ρ(T0 ) 3 T0 v He = −v H = . −1 400 ln Λ/2.2 (X + 1)(5X + 3) T ρ ∂r (B.16) Considering the modifications that must be made to DSEP, we already have the hydrogen, and thus helium, atomic diffusion coefficient defined. The variable cod2 is defined in the subroutine setup_grsett just as we have derived above, neglecting the turbulent diffusion term. Namely, · ¸ X +3 r 2 T 5/2 . cod20 = ln Λ/2.2 (X + 1)(5X + 3) Thus, to implement the turbulent diffusion coefficient, we modify the existing cod2 term, · ¸" ¶ # µ ¶5/2 µ r 2 T 5/2 X +3 ρ(T0 ) 3 T0 cod2 = 1 − 400 . ln Λ/2.2 (X + 1)(5X + 3) T ρ (B.17) We also need to know the derivative of the newly updated diffusion coefficient defined in DSEP, # · ¶ ¸" µ ¶5/2 µ ∂cod2 ρ(T0 ) 3 r 2 T 5/2 5X 2 + 30X + 21 T0 −1 . = 400 ∂X ln Λ/2.2 (5X 2 + 8X + 3)2 T ρ (B.18) These last two equations were included in the gravitational settling routine. New Control Variables ltdiff (.false.) – Logic variable controlling if turbulent mixing is included or not. turbt (6.0) – Real number variable that sets the logarithmic value for the reference temperature needed for turbulent mixing. 300 B.2 Internal Structure Constant We presented the theory behind the calculation of the internal structure constant, k 2 , in Chapter 3. The module to carry out the calculations in DMESTAR (apsidal) was designed to facilitate automatic computation of k 2 (apcon) at each model time step. This feature can be turned off, if necessary, using the new logic variable lapsid, but the code is setup to always compute k 2 . The only benefit of turning this feature off, in our opinion, is to save on computation time when computing a large grid of models where k 2 is of little interest. The value of k 2 at each model time step is printed out in the model track file (unit 37). The program also keep track of the run of the parameter η(r ), related to the deviation from sphericity as a function of the radius and density. File unit 77 contains this information. New Control Variables lapsid (.true.) – Logical variable controlling whether to compute the internal structure constant at each time step. B.3 Convective Overturn Timescale The convective overturn timescale was added to allow computation of theoretical Rossby numbers in Feiden & Chaboyer (2012a). A module was created to calculate the convective turnover time and has been fully integrated into the code. No logic variable was assigned to turn the feature on or off, it is permanently on. However, disabling this feature can be achieved by commenting our a single line of code in the primary driver main.f. Our implementation is similar to the method of Kim & Demarque (1996), who calculate the “local” convective overturn time at a distance above the base of the convection zone equal to one half of the mixing-length. This particular location was chosen based on the assumption that the tachocline (radiative-convective zone interface) is the source region of the stellar 301 magnetic dynamo (Parker 1975). However, using the stellar tachocline as our magnetic field source location for a fully convective star would be nonsensical. Therefore, in the case that the star is convective throughout, we utilize the results of Browning (2008) as a first approximation to the magnetic field source location. Browning found that the magnetic field strength within a fully convective star was at a maximum at a depth of 85% of the stellar radius. In accordance with this result, we compute the convective overturn time at one half the mixing-length above this location. The convective overturn time, in days, is output to the model track file (unit 37). B.4 Nuclear Reaction Rates An updated set of nuclear reaction cross sections was released in 2011 (Adelberger et al. 2011). All of the values recommended were rigorously vetted by the authors and should therefore represent the most accepted reaction cross sections. Table B.1 provides a comparison between the original Bahcall (1989), NACRE (Angulo et al. 1999), and the latest recommended values (Adelberger et al. 2011). We also quote the ratio between Adelberger et al. (2011) and Bahcall (1989), which is used as input for DSEP and DMESTAR (sstandard). B.5 Gough & Taylor Magnetic Inhibition Magnetic inhibition of convection has been introduced using the magnetic inhibition parameter of Mullan & MacDonald (2001), which is physically motivated by the analytical work of Gough & Tayler (1966). The radial profile is calculated in mm_mag. See also the more recent works by Mullan et al. (2007); MacDonald & Mullan (2009); Mullan & MacDonald (2010); MacDonald & Mullan (2010, 2012, 2013) for updated radial profiles and discussions of their methods. 