lattice qcd - kek...5 development of lattice qcd simulations (i) size ≈2×10−15 m lattice size...
TRANSCRIPT
1
Lattice QCD:placing a hadron in a computer
Introduction
Current focus: dynamical Nf=2+1 simulation
Nucleon structure functions
Future perspectives
Summary
Akira UkawaCenter for Computational SciencesUniversity of Tsukuba
Nikko Symposium“Creation and development of matter”6 November, 2004
2
Introduction
3
Quantum Chromodynamics
Quantum Chromodynamics (QCD)Fundamental theory of quarks and gluons and their strong interactions
Knowing
1 coupling constant and
6 quark masses
will allow full understanding of strong interactions “Yukawa’s dream(1935) in modern form”
tbcsdu
s
mmmmmm ,,,,,α
( ) ( )( )
( ) ( ) S
ffff
sQCD
eAOddAdZ
AO
miAFFTrS
−∫
∑
=
+−∂⋅+=
ψψψψψψ
ψγψπα µµµµνµν
,,1,,
81
Gross-Wilczek-Politzer 1973
4
QCD on a Lattice
Lattice QCDPowerful mean to calculate the QCD Feyman path integral
From computational point of viewRelatively simple calculation
Uniform meshSingle scale
Requires much computing power due to
4-dim. ProblemFermions essentialPhysics is at lattice spacing a=0
Precision required(<a few % error in many cases)
( ) ( )
( ) ( ) S
ffff
PsQCD
eUOddUdZ
UO
mUUUUUtrS
−∫
∑∑
=
+⋅+=
ψψψψψψ
ψγψα
,,1,,
1
QCDparameterscaleQCDaspacinglattice
Λ
Wilson 1974
5
Development of lattice QCD simulations (I)
msize 15102 −×≈
lattice size lattice spacing
L= 0.8 fm a = 0.1 fm
1981 First lattice QCD simulation
VAX
Mflopsspeed 1≈
44~ 84 latticequenched approx (no sea quarks)
Creutz-Jacobs-RebbiCreutzWilsonWeingartenHamber-Parisi
6
L(fm) a(fm)
1981 0.8 0.11985 1.2 0.11988 1.6 0.1
1980’s Taking advantage of vector supercomputers
計算機性能の進化
CRAY-1
1 GFLOPS = 毎秒10億回の浮動小数点演算
Development of lattice QCD simulations (II)
7
L(fm) a(fm)1993 2.4 0.07 QCDPAX(JPN) APE(Italy)
1 GFLOPS = 毎秒10億回の浮動小数点演算1 TFLOPS = 毎秒1兆回の浮動小数点演算
Columbia(USA) GF11(USA)
Development of lattice QCD simulations (III)
1990’s QCD dedicated computers
8
L(fm) a(fm)1998 3.0 0.05
Development of lattice QCD simulations (IV)
CP-PACS(JPN) QCDSP(USA)
2000s End of quenching and beginning of full QCD
9
Subject areas of lattice QCD
LQCD
( ) ( ) NONqqxFdxx nn nγ−− =∫ 2
1
0
21 ln,
tbcsdu
s
mmmmmm ,,,,,α
Finite-temperature/density behavior
• eta’ meson mass and U(1) problem• exotic states
glueball, dibaryon, hybrids,…• hadronic matrix elements
proton spin, sigma term, ….• structure functions/form factors
Weak interaction matrix elements
Hadron spectrum and Fundamental constanst of QCD
Hadron physics
• Strong coupling constant• Quark masses
• order of transition• critical temperature/density• equation of state
• K meson amplitudesBKK→ππ decays
• B meson amplitudesfB, BB, form factors
Physics of quark-gluon plasma
