Multipoint reweighting method and beta functions 5 for the calculation of QCD equation of state 1 0 2 n a J 6 RyoIwami∗1, S.Ejiri2, K.Kanaya3, Y.Nakagawa1,T.Umeda4, D.Yamamoto1 2 (WHOT-QCDCollaboration) ] t 1GraduateSchoolofScienceandTechnology,NiigataUniversity,Niigata950-2181,Japan a l E-mail:[email protected] - p 2DepartmentofPhysics,NiigataUniversity,Niigata950-2181,Japan e 3GraduateSchoolofPureandAppliedSciences,UniversityofTsukuba,Tsukuba,Ibaraki h [ 305-8571,Japan 4GraduateSchoolofEducation,HiroshimaUniversity,Hiroshima739-8524,Japan 1 v 1 We study a reweightingmethodaiming at numericalstudies of QCD at finite density, in which 3 3 the conventional Monte-Carlo method cannot be applied directly. One of the most important 6 problemsinthereweightingmethodistheoverlapproblem. Tosolveit, weproposetoperform 0 . simulations at several simulation points and combine their results in the data analyses. In this 1 0 report, we introduce this multipoint reweighting method and test if the method works well by 5 measuringhistogramsofphysicalquantities. Usingthismethod,wecalculatethemesonmasses 1 : as continuousfunctionsof the gaugecouplingb andthehoppingparametersk in QCD at zero v i density. Wethendeterminelinesofconstantphysicsinthe(b ,k )spaceandevaluatethederiva- X tivesofthelatticespacingwithrespecttob andk alongthelinesofconstantphysics(inverseof r a thebetafunctions),whichareneededinacalculationoftheequationofstate. The32ndInternationalSymposiumonLatticeFieldTheory 23-28June,2014 ColumbiaUniversityNewYork,NY ∗Speaker. (cid:13)c Copyrightownedbytheauthor(s)underthetermsoftheCreativeCommonsAttribution-NonCommercial-ShareAlikeLicence. http://pos.sissa.it/ MultipointreweightingmethodandbetafunctionsforthecalculationofQCDequationofstate RyoIwami 1. Introduction In a study of QCD phase diagram at finite temperature (T) and density, the complex quark determinant causes a serious problem in numerical simulations. The reweighting method is com- monlyusedtoavoidthisprobleminthelowdensityregion. However,whenweincreasethechemi- calpotential(m ),thesignproblemandtheoverlapproblembecomessevere. Inthisreport,wefocus on the overlap problem. The overlap problem is expected to be milder if one changes a couple of parametersatthesametime. Forexample,WHOT-QCDcollaborationinvestigatedthephasestruc- ture ofN-flavor QCDin the heavy-quark region, and found that the system at large quark masses f is controlled by only two combinations of parameters, b +48(cid:229) Nf k 4 and (cid:229) Nf k Ntcosh(m /T) f=1 f f=1 f f on an N3×N lattice, where b = 6/g2, and k and m are the hopping parameter and chemical s t f f potential of fth-flavor, respectively [1]. This means that, when one changes the parameters with keeping these combinations constant, the system does not change, thus the overlap problem does notarise. Weexpectsuchcombinations ofparametersexistalsointhelight-quark region. Inthatstudyintheheavy-quarkregion[1],themultipointreweightingmethod[2]forb played animportantrole: Combiningdataobtainedatdifferentb ,wecouldcalculatetheeffectivepotential inawiderangeoftheobservablevalues,thatwasmandatoryinareliableevaluationofthetransition point. In the present paper, we extend the multipoint reweighting method in the multi-parameter spaceofb andk toovercometheoverlapprobleminthelight-quark region. f Inthenextsection,themultipointreweightingmethodisintroduced,and,inSec.3,weexamine the overlap problem performing numerical simulations in two-flavor QCD. We then calculate the meson masses, thelines ofconstant physics, andthederivatives ofthelattice spacing withrespect tob andk alongthelinesofconstant physicsinSec.4. Thesection5isfortheconclusions. 