302 Table B.1 Updated DSEP nuclear reaction rates. Reaction 1 H(p, e + ν)D 3 He(3 He, 2p)4 He 3 NACRE Fusion II Ratio 4.07E-22 3.94E-22 4.01E-22 0.9853 5150 5180 5210 1.0117 0.54 0.54 0.56 1.0370 C(p, γ)13 N 1.45 1.30 1.34 0.9241 1, 2 2 He( He, γ) Be 13 C(p, γ) N 5.50 7.00 7.6 1.3818 14 N(p, γ)15 O 3.32 3.20 1.66 0.5000 9.40 9.30 10.6 1.1277 0.0243 0.0210 0.0208 0.8560 8.0E-20 2.3E-20 8.6E-20 1.0750 16 7 3 Note 7 4 12 Bachall (1988) 14 17 O(p, γ) F 8 Be(p, γ) B + 4 He(p, e ν) He 3 (1) NACRE does not publish a 12 C rate. Estimated from their Figure X. (2) Values in Solar Fusion II (Adelberger et al. 2011) are taken from Solar Fusion I (Adelberger et al. 1998). (3) The 3 He + p reaction is not available from NACRE. Fusion I value was quoted instead. New Control Variables lgtmag (.false.) – Logical variable controlling whether to modify the adiabatic gradient by a small value, δMM . B.6 Electron Conduction Opacities Thermal electron conduction opacities were revised to reflect the improvements presented by Cassisi et al. (2007). Since the opacities are in tabular form, Alexander Potekchin wrote the necessary interpolation routines (see http://www.ioffe.ru/astro/conduct/). We cleaned up his original routines and integrated them into DSEP. Whether or not a given run uses the new opacities is controlled by a new logic variable. Furthermore, DMESTAR now implements a conduction opacity module thecond with the latest Cassisi et al. (2007) interpolation routines in thecond_int. 303 New Control Variables lcond07 (.true.) – Logic variable controlling whether to use the updated electron conduction opacities. B.7 τ=100 Boundary Condition Fit Point Models of fully convective stars with masses below 0.2M⊙ require boundary conditions that are prescribed at a deeper optical depth than higher mass models. This is because atmospheres of very-low-mass stars are highly non-adiabatic (Chabrier & Baraffe 1997; Chabrier & Baraffe 2000). Therefore, the specific properties used in mixing length theory are critically important. To avoid an incompatibility between the interior models and the atmosphere, integrating the atmosphere to a level where convection becomes nearly adiabatic is recommended. DSEP was tested with this type of atmosphere configuration, but lacked a diverse set of atmospheric boundary condition tables to make it a permanent option. We are presently compiling the necessary boundary condition tables with points defined at T = Teff and also where τ = 100, an optical depth of 100. DMESTAR can automatically select which set of boundary condition tables to use. However, the user can also specify with a new logic variable, lmatm. When set to .true., the code will attach the boundary conditions at τ = 100. New Control Variables latm (.false.) – Logic variable controlling whether to use a deeper atmosphere integration, suitable for very-low-mass models. 304 Bibliography Adams W. S., Joy A. H., 1920, Publications of the Astronomical Society of the Pacific, 32, 276 Adelberger E. G., Austin S. M., Bahcall J. N., Balantekin A. B., Bogaert G., Brown L. S., Buchmann L., Cecil F. E., Champagne A. E., de Braeckeleer L., Duba C. A., Elliott S. R., Freedman S. J., Gai M., Goldring G., Gould C. R., Gruzinov A., Haxton W. C., Heeger K. M., Henley E., Johnson C. W., Kamionkowski M., Kavanagh R. W., Koonin S. E., Kubodera K., Langanke K., Motobayashi T., Pandharipande V., Parker P., Robertson R. G., Rolfs C., Sawyer R. F., Shaviv N., Shoppa T. D., Snover K. A., Swanson E., Tribble R. E., Turck-Chièze S., Wilkerson J. F., 1998, Reviews of Modern Physics, 70, 1265 Adelberger E. G., García A., Robertson R. G. H., Snover K. A., Balantekin A. B., Heeger K., Ramsey-Musolf M. J., Bemmerer D., Junghans A., Bertulani C. A., Chen J.-W., Costantini H., Prati P., Couder M., Uberseder E., Wiescher M., Cyburt R., Davids B., Freedman S. J., Gai M., Gazit D., Gialanella L., Imbriani G., Greife U., Hass M., Haxton W. C., Itahashi T., Kubodera K., Langanke K., Leitner D., Leitner M., Vetter P., Winslow L., Marcucci L. E., Motobayashi T., Mukhamedzhanov A., Tribble R. E., Nollett K. M., Nunes F. M., Park T.-S., Parker P. D., Schiavilla R., Simpson E. C., Spitaleri C., Strieder F., Trautvetter H.-P., Suemmerer K., Typel S., 2011, Reviews of Modern Physics, 83, 195 Agrawal P. C., Riegler G. R., Garmire G. P., 1980, Monthly Notices of the Royal Astronomical Society, 192, 725 Albrecht S., Reffert S., Snellen I. A. G., Winn J. N., 2009, Nature, 461, 373 305 Alfvén H., 1942, Nature, 150, 405 Allard F., Hauschildt P. H., 1995, The Astrophysical Journal, 445, 433 Allard F., Hauschildt P. H., Alexander D. R., Starrfield S., 1997, Annual Review of Astronomy and Astrophysics, 35, 137 Allard F., Hauschildt P. H., Alexander D. R., Tamanai A., Schweitzer A., 2001, The Astrophysical Journal, 556, 357 Allard F., Homeier D., Freytag B., 2011, in Astronomical Society of the Pacific Conference Series, Vol. 448, 16th Cambridge Workshop on Cool Stars, Stellar Systems, and the Sun, Johns-Krull C., Browning M. K., West A. A., eds., p. 91 Aller L. H., Chapman S., 1960, The Astrophysical Journal, 132, 461 Alpher R. A., Bethe H., Gamow G., 1948, Physical Review, 73, 803 Andersen J., 1991, Astronomy and Astrophysics Review, 3, 91 Anderson R. I., Reiners A., Solanki S. K., 2010, Astronomy and Astrophysics, 522, A81 Angulo C., Arnould M., Rayet M., Descouvemont P., Baye D., Leclercq-Willain C., Coc A., Barhoumi S., Aguer P., Rolfs C., Kunz R., Hammer J. W., Mayer A., Paradellis T., Kossionides S., Chronidou C., Spyrou K., degl’Innocenti S., Fiorentini G., Ricci B., Zavatarelli S., Providencia C., Wolters H., Soares J., Grama C., Rahighi J., Shotter A., Lamehi Rachti M., 1999, Nuclear Physics