1 / 27

Toward a fully realistic simulation of QCD – three flavor QCD project by CP-PACS/JLQCD K. Kanaya

2003/12 Nagoya. Toward a fully realistic simulation of QCD – three flavor QCD project by CP-PACS/JLQCD K. Kanaya for CP-PACS and JLQCD Collab.s kanaya@rccp.tsukuba.ac.jp. SR8000 (2000, 1.2TFlops) @ KEK. JLQCD

ata
Download Presentation

Toward a fully realistic simulation of QCD – three flavor QCD project by CP-PACS/JLQCD K. Kanaya

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. 2003/12 Nagoya Toward a fully realistic simulation of QCD – three flavor QCD project by CP-PACS/JLQCD K. Kanaya for CP-PACS and JLQCD Collab.s kanaya@rccp.tsukuba.ac.jp SR8000 (2000, 1.2TFlops) @ KEK JLQCD T.Onogi N.Tsutsui N.Yamada (H.Mino) (S.Kaya) (S.Tominaga) .... S.Aoki M.Fukugita S.Hashimoto K-I.Ishikawa N.Ishizuka Y.Iwasaki KK T.Kaneko Y.Kuramashi M.Olawa A.Ukawa T.Yoshie (R.Burkhalter) ... Y.Taniguchi K.Ide T.Ishikawa Y.Namekawa T.Yamazaki (A.AliKhan) (G.Boyd) (S.Ejiri) (V.Leske) (T.Manke) (K-I.Nagai) (J-I.Noaki) (M.Okamoto) (T.Umeda) .... CP-PACS CP-PACS (1996, 614GFlops) @ CCP, Univ.Tsukuba

  2. Introduction • Low energy properties of hadrons • difficult to calculate analytically due to strong coupling. • Lattice QCD: • The only systematic way to calculate non-perturbative properties –– directly from the first principles of QCD. • Precise calculation of hadron properties • critical test of QCD as the theory of quarks • determination of fundamental parameters (mq, a, CKM, ...) • predictions for hadron physics, QGP, ... QCD corrections important for quarks.

  3. a La Introduction (2) • Lattice QCD • Quarks and gluons formulated on space-time lattice with • finite lattice spacing a • finite lattice size L • Physics is defined in the limit • a ––> 0"continuum extrapolation" • keeping La sufficiently large (or ––> ∞) • lattice size L ––> ∞ • Core calculations: quark matrix inversion D–1 • large sparse matrix 12L4x12L4 ≈ typically 107 x 107 (12 = color x spin) • condition number µ 1/mqa • Difficult to calculate u, d quarks directly. • mq ––> mu,d "chiral extrapolation"

  4. Introduction (3) • For a precise and reliable prediction for the real world, good control of these extrapolations, based on systematic large scale simulations, is required. • Large computer power indispensable! • ß ˜ (dedicated) parallel computers • Steps toward a realistic simulation • Quenched approximation (no quark pair creation/annihilations) • Dynamical u,d quarks (degenerate) NF = 2 full QCD • Dynamical s quark NF = 2+1 fQCD • Sea u,d quark mass difference • Dynamical c quark ... Thus, "3" is the last major step toward reality. Expect only small effects in most processes

  5. Yes q q q q No! Systematic studies of LQCD by CP-PACS and JLQCD sea Step 1: NF = 0 (quenched) QCD quench: disregard pair creation/annihilation of sea quarks Important properties of QCD (confinement, AF, ChSB) retained. O(10%) error in the spectrum expected. Computer time < 1/100. valence • First study: Hamber and Parisi ('81), Weingarten ('82) • First systematic study performing all extrapolations: GF11 Collab. ('93) • Light hadron spectrum consistent with experiment within estimated errors of O(10%) • Quality of extrapolation not tested. • Quenching artifacts not identified.

  6. qQCD simulations Hamber and Parisi ('81) Weingarten ('82) GF11 ('93)

  7. Systematic studies by CP-PACS/JLQCD (2) • CP-PACS, PRL 84(00)238, PR D67(03)034503 • plaq. gauge + standard Wilson quark • lattice ba [fm] Ls a [fm] Niter • --------------------------------------------------------------- • 323´56 5.90 0.102(1) 3.26(3) 160,000 • 403´70 6.10 0.078(1) 3.10(3) 240,000 • 483´84 6.25 0.064(1) 3.08(3) 420,000 • 643´112 6.47 0.050(1) 3.18(6) 300,000 • ----------------------------------------------------------------- • 1 iter. = HB + 4 x OR • ≈100 more computations than the GF11. • First well-controlled chiral and continuum extrapolations • ––> a problem in GF11 extrapolation found. Errors in light hadron spectra ≤ 1-3% (statistical + systematic, except for quenching) • Experiment approximately • reproduced. • Deviations of O(10%) • Limitation of quenched approx.