CKM matrix and CP violation
10
CP-PACS
Energy density calculated from lattice QCD
Lattice QCD and QGP
11
experiment
Determination of
Comments Davies et al ’97(hep-lat/9703010):Involved extrapolation of Nf=0(quenched) and Nf=2 data to Nf=3
QCDSF-UKQCD ’01(hep-lat/0103023)Continuum estimate with systematic Nf=2 simulations
Davies et al ’02(hep-lat/0209121):Preliminary result based on MILC Nf=3 configurations at a=0.13fm
Systematic Nf=3 full QCD determination expected in a few years
( ) 5=fNz
SMs Mα
Lattice QCD
( ) 5=fNZ
MSs Mα
12
Constraints on the CKM matrix
mixingKK o 0−
mixingBB sdsd0,
0, − ( )
qqq BBB BfM 222 ηρ +∝∆
( ) ( ) 22055 3
811qqq BBBq
oq MBfBqbqbB =−− γγγγ µµ
( )[ ] KK BBA ˆ1 2+−∝ ρηε
( ) ( ) 22055
0
3811 KKK MBfKdsdsK =−− γγγγ µµ
d
s
d
s
B
B
B
B
BB
ff
=ξ
hHh weak'Measured weak amplitude = Known factor
including CKM matrix
QCD matrix element
X
BK
BB fB
13
Summary of lattice results for CKM matrix
240
034
06.013.0
)5(16.1
)37(227.0
87.0ˆ
+−
+−
+−
=
=
=
ξ
GeVBf
B
dd BB
K
Cf. numbers used in the figure left
05.004.021.1
1030228.0
14.006.086.0ˆ
±±=
±±=
±±=
ξ
GeVBf
B
dd BB
K
status 2004 http://www.ckmfitter.in2p3.fr/
-1.5
-1
-0.5
0
0.5
1
1.5
-1 -0.5 0 0.5 1 1.5 2
sin 2β
B → ρρ
B → ρρ
∆md
∆ms & ∆md
εK
εK
|Vub/Vcb|
sin 2β
α
βγ
ρ
η
excluded area has CL < 0.05
C K Mf i t t e r
ICHEP 2004
14
Current focus
15
Current focus: Nf=2+1 dynamical simulation (I)
Quenched approximation : ignore sea quark effects
Acceptable for heavy quarks : c, b, t m>1GeV
Misses important effects with light quarks : u, d, s m<0.1GeV
Dynamical simulation (“unquenching”) is not easy
Need O(10)-O(100) more computer power
Algorithm for odd number of fermions developed only recently
16
Quenching effects on the hadron specturm
Sea quark effects ignored
General pattern reproduced, but clear systematic deviation beyond 10% precision
Calculated quenched spectrum
CP-PACS Collaboration 1998
17
Recent progress due to Development of Tflops-class computersAlgorithms for odd quark flavors (PHMC, RHMC, …)
Two-flavor full QCD (since around 1996)u and d quark dynamical and degenerates quark quenched (no sea quark)
Number of studies: SESAM/UKQCD/MILC/CP-PACS/JLQCD
Two+one-flavor full QCD (since around 2000)u and d quark dynamical and degenerates quark dynamical and heavier
Extensive studies have begun : MILC/CP-PACS-JLQCD
2=fN
12 +=fN
Current focus: Nf=2+1 dynamical simulation (II)
18
Tsukuba/KEK joint effort toward Nf=2+1
1
23
21 2a
β=1.90a ~ 0.10fm20^3 x 408000 trajectory
already finished
β=1.83a ~ 0.12fm16^3 x 325000 trajectory
almost finishedβ=2.05a ~ 0.07fm28^3 x 562000 trajectory
on-going
Fixed physical volume~ (2.0fm)^3 Lattice spacing
Earth simulator@ Jamstec