A, 656, 3 Anosova Z. P., Orlov V. V., Chernyshev M. V., 1989, Soviet Astronomy Letters, 15, 3 Antia H. M., Chitre S. M., Thompson M. J., 2000, Astronomy and Astrophysics, 360, 335 —, 2003, Astronomy and Astrophysics, 399, 329 Antonopoulou E., 1987, Astrophysics and Space Science, 135, 335 306 Asplund M., Grevesse N., Sauval A. J., Scott P., 2009, Annual Review of Astronomy and Astrophysics, 1 Ayres T. R., 2009, The Astrophysical Journal, 696, 1931 Babcock H. D., 1959, The Astrophysical Journal, 130, 364 Babcock H. W., Babcock H. D., 1955, The Astrophysical Journal, 121, 349 Bahcall J., Pinsonneault M., Basu S., Christensen-Dalsgaard J., 1997, Physical Review Letters, 78, 171 Bahcall J. N., 1989, Neutrino astrophysics. Cambridge Univ. Press, New York, NY Bahcall J. N., Loeb A., 1990, The Astrophysical Journal, 360, 267 Bahcall J. N., Serenelli A. M., Basu S., 2005, The Astrophysical Journal, 621, L85 Baraffe I., Chabrier G., Allard F., Hauschildt P. H., 1997, Astronomy and Astrophysics, 327, 1054 Baraffe I., Chabrier G., Allard F., Hauschildt P. H., 1998, Astronomy and Astrophysics, 412, 403 Barden S. C., 1985, The Astrophysical Journal, 295, 162 Barnes J. R., Jeffers S. V., Jones H. R. A., 2011, Monthly Notices of the Royal Astronomical Society, 412, 1599 Barnes S. A., 2010, The Astrophysical Journal, 722, 222 Barrado y Navascués D., Fernández-Figueroa M. J., García-López R. J., De Castro E., Cornide M., 1997, Astronomy and Astrophysics, 326, 780 Bass G., Orosz J. A., Welsh W. F., Windmiller G., Ames Gregg T., Fetherolf T., Wade R. A., Quinn S. N., 2012, The Astrophysical Journal, 761, 157 307 Basu S., 1998, Monthly Notices of the Royal Astronomical Society, 298, 719 Basu S., Antia H. M., 1997, Monthly Notices of the Royal Astronomical Society, 287, 189 —, 2004, The Astrophysical Journal Letters, 606, L85 Basu S., Verner G. a., Chaplin W. J., Elsworth Y., 2012, The Astrophysical Journal, 746, 76 Batalha N. M., Rowe J. F., Bryson S. T., Barclay T., Burke C. J., Caldwell D. A., Christiansen J. L., Mullally F., Thompson S. E., Brown T. M., Dupree A. K., Fabrycky D. C., Ford E. B., Fortney J. J., Gilliland R. L., Isaacson H., Latham D. W., Marcy G. W., Quinn S. N., Ragozzine D., Shporer A., Borucki W. J., Ciardi D. R., Gautier III T. N., Haas M. R., Jenkins J. M., Koch D. G., Lissauer J. J., Rapin W., Basri G. S., Boss A. P., Buchhave L. A., Carter J. A., Charbonneau D., Christensen-Dalsgaard J., Clarke B. D., Cochran W. D., Demory B.O., Desert J.-M., Devore E., Doyle L. R., Esquerdo G. A., Everett M., Fressin F., Geary J. C., Girouard F. R., Gould A., Hall J. R., Holman M. J., Howard A. W., Howell S. B., Ibrahim K. A., Kinemuchi K., Kjeldsen H., Klaus T. C., Li J., Lucas P. W., Meibom S., Morris R. L., Prša A., Quintana E., Sanderfer D. T., Sasselov D., Seader S. E., Smith J. C., Steffen J. H., Still M., Stumpe M. C., Tarter J. C., Tenenbaum P., Torres G., Twicken J. D., Uddin K., Van Cleve J., Walkowicz L., Welsh W. F., 2013, The Astrophysical Journal Supplement Series, 204, 24 Bayless A. J., Orosz J. A., 2006, The Astrophysical Journal, 651, 1155 Bean J. L., Benedict G. F., Endl M., 2006, The Astrophysical Journal Letters, 653, L65 Bender C. F., Mahadevan S., Deshpande R., Wright J. T., Roy A., Terrien R. C., Sigurdsson S., Ramsey L. W., Schneider D. P., Fleming S. W., 2012, The Astrophysical Journal, 751, L31 Berger D. H., Gies D. R., McAlister H. A., ten Brummelaar T. A., Henry T. J., Sturmann J., Sturmann L., Turner N. H., Ridgway S. T., Aufdenberg J. P., Merand A., 2006, The Astrophysical Journal, 644, 475 308 Bessell M. S., 1991, The Astronomical Journal, 101, 662 Bjork S. R., Chaboyer B., 2006, The Astrophysical Journal, 641, 1102 Böhm-Vitense E., 1958, Zeitschrift für Astrophysik, 46, 108 Bonaca A., Tanner J. D., Basu S., Chaplin W. J., Metcalfe T. S., Monteiro M. J. P. F. G., Ballot J., Bedding T. R., Bonanno A., Broomhall A.-M., Bruntt H., Campante T. L., ChristensenDalsgaard J., Corsaro E., Elsworth Y., García R. A., Hekker S., Karoff C., Kjeldsen H., Mathur S., Régulo C., Roxburgh I., Stello D., Trampedach R., Barclay T., Burke C. J., Caldwell D. A., 2012, The Astrophysical Journal Letters, 755, L12 Bonfils X., Delfosse X., Udry S., Santos N. C., Forveille T., Ségransan D., 2005, Astronomy and Astrophysics, 442, 635 Borucki W. J., Koch D. G., Basri G., Batalha N., Boss A., Brown T. M., Caldwell D., Christensen-dalsgaard J. r., Cochran W. D., Devore E., W E., Dupree A. K., Iii T. N. G., Geary J. C., Gilliland R., Welsh W. F., Allen C., Bryson S. T., Buchhave L., Chandrasekaran H., Fressin F., Haas M., Horch E., Howard A., Isaacson H., Kolodziejczak J., Li J., Macqueen P., Meibom S. r., Prsa A., Quintana E. V., 2010, arXiv:1006.2799 Borucki W. J., Summers A. L., 1984, Icarus, 58, 121 Boyajian T. S., von Braun K., van Belle G., McAlister H. a., ten Brummelaar T. a., Kane S. R., Muirhead P. S., Jones J., White R., Schaefer G., Ciardi D., Henry T., López-Morales M., Ridgway S., Gies D., Jao W.