2. Multipoint reweighting method We extend the multipoint reweighting method to the multi-parameter space such that both gaugeandquarkparametersarechangedsimultaneously. Letusfirstdefineahistogramforasetof physical quantities X =(X ,X ,···)as 1 2 w(X;b ,~k ,~m ) = DU (cid:213) d (X −Xˆ)e−Sg (cid:213)Nf detM(k ,m ). (2.1) i i f f Z i f=1 where S is the gauge action, M is the quark matrix and Xˆ =(Xˆ ,Xˆ ,···) is the operators for X. g 1 2 WedenoteS≡S −(cid:229) lndetM. Thecouplingparametersofthetheoryareb ,~k =(k ,k ,···,k ), g f 1 2 Nf and~m =(m ,m ,···,m ). Then, thepartition function isgivenbyZ(b ,~k ,~m )= w(X;b ,~k ,~m )dX 1 2 Nf withdX =(cid:213) dX,andtheprobability distribution functionofX isgivenbyZ−1w(X;b ,~k ,~m ). The i i R expectation valueofanoperator O[Xˆ]whichiswrittenintermsofXˆ iscalculated by 1 hO[Xˆ]i(b ,~k ,~m )= Z(b ,~k ,~m ) O[~X]w(~X;b ,~k ,~m )dX. (2.2) Z For simplicity, we denote the set of coupling parameters (b ,~k ,~m ) as b in the following of this section. 2 MultipointreweightingmethodandbetafunctionsforthecalculationofQCDequationofstate RyoIwami We perform a simulation at b and calculate the histogram at b . We fix three quantities X, 0 S(b )≡S and S(b )≡S to construct the histogram, where S is the value of the action with the 0 0 coupling parameters b ontheconfiguration generated atb . FromEq.(2.1),wefind 0 w(X,S,S ;b )=e−(S−S0)w(X,S,S ;b ). (2.3) 0 0 0 Thehistogram ofX isgivenbyw(X;b )= w(X,S,S ;b )dSdS . 0 0 When X has large correlation with the difference of the actions, S−S , the peak position of R 0 the distribution may change appreciably, causing the overlap problem. (SeeSec.3.) Toovercome the overlap problem and to obtain w and V that are reliable in a wide range of X, we extend eff thereweighing formulas tocombine dataobtained atdifferent simulation points [1,2]forthecase including thefermionaction. We combine a set of N simulations performed at b with the number of configurations N sp i i wherei=1,···,N . UsingEq.(2.3),theprobability distribution function atb isrelatedtothatat sp i b as Z−1(b )w(X,S,~S;b )=Z−1(b )e−(Si−S)w(X,S,~S;b ). (2.4) i i i Summinguptheseprobability distribution functions withtheweightN, i Nsp Nsp (cid:229) N Z−1(b )w(X,S,~S;b )=eS(cid:229) N Z−1(b )e−Siw(X,S,~S;b ), (2.5) i i i i i i=1 i=1 weobtain Nsp w(X,S,~S;b )=G(S,~S;b ,~b ) (cid:229) N Z−1(b )w(X,S,~S;b ) (2.6) i i i i=1 where~b =(b ,···,b )and 1 Nsp e−S G(S,~S;b ,~b )= . (2.7) (cid:229) Ni=sp1Nie−SiZ−1(b i) Notethat theleft-hand sideofEq.(2.5) givesanaivehistogram using allthe configurations disre- garding the difference in the simulation parameter (b ,~k ). Thehistogram w(X,S,~S;b ) at (b ,~k ) is givenbymultiplying G(S,~S;b ,~b )tothenaivehistogram. Thepartition functionisgivenby Nsp Nsp Z(b )= (cid:229) N G(S,~S;b ,~b )Z−1(b )w(X,S,~S;b )dXdSd~S= (cid:229) N G(S,~S;b ,~b ) . (2.8) i i i i i=1 Z i=1 (bi) D E The right-hand side is just the naive sum of G(S,~S;b ,~b ) obtained on all the configurations. The partition function atb