0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 r 0 M
(a/r 0 ) 2
MTM Nucleon Overlap Nucleon MTM Delta Overlap Delta
Figure 5.8: Continuum limit s aling of the MTM and overlap light baryon
masses (nu leon, delta). The overlap masses are omputed at the mat hing
mass. The MTMvales are slightly shiftedtothe leftand theoverlap ones to
the right, for learerpresentation.
In parti ular, even if the baryon massesare not onsiderably ae ted by the
zero modes, the matrix elements of the baryoni orrelation fun tions still
might hange signi antly, as is anti ipated in [124℄. Therefore, this issue
will be investigated further in the future [112℄. In parti ular, the role of
the zero modes an be assessed by usingdierentinterpolatingoperatorsfor
baryoni orrelationfun tions, sin e dierentoperators ouple ina dierent
waytozeromodes[124℄. Moreover,theexpli itsubtra tionpro eduremaybe
followedandtheoverlapofthesour esandthezeromodesmaybe omputed.
However,thepresentanalysisalreadyallowsusto on ludethatthe
mag-nitudeof thezero modesee ts indierentobservables maybedierentand
that some observables maybe mu hmore vulnerabletothe zeromodes
on-tribution(e.g.
f π
)than some other (e.g. the baryon masses).5.3 Topologi al harge and sus eptibility
Inthisse tion,wereporttheresultsofinvestigationofsometopologi alissues
relatedtogaugeeld ongurationsthatwehaveused. Forsomeofthem,we
have omputedthezeromodes, whi hallowsusto al ulatetheirtopologi al
-10
β=3.9, L/a=16, aµ=0.004, 544 confs -10 -5 β=3.9, L/a=16, aµ=0.0074, 260 confs
-10
β=4.05, L/a=20, aµ=0.003, 300 confs -10 -5 β=4.2, L/a=24, aµ=0.002, 396 confs
Figure 5.9: Monte Carlo history of the index of the overlap operator for
dierent ensembles. The verti al axiss ale is the same for allplots.
harge as the dieren e of the numbers of zero modes in the negative and
positive hirality se tors (eq. (1.79)), i.e. the index of the overlap Dira
operator. Inpra ti e, zero modes ona given ongurationo uronly inone
hirality se tor or, in other words, the probability of having zero modes in
both se torsfor the same ongurationiszero [122℄.
Fig. 5.9 shows the Monte Carlo histories of the index of the overlap
op-eratorforfourdierentensemblesof ongurations:
β = 3.9, L/a = 16, aµ = 0.004
;β = 3.9, L/a = 16, aµ = 0.0074
;β = 4.05, L/a = 20, aµ = 0.003
;β = 4.2, L/a = 24, aµ = 0.002
; all of them orresponding to linear latti e extent ofL ≈ 1.3
fm. The plots indi atethat the auto orrelations inMonte Carlotime arerathernot largeanddierenttopologi alse torsare sampled.The verti al s aleon ea hof the plots is the same and hen eit isnoti eable
that topologi al harge u tuationsare the largestfor
β = 3.9
andonsider-ably smallerand omparable to ea h other for
β = 4.05
andβ = 4.2
.The histograms of the index are shown in Fig. 5.10. To allow for
om-parison, the number of ongurations whi h orresponds to the given index
N(index)
hasbeennormalizedby thetotalnumberof ongurationsforea h0 β=3.9, L/a=16, aµ=0.004, 544 confs
0 β=3.9, L/a=16, aµ=0.0074, 260 confs
0 β=4.2, L/a=24, aµ=0.002, 396 confs
Figure 5.10: Histograms of the index of the overlap operator for dierent
ensembles. Also shown are Gaussian ts of the distributions. The axes
s ales are the same for allplots.
ensemble
N(total)
and the axesare plotted inthe same s ale. We alsoshowts tothe Gaussian probabilitydistribution:
p(Q) = 1
√ 2πσ e − Q
2
2σ2 ,
(5.5)where
Q
isthe index andσ
isthe onlytting oe ient, i.e. we enfor e themean ofthedistribution tobe0,sin ethe probabilitydistributionshouldbe
symmetri withrespe tto ongurations withnegativeandpositive
topolog-i al harge.