-C., Rojas-Ayala B., Parks J. R., Sturmann L., Sturmann J., Turner N. H., Farrington C., Goldfinger P. J., Berger D. H., 2012, The Astrophysical Journal, 757, 112 Brandenburg A., Subramanian K., 2005, Physics Reports, 417, 1 Brogaard K., VandenBerg D., Bruntt H., Grundahl F., Frandsen S., Bedin L., Milone A., 309 Dotter A., Feiden G., Stetson P., Sandquist E., Miglio A., Stello D., Jessen-Hansen J., 2012, Astronomy and Astrophysics, 543, A106 Brooker R., Olle T., 1955, Monthly Notices of the Royal Astronomical Society, 115, 101 Brown B. P., Browning M. K., Brun A. S., Miesch M. S., Toomre J., 2010, The Astrophysical Journal, 711, 424 Browning M. K., 2008, The Astrophysical Journal, 676, 1262 Burrows A., Heng K., Nampaisarn T., 2011, The Astrophysical Journal, 736, 47 Caillault J.-P., 1982, The Astronomical Journal, 87, 558 Canuto V., 1970, The Astrophysical Journal, 159, 641 Canuto V. M., Mazzitelli I., 1991, The Astrophysical Journal, 370, 295 Carr R. B., 1967, The Astronomical Journal, 72, 788 Carter J. A., Fabrycky D. C., Ragozzine D., Holman M. J., Quinn S. N., Latham D. W., Buchhave L. a., Van Cleve J., Cochran W. D., Cote M. T., Endl M., Ford E. B., Haas M. R., Jenkins J. M., Koch D. G., Li J., Lissauer J. J., MacQueen P. J., Middour C. K., Orosz J. a., Rowe J. F., Steffen J. H., Welsh W. F., 2011, Science, 331, 562 Casagrande L., Flynn C., Portinari L., Girardi L., Jimenez R., 2007, Monthly Notices of the Royal Astronomical Society, 382, 1516 Cassisi S., Potekhin A. Y., Pietrinferni A., Catelan M., Salaris M., 2007, The Astrophysical Journal, 661, 1094 Catalano S., Biazzo K., Frasca A., Marilli E., 2002, Astronomy and Astrophysics, 394, 1009 Chaboyer B., Fenton W. H., Nelan J. E., Patnaude D. J., Simon F. E., 2001, The Astrophysical Journal, 562, 521 310 Chaboyer B., Kim Y.-C., 1995, The Astrophysical Journal, 454, 767 Chabrier G., Baraffe I., 1995, The Astrophysical Journal, 451, L29 —, 1997, Astronomy and Astrophysics, 327, 1039 Chabrier G., Baraffe I., 2000, Annual Review of Astronomy and Astrophysics, 38, 337 Chabrier G., Gallardo J., Baraffe I., 2007, Astronomy and Astrophysics, 472, L17 Chabrier G., Küker M., 2006, Astronomy and Astrophysics, 446, 1027 Chandrasekhar S., 1933, Monthly Notices of the Royal Astronomical Society, 93, 390 —, 1961, Hydrodynamic and hydromagnetic stability, International Series of Monographs on Physics. Clarendon, Oxford Charbonneau D., 2009, Proceedings of the International Astronomical Union, 253, 1 Charbonneau D., Berta Z. K., Irwin J., Burke C. J., Nutzman P., Buchhave L. A., Lovis C., Bonfils X., Latham D. W., Udry S., Murray-Clay R. A., Holman M. J., Falco E. E., Winn J. N., Queloz D., Pepe F., Mayor M., Delfosse X., Forveille T., 2009, Nature, 462, 891 Claret A., 1999, Astronomy and Astrophysics, 350, 56 Claret A., Giménez A. C., 1993, Astronomy and Astrophysics, 277, 487 Conroy C., van Dokkum P., 2012, The Astrophysical Journal, 747, 69 Coughlin J. L., López-Morales M., Harrison T. E., Ule N., Hoffman D. I., 2011, The Astronomical Journal, 141, 78 Cutri R. M., Skrutskie M. F., van Dyk S., Beichman C. A., Carpenter J. M., Chester T., Cambresy L., Evans T., Fowler J., Gizis J., Howard E., Huchra J., Jarrett T., Kopan E. L., Kirkpatrick J. D., Light R. M., Marsh K. A., McCallon H., Schneider S., Stiening R., Sykes 311 M., Weinberg M., Wheaton W. A., Wheelock S., Zacarias N., 2003, VizieR Online Data Catalog, 2246, 0 D’Antona F., Ventura P., Mazzitelli I., 2000, The Astrophysical Journal, 543, L77 Delfosse X., Forveille T., Mayor M., Burnet M., Perrier C., 1999, Astronomy and Astrophysics, 341, L63 Delfosse X., Forveille T., Perrier C., Mayor M., 1998, Astronomy & Astrophysics, 331, 581 Delfosse X., Forveille T., Ségransan D., Beuzit J.-L., Udry S., Perrier C., Mayor M., 2000, Astronomy and Astrophysics, 364, 217 Demarque P., Woo J., Kim Y., Yi S. K., 2004, The Astrophysical Journal Supplement Series, 155, 667 Demidov M. L., Zhigalov V. V., Peshcherov V. S., Grigoryev V. M., 2002, Solar Physics, 209, 217 Donati J.-F., Landstreet J. D., 2009, Annual Review of Astronomy and Astrophysics, 47, 333 Dotter A., Chaboyer B., Jevremović D., Baron E., Ferguson J. W., Sarajedini A., Anderson J., 2007, The Astronomical Journal, 134, 376 Dotter A., Chaboyer B., Jevremović D., Kostov V., Baron E., Ferguson J. W., 2008, The Astrophysical Journal Supplement Series, 178, 89 Dotter A. L., 2007, PhD thesis, Dartmouth College (Hanover, NH). Doyle J. G., Butler C. J., van den Oord G. H. J., Kiang T., 1990, Astronomy and Astrophysics, 232, 83 Doyle J. G., Mathioudakis M., 1990, Astronomy and Astrophysics, 227, 130 312 Doyle L. R., Carter J. A., Fabrycky D. C., Slawson R. W., Howell S. B., Winn J. N., Orosz J. A., PrËĞsa A., Welsh W. F., Quinn S. N., Torres G., Buchhave L. A., Marcy G. W., Fortney J. J., Jenkins J. M., Borucki W. J., Koch D., Middour C. K., Hall J. R., Mccauliff S., Fanelli M. N., Quintana E. V., Holman M. J., A D., Still M., Stefanik R. P., Brown W. R., Esquerdo G. A., 2011, Science, 333, 1602 Dressing C. D., Charbonneau D., 2013, The Astrophysical Journal, 767, 95 Eggen O. J., Sandage A., 1967, The Astrophysical Journal, 148, 911 Feiden G. A., Chaboyer B., 2012a, The Astrophysical Journal, 757, 42 —, 2012b, The Astrophysical Journal, 761, 30 Feiden G. A., Chaboyer B., Dotter A., 2011, The Astrophysical Journal Letters, 740, L25 Feigelson E. D., Babu G. J., 1992, The Astrophysical Journal, 397, 55 Ferguson J. W., Alexander D. R., Allard F., Barman T., Bodnarik J. G., Hauschildt P. H., Heffner-Wong A., Tamanai A., 2005, The Astrophysical Journal, 623, 585 Fisher G. H., Longcope D. W., Metcalf T. R., Pevtsov A. A., 1998, The Astrophysical Journal, 508, 885 Fleming T. A., Gioia I. M., Maccacaro T., 1989, The Astrophysical Journal, 340, 1011 Foukal P., Fowler L. A., Livshits M., 1983, The Astrophysical Journal, 267, 863 Foukal P., Fröhlich C., Spruit H., Wigley T. M. L., 2006, Nature, 443, 161 Gaidos E., 2013, arXiv:1301.2384 Gillon M., Bonfils X., Demory B.