canbedetermined, uptoanoverallfactor, bytheconsistency relations, i Z(b )= (cid:229)Nsp N G(S,~S;b ,~b ) = (cid:229)Nsp N e−Si (2.9) i k=1 kD i E(bk) k=1 k*(cid:229) Nj=sp1Nje−SjZ−1(b j)+(bk) 3 MultipointreweightingmethodandbetafunctionsforthecalculationofQCDequationofstate RyoIwami fori=1,···,N . Denoting f =−lnZ(b ),theseequations canberewrittenby sp i i 1=(cid:229) Nsp N (cid:229) Nsp N exp[S −S − f + f ] −1 . Starting from appropriate initial values of k=1 k j=1 j i j i j (cid:28)(cid:16) (cid:17) (cid:29)(bk) f,wesolve theseequations numerically byaniterative method. Notethat, inthiscalculation, one i of f mustbefixedtoremovetheambiguitycorresponding totheundetermined overallfactor. i Theexpectation valueofanoperatorXˆ atb canbeevaluatedas hXˆi(b ) = Z(1b )Z Xw(X,S,~S;b )dSd~S = Z(1b )i(cid:229)N=sp1Ni XˆG(S,~S;b ,~b ) (bi), (2.10) D E andthehistogram ofX isobtained by Nsp w(X;b ) = (cid:229) N d (X−Xˆ)G(S,~S;b ,~b ) , (2.11) i i=1 (bi) D E Note that, again, (cid:229) Nsp N XˆG in the right-hand side is just the naive sum of XG over all the i=1 i (bi) configurations disregarding thedifferenceinthesimulationpointb . (cid:10) (cid:11) i 3. Overlapproblem andhistograms To study if the multipoint reweighting method is useful in avoiding the overlap problem, we perform simulations ofQCDwithdegenerate 2flavors ofclover-improved Wilson quarks coupled withRG-improvedIwasakiglue,atm =0. Theimprovementparametersoftheactionarethesame asthoseadopted inRef.[4]. Thesimulations arecarried outonan84 latticeat9simulation points (3b ’s×3k ’s)fortheteststudyinthissection, andona164 latticeat30points(6b ’s×5k ’s)in Sec.4. Thenumberofconfigurations forthemeasurementis200ateachsimulationpoint. The action S required in the reweighting method is composed by Wilson loops and lndetM. The calculation of the fermion part requires large computational cost. In this study, we evaluate detM by measuring the first and second derivatives of lndetM at several k ’s, where i=1,2,···, i on each configuration, and interpolate the determinant between k and k assuming a quadratic i i+1 function, lndetM(k )=lndetM(k )+C (k −k )+C (k −k )2+C (k −k )3+C (k −k )4. The i 1 i 3 i 3 i 4 i coefficients C are determined such that the first and second derivatives are consistent with the a measured valuesatk andk . Theclovertermwiththecoefficient c isalsoevaluated interms i i+1 SW of its derivatives. Here, because c depends on b in our choice, it affects the b -dependence of SW observables. However, we find that the effects from c are very small in the range of b rele- SW vant in our study. We thus approximate the b -dependence of the action by a linear function, i.e. lndetM(b ,~k )=lndetM(b ,~k )+(b −b )(dc /db )(¶ lndetM/¶ c )(b ,~k ) [3]. Note that the 0 0 SW SW 0 overallconstant oflndetM isnotneeded. In the left panel of Fig. 1, we show the results for the improved plaquette P = c W1×1+ 0 2c W1×2 of the Iwasaki action at b =1.825, where c =−0.331, c =1−8c , and Wi×j is the 1 1 0 1 (i× j)Wilsonloop. Blackdotsrepresenttheexpectation valuesofPatthethreesimulation points withoutreweighting. Blue,greenandpurplecurvesaretheresultsofthenaivereweightingmethod usingthedataatk =0.1400,0.1425,and0.1440,respectively. Eachresultofthenaivereweighting method is reliable around the corresponding simulation point, but fails reproducing far away sim- ulation