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
Introduction to exoplanetology Lesson #6: Direct exoplanet detection methods (1/2) Olivier Absil [email protected] Outline I. Direct detection: why and how? II. High contrast imaging from the ground III. High contrast imaging from space IV. Main results from high-contrast imaging V. Stellar interferometry I. Direct detection Why and how? Why direct detection? • Characterization of planetary atmospheres • Needs spectroscopy on actual planetary photons → isolate planet from star Why direct detection? Transits mostly here • Access non-transiting planets • Opens up a wide range of separations (beyond a few 0.1 AU) • Complementary with transit spectroscopy Why direct detection? • Get full orbital solution • 3D orbit —> direct, model-independent access to the planet mass • Study dynamics of planetary systems, interactions with dust disks, etc. Challenge #1: contrast • • Visible: reflected light Infrared: thermal (blackbody-type) emission Challenge #2: angular separation • • • 1 AU @ 10 pc = 0.1 arcsec Theoretically within reach of 10-m class telescope But … watch out for turbulence! Challenge summary A firefly close to a lighthouse … 1000 miles away! (note: the star never turns off) Technique #1: imaging • • Diffraction in circular aperture → Airy pattern Angular resolution = size of Airy disk: θ = 1.22 λ/D • λ = 2 µm, D = 10 m → θ = 0.05" (50 mas) = 1 AU at 20 pc • Extended pattern ≡ noise! Rayleigh criterion The Airy pattern : aperture radius θ = 1.22 λ/D Technique #2: interferometry • Two separated telescopes → fringes • Angular resolution set by baseline (B): θ = 0.5 λ/B • Crests of 1st packet fall on troughs of 2nd packet • λ = 2 µm, B = 100 m → θ = 2 mas B ang58 m. an d a ime tion 8–30 The fringe pattern unresolved binary resolved binary nnel ð19Þ (fringe packets offset for clarity) Fig. 5 An illustration of the optical Ramsey method. A wave packet oscillates between the Rydberg states, n, an II. High contrast imaging from the ground Atmospheric windows V IJ HK L M N Imaging through the Earth atmosphere • Temperature variations Short exposure • Distorted wavefront • Short exposure: speckles • Long exposure: wide PSF Long exposure Loss of angular resolution La Palma • Fried parameter r0: diameter of a circular area over which the rms wavefront aberration due to passage through the atmosphere is equal to 1 radian • r0 ~ 10 cm at good astronomy site • Same resolution as 10 cm telescope! D = 5 cm 10 cm 60 cm 1.2 m Atmospheric turbulence • • Wavelength dependence: r0 ∝ λ6/5 • 10 cm @ 500 nm • 50 cm @ 2 µm • 4 m @ 10 µm Seeing = FWHM of long exposure image • Equal to 0.98 λ / r0 • 1" seeing for r0 = 10 cm @ 500 nm • Varies slowly with wavelength (λ−1/5) Atmospheric turbulence • • Coherence time: t0 = 0.31 r0 / ⟨vwind⟩ • Valid under Taylor’s « frozen turbulence » hypothesis • t0 ≃ 3 msec for r0 = 10 cm and ⟨vwind⟩ = 10 m/s Isoplanatic angle: θ0 = 0.31 r0 / ⟨h⟩ • θ0 ≃ 1.3" for r0 = 10 cm and ⟨h⟩ = 5 km • Stars separated by θ0 have different short-exposure PSFs Correction needed! Adaptive optics Adaptive optics Wavefront sensing Shack-Hartmann wave front sensor Strehl ratio • • S = |⟨exp(iφ)⟩|2 ≃ exp(−σφ2) • φ = wavefront phase • σφ = rms phase on pupil Quantifies image quality • ≃ peak intensity ratio wrt perfect image • Perfect image → S = 1 • D = r0 → S = 0.36 AO correction MACAO on VLT 50% 20% Detection in speckle noise? ESO-3.6m/ComeOn+ (1994) S=0.21 Planet Speckles Getting rid of speckles 1. Observing strategies Solution: PSF subtraction • • aka Reference-star Differential Imaging (RDI) β Pic Observe a reference star • Similar to science target • Scale its PSF (flux) • Subtract it from science observation α Pic β Pic − α Pic Limitations of RDI • No perfect reference star • Should be of same magnitude, color and position as science target • Time spent on reference star is « lost » (no planetary photon) • Atmospheric conditions change with time • Telescope/instrument aberrations also change with time • Quasi-static speckles, can mimic planetary signal Four differential solutions • Keep observing the same target (no reference star) • Tune an observing parameter to discriminate between stellar PSF and planet • Spectral differential imaging (SDI) • Spectral deconvolution (SD) • Angular differential imaging (ADI) • Polarimetric differential imaging (PDI) Spectral differential imaging SDI in practice λ1 λ2 λ1−λ2 λ3 (λ1−λ2) − k(λ1−λ3) Pros and cons of SDI + Detection down to diffraction limit (1 λ/D) − Companion needs strong molecular absorption (e.g., methane) − Differential aberrations between the three wavelength optical paths − Narrow filters → lower sensitivity Spectral deconvolution • Integral Field Spectrograph (IFS) • • Diffraction and speckle pattern scale as function of wavelength • • Provides field image as function of wavelength (« image cube ») Pattern moves out from star with increasing wavelength Exoplanet position is fixed • Can be distinguished from speckles SD in practice Similar principle as SDI, on much wider wavelength band Observed (x,λ) slice Rescaled (x,λ) slice Pros and cons of SD + Works with any type of planet (no specific feature) + No differential aberrations / simultaneous observations + Does not rely on specific feature in planet spectrum + End product = spectrum of the planet! • Detect and characterize planet at the same time − Speckle pattern not perfectly constant over wavelength − Limited inner and outer working angles (depend on wavelength range and spectral resolution) Angular differential imaging • Use field rotation while keeping telescope fixed • Usually done by switching off derotator • Planet moves around star as a function of parallactic angle • Quasi-static speckles stay at fixed position ADI in practice Pros and cons of ADI + Works with any type of planet (no specific feature) + Does not require specific hardware − Does not work well for stars far from zenith (small variation of the parallactic angle) − Limited inner working angle (planet must move by more than 1 λ/D in the field for ADI to work) − Speckle pattern evolves with time Speckle decorrelation t0 t0 + 10 min (t0 + 10 min) − t0 t0 + 100 min (t0 + 100 min) − t0 Polarimetric differential imaging • Reflected light from planet is partially polarized • Typically 10% polarisation • Star produces unpolarized light • Can be exploited by polarimetric imager 8 PDI in practice Chapter 1 H. Canovas et al.: Near-infrared imaging polarimetry of HD 142527 Figure 1.4: Schematic view of the polarization measurement process using temporal and spatial modulation used by ExPo. The modulator switches between orthogonal directions of the polarisation to be measured, defining states A and B. The polarising beamsplitter then separates the two orthogonal polarisation states into left (L) and right (R) beams. Double subtraction of (AL-AR)-(BL-BR) minimises (systematic) spatial and temporal differences (Canovas et al. 2012a). Image credit: M. Rodenhuis way of modulating but one can design, and optimise, many different modulation schemes depending on what Stokes component (or cobination of them) one is interested in measuring. More generally, instead of using a rotating polariser, the modulation is achieved by means of a retarding element (i.e. modulator) and a fixed polariser (i.e. analyser). A retarding element imparts a phase lag to one of the orthognal components of the electric field, thus, modifying the polarisation state of the wave passing through. by carefully orienting the retarder in the required direction we can transform the polarisation state we want to measure in the direction fixed polariser filter the Fig.the 1. Processed images of HDwould 142527 (top row)toand HDdetector. 161743 (bottom row) in H band. From left to right: intensity image (I) in logarithmic Pros and cons of PDI + Speckle subtraction can be very good + No limitation in inner or outer working angle − Small fraction of planet light is polarized → low sensitivity − Works only in reflected light (best in visible range) − Requires specific, non-standard hardware Getting rid of speckles 2. Image processing The LOCI algorithm coefficients to be determined • Locally Optimized Combination of Images • Goal: make best use of a set of reference images • Can be used with various observing techniques • Solve a linear system to minimize residuals ref PSF to be minimized pixels target PSF individual frames LOCI working with ADI Practical use of LOCI • To be applied locally because correlation between frames expected to depend on position • Optimize LOCI coefficient on « optimization » zone • Perform PSF subtraction on « subtraction » zone • Optim. zone > subtr. zone to avoid planet signal subtraction Pros and cons of LOCI • − − Several free parameters + Can be fine tuned to any particular data set − Not very straightforward to use Strong self-subtraction of planetary signal • LOCI tries to remove everything → bias on photometry • Can be evaluated with fake companions, or mitigated by introducing even more parameters (masking) CPU intensive Principal Component Analysis (PCA) • Method to reduce the number of variables needed to describe a data set • Build an orthogonal basis on which to decompose the images • Truncated basis to perform PSF subtraction How PCA works • Subtract mean (M) from data set (D): A = D − M • Calculate covariance matrix: C = AAT • Calculate eigenvectors (Q) and eigenvalues • CQ = QΛ, with Λ a diagonal matrix of eigenvalues • Choose first K components in Q → QK • Project data onto these components • F = QKA → new data set of reduced dimensions PCA with a stack of images • Use the women’s face to compute orthogonal basis • Project man’s face on first 50 coefficients and add • a stack of images Describe PCA man’s with face with 50 coefficient instead of (same method, but every pixel is a dimension) thousands of pixels 8 new components per image Take the best N PCA eigenfaces and dot product a real face with Practical use of PCA Pros and cons of PCA + − Can be applied to whole image at once + CPU time can be drastically reduced! − In practice, annulus-wise version is generally better Self-subtraction reduced, but still present + Reduced bias wrt LOCI, generally more linear • Fake companion injection still used in most cases + Forward modeling also possible A « real-life » example HR 8799 2008: ADI detection Contrast ~ 14 mag (~10−6) th 4 planet with ADI+LOCI 2010: long slit spectroscopy Figure 2. Image with the HR 8799 c spectrum before extraction. The spatial direction along the star–planet axis is vertical. The spectral direction is horizontal, with wavelength increasing from left to right. ANSON ET AL. Vol. 710 15% 400 s, taken nt. A plied 799 c re 1). re an on of c and his is tions. ound ation were l sky. dures ected ringe e flat Since on on to be aging speco this ectrorame on of Figure 4. into three (circles, d dash-dotte No. 1, 2010 SPECTROSCOPY OF HR 8799 c Figure 2. Image with the HR 8799 c spectrum before extraction. The spatial direction along the star–planet axis is vertical. The spectral direction is horizontal, with wavelength increasing from left to right. L37 Figure 3. Upper panel: spectrum of HR 8799 c. The dashed lines and faintly shaded area (light blue in the online version) denote the errors. Lower panel: log g = detailed that non Non-equ large dif particula presence 4 µm fo Fortney (2008) t being pr behavio of whet thereby The r planet de ing purp narrowe since it models. true also in the sp 8.66 ma 9.50 ma and is ex In the 2011: HK band IFS The Astrophysical Journal, 733:65 (18pp), 2011 May 20 Ba The Astrophysical Journal, 733:65 (18pp), 2011 May 20 Barman et al. Table 1 OSIRIS Observations Target HR8799b HD210501 BD+14 4774 #exp 9 5 4 Exp. Time (s) 900 30 30 Band Pass K UT Date 2009 Jul 22 HR8799b HD 210501 BD+14 4774 6 7 5 900 30 30 H 2009 Jul 23 HR8799b BD+14 4774 7 7 900 30 H 2009 Jul 30 HR8799b HD210501 9 3 900 30 K 2010 Jul 11 HR8799b HD210501 8 3 900 30 H 2010 Jul 13 and no artificial wavelength shifts were discovered; a potential concern given the temperature variations encountered during the observations.4 While the planet is easily detected in all of the BCD cubes, contamination by scattered starlight remains a concern. Figure 1 shows the FOV and orientation for the OSIRIS observations Figure 1. Unprocessed NIRC2 image (20 s exposure) of HR8799. The rectangle and the extent of speckle contamination in the observed region. shows the orientation and size of the OSIRIS spectrograph FOV (0.