-O., Seager S., Deming D., Triaud A. H. M. J., 2010, Astronomy and Astrophysics, 525, A32 Giménez A., 1985, The Astrophysical Journal, 297, 405 313 Gizis J. E., 1997, The Astronomical Journal, 113, 806 Gough D. O., Tayler R. J., 1966, Monthly Notices of the Royal Astronomical Society, 133, 85 Grevesse N., Sauval A. J., 1998, Space Science Reviews, 85, 161 Güdel M., Audard M., Magee H., Franciosini E., Grosso N., Cordova F. A., Pallavicini R., Mewe R., 2001, Astronomy and Astrophysics, 365, L344 Güdel M., Schmitt J. H. M. M., Bookbinder J. A., Fleming T. A., 1993, The Astrophysical Journal, 415, 236 Guenther D. B., Demarque P., Kim Y., Pinsonneault M. H., 1992, The Astrophysical Journal, 387, 372 Gurnett D. A., Bhattacharjee A., 2005, Introduction to Plasma Physics. Cambridge University Press, Cambridge, UK Hale G. E., 1908, The Astrophysical Journal, 28, 315 Haro G., Chavira E., Gonzalez G., 1975, Information Bulletin on Variable Stars, 1031, 1 Harrington R. S., Dahn C. C., 1980, The Astronomical Journal, 85, 454 Hauschildt P. H., Allard F., Baron E., 1999a, The Astrophysical Journal, 512, 377 Hauschildt P. H., Allard F., Ferguson J., Baron E., Alexander D. R., 1999b, The Astrophysical Journal, 525, 871 Henry T. J., Jao W.-C., Subasavage J. P., Beaulieu T. D., Ianna P. A., Costa E., Méndez R. A., 2006, The Astronomical Journal, 132, 2360 Henyey L. G., Forbes J. E., Gould N. L., 1964, The Astrophysical Journal, 139, 306 Henyey L. G., Vardya M. S., Bodenheimer P., 1965, The Astrophysical Journal, 142, 841 314 Hoxie D. T., 1970, The Astrophysical Journal, 161, 1083 —, 1973, Astronomy and Astrophysics, 26, 437 Hubbard W. B., Lampe M., 1969, The Astrophysical Journal Supplement Series, 18, 297 Hussain G. A. J., Brickhouse N. S., Dupree A. K., Reale F., Favata F., Jardine M. M., 2012, Monthly Notices of the Royal Astronomical Society, 423, 493 Hut P., 1981, Astronomy and Astrophysics, 99, 126 Ibanoglu C., 1987, Astrophysics and Space Science, 139, 139 Iglesias C. A., Rogers F. J., 1996, The Astrophysical Journal, 464, 943 Irwin J., Charbonneau D., Berta Z. K., Quinn S. N., Latham D. W., Torres G., Blake C. H., Burke C. J., Esquerdo G. a., Fürész G., Mink D. J., Nutzman P., Szentgyorgyi A. H., Calkins M. L., Falco E. E., Bloom J. S., Starr D. L., 2009, The Astrophysical Journal, 701, 1436 Irwin J. M., Quinn S. N., Berta Z. K., Latham D. W., Torres G., Burke C. J., Charbonneau D., Dittmann J., Esquerdo G. A., Stefanik R. P., Observatory H., 2011, The Astrophysical Journal, 742, 123 Isobe T., Feigelson E. D., Akritas M. G., Babu G. J., 1990, The Astrophysical Journal, 364, 104 Jackson J. D., 1999, Classical Electrodynamics, 3rd edn. John Wiley & Sons, Hoboken, NJ Jackson R. J., Jeffries R. D., 2013, Monthly Notices of the Royal Astronomical Society Jackson R. J., Jeffries R. D., Maxted P. F. L., 2009, Monthly Notices of the Royal Astronomical Society, 339, L89 Johns-Krull C. M., 2007, The Astrophysical Journal, 664, 975 Johnson J. A., Apps K., 2009, The Astrophysical Journal, 699, 933 315 Joy A. H., Sanford R. F., 1926, The Astrophysical Journal, 64, 250 Kasting J. F., 1988, Icarus, 74, 472 —, 1991, Icarus, 94, 1 Kasting J. F., Whitmire D. P., Reynolds R. T., 1993, Icarus, 101, 108 Kim Y.-C., Demarque P., 1996, The Astrophysical Journal, 457, 340 Kirkpatrick J. D., Henry T. J., McCarthy Jr. D. W., 1991, The Astrophysical Journal Supplement Series, 77, 417 Kjurkchieva D. P., Marchev D. V., Heckert P. A., Ordway J. I., 2005, The Astronomical Journal, 129, 1084 Knigge C., Baraffe I., Patterson J., 2011, The Astrophysical Journal Supplement Series, 194, 28 Kochukhov O., Makaganiuk V., Piskunov N., Snik F., Jeffers S. V., Johns-Krull C. M., Keller C. U., Rodenhuis M., Valenti J. a., 2011, The Astrophysical Journal, 732, L19 Kopal Z., ed., 1978, Astrophysics and Space Science Library, Vol. 68, Dynamics of close binary systems. Reidel, Dordrecht Kopparapu R. K., Ramirez R., Kasting J. F., Eymet V., Robinson T. D., Mahadevan S., Terrien R. C., Domagal-Goldman S., Meadows V., Deshpande R., 2013, The Astrophysical Journal, 765, 131 Korn A. J., Grundahl F., Richard O., Mashonkina L., Barklem P. S., Collet R., Gustafsson B., Piskunov N., 2007, The Astrophysical Journal, 671, 402 Kraus A. L., Tucker R. A., Thompson M. I., Craine E. R., Hillenbrand L. A., 2011, The Astrophysical Journal, 728, 48 316 Kron G. E., 1952, The Astrophysical Journal, 115, 301 Kulsrud R. M., 2004, Plasma Physics for Astrophysics. Princeton Univ. Press, Princeton, NJ Kumar S. S., 1963, The Astrophysical Journal, 137, 1121 Kuznetsov M. K., Pavlenko Y. V., Jones H., Pinfield D. J., 2012, Advances in Astronomy and Space Physics, 2, 15 Lacy C. H., 1977, The Astrophysical Journal, 218, 444 Lastennet E., Fernandes J., Oblak E., 2003, Astronomy & Astrophysics, 409, 611 Laughlin G., Bodenheimer P., Adams F. C., 1997, The Astrophysical Journal, 482, 420 Lee J. W., Kim S.