SR8000/F1@KEK
CP-PACS@Tsukuba
SR8000/G1@Tsukuba
VPP5000@Tsukuba
A three year project 2003-2005
19
Hyperfine splitting at a ~ 0.1fm
Progressively closer agreement with experiment:
Nf=0quenched approxNf=2 u&d quarks dynamicalNf=2+1 u,d,s quarks dynamical
Remaining difference due to finite lattice spacing?-10
-5
0
5
10
(m-m
exp)/
mex
p [%]
mK* (K-input)
Nf=2+1 (RG+clover(NPT))Nf=2 (RG+clover(PT))Nf=0 (RG+clover(PT))
mφ (K-input) mK (φ-input) mK* (φ-input)
a ~ 0.1 fm a ~ 0.1 fma ~ 0.1 fma ~ 0.1 fm
mK* mφ
mK
mK*
examine a dependence
20
Hyperfine splitting toward the continuum limit
0 0.1 0.2 0.3 0.4 0.5
a2 [GeV
-2]
0.85
0.9
0.95
1
1.05
mes
on m
ass
[GeV
]
φ
K*
experiment
K-input
β=1.83
β=1.83
β=1.90
β=1.90
β=2.05
β=2.05
0 0.1 0.2 0.3 0.4 0.5
a2 [GeV
-2]
0.4
0.5
0.6
0.7
0.8
0.9
1
mes
on m
ass
[GeV
]
K
K*
experiment
φ-input
β=1.83β=1.90β=2.05
β=2.05
β=1.90 β=1.83
preliminary
21
Light quark mass results at a ~ 0.1fm
2.0
3.0
4.0
mud
MS(µ
=2G
eV)
Nf=2+1 Nf=2 Nf=0
a ~ 0.1 fm
AWI
a ~ 0.1 fma ~ 0.1 fm
RG + clover(PT) RG + clover(PT)
AWI AWI
K-input
φ-input
RG + clover(NPT)
60
80
100
120
140
160
msM
S(µ
=2G
eV)
Nf=2+1 Nf=2 Nf=0
RG + clover(PT) RG + clover(PT)
a ~ 0.1 fm a ~ 0.1 fmAWI AWI
RG + clover(NPT)
K-input
AWIa ~ 0.1 fm
φ-input
φ-input
K-input
φ-input
K-input
MeVGeVm
MeVmmGeVm
MSs
duMS
385)2(
06.005.32
)2(
±==
±=+
==
µ
µ
Sizably small compared to folklore,
e.g. mud ~ 5MeV, ms ~ 150MeV
22
Nucleon structure functions
23
Nucleon structure functions (I)
Lattice QCD provides normalization of moments in terms of nucleon matrix elements of a set of operators
Progressively difficult for higher moments due to complicated operator structure
Series of calculations by the German group Nf=0 quenched calculationNf=2 u&d dynamical
( ) ( ) NONqqxFdxx nn nγ−− =∫ 2
1
0
21 ln,
qDDqiO nn nn
µµµµµµ γt
Lt
L 2121 1−= etc
QCDSF: M. Goeckeler et al
24
QCDSF quenched results
0.00 0.01 0.02 0.03 0.04(a/r0)
2
0.0
0.1
0.2
0.3
0.0
0.1
0.2
0.3
v3
rgi
v4
rgi
0.00 0.01 0.02 0.03 0.04(a/r0)
2
0.2
0.3
0.4
0.5
0.2
0.3
0.4
0.5
0.2
0.3
0.4
0.5
v2a
rgi
v2b
rgi (p=p1)
v2b
rgi (p=0)
3x
x
2x
Flavor non-singlet moments for n=2,3,41st moment shows a sizable deviation from experiment ( a long-standing puzzle)
calculated with 3 types of lattice operators
experiment
hep-ph/0410187
25
QCDSF Nf=2 results
0 1 2 3 4 5 6 7 8
(r0 m
PS)2
0
0.2
0.4
√2 mK
v2b
(u-d) RGI
0 0.02 0.04 0.06
(a / r0)2
0
0.2
0.4
x
2x
3x
Similar trend as the quenched case, i.e., Reasonable agreement for n=3,4Significant deviation for n=2
experiment
hep-lat/0409162
26
Moments of nucleon structure functions (II)
Linear chiralextrapolation misses experimentSmall quark mass behavior is the issuePion not light enough to see curvature?