results. The reason can be easily understood by consulting the histogram of P: Red curve 4 MultipointreweightingmethodandbetafunctionsforthecalculationofQCDequationofstate RyoIwami 0.5 0.5 k =0.1400 k =0.1400 1.7 k =0.1400 k =0.1412 k =0.1412 kk ==00..11442450 0.4 kk ==00..11442346 0.4 kk ==00..11442346 simulation point k =0.1440 k =0.1440 1.69 multi point 0.3 0.3 P 1.68 0.2 0.2 0.1 0.1 1.67 0 0 1.64 1.66 1.68 1.7 1.72 1.64 1.66 1.68 1.7 1.72 1.66 0.14 0.1k 42 0.144 P P Figure1: Left: TheexpectationvalueofP≡c W1×1+2c W1×2 asafunctionofk atb =1.825. Middle: 0 1 The histogram of P at various k ’s obtained by the naive reweighting method using the configurations at k =0.140.Right:ThehistogramofPbythemultipointreweightingmethod. (b ,k )=(1.8325,0.14000) (b ,k )=(1.8250,0.14000) (b ,k )=(1.8400,0.14000) (b ,k )=(1.8250,0.14160) (b ,k )=(1.8475,0.14000) (b ,k )=(1.8250,0.14240) 0.02 0.02 0.016 0.016 0.012 0.012 0.008 0.008 0.004 0.004 0 0 4 9.6 4 9.6 3 3 9.8 9.8 2 2 [dS / dk ] 1 10 dS / db [dS / dk ] 1 10 dS / db SUB 0 SUB 0 -1 -1 Figure 2: The b -dependence (left) and k -dependence (right) of the histogram for N−1(¶ S/¶b ) and site N−1[¶ S/¶k ] ≡N−1[¶ S/¶k −(288Nk 4/c )(¶ S/¶b )],whereN =84. site SUB site f 0 site in the middle panel ofFig. 1 isthe original histogram at(b ,k )=(1.825,0.140), and green, blue, magentaandlightbluecurvesarethehistogramsatk =0.1412,0.1424,0.1436,and0.144,respec- tively, estimated by the naive reweighting method Eq. (2.3) using the data at k =0.140. Because the histograms atk other than the simulation point is calculated asthe product of the reweighting factor and the original histogram, the histograms are not reliable out of the range of the original distribution. Infact,thevalueofPdistributes between1.64and1.69. Evenwhenk ischanged by the reweighting method, the upper and lower bounds of the distribution does not change in Fig. 1 (middle). Since the expectation value is approximately the peak position of the histogram, the expectation valuealsocannot gooutoftherangeofthedistribution, asshowninFig.1(left). To enlarge the range of the distribution, we combine the simulation data obtained at k = 0.1400, 0.1425 and 0.1440 using the multipoint reweighting method explained in the previous section. TheredcurveintheleftpanelofFig.1istheresultofthemultipointreweightingmethod. Wefindthattheredcurvesmoothlyconnectsallthedirectsimulationresultswithsmallerrorbars. 5 MultipointreweightingmethodandbetafunctionsforthecalculationofQCDequationofstate RyoIwami b =1.806 mPS / mV = 0.70 0.8 b =1.8125 0.1435 mPS / mV = 0.72 b =1.819 mPS / mV = 0.74 b =1.825 mPS / mV = 0.76 b =1.831 b =1.837 0.143 0.75 V m / PS k0.1425 m 0.7 0.142 0.65 0.1415 6.95 7 7.05 7.1 7.15 1.815 1.82 1.825 1.83 1.835 1 / k b Figure3:Left:Thepseudoscalarandvectormesonmassratiom /m .Right:Thelinesofconstantphysics PS V foreachm /m inthe(b ,k )plane. PS V In the right panel of Fig. 1, histograms from the multipoint reweighting method are plotted for k =0.1412, 0.1424,0.1436, and0.144. Themethodisapplicable tootherobservables. Theexpectation valuesof¶ S/¶b and [¶ S/¶k ] ≡¶ S/¶k −(288N k 4/c )(¶ S/¶b ) SUB f 0 are needed in the calculation of the equation of state in the integral method. In the right and left panels of Fig. 2, we show the b - and k -dependence of the 2-dimensional histogram of these quantities using the multipoint reweighting method. The histogram moves as b and k are varied. We can compute the expectation values of ¶ S/¶b and [¶ S/¶k ] as continuous functions of b SUB andk withouttheoverlapproblem. Onceweobtainphysical quantities ascontinuous functions of couplingparameters,wecancalculatethelinesofconstantphysicsinthecouplingparameterspace aswellasthebetafunctions. 