-L., Lee C.-U., Lee B.-C., Park B.-G., Hinse T. C., 2013, The Astrophysical Journal, 763, 74 Leggett S. K., Allard F., Hauschildt P. H., 1998, The Astrophysical Journal, 509, 836 Lépine S., Gaidos E., 2011, The Astronomical Journal, 142, 138 Leung K.-C., Schneider D. P., 1978, The Astronomical Journal, 83, 618 Li L. H., Sofia S., 2001, The Astrophysical Journal, 549, 1204 Libbrecht K. G., Woodard M. F., 1990, Nature, 345, 779 Limber D. N., 1958a, The Astrophysical Journal, 127, 363 —, 1958b, The Astrophysical Journal, 127, 387 Liu Q., Yang Y., Gu S., Hang Z., 1996, Information Bulletin on Variable Stars, 4326, 1 López-Morales M., 2007, The Astrophysical Journal, 660, 732 López-Morales M., Orosz J. A., Shaw J. S., Havelka L., 2006, NSVS01031772: A New 0.50+0.54 M ⣏ Detached Eclipsing Binary, arXiv:astro-ph/0610225 317 López-Morales M., Ribas I., 2005, The Astrophysical Journal, 631, 1120 López-Morales M., Shaw J. S., 2007, in ASP Conference Series, Vol 362: The Seventh Pacific Rim Conference on Stellar Astrophysics, Vol. 362, pp. 26–32 Lydon T. J., Sofia S., 1995, The Astrophysical Journal, 101, 357 MacDonald J., Mullan D. J., 2009, The Astrophysical Journal, 700, 387 —, 2010, The Astrophysical Journal, 723, 1599 —, 2012, Monthly Notices of the Royal Astronomical Society, 421, 3084 —, 2013, The Astrophysical Journal, 765, 126 Mann A. W., Brewer J. M., Gaidos E., Lépine S., Hilton E. J., 2013a, The Astronomical Journal, 145, 52 Mann A. W., Gaidos E., Kraus A., Hilton E. J., 2013b, arXiv: 1304.7269 Martins D. H., 1975, Publications of the Astronomical Society of the Pacific, 87, 163 Mathieu R. D., Baraffe I., Simon M., Stassun K. G., White R., 2007, in Protostars & Planets V, pp. 411–425 Mathis S., Zahn J.-P., 2005, Astronomy and Astrophysics, 440, 653 McLaughlin D. B., 1924, The Astrophysical Journal, 60, 22 McLean M., Berger E., Reiners A., 2012, The Astrophysical Journal, 746, 23 Mestel L., 1999, International Series of Monographs on Physics, Vol. 154, Stellar magnetism, 2nd edn. Clarendon, Oxford, UK Metcalfe T. S., Mathieu R. D., Latham D. W., Torres G., 1996, The Astrophysical Journal, 456, 356 318 Michaud G., Fontaine G., Beaudet G., 1984, The Astrophysical Journal, 282, 206 Mihalas D., 1978, Stellar atmospheres, 2nd edn. W. H. Freeman and Co., San Francisco, CA Mohanty S., Basri G., Shu F., Allard F., Chabrier G., 2002, The Astrophysical Journal, 571, 469 Montes D., de Castro E., Fernandez-Figueroa M. J., Cornide M., 1995a, Astronomy and Astrophysics Supplement, 114, 287 Montes D., Fernandez-Figueroa M. J., de Castro E., Cornide M., 1995b, Astronomy and Astrophysics, 294, 165 Morales J. C., Gallardo J. J., Ribas I., Jordi C., Baraffe I., Chabrier G., 2010, The Astrophysical Journal, 718, 502 Morales J. C., Ribas I., Jordi C., 2008, Astronomy & Astrophysics, 478, 507 Morales J. C., Ribas I., Jordi C., Torres G., Gallardo J., Guinan E. F., Charbonneau D., Wolf M., Latham D. W., Anglada-Escudé G., Bradstreet D. H., Everett M. E., O’Donovan F. T., Mandushev G., Mathieu R. D., 2009a, The Astrophysical Journal, 691, 1400 Morales J. C., Torres G., Marschall L. A., Brehm W., 2009b, The Astrophysical Journal, 707, 671 Morin J., Donati J.-F., Petit P., Delfosse X., Forveille T., Albert L., Aurière M., Cabanac R., Dintrans B., Fares R., Gastine T., Jardine M. M., Lignières F., Paletou F., Ramirez Velez J. C., Théado S., 2008, Monthly Notices of the Royal Astronomical Society, 390, 567 Morin J., Donati J.-F., Petit P., Delfosse X., Forveille T., Jardine M. M., 2010, Monthly Notices of the Royal Astronomical Society, 407, 2269 Muirhead P. S., Edelstein J., Erskine D. J., Wright J. T., Muterspaugh M. W., Covey K. R., Wishnow E. H., Hamren K., Andelson P., Kimber D., Mercer T., Halverson S. P., Vander319 burg A., Mondo D., Czeszumska A., Lloyd J. P., 2011, Publications of the Astronomical Society of the Pacific, 123, 709 Muirhead P. S., Hamren K., Schlawin E., Rojas-Ayala B., Covey K. R., Lloyd J. P., 2012, The Astrophysical Journal Letters, 750, L37 Mullan D. J., MacDonald J., 2001, The Astrophysical Journal, 559, 353 —, 2010, The Astrophysical Journal, 713, 1249 Mullan D. J., MacDonald J., Townsend R. H. D., 2007, The Astrophysical Journal, 670, 1420 Neckel H., 1995, Solar Physics, 156, 7 Noyes R. W., Hartmann L. W., Baliunas S. L., Duncan D. K., Vaughan A. H., 1984, The Astrophysical Journal, 279, 763 Nutzman P., Charbonneau D., 2008, Publications of the Astronomical Society of the Pacific, 120, 317 O’Neal D., 2006, The Astrophysical Journal, 645, 659 O’Neal D., Neff J. E., Saar S. H., 1998, The Astrophysical Journal, 507, 919 O’Neal D., Neff J. E., Saar S. H., Cuntz M., 2004, The Astronomical Journal, 128, 1802 Orosz J. A., Welsh W. F., Carter J. A., Brugamyer E., Buchhave L. A., Cochran W. D., Endl M., Ford E. B., MacQueen P., Short D. R., Torres G., Windmiller G., Agol E., Barclay T., Caldwell D. A., Clarke