Computational Tools for Multimessenger Astronomy in the Gravitational Wave Era
•
1 j'aime•296 vues
Plenary talk at 2nd meeting of the Black Hole and Gravitational Wave Thematic Network funded by CONACYT in Guadalajara, Mexico. November 6th, 2017.
Recommandé
Recommandé
Contenu connexe
Tendances
Tendances (20)
Similaire à Computational Tools for Multimessenger Astronomy in the Gravitational Wave Era
Similaire à Computational Tools for Multimessenger Astronomy in the Gravitational Wave Era (20)
Plus de University of Illinois at Urbana-Champaign
Plus de University of Illinois at Urbana-Champaign (7)
Dernier
Dernier (20)
Computational Tools for Multimessenger Astronomy in the Gravitational Wave Era
- 3. University of Illinois • Original US public landgrant institution, founded 1867 • US $620M/yr in R&D • 47K students • 12K international students • 425K living alumni • #5 CS • #6 ECE • #9 Physics • Astronomy • NCSA
- 4. National Center for Supercomputing Applications (NCSA) • Founded 1986 – one of first US NSF Supercomputing Centers • Founding Director Larry Smarr, Last Director Ed Seidel • Over 200 staff, ~80 affiliated faculty, also students, postdocs
- 8. Observing the Universe • Seeing: Most of what we know from photons • Optical/IR, Radio, frequencies > 10Mhz • Incoherent superposition, surface processes • Interact very strongly with matter: can’t see much! • Take years for photons to emerge from stellar core • Detecting: Neutrinos and Cosmic Rays • Neutrinos weakly interacting, completely different processes • Yet even neutrinos trapped in stellar core collapse! • Cosmic rays • High energy p or nuclei from SN, AGN, ??? • Listening: Gravitational waves • Predicted by Einstein, only recently directly detected • Respond to coherent bulk motion of matter • Omnidirectional, frequencies below 10Khz Camland Project, Japan
- 9. Observing the Universe • Seeing: Most of what we know from photons • Optical/IR, Radio, frequencies > 10Mhz • Incoherent superposition, surface processes • Interact very strongly with matter: can’t see much! • Take years for photons to emerge from stellar core • Detecting: Neutrinos and Cosmic Rays • Neutrinos weakly interacting, completely different processes • Yet even neutrinos trapped in stellar core collapse! • Cosmic rays • High energy p or nuclei from SN, AGN, ??? • Listening: Gravitational waves • Predicted by Einstein, very recently directly detected • Respond to coherent bulk motion of matter • Omnidirectional, frequencies below 10Khz Each field undergoing revolution in next decade… Imagine combining them all to observe the Universe! Each results from completely different process, on different time scales…
- 10. Key Questions • Physics: • Do gravitational waves exist? Do they travel at speed of light? • Are other theories of gravity possible? • How do black hole binaries form with observed masses, spins, etc? • What happens in a core collapse supernova? • What is the equation of state of dense matter? • What are various classes of gamma-ray burst progenitors? • What is the nature of the expanding universe? • Compute: • How to simulate dynamical gravity, magnetized plasma, realistic EOSs, nuclear reactions, neutrinos, etc. • How can we measure, analyze, compute, serve information to communities about all this, with the “exascale” crises at hand? • How can appropriate software tools be built and verified? • Collaboration: • How can grand challenge communities work together? • How can we make our science reproducible? • How can we provide data sharing and rapid data analysis?
- 14. GW170817
- 15. Multi-Messenger Basics • GWs carry very different information, at different frequencies and times, from different processes, e.g.: • Core collapse (SN II, Long GRBs, etc): GWs promptly from inner core bounce, neutrinos trapped later, photons later (if it explodes!) • NS-BH/NS-NS inspirals • GR170817: Short GRB seen as EM counterpart to GW source! • GWs can probe inner GRB engine; EOS etc • Consider EM transients… • EM Trigger: increase confidence of signal of certain type at certain time • EM follow-ups: GWs may come first! 1-10min triggers! • Trigger allows one catch SN, GRB, kilonova, or ??? Short GRBs are (almost certainly) binary systems Deep optical followup of GRBs does not show evidence for supernovae/stellar collapse GRBs are not associated with star formation GRBs are found far from the centers of their host galaxies Timescales are consistent Simulations produce GRBs Koppitz & Rezzolla
- 16. Complex to Model • Multiscale, multiphysics, data- driven • To model SN or neutron stars merger need general relativity, magneto-hydrodyamics, neutrino & radiation transport, complex equations of state, chemical reactions all integrated together • Need simple but effective interfaces that can be implemented in software He O/Ne/Mg Si Fe-group nuclei ~107 km C (not drawn to scale) Iron Core Collapse AccretionAccretion Collapse to a Black Hole Protoneutron Star with Accretion Disk Jet Formation and Sustainment Jet Propagation / Breakout Afterglow Emission Disruption of Star Schnetter et al, PetaScale Computing: Algorithms and Applications, 2007
- 17. E.g. Multi-Messenger Process: Core Collapse Photons, GWs, Neutrinos • Axisymmetry • Rapidly rotating stellar collapse • Can now compute GWs and neutrino signals • Full 3D on Blue Waters • Full 3D GR models for GRB central engine understanding • AMR, MHD • Exascale simulations • Track explosion to stellar surface; 10000x more expensive; want GW, n, g 0 10 20 0 5 10 15 20 25 30 ne/10 ¯ne ⇥ 4 nx 300 150 0 150 2.5 3.0 3.5 s12WH07j4 Central density GW strain Neutrino luminosity t tbounce [ms] rc[1014gcm3]h+D[cm]L[B/s] Christian Ott, Caltech, computed on Blue Waters
- 18. Tools For New MMA Science • Need to Model Multimessenger Sources • Need Rapid Data Reduction and Data Sharing
- 19. Cactus Framework (1997-now) www.CactusCode.org • Open source component framework for HPC • Modular system with high level abstractions • Components (“thorns”) defined by their parameters, variables, methods • Cactus “flesh” glues components together • Cactus Computational Toolkit: general thorns • Multiple application areas develop toolkits • Numerical relativity, CFD, coastal science, petroleum, quantum gravity, cosmology, …
- 20. Key Features Ü Cactus framework provides scheduling, application APIs Ü Driver thorn provides load balancing, parallelization Ü Application thorns deal only with local part of parallel mesh Ü Different thorns (with same interface) can be used to provide the same functionality, easily swapped.