4. Lines ofconstant physics andbeta functions In this study, we define the lines of constant physics by fixing the dimension-less ratio of pseudoscalar and vector meson masses m /m in the (b ,k ) space. On the lattice, we measure PS V dimension-less observables m a and m a, where a is the lattice spacing. Because these lattice PS V observables varyaswechange b ork ,aisvariedwhenwemovealongalineofconstant physics. Thebeta functions ¶b /¶ a and¶k /¶ aare defined through thevariation aalong alineofconstant physics. Thebetafunctions areneededinthecalculation oftheequation ofstate. Themultipoint reweighting methodisuseful forthecalculation ofthebeta functions because wecancalculate observables ascontinuous functions ofb andk . Combining thedataat30simu- lationpoints(6b ’s×5k ’s)onthe164 latticebythemultipointreweightingmethod,weobtainthe massratiom /m plotted intheleftpanel ofFig.3. Fromthese data, determine thelinesofcon- PS V stantphysicsform /m =0.70,0.72,0.74and0.76,asshowninFig.3(right). Wethencalculate PS V thederivatives,(m a)¶b /¶ (m a)and(m a)¶k /¶ (m a)alongeachlineofconstantphysics. The V V V V 6 MultipointreweightingmethodandbetafunctionsforthecalculationofQCDequationofstate RyoIwami 0 0.04 -0.5 0.02 a) a) mv mV /d -1 /d 0 bma(dv kma(dV m / m = 0.70 m / m = 0.70 PS V -1.5 mPS / mV = 0.72 -0.02 mPS / mV = 0.72 PS V m / m = 0.74 m / m = 0.74 PS V PS V m / m = 0.76 m / m = 0.76 PS V PS V -2 -0.04 1.816 1.82 1.824 1.828 1.832 1.816 1.82 1.824 1.828 1.832 b b Figure4: Thebetafunctions:(m a)¶b /¶ (m a)(left)and(m a)¶k /¶ (m a)(right). V V V V resultsareshowninFig.4. Combiningthesebetafunctionswiththemeasurementofh¶ S/¶k iand h¶ S/¶b ionfinite-temperature lattice,wecancalculate theequation ofstate. 5. Conclusions andoutlook We discussed the multipoint reweighting method in a multi-dimensional parameter space to avoid the overlap problem. Using the method, we can reliably calculate histograms of physical quantities as well as the expectation values of the physical quantities, as continuous functions of coupling parameters. These enable us to compute the lines of constant physics and the beta functions, whichareneededinacalculation oftheequationofstate. Ourfinalobjective isastudyoffinitedensity QCD.Usingthemultipoint reweighing method, we may absorb the main effect of the chemical potential by a change of b and k . If so, we may investigate finitedensity QCDavoiding thesignproblem. Acknowledgments Thiswork isinpart supported byGrants-in-Aid ofthe Japanese Ministry of Education, Culture,Sports,ScienceandTechnology (Nos.26400244, 26400251). References [1] H.Saito,S.Ejiri,S.Aoki,K.Kanaya,Y.Nakagawa,H.Ohno,K.Okuno,T.Umeda(WHOT-QCD Collaboration),Phys.Rev.D89,014508(2014). [2] A.M.FerrenbergandR.H.Swendsen,Phys.Rev.Lett.63,1195(1989). [3] B.B.Brandt,H.Wittig,O.PhilipsenandL.Zeidlewicz,PoS(LATTICE 2010)172. [4] S.Ejiri,Y.Maezawa,N.Ukita,S.Aoki,T.Hatsuda,N.Ishii,K.Kanaya,andT.Umeda(WHOT-QCD Collaboration),Phys.Rev.D82,014508(2010). 7