The plots leadto the on lusion that the probabilitydistributions of the
index
Q
areapproximatelyGaussian. However, thequalityofthetsisrather poor the most frequent value of the index is intwo ases atQ 6= 0
and ingeneral the symmetry with respe t to
Q = 0
is not very good. This resultsfrom the fa t that the total number of ongurations for ea h ensemble is
too smalland an order of magnitude in rease in statisti s would be needed
to obtaina reliable distribution.
0 0.02 0.04 0.06 0.08 0.1
0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 r 0 4 χ top
(a/r 0 ) 2
β=4.2 β=4.05 β=3.9
m π ≈300 MeV m π ≈450 MeV
Figure 5.11: Topologi al sus eptibilityfor 4 ensembles of gauge eld
ong-urations. The
m π ≈ 450
MeV ensemble is slightly shifted to the right forbettervisibility.
The u tuations of the topologi al harge determine the value of
topo-logi al sus eptibility
χ top
for ea hensemble. This quantity an be omputedfrom eq. (1.81), i.e. as the mean value of the topologi al harge squared,
normalized bythe volume. Alternatively, it an alsobe determinedfromthe
Gaussiant (5.5)as the varian e
σ 2
, againnormalizedby the volume. Eventhough the probability distributions are ratherfar away fromGaussian, the
statisti al errors of
hQ 2 i
and ofσ 2
are large and we have he ked that bothmethods lead to onsistent results. In Fig. 5.11 we plot the results of the
former method for three light-quark ensembles with sea quark masses
or-responding to the pion mass of around 300 MeV and one ensemble at the
heavier pion mass of approximately 450 MeV. However, sin e the statisti al
errors are large, a meaningful extrapolation to the ontinuum limit is not
possible one would learly need more statisti s. The same holds true with
regard tothe sea quark mass dependen e of the topologi alsus eptibility
the expe ted in rease of
χ top
for larger quark mass is observed, but it is notstatisti ally signi ant.
Therefore, the topologi alaspe ts alsoneed to beinvestigated further in
thefuturea onsiderablein reaseinpre isionisneededtodrawmeaningful
physi al on lusions.
Chiral symmetryis ofutmostimportan eforlow-energyproperties ofQCD.
Therefore, when dis retizingQCDona4-dimensionalspa e-timegrid to
ad-dress non-perturbativephenomena, retaining hiral symmetry inthis latti e
version of QCD (LQCD) is anessential element. Hen e, hirally-symmetri
fermion dis retizationsare needed to fully explore the lowenergy regime of
QCD. A very appealing kind of hiral fermions are the overlap fermions.
However, theiruse indynami alLatti eQCDsimulationsisstilla hallenge,
sin e they are very demandingfrom the omputationalpoint of view.
Chiralpropertiesof fermions are espe iallyimportantinthe valen e
se -tor. At the same time, the most expensive part of a simulation is the
gen-eration of gauge eld ongurations. Hen e, apossible way toover ome the
ost problemofdynami aloverlapsimulations,whi hatthesametimekeeps
theirgood hiralproperties,istofollowamixeda tionapproa hwheregauge
eld ongurations are generated using a omputationally heaper fermion
dis retization and the overlap operator isused only inthe valen e se tor.
The main aimof this thesis was to investigate a parti ular mixed a tion
setup of overlap valen e and maximally twisted mass (MTM) sea quarks.
In this way, we ould prot from a wide set of gauge eld ongurations
generated by the European Twisted MassCollaboration(ETMC).