′′ 02 scale). Since the spatial scaling of the speckles, radially from the star, HR8799b is barely visible in the center of this rectangle and comparable in proportional toofλ,HR8799b speckles(scaled can intersect spatial of Figure 3. OSIRIS H -isand K-band fluxes to 10 pc)the plotted withlocation 1σ uncertainties. The location of prominent water, CH4 , and CO abs brightness to the speckle contamination. Dithering was done along the long axis are indicated. The fluxes extracted without the speckle-suppression algorithm are shown as dotted lines. The bottom two curves are the mean residuals o the planet in a wavelength-dependent manner. To suppress the of the FOV. with flat spectra extracted from the same data cubes (see the text for details). The Kn3 spectrum of Bowler et al. (2010) is shown as green pluses (sca speckle pattern, each BCD cube was processed with custom IDL down) overplotted with the broadband spectrum (black points) interpolated the Bowler et al. Kn3 wavelength points. The mean 1σ uncertainties a routines that first subtracts a (spatially) low-passonto filtered version range are shown at either end for each data set. The larger red filled symbols are the NICI CH short/long (boxes), NIRC2 narrow (circles), and NIRC wavelength and dispersion solutions, sky subtraction, cosmic of the image slice in each wavelength channelphotometry and rebins the 4from (stars) photometry taken from Table 3. Blue symbols are the corresponding derived the OSIRIS spectra. ray rejection, and so forth (Krabbe et al. 2004). On all three data cube (by taking the median in the λ dimension) to 25 (A color version of this figure is available in the online journal.) nights in 2009, OSIRIS was operating above its normal detector wavelength channels (R ∼ 60). Each monochromatic image 2013: full IFS study with SD The Astrophysical Journal, 768:24 (16pp), 2013 May 1 Oppenheimer et al. Figure 4. Spectra of the b, c, d, and e components using the S4 algorithm. 1σ error bars are indicated on either side of each point. Spectra are shown for comparative purposes in normalized fλ . The dotted spectrum of component c is extracted from the 2012 October data and shows poorer detection in the J band due to seeing conditions. All other spectra are from the 2012 June epoch. The black dashed lines are spectra of the background extracted at the same radial distance from A for each component. They consist of the average of six different randomly-selected locations in the azimuthal direction; they represent comparative “noise level” estimates. Tentative identification of some molecular features (excluding water) are indicated at the top of the plot. CO2 is listed with a question mark as explained in Section 8.3 and could also be attributable to HCN. Getting rid of speckles 3. Hardware solutions Coronagraphy Coronagraphy With central obscuration Phase mask coronagraph • Proposed by Roddier (1997) • Apply 180° phase shift to PSF center • Ideal mask size = 43% of 1st Airy ring • Very chromatic design Four Quadrant Phase Mask Beyond the FQPM • Quadrants → octants → continuous phase ramp Vortex phase mask • Continuous phase ramp from 0 to 4π • Main problem: chromaticity of the phase ramp December 15, 2005 / Vol. 30, No. 24 / OPTICS LETTERS 3309 The Astrophysical Journal Supplement Series, 209:7 (8pp), 2013 November y profile, 'U!x! , y!"'2 of a beam containx. (b) Surface profile of a VPM. in Fig. 3(a). Images for vortex coronagraphs having topological charges m= 0, 1, 2, and 3 are shown in Figs. 3(b)–3(d), respectively. The m = 2 case provides a bright image of the weak off-axis source, while eliminating light from the on-axis image. The superior performance of the m = 2 case is attributed to zero intensity values for the starlight extending over the entire exit pupil of radius REP. This remarkable result may be found analytically by computing a second-order Hankel transform of the Airy disk.13 Assuming paraxial rays, we find19 'U1!u!,v!"' = ( 2 2 A1REP /%! , %! ) REP 0, %! & REP ) . !5" Its (o tra ax th W de fin in to a be ca Je Achromatic phase mask? • 1194 Sub-wavelength gratings → achromatic half wave plate MAWET ET AL. Fig. 6.—Pancharatnam phase ramp of the AGPM "p ¼ complex 2!. Fig. 3. 4QZOG implementation: s and coronagraph: p are the vectorial 2. Within one light revolution around thek. In The associatedamplitude topologicalcomponents charge is lp of¼ the incoming of wave vector meters of the 2(2!).polarization states optical axis, i.e., ! of ¼ the 2!, one confirms "p p¼global each four easily quadrants, the that s and r to the grath h, and the are decomposed in the corresponding TEi and TMi vectorial com- Vol. 633 Fig. 7.