From D. Dolgov et al hep-lat/0201021
( ) ⎟⎟⎠
⎞⎜⎜⎝
⎛+
+Λ+
−= 222
22
2
2
ln4
131 π
πχ
πππ
dmm
mm
fg
ax An
n
W. Detmold et al hep-lat/0103006
dux
−
2πm
27
Future perspectives
28
Next generation of computers
10 TFLOPS class machines for QCD
QCDOC (USA) 2004-2005
Well advanced
1st shipment to UK this month
Large installation in USA early next year
ApeNEXT (Italy) 2005
Some delay in R&D
Large installation in Italy and possibly Germany
CP-PACS II (Japan) 2006
Development to start in fiscal 2005
2/3 installation in early 2006, completion in 2007
29
CP-PACSに至る歴史:20年間の発展
1978 1980 1989 1996
世界最高速を達成したCP-PACS
第1号機PACS-9
22億円 (5年間)競争的資金(科研費創成基礎研究)614億回/秒CP-PACS1996年
3億円 (3年間)競争的資金(科研費特別推進)14億回/秒QCDPAX1989年
3000万円基盤的研究費+学内裁量経費3百万回/秒PAX-32J1984年
600万円基盤的研究費4百万回/秒PAX-1281983年
150万円基盤的研究費50万回/秒PAXS-321980年
150万円基盤的研究費7千回/秒PACS-91978年
金額開発・制作費の種類計算速度名称完成年
第2号機PAXS-32
第5号機QCDPAX
30
0.1
1
10
100
1000
104
105
106
1975 1980 1985 1990 1995 2000 2005 20100.0001
0.001
0.01
0.1
1
10
100
1000
GFLOPS
year
TFLOPS
CRAY-1
CP-PACS
Earth Simulator
CP-PACS-II
cluster systemsQCD-PAX
実機設計
試作・検証・確定
実機製作
計算開始
システムソフトウェア開発
計算物理プログラム開発
全システムによる計算の実施
25% 100%
平成15年4月 平成16年4月 平成17年4月 平成18年4月 平成19年4月
基本仕様検討
テストシステム構築・テスト実施
平成20年4月
CP-PACS-II project (I)
計算科学研究センター発足
CP-PACS稼動10周年
平成18年10月
75%
計画開始
全システム完成
準備研究実施中
高性能化の著しいコモディティ(汎用品)のプ
ロセッサ技術、ネットワーク技術を用いて超
並列システムを短期に開発・製作
CP-PACS-II(仮称) CP-PACS
ピーク性能 24.6TFLOPS 0.61TFLOPS主記憶容量 6.1TByte 0.128TByteディスク容量 1.05PByte 1.05PByte開発期間 2年(H17~18年度) 4.5年(H4年度~8年度)完成時期 平成18年度末 平成8年9月
1TFLOPS=1兆回/秒の計算性能
31
計算物理学研究センターの計算科学研究センターへの
改組・拡充計算物理学研究センター
計算素粒子物理学研究部門
計算宇宙物理学研究部門
計算物性物理学研究部門
計算生命物理学研究部門
並列計算機工学研究部門
計算科学研究センター
素粒子宇宙研究部門
物質生命研究部門
地球生物環境研究部門
超高速計算システム研究部門
計算情報学研究部門
平成4年度~15年度 平成16年度発足
2
2
2
2
3
教員 11名(教授4 助教授4 講師3)
6
11
3
5
6
教員 31名(教授11 助教授12 講師8)
計算科学研究センター全体の
事業として実施
事業統括責任者(センター長)
物質・生命プロジェクト責任者
素粒子・宇宙プロジェクト責任者
クラスタ計算機プロジェクト責任者
平成16年4月改組・拡充
CP-PACS-II project (II)
32
International Research Network for Computational Particle Physics
UK core institution:University of Edinburgh
Dept. of PhysicsEPCC
Germany core institution:
DESYVon Neumann
Inst.for computing
USA core institution:Fermi National Accelerator
Laboratory(FNAL)
Japan core institution:
University of Tsukuba
Center for Computational
Sciences
SciDACNetworkin USA
Edinburgh
GlasgowLiverpool
Southampton
Swansea DESY/NeumannBerlin/Zeuthen
BielefeldRegensburg LatFor
Networkin Germany
KEK
Hiroshima U
LFT ForumNetworkin Japan
Future expansion to EU NetworkItaly, France, Spain, Denmark,…
Main supercomputer sites
International Lattice Data Grid (ILDG)database of QCD gluon configurations at major supercomputer facilities acceleration of research via mutual usage of QCD gluon configurations via fast internetfuture international sharing of supercomputingand data storage resources
Future expansion to Asia/Oceania
Kyoto U
UKQCDNetworkin United Kingdom
U. Tsukuba
FNAL
Washinghon U
UCSB
MIT/Boston U
BNL/Columbia
JLAB
Arizona
Utah
Indiana
St.Louise
JSPS core-to-core program
QCDOC x 2
QCDOC
APENEXT
CP-PACS II
KEK supercomputer
33
Summary
Definitive prospect toward exact QCD predictions with realistic quark spectrum over the next few years
Firm numbers to our phenomenology/experiment colleaguesQuantitative understanding of the full range of strong interactions
1935 meson theory (Yukawa)1951 strangeness (Gell-Mann-Nishijima) 1961 chiral symmetry and pion(Nambu-Jona-Lasinio)1973 QCD and asymptotic freedom(Gross-Wilczek-Politzer)1974 Lattice QCD(Wilson)1981 Monte Carlo simulation of QCD(Creutz-Jacobs-Rebbi)
Lattice QCD: a prototype of computational research in which theory and instrument making go hand in hand.