In parti ular, we wanted to perform a ontinuum limit s aling test of
overlap fermions, a study that has not been done before. However, to
per-formsu h investigationwith atypi allinearlatti e extent of2fm verylarge
omputer resour es wouldberequired,even ifthe overlapoperator wasused
only in the valen e se tor. Therefore, we de ided to employ a small volume
with
L ≈ 1.3
fm. Su hvolumeissu ienttotestthe ontinuumlimits alingbehaviour. Wede ided totake thepion de ay onstantas ourmain physi al
observabletostudythelatti eartefa tsoftheoverlapdis retization. Usinga
suitable mat hing ondition ofoverlap and twisted mass fermions,for whi h
we have taken the pion mass, and assuming universality, the same
ontin-uum limitvalue for
f π
shouldberea hedwithbothkindsoflatti efermions.Weexpli itly he ked inthe freetheory thatthis expe tationisfullled. We
learlyobservedthe
O(a 2 )
leading ut-odependen eanddemonstratedthatf π
agrees in the ontinuum limit.However, when moving to the intera ting ase, we en ountered a puzzle
in that the ontinuum limits of the two latti e fermions used ame out to
be in onsistent with ea h other. It is one of the main results of this thesis
that the solution of this puzzle ould be identied as the exa t hiral zero
modes of the overlap Dira operator. Being hiral, this operator admits
zero modes at any value of the latti e spa ing. This is in ontrast to the
non- hiral twisted mass Dira operator whi h does not admit su h hiral
zero modes, at least not at our urrent values of the latti e spa ing. In
order to demonstrate that the hiral zero modes are indeed the ause of
the mismat h of
f π
in the ontinuum limit, we used the fa t that the zeromodes ouple inan identi al way tothe pseudos alar and s alar orrelation
fun tions. Hen e, in the dieren e of these orrelation fun tions (the
so- alled PP-SS orrelator), the zero modes ontribution is exa tly an elled.
Performing now a ontinuum limit s aling test of the pion de ay onstant
as omputed from the overlap PP-SS orrelator, whi h is not ae ted by
the zero modes, weobtained indeed onsistent ontinuum limitvalues for
f π
omputed from the two fermion dis retizations.
Wealso ross- he ked thisresult byexpli itlysubtra tingthezero modes
at the level of overlap propagators. This further onrmed the pi ture that
the hiral zeromodesneed tobetreatedspe ially,atleast inthe smallnite
volumeused here. However, the modi ationofpropagators byexpli it
sub-tra tionof a partof eigenmodes of theDira operatoris aeld-theoreti ally
not well dened pro edure and may lead to un ontrollable systemati
un- ertainties. Therefore, we interpret our ndings when subtra ting the zero
modes expli itly only as a plausibility he k, whi h however points in the
right dire tion.
The use of the PP-SS orrelator is, in ontrast, safe from the
eld-theoreti al point of view. However, it leads to another di ulty. The
sub-tra tion of the zero modes from the s alar orrelator introdu es signi ant
O(a 2 )
ee ts related to the enhan ed doublepole ontributionto the s alar orrelation fun tion, as suggested by results from hiral perturbationthe-ory. This ee t results from the fa t that the sea and valen e quarks are
dis retized in a dierent way and thus unitarity is violated at any non-zero
valueofthelatti espa ing. Althoughbeingadis retizationee t itvanishes
in the ontinuum limit,itmay render the approa h tothis limitdi ult.
Therefore,the on eptually leanestwaytota klethezeromodeproblem
is to avoid the region of parameters where the zero modes ontribution is
signi ant. To nd this region, we analyzed the dependen e of the zero
mode ee ts on the latti e volume and the sea quark mass. In this way, we
modes ontribution, a hazardous and a non-safe regime. We onsider
the identi ation of these regions to be the most important result of this
work. It allows to provide parameter values for future simulations where
problems with the zero modes will be ompletely absent. The situation is
best illustrated in Fig. 4.26. Letus givetwo expli itexamples of the values
of pion masses and latti e sizes for safe simulations:
•
atm π ≈ 300
MeV, the safe linear latti eextent isL ≈ 2.6
fm,•
atm π ≈ 450
MeV, the safe linear latti e extent is down toL ≈ 2.0
fm.