  8. qQCD simulations CP-PACS ('98)

  9. CP-PACS Systematic studies by CP-PACS/JLQCD (3) • Step 2: NF = 2 full QCD • dynamical degenerate u, d, but quenched s • Naively, several hundreds more computer power required. • ß • CP-PACS PRL 85(00)4674 [E: 90(03)029902], PR D65(02)054505 [E: D67(03)059901] • JLQCD PR D68(03)0504502 • Improved lattice action • computation ≈ 1/10 Not enough => lattice ≈ 2.5 fm CPPACS: a = 0.2–0.1 fm mqsea ≥ 70 MeV [Mps/Mv ≥ 0.6] mqval ≥ 40 MeV [Mps/Mv ≥ 0.5] • The 1st cont. extrap. • ~ GF11 in qQCD. •  Deviations largely reduced ◇ :Experiment ●: NF = 2 QCD ■□: quenched QCD

  10. Toward a realistic simulation But still small gaps?? Among many possible origins (e.g. syst. err. from chiral extrap., small volume etc.) quarks:udscbt 3 MeV 7 MeV 70-100 MeV 1.2 GeV 4 GeV 175 GeV ÝÝ QCD scalesLQCD, TC Step 3: NF = 2+1 full QCD dynamical s-quark Large computer power required: * Calculation of s quark * More parameters to be scanned/adjusted as a joint project of CP-PACS and JLQCD collabs.

  11. Computing Facilities machines full machine for LQCD GF/node #Node GFlops #Node GFlops SR8000/F1 12 100 1200 ~ 64 ~768 @ KEK CP-PACS 0.3 2048 614 2048 614 @ CCP, U.Tsukuba SR8000/G1 14.4 12 173 12 173 @ CCP, U.Tsukuba VPP5000 9.6 80 768 ~ 24 ~230 @ SIPC, U.Tsukuba Earth Simulator 64 640 40960 ~ 10 ~640 @ ES Center

  12. Algorithm Conventional Hybrid MC algorithm exact only for NF = 2 n for Wilson-type quarks NF = 4 n for staggered-type quarks mu ~ md << ms single (odd) flavor algorithm needed to avoid unexpected systematic errors. (cf.) other groups adopting approximate algorithms. • Exact algorithms based on polynomial approx. for quark matrix D or det D • multiboson algorithm (Alexandrou et al., PR D60(99)034504) • HMC (Takaishi and de Forcrand, Int. J. Mod. Phys. C13(02)343) A systematic test of various alternativesAoki et al., PR D65(02)094507 We adopt: HMC for u, d PHMC for s ~ 1.5 times more CPU time than NF = 2 HMC Implementation+optimization: performance ~ 20-44% in production runs!

  13. Computing Facilities machines full machine for LQCD GF/node #Node GFlops #Node GFlops / [*] SR8000/F1 12 100 1200 ~ 64 ~768 @ KEK 35% CP-PACS 0.3 2048 614 2048 614 @ CCP, U.Tsukuba20% SR8000/G1 14.4 12 173 12 173 @ CCP, U.Tsukuba44% VPP5000 9.6 80 768 ~ 24 ~230 @ SIPC, U.Tsukuba44% Earth Simulator 64 640 40960 ~ 10 ~640 @ ES Center31% [*] performance of PHMC code on 203x40 (T. Yoshié, Lattice 2003)

  14. Action Test study: a-1 ~ 1.5--2.5 GeV, 83x16, 123x32 S. Aoki et al., Nucl. Phys. B(PS)106(02)263 Plaquette gauge + Clover v.s. RG-improved gauge + Clover • b = 4.6, 4.8, ... 6.0; 83x16 b = 1.5, 1.55, ..., 2.25; 83x16 • With clover-improved Wilson quarks, •  severe lattice artifact with the plaquette gauge action • at a-1 < 2 GeV • Effective adjoint coupling due to the clover term? (No such phenomena with naive Wilson quarks.) •  improvement of the gauge action (RG or Symanzik) important ~

  15. Action (2) We adopt: • RG-improved gauge action (Iwasaki, '83) Sg = (b/6) {c0S + c1S } c0 + 8c1 = 1, c1 = –0.331 • clover-improved Wilson quarks with cSW non-perturbatively evaluated using the Schrödinger functional method (Lüscher et al., '96, Jansen and Sommer, '98)  at fixed physical lattice size L* = 6 x ab=1.9 to ensure full O(a)-improvement K.-I. Ishikawa, Lattice 2002; 2003

