Download p - CEA-Irfu

Visualization of large astrophysical
simulations datasets
D. Pomarède 1,*, E. Audit 2, R. Teyssier 2, B. Thooris 1
1
Service d’Electronique, des Détecteurs, et de l’Informatique, LILAS
2 Service d’Astrophysique
CEA/DAPNIA, Saclay, 91191 Gif-sur-Yvette, France
*Email : [email protected]
The SDvision graphical interface is designed to visualize large datasets produced in numerical simulations of astrophysical plasmas. It is conceived for the visualization of 2D
and 3D scalar and vector fields distributed over regular mesh grids or more complex structures such as Adaptive Mesh Refinement data, as well as N-body systems. Various
implementations of the visualization of the objects are simultaneously proposed, such as 3D isosurfaces, volume projections, hedgehog and streamline displays, surface and
image of 2D subsets, profile plots, particle clouds. It is part of the Numerical Simulations Software Project [1,2] engaged at the Saclay/DAPNIA Laboratory with the objective to
provide a core of software modules useable by the astrophysical simulation tools in development.
This application is essentially used to visualize large
The baseline technology is the object-oriented
data sets generated by parallelized code running on
programming offered by IDL’s Object Graphics [3].
massively parallel mainframes. The data are written
The interface is implemented as a graphical widget
in the HDF5 Hierarchical Data Format developed by
providing interactive and immersive 3-dimensional
the NCSA [4], allowing for efficient data access. Two
navigation capabilities. The user acts on the objects
methods of rendering objects are supported: via a
attributes through an ensemble of menus, drop lists,
hardware graphics accelerator or via a software
buttons and dialog fields. Complex objects such as
rendering package. The first solution, based on the
isosurfaces are implemented as polygon and polyline
use of the OpenGL libraries, is highly efficient when
objects with the help of optimized procedures of the
operating on a local machine. The second solution
IDL library. Visualization of 3D objects benefits from
applies when the application runs on a remote
the implementation of light objects that represents
the Saclay/DAPNIA Visualization Interface
computer.
sources of illuminations.
SDvision
The SDvision program is employed to visualize data from RAMSES [5], an AMR-based hybrid N-body and hydrodynamical code which solves the interplay of dark matter and
the baryon gas in the study of cosmological structures formation, and from HERACLES [6], a radiation hydrodynamics code used in particular to investigate turbulences in
interstellar molecular clouds.
The SDvision
SDvision widget used to visualize a surface of the density distribution
distribution in the
Interstellar Medium obtained in an HERACLES simulation. The navigation
navigation
capabilities of the widget make it possible to rotate, displace and scale the scene at
will through mouse actions. The distribution is also projected onto
onto an image on which
profiles can be interactively obtained. Menus are used to modify the attributes of the
objects on display (palette, shading, texture, lighting). Sliders
Sliders can be used to scan
through the volume in either direction.
a)
The SDvision
SDvision widget allows to visualize complex, composite scenes. The graphical objects lifecycles are managed
through the use of controllers (displayed on the right hand side).
side). Such controllers are implemented for « data
objects » (isosurface, volume projection, image, vectors, streamlines), « geometry objects » (boundaries, axis), and
« scene objects » (illuminations). Objects are configured through dedicated interfaces.
interfaces. In the example above,
obtained in a 200×200×200 regulargrid
HERACLES
simulation,
the scene includes two isosurfaces, one
regular
one image and
one streamline objects. Isosurfaces are 3D contours. An histogram
histogram of the associated data is displayed and can be
used to set interactively the desired contour value.
b)
c)
d)
The SDvision
SDvision widget provides simultaneous access to the NN-body data and the complex
AMR octree structures generated by RAMSES. In the example above, the 128×128×128
Dark Matter particles are viewed together with an image of the hydrodynamical
hydrodynamical baryon
density. An histogram of either the DM density or velocity is displayed,
displayed, used to select a
subset of the full particle complement. On the leftleft-hand side, an interface is dedicated to
loading data and select higher resolution levels. The highest resolution
resolution is reached at
level 14 of the AMR, equivalent to that of a 8192×
8192×8192×
8192×8192 Cartesian grid and
4.1×
4.1×107 cells . The AMR data are projected onto regular Cartesian grids.