Clearly, the identi ation of safe simulation regions for valen e overlap
fer-mions is not onlyimportant for extensions of the present work, but alsofor
other ollaborationsworldwidewhoareusingoverlapfermionsinthevalen e
se tor.
Letus nish by givingsome dire tions for further work. We group these
intwoareas. The rstare possiblephysi stargetswith thesafe simulation
parameters. Withour knowledgeof these parameters, we plan to:
•
omputeobservablesforwhi hgood hiralpropertiesofvalen efermions are essential e.g. the kaon bagparameterB K
, orthe de ayK → ππ
;•
investigatequestionsthat arerelatedtotopology,i.e. the omputation oftopologi alsus eptibilityandthe determinationofthesingletmesonmass
η ′
;•
analyze in the mixed a tion setup unitarity violations in the s alarorrelatorandinmixed orrelationfun tions(withonevalen eandone
sea quark)this needsasetupwithnegligiblezero modes ontribution
to isolatethis ee t;
•
onfrontthesimulationresultswith(MixedA tion)PartiallyQuen hedChiral Perturbation Theoryformulas toextra t the orrespondinglow
energy onstants;
•
perform a ontinuumlimits alingtestof thepion de ay onstant(andother observables) at larger volume in order to he k for the size of
quadrati latti espa ing dependen e.
Moreover, it would also be interesting to further investigate the role of the
zero modes torea ha better understanding. To this end, we plan to:
•
test alternative mat hing onditions, dierent from the mat hing of the pion mass. In parti ular, weplan to ompute the ne essaryrenor-malization onstants in order to use the mat hing ondition of equal
renormalized quarkmasses;
•
investigate the role of the zero modes in baryoni observables;•
perform ananalysis of topologi alaspe ts by expli itly omputing the zero modes.Summarizing,webelievethattheresultsofthisworkprovideanessential
andsofarmissingbasisforfuturelarges alesimulationsusingmixeda tions.
Inparti ular,foroursetupofoverlapvalen eandmaximallytwistedmasssea
quarks we have determined simulation parameters for safe simulationson a
quantitativelevel. Thus, respe tingthe limitsontheparametersdetermined
here and performing simulations on large enough latti e volume at a given
pion mass, it will be possible to prot from the good hiral properties of
overlapfermionsandobtainpre isephysi alresultsforquantitiesthatwould
behardtoaddresswithnon hirally-symmetri versionsof latti efermions.
Firstofall,IwouldliketothankmysupervisorKarlJansen,whointrodu ed
metoLatti eQCDandwasalwayspatientinansweringallmyquestionsand
sharing his great experien e. Thank you for your onstant support, many
fruitfulandinspiringdis ussionsandthefriendlyatmospherethatyoualways
reate.
Ithank my supervisor PiotrTom zak whoalsosupported mewhile Iwas
working on this proje t and from whom I learned a lot over many years,
espe iallyabout s ienti programming and statisti alphysi s.
Very spe ial thanks go to Gregorio Herdoiza for many important and
insightful dis ussions, aswellas for numerous areful ross- he kings of the
results. Thank you for answering a lot of my naive questions and tea hing
me the right (patient)attitude tophysi s problems. Thank you alsofor the
very pleasant atmosphere ofour ommonwork.
Iwouldliketothankallthepeoplewith whomI haveworked ondierent
aspe ts related to this thesis: Vin ent Dra h, Elena Gar ia Ramos, Jenifer
Gonzalez Lopez, Agnieszka Kujawa (whois now my wife), Andrea Shindler.
Thank youfor many stimulatingdis ussions and for the ni e working
atmo-sphere.
IamalsoindebtedtoKarolinaAdamiakthanks towhomI metKarl and
ame to Zeuthen for the rst time.
