x
Tr
(S(~x, t;~0, 0) S † (~x, t;~0, 0)).
(B.40)Weintrodu e the Fourier transformof the positionspa e propagator:
S(~x, t;~0, 0) = 1
where
N
andN 4
are the number of latti e sites in the spatial and temporaldire tions, respe tively. This yields:
C P P (t) = 1
Using the following expression forthe Dira -deltafun tion:
δ(~p − ~p ′ ) = 1
S(p) =
4
or5
X
ξ=0
S ξ (p)γ ξ ,
(B.45)where
γ 0 ≡
1 and the indexξ
runs from 0 to 4 in the ase of overlap andCreutz fermions orfrom 0to 5in the ase of Wilson twistedmass fermions.
Hen e, we obtain:
C P P (t) = N c N d
N 3 N 4 2 X
~ p
X
p 4 ,p ′ 4 4
or5
X
ξ=0
S ξ (~p, p 4 ) S ξ ∗ (~p, p ′ 4 ))e i(p 4 −p ′ 4 )t ,
(B.46)where
N d =
Tr(
1)
isthe numberof Dira omponents (i.e. the dimension ofspa e-time) and
N c =
Tr(
1c )
is the number of olours (in the free ase thestru ture in olour spa e istrivial).
Improvements of the HMC
algorithm
The gaugeeld ongurations that wehave usedforthis proje t were
gener-atedwiththetwistedmassLatti eQCDprogramsuite(tmLQCD)ofJansen
and Urba h [83℄. A detailed des ription of all the te hni al details is given
in this referen e. Here we shortly dis uss a few improvements of the HMC
algorithmthat are relevant fromthe pointof view of this thesis.
Forsomelatti eDira operators(e.g. Wilsontwistedmass),itispossible
to de ompose the Dira matrix intosubspa es of even and odd latti e sites,
thus redu ing the dimension of the problem. Su h te hnique is alled
even-odd pre onditioning [126℄.
Another approa h is to use more than one set of pseudo-fermion elds,
i.e. split the fermion determinant into two (or more) parts. One of the
widely used methods of this kind is alled the Hasenbus h tri k (or mass
pre onditioning)[127, 128℄and onsistsinutilizingtheidentity(examplefor
the
N f = 2
ase with degenerate quark massesµ
):| det( ˆ D)| 2 = det( ˆ D ˆ D † + µ 2 ) det D ˆ ˆ D † D ˆ ˆ D † + µ 2
!
.
(C.1)Su h de omposition splits the ontribution of the low-frequen y and
high-frequen y modes of
D ˆ ˆ D †
and thus redu es the ondition number of theproblem. It also allows for integration of dierent parts of the a tion on
dierenttime s ales, su hthat the most expensive part an besimulated on
the oarsest time s ale. A general version ofan HMC algorithm
in orporat-ing even-odd pre onditioning, mass pre onditioning and multiple time s ale
integration waspresented by Urba h,Jansen, Shindlerand Wenger[129℄. It
wasalsoshownby numeri alinvestigationthatsu hversionof thealgorithm
the smallquark mass limit. This version of the algorithmis the base of the
tmLQCD suite initspart that wasrelevant forthe generation ofgauge eld
ongurations used inthis proje t.
Anotherimportantimprovementof theHMC algorithm omesunder the
name ofPolynomialHMC (PHMC).It wasintrodu edand analyzedin[115,
116, 117, 118℄. This version of the algorithm an be applied to simulate
non-degenerate quarks. It is used e.g. in the tmLQCD suite in simulations
in ludingthe strangeand harm quark.
The number of other improvements of the HMC algorithm is very large
and isstillin reasing. Tonalizethis appendix wejustmention afewmore.
Fortheirdes ription werefertooriginalpapers. A wideand important lass
of improvements on ern integration s hemes and ome under the name of
multiple time-s ale integration. The generalization of the leap-frog s heme
to multiple time s ales was originally proposed by Sexton and Weingarten
[130℄. Another approa h is the so- alledse ond order minimalnorm (2MN)
integrator[131,132℄. Avariantof the HMCalgorithm alledRationalHMC
(RHMC) was dis ussed in [133, 134, 135, 136℄. Domain-de omposed HMC
was introdu ed in a series of papers by Lüs her [137, 138, 139℄ and later
augmented by low-mode deation [140℄.
Tree-level test of zero modes
subtra tion
In this appendix, we show the results of a free-eld test of routines used to
subtra t the zero modes (we will refer tothem as subtra tion routines) at
the level of propagators. The test is performed on a small latti e of
4 3 × 8
,with quarkmass set to
am = 0.2
. Weperform thesubtra tion inthree ways for the pseudos alar (PP) and s alar (SS) orrelation fun tion, using:•
formula (2.16) for the PP orrelator and an analogous formula for SS(analyti alformula),
•
GWC ode withsubtra tion routines for pointsour es,•
GWC ode withsubtra tion routines for sto hasti sour es.Usingnotation of Se tion 4.3, we write the mesoni orrelation fun tion
as:
C(t) = C 00 (t) + 2C 0N (t) + C N N (t),
(D.1)where the rst two terms involve the zero modes. Computing orrelation
fun tionsfromthefullpropagator(withallmodes)leadsto
C(t)
,whileifthezero modes are subtra ted atthe levelof propagators, onlythe part
C N N (t)
isobtainedbyperforming ontra tions,i.e. subtra tionofzeromodes an els
both the diagonal ontribution
C 00 (t)
and the mixed oneC 0N (t)
.We remind here the formulafor the pseudos alar orrelator (2.16) and
gen-eralize itto in ludethe s alar ase:
C(t) = N c N d
where we obtain the pseudos alar orrelator by hoosing:
s(ξ) = 1
for allξ
and the s alar orrelator if we take
s(ξ) = −1
forξ = 0
ands(ξ) = 1
forξ = 1, 2, 3, 4
.To isolate the ontribution of the zero-modes, we have to al ulate the
diagonalpart
C 00 (t)
and the mixed partC 0N (t)
:latter orresponds to non-zero modes in the mixed term).
The ontribution of the zero modes is
C 00 (t) + 2C 0N (t)
and it is for the4 -0.046231202662437964612252 02
5 -0.048644806504103299538144 26
6 -0.055983137376237557258917 83
7 -0.068542693495896617195128 90
As we have shown analyti ally in Se tion 4.3, the ontribution of the zero
modes isthe sameinboththepseudos alarandthes alar orrelator,up toa
sign,whi hisamatterof onvention. Withsu h onvention,the ontribution
of the zero modes exa tly an les inthe sum
C P P +SS = C P P + C SS
.Wealsoshowthe part
C N N (t)
ofthe PPand SS orrelators,i.e. the part with zero modes subtra ted:t C_PP(t)
0 3.2070495598256036906548160 9
1 0.0538755232498468739077424 1
2 0.0118515945140288403436557 0
3 0.0043685036209793262629297 3
4 0.0025747032878168582548106 3
5 0.0043685036209795205519590 4
6 0.0118515945140288542214435 1
7 0.0538755232498471792190741 8
t C_SS(t)
0 -2.899846672457193630378924 35
1 0.0534957243448562630439013 8
2 0.0097986066820899761498076 0
3 0.0002349545374039987977177 2
4 -0.002407409019565424912467 93
5 0.0002349545374036310363408 1
6 0.0097986066820900039053832 1
7 0.0534957243448555622156170 8
Wewill use these numbers to ompare with the subtra tionroutines.