e)
Illustration of the immersive capabilities of the SDvision
SDvision widget. The viewing point is located within the simulation
simulation volume and the scene is observed with a widewide-angle focal. This simulation of cosmological structures formation
formation for a volume of size 100 h-1 Mpc is obtained with RAMSES with a
resolution up to the level 8 of the AMR, equivalent to a 256×256×256 regular Cartesian grid. From this internal viewpoint, the scene
scene is augmented with various graphical objects allowing to better
better interpret, analyze and validate the simulation. An important aspect is the transparency of the objects
obtained by tuning their associated alphaalpha-channels. In a) are displayed the highhigh-density Dark Matter cores embedded in an a gray isosurface of the
the baryon density. A red and a yellow lowerlower-density isosurfaces are also added together with a transparent image
image of the density. In b) the image is kept and
the view is augmented with a display of a maximummaximum-intensity projection of the baryon density. Filamentary structures
structures are observed, wellwell-correlated to the grey isosurface topology in a). The effect of the AMR algorithm can be inferred from the varying granularity of
of the graphical objects. The red Dark
Matter highhigh-density cores displayed in c) are correlated with the most dense baryonic regions, where galaxy clusters are in development. In d) the baryonic hydrodynamical velocity field is added; the streamlines
streamlines are seen plunging toward the DM cores. In e) the transparent
transparent image gives another profile
of the baryonic density which exhibits filaments and halos. This immersive investigation of the structures gives strong indication
indication of the validity of the simulation.
HERACLES simulation of turbulences in the interstellar medium on a highhigh-resolution
1200×1200×1200 grid with dimension 15 pc. Left: image of the plasma density
density on a slice. In
this very turbulent simulation, a cold dense phase and a diluted hot phase are tightly
interwoven. Right: volume projection of the density. The brightest
brightest spots are dense protostellar
cores formed by the thermal instabilities.
Simulation of turbulences in the interstellar medium with HERACLES.
HERACLES. Left : distribution of
internal energy and hydrodynamical velocity field on a slice. Interactively
Interactively adjustable sliders
can be used to scan through the volume box. Right : an isosurface
isosurface of the density field is
displayed as a semisemi-transparent polygon object. The velocity field displayed as streamlines
streamlines
can thus be viewed in and out of the isosurface.
Simulation of cosmological structures formation for a volume of size 100 h-1 Mpc with RAMSES.
On the left, the density distribution is visualized by volume projection.
projection. The hydrodynamical
baryon density is distributed in halos and filaments. The brightest
brightest areas are associated to the
development of galaxy clusters. On the right, Dark Matter is visualized
visualized as a particle cloud. DM
particles concentrate in halos where regions of high baryon density
density develop.
Perspective of developments :
• optimization of memory management to access larger data sets : synchronous spatial and resolution zoom.
• parallelism to improve speed for some specific processes (AMR projection into regular grids, computation of volume projections) using the fastDL/mpiDL solutions for IDL [3].
• application to new simulation algorithm (multiple-grid HERACLES) and in new fields (solar dynamics, proto-planetary systems).
References:
[1] “Numerical Simulations of Astrophysical Plasmas : status and perspectives of the Saclay/DAPNIA software project”, E. Audit, D. Pomarède, R. Teyssier, B.Thooris, Proceedings of the First CalSpace-IGPP International
Conference on Numerical Modeling of Space Plasma Flows, Palm Springs CA, USA, March 27-30, 2006, to appear in the Astronomical Society of the Pacific Conference Series.
[2] The SNOOPY Project Web Site http://www-dapnia.cea.fr/Projets/SNOOPY/
[3] IDL The Data Visualization & Analysis Platform, http://www.ittvis.com/idl/
[4] The National Center for Supercomputing Applications HDF Home Page, http://hdf.ncsa.uiuc.edu/
[5] “Cosmological Hydrodynamics with Adaptive Mesh Refinement – A New High Resolution Code Called RAMSES”, R. Teyssier, Astronomy and Astrophysics, 385, 2002, 337-364
[6] “HERACLES : a new, parallelized, multi geometry and tridimensional RHD code”, M. González, E. Audit, P. Huynh, to be submitted to Astronomy and Astrophysics (2006).