- 21. Building a Computational Numerical Relativity Community • Cactus came from the relativity community (USA GC) • EU project with 10 sites developed community open code • Each group had different expertise • Cactus allowed developing shared interfaces/standards • Easy to add a component, share components • Supports both collaboration and competition EU Network for Gravitational Wave Sources: 2001
- 22. Evolution of a Software Community • Cactus • initiated at AEI to facilitate interaction between GR codes from different groups • defines data model, restricts itself to glue between modules • CactusEinstein • introduces “Base” thorns • horizon finding and analysis, • EU network “Sources of Gravitational Waves” • Whisky hydro code • LORENE ID solver • PostNewtonian equations matched to Numerial Relativity • XiRel • McLachlan open BSSN code • GRHydro open Hydro code • scale Carpet beyond 1000 MPI ranks • define reproducible benchmark
- 23. 113 users in 56 groups, 20 countries, 6 continents! http://www.einsteintoolkit.org
- 24. Einstein Toolkit Goals • Goals • community-driven and community-building • supported and reliable • mailing list • wiki, bug tracker • core computational tools for relativistic astrophysics and gravitational physics • Components • Cactus thorns for GW science • Simulation factory • GetComponents component retrieval tool • Kranc • Waveform analysis
- 25. Einstein Toolkit Guiding Principles • Open, community-driven software development • Well thought out and stable interfaces • Separation of physics software from computational science infrastructure • Complete working production codes • Does not itself develop codes • codes are proposed for inclusion, then reviewed • must be of current interest for the community
- 26. Main Einstein Toolkit Cactus Modules • Base thorns with hooks for thorns to interact: • ADMBase: gij, a. bj, ... • HydroBase: r0, Pi, … • TmunuBase: Tµn • “Default” evolution • McLachlan • GRHydro, IllinoisGRMHD • PITTNull code • Tools • Analysis, IO, etc
- 27. Einstein Toolkit Based Results • INSPIRE lists 136 citations to 2006 Einstein Toolkit paper • binary black hole mergers • binary neutron star mergers • mixed binaries • supernovae • accretion disks • boson stars • cosmology • not so well known results • collected at Einstein Toolkit workshops • GRB jets • high spin ID • DG methods • visualization • more postprocessing
- 28. Einstein Toolkit Gallery : GW150914 • Parameter file and data • Instructions for running and analyzing • Horizon trajectories • Waveform • Curvature scalars • Gravitational waves • Data, parameter file is citable (DOI)
- 34. How to Get Involved? • Join the Einstein Toolkit Consortium • Fill out web interface on our web page (adds you to user mail list) • Try out the tutorials and examples • Join weekly phone calls • Post to users group if you have a problem or a question • Join one of the new working groups • Come to the next Einstein Toolkit School/Workshop • Talk to me!
- 35. Deep Learning Revolution Deep Learning for Time-Series Signal Processing for Gravitational Wave Astrophysics
- 42. • Very long networks of artificial neurons (dozens of layers) • State-of-the-art algorithms for face recognition, object identification, natural language understanding, speech recognition and synthesis, web search engines, self-driving cars, games (Go) etc. • Does not require hand-crafted features to be extracted first • Automatic end-to-end learning • Deeper layers can learn highly abstract functions 42 Source: https://cs231n.github.io/ Deep Learning
- 44. • Very easy to use, highly automated • All are optimized for accelerated training on GPUs • Developed and/or sponsored by industries (open- source) • Comparisons between these frameworks: https://en.wikipedia.org/wiki/Comparison_of_deep_l earning_software Deep Learning Frameworks
- 48. 48 But Does It Really Work? Try out on Real LIGO Data 1. Real noise from LVT151210 and GW151226 for training (4096s each). Open source, data taken from https://losc.ligo.org/events/GW150914/ 2. Tested on noise from GW150914 and GW170104 3. Same architecture, same nets work without re-training
- 53. Conclusion
- 54. See electromagnetic waves Feel astro-particlesHear gravitational waves LIGO, VIRGO, KAGRA, eLISA DES, LSST, JWST, WFIRST IceCube Real-time Multi-messenger Astronomy Simulations – Observations – Data Analysis – Theory
- 55. Thank you! Credits • Multi-messenger Astronomy: Ed Seidel • Einstein Toolkit: Roland Haas • LIGO: Eliu Huerta • Deep Learning: Daniel George • Neutron Star Simulation: Shawn Rosofsky
- 56. CONACYT Partnership With Illinois CONACYT Graduate Fellowship Program for Mexican National Students Admitted to U. Illinois PhD Programs • University of Illinois System (UIC, UIUC, UIS) has entered into an agreement with CONACYT to support up to 10 Mexican nationals who have been admitted into PhD programs at UI. • CONACYT will support students for the first four years. The host campus will support a fifth year. • Competitive - students apply for the CONACYT award after being accepted into a UI System PhD program. • Call for applications open in February 2018 (?)