B. D., Doyle L. R., Fabrycky D. C., Geary J. C., Haghighipour N., Holman M. J., Ibrahim K. A., Jenkins J. M., Kinemuchi K., Li J., Lissauer J. J., Prša A., Ragozzine D., Shporer A., Still M., Wade R. A., 2012, The Astrophysical Journal, 758, 87 Osten R. A., Hawley S. L., Allred J. C., Johns-Krull C. M., Roark C., 2005, The Astrophysical Journal, 621, 398 320 Osterbrock D. E., 1953, The Astrophysical Journal, 118, 529 Pallavicini R., Golub L., Rosner R., Vaiana G. S., Ayres T., Linsky J. L., 1981, The Astrophysical Journal, 248, 279 Parker E. N., 1955, The Astrophysical Journal, 122, 293 —, 1975, The Astrophysical Journal, 198, 205 Parker E. N., 1979, Cosmical magnetic fields: Their origin and their activity. Oxford Univ. Press, New York, NY Parsons S. G., Marsh T. R., Gänsicke B. T., Rebassa-Mansergas A., Dhillon V. S., Littlefair S. P., Copperwheat C. M., Hickman R. D. G., Burleigh M. R., Kerry P., Koester D., Nebot Gómez-Morán A., Pyrzas S., Savoury C. D. J., Schreiber M. R., Schmidtobreick L., Schwope A. D., Steele P. R., Tappert C., 2012, Monthly Notices of the Royal Astronomical Society, 420, 3281 Peimbert M., Luridiana V., Peimbert A., 2007, The Astrophysical Journal, 666, 636 Pevtsov A. A., Fisher G. H., Acton L. W., Longcope D. W., Johns-krull C. M., Kankelborg C. C., Metcalf T. R., 2003, The Astrophysical Journal, 598, 1387 Phan-Bao N., Lim J., Donati J.-F., Johns-Krull C. M., Martín E. L., 2009, The Astrophysical Journal, 704, 1721 Piotto G., Villanova S., Bedin L. R., Gratton R., Cassisi S., Momany Y., Recio-Blanco A., Lucatello S., Anderson J., King I. R., Pietrinferni A., Carraro G., 2005, The Astrophysical Journal, 621, 777 Piskunov N. E., Kupka F., Ryabchikova T. A., Weiss W. W., Jeffery C. S., 1995, Astronomy and Astrophysics Supplement, 112, 525 321 Planck Collaboration, Ade P. A. R., Aghanim N., Armitage-Caplan C., Arnaud M., Ashdown M., Atrio-Barandela F., Aumont J., Baccigalupi C., Banday A. J., et al., 2013, arXiv: 1303.5076 Popper D. M., 1976, Information Bulletin on Variable Stars, 1083, 1 —, 1980, Annual Review of Astronomy and Astrophysics, 18, 115 —, 1984, The Astronomical Journal, 89, 132 —, 1997, The Astronomical Journal, 114, 1195 Prather M. J., 1976, PhD thesis, Yale University (New Haven, CT). Proffitt C. R., Michaud G., 1991, The Astrophysical Journal, 380, 238 Prša A., Batalha N., Slawson R. W., Doyle L. R., Welsh W. F., Orosz J. A., Seager S., Rucker M., Mjaseth K., Engle S. G., Conroy K., Jenkins J., Caldwell D., Koch D., Borucki W., 2011, The Astronomical Journal, 141, 83 Qian S.-B., Zhang J., Zhu L.-Y., Liu L., Liao W.-P., Zhao E.-G., He J.-J., Li L.-J., Li K., Dai Z.-B., 2012, Monthly Notices of the Royal Astronomical Society, 423, 3646 Reid I. N., Hawley S. L., Gizis J. E., 1995, The Astronomical Journal, 110, 1838 Reiners A., 2012, Living Reviews in Solar Physics, 8, 1 Reiners A., Basri G., 2007, The Astrophysical Journal, 656, 1121 —, 2009, Astronomy and Astrophysics, 496, 787 —, 2010, The Astrophysical Journal, 710, 924 Reiners A., Basri G., Browning M., 2009, The Astrophysical Journal, 692, 538 Reiners A., Joshi N., Goldman B., 2012, The Astronomical Journal, 143, 93 322 Reiners A., Mohanty S., 2012, The Astrophysical Journal, 746, 43 Ribas I., 2003, Astronomy and Astrophysics, 398, 239 Ribas I., 2006, Astrophysics and Space Science, 304, 89 Ribeiro T., Baptista R., Kafka S., 2011, The Astronomical Journal, 142, 106 Richard O., Michaud G., Richer J., 2005, The Astrophysical Journal, 619, 538 Richer J., Michaud G., Turcotte S., 2000, The Astrophysical Journal, 529, 338 Rogers F. J., Nayfonov A., 2002, The Astrophysical Journal, 576, 1064 Rojas-Ayala B., Covey K. R., Muirhead P. S., Lloyd J. P., 2010, The Astrophysical Journal Letters, 720, L113 —, 2012, The Astrophysical Journal, 748, 93 Rossiter R. A., 1924, The Astrophysical Journal, 60, 15 Saar S. H., 1990, in IAU Symposium 138: Solar Photosphere: Structure, Convection, and Magnetic Fields, Stenflo J. O., ed., pp. 427–441 —, 1996, in IAU Symposium 176: Stellar Surface Structure, pp. 237–243 —, 2001, in Astronomical Society of the Pacific Conference Series, Vol. 223, 11th Cambridge Workshop on Cool Stars, Stellar Systems and the Sun, Garcia Lopez R. J., Rebolo R., Zapaterio Osorio M. R., eds., p. 292 Saar S. H., Donahue R. A., 1997, The Astrophysical Journal, 485, 319 Saumon D., Chabrier G., Van Horn H., 1995, The Astrophysical Journal Supplement Series, 99, 713 Schlaufman K. C., Laughlin G., 2010, Astronomy and Astrophysics, 519, A105 323 Schmitt J. H. M. M., Fleming T. A., Giampapa M. S., 1995, The Astrophysical Journal, 450, 392 Schrijver C. J., Cote J., Zwaan C., Saar S. H., 1989, The Astrophys, 337, 964 Seager S., Mallén-Ornelas G., 2003, The Astrophysical Journal, 585, 1038 Ségransan D., Delfosse X., Forveille T., Beuzit J.