The NINJA2 Waveform Model Data Set

Transcription

1 The NINJA2 Waveform Model Data Set Sascha Husa Universitat de les Illes Balears For the Ninja Collaboration ninja-project.org LLama BAM T LLama BAM T4 M M SpEC BAM T1 SpEC BAM T SpEC Llama Sascha Husa (Palma) NR-DA NRDA, Cardiff, July / 28

2 The NINJA2 Waveform Model Data Set Sascha Husa Universitat de les Illes Balears For the Ninja Collaboration ninja-project.org Sascha Husa (Palma) NR-DA NRDA, Cardiff, July / 28

3 Why Ninja? Numerical relativists claimed for decades that they want to be useful for GWDA! Advanced detectors put NR on a tight timeline to show its worth! Actually, direct observation of BH geometry is one holy grail of physics a good place to look for new physics! Reality check: Can NR really make a difference for GW-DA? Detection without source identification is boring. Squeeze maximal information from detections! Use all your physics knowledge! Sascha Husa (Palma) NR-DA NRDA, Cardiff, July / 28

4 Numerical INJection Analysis Project Core Ideas Science of NR-enhanced DA has to be studied jointly by both communities! Both deal with imperfect numerical data, both have to make tradeoffs regarding computational resources, human resources,... Injections of IMR waveforms has been identified early on as an important application of NR WFs test & tune actual search codes. Many issues can only be understood when working with noise from actual GW detectors. NINJA conceived as a community KITP in 2008 want to achieve critical mass to explore collaboration of LSC/Virgo with NR community. Even ninjas have rules (guidelines): Flat hierarchy; leadership by initiative just do it ; promote young scientists Small authorlist papers encouraged (but there are rules!) Act fast! (violated completely) Sascha Husa (Palma) NR-DA NRDA, Cardiff, July / 28

5 NINJA-1 NINJA-1: open call to NR and GW-DA communities to provide NR data sets, inject into simulated detector data and recover with standard search codes. Goal: create communication channel, common language, identify obstacles. Success: Broad participation from NR community (all 10 groups) and from the GW-DA (9 groups). number of detected signals log 10 B NR groups agreed immediately. Figure: Bayes factors for TaylorF2 and IMRPhenA, Birmingham group. Success: wrote 2 CQG papers [Aylott + 80 others, Class.Quant.Grav. 26 (2009) ] [Aylott + 80 others, Class.Quant.Grav. 26 (2009) ] BUT: In order to keep the threshold low, NINJA-1 was not designed to do real science. E.g., number of submitted WFs/group was restricted to 2(5). Sascha Husa (Palma) NR-DA NRDA, Cardiff, July / 28

6 NINJA-2 for real science 1 Use real noise through LSC/Virgo MOU (plus simulated noise for test runs). 2 More WFs, more injections (127 in NINJA-1). 3 PN-NR stitching to cover mass down to 10M. 4 Accuracy requirements, motivated by Samurai, NR-AR preparation,... : 5 usable orbits before merger (without junk bursts etc.) GW Mω 2, Hybrid WF usable for LIGO/Virgo down to 10 M at 20 Hz, i.e. GW Mω 2,2 = NR (2,2) amplitude accuracy below 5% NR (2,2) accumulated phase error 0.5 radian. Highest PN order available for phase and amplitude for hybridisation 5 All spherical harmonic modes welcome (only l = m = 2 required). Sascha Husa (Palma) NR-DA NRDA, Cardiff, July / 28

7 Size of the data set Just as NINJA-1 owed much of its success to the project s concrete scope fixed at the outset, we limit the scope of NINJA-2 to exclude precessing systems in order to simplify this project. [MOU] configurations: 48, hybrids: 61 NR strain, h(t) format: 24 NR strain, amp/phase format: 9 NR ψ 4 : 28 Fresh svn checkout: 2.8 GByte 10 people have submitted WFs from 6 different codes. 13 NR groups (by institution) have signed the MOU: AEI, Barcelona CSIC-IEEC, Caltech, Cardiff, CITA, Cornell, Florida, Georgia Tech, Jena, Palma (UIB), Perimeter Institute, Rochester (RIT), UIUC. du -ms directory N hybrid 875 LEAN FAU SpEC BAM Llama RIT 2 70 GATech UIUC 2 Sascha Husa (Palma) NR-DA NRDA, Cardiff, July / 28