  16. Simulation Parameters • a-1 ~ 2 GeV (b = 1.9, cSW = 1.715) as the 1st point towards the continuum limit • 163x32 (La ~ 1.6 fm) 203x40 (La ~ 2.0 fm) on-going • mud: 5–6Kud points (L) mPS(LL) / mV(LL) ~ 0.64-0.77 • ms: 2Ks points (S) mPS(SS) / mV(SS) ~ 0.72, 0.77 slightly missed the phys. pt. (0.68) • 3000 traj. at each (Kud, Ks) on 163x32 ≥5000 203x40 had. meas. every 10 traj., jackknife-bin = 50 traj. for 16, 100 traj. for 20 163x32

  17. Simulation Parameters (2) 163x32, b = 1.9, cSW = 1.715 (3000 traj. each) Kud Ks dt Npoly mPS/mV(LL) mPS/mV(SS) 0.1358 0.1358 1/100 110 0.768(4) 0.768(4) 0.1361 1/100 100 0.745(5) 0.766(4) 0.1364 1/125 100 0.717(5) 0.766(4) 0.1366 1/125 100 0.691(4) 0.761(4) 0.1368 1/125 100 0.663(6) 0.762(4) 0.1370 1/140 100 0.639(6) 0.769(4) 0.1358 0.1364 1/100 130 0.763(5) 0.716(5) 0.1361 1/100 130 0.743(4) 0.718(4) 0.1364 1/125 130 0.716(5) 0.716(4) 0.1366 1/125 130 0.694(5) 0.716(5) 0.1368 1/125 130 0.662(7) 0.712(6) 0.1370 1/140 130 0.636(7) 0.715(6)

  18. Simulation Parameters (3) 203x40, b = 1.9, cSW = 1.715 (5000 traj. each) Kud Ks dt Npoly mPS/mV(LL) mPS/mV(SS) 0.1358 0.1358 1/125 110 0.767(2) 0.1361 1/125 110 0.742(2) 0.764(2) 0.1364 1/140 110 0.721(3) 0.770(2) 0.1368 1/160 110 0.675(4) 0.771(2) 0.1370 1/180 110 0.639(3) 0.769(2) 0.1358 0.1364 1/125 140 0.767(2) 0.722(2) 0.1361 1/125 140 0.746(2) 0.720(2) 0.1364 1/140 140 0.714(3) 0.1368 1/180 140 0.665(4) 0.715(3) 0.1370 1/180 140 0.623(4) 0.708(3) Preliminary ! 163x32 data will be discussed, unless otherwise stated.

  19. Meson masses In this study, Kval = one of Ksea(L or S) only "diagonal" mPS(LL) obtained atKud = 0.1366, Ks=0.1364 [mPS(LL)/mV(LL)=0.694(5), mPS(SS)/mV(SS) =0.716(4)]  Doubly-smeared source & single exp. fit mV(LL)

  20. Meson masses (2) • Chiral extrapolation: quadratic polynomial fits • (PS) • (V1) • (V2) • [1] = (PS)+(V1) • [2] = (PS)+(V2) • central value = weighted average of [1] & [2]; syst. error = [1] - [2] • Identification of the physical point a , Kud , Ks • K-input:Mp , Mr , MK • f-input:Mp , Mr , Mf • 

  21. Meson masses (3) K-input K-input At a ~ 0.1 fm  consistent with Experiment  K- and f-inputs agree Similar results also for J. Assuming small a-dep. (due to non-pert. improvement)  quenching artifacts removed in NF=3 QCD

  22. J : influence of the chiral extrapolation not strong • consistent with Experiment

  23. J Preliminary ! • consistent with Experiment

  24. Light meson decay constants Using MF-improved 1-loop ZA, cA, bA • Systematic errors? • scaling violation • chiral extrapolation • non-perturbative effects in Z 's

  25. MF-improved 1-loop matching with MS at m = a-1 • 4-loop running to m = 2 GeV Quark masses K-input K-input  mq(NF=3) < mq(NF=2) < mq(NF=0) VWI and AWI disagree  K- and f-inputs agree

  26. Assuming small a-dep. in mqAWI (as seen in the NF=2 case), in the MS scheme at m = 2 GeV, Quark masses (2)  10-15% smaller than NF=2  consistent with 1-loop ChPT 24.4(1.5)(Leutwyler, PL B378(96)313)

  27. Tentative conclusions and future plans • NF=2+1 QCD • RG-improved gauge + clover-improved Wilson quark with non-perturbative cSW • exact NF=2+1 algorithm (PHMC) • 163x32, a-1 ~ 2 GeV, 6Kud x 2Ks x 3000traj.'s finished  light meson spectrum ~ experiment already at a-1~2GeV assuming small scaling violation in AWI-mq  mq 10-15% lighter than NF = 2  decay constants ?  further study needed • next • 203x40, a-1 ~ 2 GeV jobs  finite volume effects • two more a-1 points  continuum extrapolation

More Related