-L., Udry S., Perrier C., Mayor M., 2000, Astronomy and Astrophysics, 364, 665 Shulyak D., Seifahrt A., Reiners A., Kochukhov O., Piskunov N., 2011, Monthly Notices of the Royal Astronomical Society, 418, 2548 Skumanich A., 1972, The Astrophysical Journal, 171, 565 Skumanich A., Smythe C., Frazier E. N., 1975, The Astrophysical Journal, 200, 747 Smith M. A., 1974, The Astrophysical Journal, 189, 101 Soderblom D. R., 2010, Annual Review of Astronomy and Astrophysics, 48, 581 Spada F., Demarque P., 2012, Monthly Notices of the Royal Astronomical Society, 422, 2255 Spruit H. C., 1977, Solar Physics, 55, 3 —, 1982a, Astronomy and Astrophysics, 108, 348 —, 1982b, Astronomy and Astrophysics, 108, 356 Spruit H. C., Weiss A., 1986, Astronomy and Astrophysics, 166, 167 Stein R. F., Brandenburg A., Nordlund A., 1992, in Astronomical Society of the Pacific Conference Series, Vol. 26, Cool Stars, Stellar Systems, and the Sun, Giampapa M. S., Bookbinder J. A., eds., p. 148 Stelzer B., Burwitz V., Audard M., Güdel M., Ness J.-U., Grosso N., Neuhäuser R., Schmitt J. H. M. M., Predehl P., Aschenbach B., 2002, Astronomy and Astrophysics, 392, 585 324 Stempels H. C., Hebb L., Stassun K. G., Holtzman J., Dunstone N., Glowienka L., Frandsen S., 2008, Astronomy and Astrophysics, 755, 747 Stempels H. C., Piskunov N., Barklem P. S., 2001, in Astronomical Society of the Pacific Conference Series, Vol. 223, 11th Cambridge Workshop on Cool Stars, Stellar Systems and the Sun, Garcia Lopez R. J., Rebolo R., Zapaterio Osorio M. R., eds., p. 878 Stothers R. B., 1974, The Astrophysical Journal, 194, 651 Sweigart A. V., 1973, Astronomy and Astrophysics, 24, 459 Terrien R. C., Fleming S. W., Mahadevan S., Deshpande R., Feiden G. A., Bender C. F., Ramsey L. W., 2012, The Astrophysical Journal Letters, 760, L9 Thompson W. B., 1951, Philosophical Magazine, 42, 1417 Thoul A. A., Bahcall J. N., Loeb A., 1994, The Astrophysical Journal, 421, 828 Torres G., 2007, The Astrophysical Journal Letters, 671, L65 —, 2013, Astronomische Nachrichten, 334, 4 Torres G., Andersen J., Giménez A., 2010, The Astronomy and Astrophysics Review, 18, 67 Torres G., Ribas I., 2002, The Astrophysical Journal, 567, 1140 Trampedach R., Stein R. F., 2011, The Astrophysical Journal, 731, 78 Vaiana G. S., Cassinelli J. P. J., Fabbiano G., Giacconi R., Golub L., Gorenstein P., Haisch B. M., Harnden F. R., Johnson H. M., Linsky J. L., Maxson C. W., Mewe R., Rosner R., Seward F., Topka K., Zwaan C., 1981, The Astrophysical Journal, 245, 163 Valenti J. A., Piskunov N., 1996, Astronomy and Astrophysics Supplement, 118, 595 Van Cleve J. E., Caldwell D. A., 2009, Kepler Instrument Handbook No. KSCI-19033. NASA / Ames Research Center, Moffett Field, CA 325 van Dokkum P. G., Conroy C., 2010, Nature, 1 van Gent H., 1926, Bulletin of the Astronomical Institutes of the Netherlands, 3, 121 van Leeuwen F., 2007, Astronomy and Astrophysics, 474, 653 van Saders J. L., Pinsonneault M. H., 2012, The Astrophysical Journal, 751, 98 Vida K., Oláh K., Kővári Z., Korhonen H., Bartus J., Hurta Z., Posztobányi K., 2009, Astronomy and Astrophysics, 504, 1021 Vitense E., 1953, Zeitschrift für Astrophysik, 32, 135 Viti S., Jones H., Maxted P., Tennyson J., 2002, Monthly Notices of the Royal Astronomical Society, 329, 290 Viti S., Jones H. R. A., Schweitzer A., Allard F., Hauschildt P. H., Tennyson J., Miller S., Longmore A. J., 1997, Monthly Notices of the Royal Astronomical Society, 291, 780 Voges W., Aschenbach B., Boller T., Bräuninger H., Briel U., Burkert W., Dennerl K., Englhauser J., Gruber R., Haberl F., Hartner G., Hasinger G., Kürster M., Pfeffermann E., Pietsch W., Predehl P., Rosso C., Schmitt J. H. M. M., Trümper J. R., Zimmermann H. U., 1999, Astronomy & Astrophysics, 349, 389 Voges W., Aschenbach B., Boller T., Brauninger H., Briel U., Burkert W., Dennerl K., Englhauser J., Gruber R., Haberl F., Hartner G., Hasinger G., Pfeffermann E., Pietsch W., Predehl P., Schmitt J., Trumper J., Zimmermann U., 2000, VizieR Online Data Catalog, 9029, 0 von Braun K., Boyajian T. S., Kane S. R., Hebb L., van Belle G. T., Farrington C., Ciardi D. R., Knutson H. A., ten Brummelaar T. A., López-Morales M., McAlister H. A., Schaefer G., Ridgway S., Collier Cameron A., Goldfinger P. J., Turner N. H., Sturmann L., Sturmann J., 2012, The Astrophysical Journal, 753, 171 326 Vos J., Clausen J. V., Jorgensen U. G., Ostensen R. H., Claret A., Hillen M., Exter K., 2012, Astronomy and Astrophysics, 540, 64 Walkowicz L. M., Hawley S. L., 2009, The Astronomical Journal, 137, 3297 Weis E. W., 1991, The Astronomical Journal, 101, 1882 Weiss A., Hillebrandt W., Thomas H.-C., Ritter H., 2004, Cox and Giuli’s Principles of Stellar Structure, 2nd edn. Cambridge Scientific Publishers, Cambridge, UK Wende S., Reiners A., Ludwig H.-G., 2009, Astronomy and Astrophysics, 508, 1429 Windmiller G., Orosz J. A., Etzel P. B., 2010, The Astrophysical Journal, 712, 1003 Winn J. N., Albrecht S., Johnson J. A., Torres G., Cochran W. D., Marcy G. W., Howard A., Isaacson H., 2011, The Astrophysical Journal Letters, 741, L1 Woolf V. M., Wallerstein G., 2005, Monthly Notices of the Royal Astronomical Society, 356, 963 Wright N. J., Drake J. J., Mamajek E. E., Henry G. W., 2011, The Astrophysical Journal, 743, 48 Yang H., Johns-Krull C. M., 2011, The Astrophysical Journal, 729, 83 Young A., Skumanich A., Stauffer J. R., Harlan E., Bopp B. W., 1989, The Astrophysical Journal, 344, 427 Zahn J.-P., 1977, Astronomy & Astrophysics, 57, 383 Zasche P., 2012, Acta Astronomica, 62, 97 Zwicky F., 1966, Science, 153, 53 327