8 Data format - I A gravitational waveform is a time-dependent complex function on the celestial sphere (i.e. a function of 2 angles), definied at infinite separation from the source. It is useful to describe these WFs in terms of spin-2 spherical harmonics h lm (t), where typically only a smallish number of modes contributes: l 2, m = l,.., l (2l + 1). E.g. for l max = 2, 4, 6 we have in total 5, 21, 45 complex mode functions, many of which can vanish by symmetry or can be neglected due to smallness. m = 0 are real, non-oscillatory, but show ringdown (GW memory) For NINJA-2 submitters decide which modes they want to submit, only l = 2, m = ±2 is required. Most submissions contain other modes. Sascha Husa (Palma) NR-DA NRDA, Cardiff, July / 28

9 Data format - II NINJA-1: 3-column format (t, Re[h], Im[h]), equispaced in time. For long hybrids the actual data need to be represented more efficiently than with equispaced time series of the complex strain (higher modes!). NINJA-2: h = Ae I Φ, new format (t, A, Φ) unequally spaced, use interpolation t M... the data repository will be accepting data in a new format, as well as the old. We ask that people please submit waveforms (especially hybrids) using this new format. Sascha Husa (Palma) NR-DA NRDA, Cardiff, July / 28

10 Data format: NINJA-1 metadata very limited Sascha Husa (Palma) NR-DA NRDA, Cardiff, July / 28

11 Data format: NINJA-2 metadata Based on NINJA-1 format, and described in detail on arxiv: v3, Data formats for numerical relativity waves, P. Ajith, M. Boyle, D. A. Brown, S. Fairhurst, M. Hannam, I. Hinder, S. Husa, B. Krishnan, R. A. Mercer, F. Ohme, C. D. Ott, J. S. Read, L. Santamaria, J. T. Whelan. This data format has now been adapted to also describe matter sources! Include more information on waveforms, e.g.: their physical properties (eccentricity, initial BH momenta/speeds) numerical relativity information: initial data, extraction radius, code-version, simulation uuid,... methods used to produce hybrid waveforms accuracy availability (license tag, publication info) bibtex keys for waveforms, codes, initial data,... Sascha Husa (Palma) NR-DA NRDA, Cardiff, July / 28

12 NINJA-2 metadata, details I Key ideas: Simulation data are distributed together with a metadata file that describes the data set and contains the file names of the actual simulation data: Simple text file that is easy to read and edit for humans, and contains key = value pairs, organized in [sections] (config-file format). Sections used in NINJA-2: metadata, ht-data, ht-amphi-data, psi4t-data, psi4t-amphi-data. Principal design issues: allow automated production of waveform tables, simple to read and edit for humans, simple to parse with scripts, easy to extend (by adding or extending keys, sections, otpions for keys). Another option would have been an XML- format. Sascha Husa (Palma) NR-DA NRDA, Cardiff, July / 28

13 NINJA-2 metadata, details II Purpose: Provide an extendable metadata format to exchange data about black-hole-binary simulations, designed in particular to serve efforts such as the ninja waveform catalog, ninja2 (and follow-up) projects, data format for injections, projects such as NRAR, any collaboration that exchanges BBH simulation data. Sascha Husa (Palma) NR-DA NRDA, Cardiff, July / 28

14 Data format: NINJA-2 metadata example I [metadata] license = ninja-catalog/ninja2/thirdparty/private/public comments = documentation = short.txt myfile.pdf publication = arxiv: /unpublished/inprint simulation-bibtex-keys = Aylott:2009ya simulation-name = spp50 simulation-uuid = 373b1340-db3d-11de-aef4-0002a5d5c51b authors-tag = SubCorrIts submitter- = Sub Mitter <submitting@author.edu> authors- s = Corr Espondent<corresponding@author.edu>, Its Me <somebody@else.e data-type = NR/PN/hybrid hybrid-nr-uuid = 373b1340-db3d-11de-aef4-0002a5d5c51ba hybrid-approximant = TaylorT1 hybrid-method = hybrid-doc.txt # optional # only for hybrids # only for hybrids simulation-relative-resolution = 3d18fda3-c58d-4a95-9a87-b1275da258e6 2 code = mycode Sascha Husa (Palma) NR-DA NRDA, Cardiff, July / 28

15 Data format: NINJA-2 metadata example II code-version = 247 code-bibtex-keys = initial-data-type = Bowen-York quasicircular initial-data-uuid = initial-data-bibtex-keys = quasicircular-bibtex-keys = initial-separation = 11 initial-separation-angle = 90 # only for aligned spins initial-bh-momentum1x = initial-bh-momentum1y = initial-bh-momentum1z = 0 eccentricity = eccentricity-error-range = mass-ratio = 1.0 spin1x = 0.0 spin1y = 0.0 spin1z = 0.5 Sascha Husa (Palma) NR-DA NRDA, Cardiff, July / 28