—AGPM scheme and analogy with the FQ-PM coronagraph. The AGPM can be seen as a polarization FQ-PM. The parallel potentially interfering polarization states are out of phase according to the FQ-PM focal plane phase shift distribution. Here " TE and " TM are the output phases of the polarization components TE and TM such that !" TE"TM ¼ j" TE " " TM j ¼ !. Annular Groove Phase Mask (AGPM) The AGPM performance Peak%rejec)on% 1000" 100" 10" 1" août(10" nov.(10" févr.(11" juin(11" sept.(11" déc.(11" avr.(12" juil.(12" oct.(12" Fig. 8. Azimuthally averaged coronagraphic profile (CP) obtained with the OVVC2 u LCP, measured at 1550 nm and with a virtual defect size 10% of the beam size F λ . PSF is also shown in dashed line. The camera noise floor (dotted straight line) of a 3 × 10−5 is quickly reached at 3.5 λ /d. On-sky example Raw image Post-processed (ADI+PCA) Sensitivity to planets β Pic b Apodization • « Pupil plane coronagraphy » • Act on amplitude or phase of wavefront in pupil • Redistribute the intensity in the focal plane to make it more compact or create a dark region Amplitude apodization • Goal: reduce PSF side lobes • Degraded angular resolution Apodized coronagraph • Apodization makes PSF more compact on the focal plane mask Shaped pupils How it’s done P. Martinez et al.: Design, analysis, and testing of a microdot apodizer for the Apodized Pupil Lyot Coronagraph • Microdot mask 5. Left: simulation map of the binary apodizer with 5 × 5 µm dots. Right: shadowgraph inspection of the manufactured microdots apodizer 0). For the sake of clarity, only a quarter of the apodizer is shown. Phase apodization • Act on wavefront phase instead of amplitude • Can produce asymmetric PSF APP: Apodizing Phase Plate On-sky example L50 QUANZ ET AL. Vol. 722 Figure 1. Left: median combination of one cube of unsaturated exposures of β Pictoris used to determine the photometry. The effect of the APP can be seen in the left side of the PSF where the diffraction rings are effectively suppressed increasing the contrast between 2 and 7 λ/D. Middle: median combination of one cube of saturated exposures of the PSF reference star HR2435. Right: median combination of all PSF-subtracted science exposures. β Pictoris b is indicated by the arrow. The right side suffers from subtraction residuals as does the central region of the PSF, which has been masked out. Table 1 Summary of Observations on 2010 April 3 emission, we subtracted from each individual frame the cor- Phase-induced amplitude apodization Other hardware solutions • Low-order wavefront sensing Adaptive Optics for Extremely Large Telescopes II • • • Performed at instrument level (science camera, unexploited suggesting that we are not in a Fraunhofer case. Hence, in the current configuration, the bench is not ready for the demonstration of the Fresnel effects we plan to do. stellar light, etc) The presence of very bright off-axis ghosts (carrying each its speckle structure) seems to have a significant contribution to the non–symetry of the speckle structure. These ghosts are due to the coating of the DM that is not optimized for NIR. Another source of errors may come from amplitude errors of the apodizer itself. Finally, the source being highly coherent, even faint unwanted light may interfere with the speckle field. Corrects for non-common path errors between AO and instrument Speckle nulling • Fig. 5. Preliminary results of speckle nulling with the FFREE bench. Left: Coronagraphic image without Measure the complex amplitude quasi and correction. Speckle intensity isof between 10 and 10static in the probespeckles zones.. Middle: correction on both sides: speckle rejection factor (rms intensity) = 5×. Right: correction on one side only: speckle rejection factor = 30× (rms intensity). correct with deformable mirror -4 7. Conclusion -3 Tree of life All steps in one picture Telescope pointed (i) (i) (i) (i)(i) AO turned on (ii) (ii) (ii) Speckle calibration Centered on mask (ii)(ii) (iii) (iii) (iii) (iii) (iii) ADI + postprocessing b" b" b"b"b" c" c" Project Project 1640 1640 Project Project Project1640 1640 1640 c"c"c" Toward Toward Exploration Exploration of Other of Other Worlds Worlds Toward Exploration of Other Worlds Toward Toward Exploration Exploration of of Other Other Worlds Worlds e" e" d" d" e"e"e" d"d"d" N" N" N" N" N" E" (iv) (iv) (iv) (iv) (iv) (v) (v) (v) (v)(v) E"E"E" 0.5”" 0.5”" 0.5”" E" 0.5”"0.5”"