16 Data format: NINJA-2 metadata example III spin2x = 0.0 spin2y = 0.0 spin2z = 0.5 freq-start-22/freqstart22 = 0.05 number-of-cycles-22 = 19.3 extraction-radius = somenumber/infinity/extrapolated extraction-techniques = number-of-cycles-22-error-relative = amplitude-error-relative = 0.02 [ht-data] 2,2 = hmod.r5.l3.l2.m2 2,1 = hmod.r5.l3.l2.m1 2,-1 = hmod.r5.l3.l2.m-1 2,-2 = hmod.r5.l3.l2.m-2 Sascha Husa (Palma) NR-DA NRDA, Cardiff, July / 28

17 Tools to process/check NR data Conversion scripts to/from Amp-Phase format, Mike Boyle. Python scripts to parse and check data, etc.; Larne,...? Standard implementations of PN amplitude and phase functions, standard integrator (Mike, Ajith, SH, Frank Ohme,... ) NRCore Mathematica package, svn co svn:// /nrcore; soon at parse metadata, NINJA-2 data file reader, produce DokuWiki tables,... Mathematica notebooks to check PN test WFs, submitted NR and hybrid WFs, submit BAM WFs.... Sascha Husa (Palma) NR-DA NRDA, Cardiff, July / 28

18 Submission statistics: author number Sascha Husa (Palma) NR-DA NRDA, Cardiff, July / 28

19 Submission statistics: parameter space BAM MayaKranc 10 LEAN 8 q Llama Χ2 LazEv 0.0 Χ SpEC Sascha Husa (Palma) NR-DA NRDA, Cardiff, July / 28

20 WF lengths from metadata I ,2 NR start frequencies MΩ Llama_u2u2-q1--D8-h64-r250 : freq-start-22 = Llama_u4u4-q1--D8-h64-r250 : freq-start-22 = Sascha Husa (Palma) NR-DA NRDA, Cardiff, July / 28

21 WF lengths from metadata II injectable Mass 20 Hz Sascha Husa (Palma) NR-DA NRDA, Cardiff, July / 28

22 WF lengths from metadata III Sascha Husa (Palma) NR-DA NRDA, Cardiff, July / 28

23 WF lengths from metadata IV Sascha Husa (Palma) NR-DA NRDA, Cardiff, July / 28

24 PN comparisons PN approximants used for aligned-spin hybrids: TaylorT1, T2, F2, T4 Data format document (arxiv: v3) contains thoroughly checked PN expressions, and list of errata in literature. Compared analytical expressions (energy, fluxes, ODE RHSs), and numerical data. Bugs in literature fixed (Mike & others) Mathematica notebook for PN data comparisons: ninja2/postnewtonian/code/sascha svn commit Ninja2PNCheck Full.nb. (my) BAM hybrids have been submitted with both a TaylorT1 and a TaylorT4 version. Sascha Husa (Palma) NR-DA NRDA, Cardiff, July / 28

25 Hybrid checks I Time and Fourier domain plots of all WFs have been generated for the internal Wiki (Larne): Sascha Husa (Palma) NR-DA NRDA, Cardiff, July / 28

26 Hybrid checks II Overlap calculation have revealed some hybridization problems, in particular a bug in producing the BAM and some Llama WFs. Deviations in improved hybrids due to different PN approximants, and proably occasionally non-optimal hybridization. Sascha Husa (Palma) NR-DA NRDA, Cardiff, July / 28

27 Hybrid checks III Matches with Llama WF, comparing with mass-optimization: GT GT 0.99 SpEC SpEC BAM T1 BAM T1 match 0.98 BAM T4 match BAM T4 BAM T1 OPT BAM T1 OPT 0.97 BAM T4 OPT BAM T4 OPT SpEC OPT SpEC OPT Sascha Husa (Palma) NR-DA NRDA, Cardiff, July / 28

28 Hybrid checks IV Parameter bias from mass optimization ( M/M) is very small LLama BAM T LLama BAM T4 M M SpEC BAM T1 SpEC BAM T SpEC Llama Sascha Husa (Palma) NR-DA NRDA, Cardiff, July / 28

29 Conclusions NINJA-2 is a heroic and so far successful undertaking, but there is also still a lot of imperfection! Data set is very sparse! Few spinning WFs for unequal masses, few WFs with unequal spins. NINJA-2 is another lesson in: things take a lot of time. Data set is ready for usage by DA community, but understanding of hybrid accuracy, effect of PN approximants, etc., is still building up. A number of tools have been built to investigate the data set, there has been significant activity of looking into this treasure of data. What science can we do with this? Sascha Husa (Palma) NR-DA NRDA, Cardiff, July / 28