I thank allthe people who ontributed tothe omputer ode that I have
been using and who have helped me in the use of this ode, espe ially to:
Remi Baron, Vin ent Dra h, Luigi S orzato, Andrea Shindler, Carsten
Ur-ba h, Mar Wagner.
I a knowledge useful dis ussions with: Mariane Brinet, Maarten
Golter-man, Dru Renner, Luigi S orzato, Stefan S haefer, Carsten Urba h, Urs
Wenger.
I would also like to thank the Organizers of Les Hou hes 2009 Summer
S hool Modern perspe tives in Latti e QCD: Quantum eld theory and
high performan e omputing Laurent Lellou h,Rainer Sommer,Benjamin
Svetitsky,AnastassiosVladikasformakingitpossibleformetoparti ipate
s hool for their very lear presentation of di ult topi s that allowed me to
learn many importantaspe ts for this work.
I have alsoproted alot from the Latti e Pra ti es workshop in 2008. I
thank the Organizers ofthis s hool Karl Jansen,Dirk Pleiterand Carsten
Urba h.
I a knowledge the use of omputer resour es of the Leibniz
Re henzen-trum inMuni hand Pozna«Super omputingandNetworking Centre. I also
thank the staof these institutions for te hni al support.
I thank DESY Zeuthen for hospitality and nan ial support during my
stays in Zeuthen.
Thiswork waspartly nan ed fromMinistryofS ien e and Higher
Edu- ation grant nr. N N202 237437. I alsoa knowledge nan ialsupportfrom
the Foundation for Polish S ien e who granted me the START s holarship
(2009, 2010).
Lastbut notleast, Ithankmyfamilywhosupported meovermanyyears
espe iallymy wife Agnieszka and my parents.
Wilson gauge a tion
We showhere that the expression for the Wilson gauge a tion has the right
QCD ontinuum limit. We willuse the Baker-Campbell-Hausdorformula:
e aA e aB = e aA+aB+ a
2
2 [A,B]+O(a 3 ) ,
(A.1)generalized to:
e aA e aB e aC e aD = e a(A+B+C+D)+ a 2 2 ([A,B]+[A,C]+[A,D]+[B,C]+[B,D]+[C,D])+O(a 3 ) ,
(A.2)
setting:
A = igA µ (x)
,B = igA ν (x+aˆ µ)
,C = igA µ (x+aˆ ν)
andD = igA ν (x)
.Now, inserting(1.30) in (1.33)and using (A.2), weobtain:
U x,µν = exp h
iga A µ (x) + A ν (x + aˆ µ) − A µ (x + aˆ ν) − A ν (x) +
− g 2 a 2 2
[A µ (x), A ν (x + aˆ µ)] − [A µ (x), A µ (x + aˆ ν)] +
−[A µ (x), A ν (x)] − [A ν (x + aˆ µ), A µ (x + aˆ ν)] +
−[A ν (x + aˆ µ), A ν (x)] + [A µ (x + aˆ ν), A ν (x)] + + O(a 3 ) i
.
(A.3)WeTaylor-expand terms like:
A µ (x + aˆ ν) ≈ A µ (x) + a∂ ν A µ (x)
(A.4)to order
a
and this implies:From (1.8) and (1.6), the ommutatorof the gauge elds an berearranged
as:
[A µ (x), A ν (x)] = A b µ (x)A d ν (x)[t b , t d ] = if bdc A b µ (x)A d ν (x)t c ,
(A.6)nally yielding(negle ting
O(a 3 )
terms):U x,µν = exp h
Expanding to
O(a 4 )
, we obtainfor the Wilson a tion (1.31):S gauge [U] = β X
sin e
O(a 2 )
terms are purely imaginary. Finally, we useP
x
Tree-level s aling test
B.1 Overlap fermions
We show here expli itly the omputation of the overlap Dira operator in
momentum spa e, whi h was given by Lüs her [29℄. We begin with the
derivation of the kernel operator the massless Wilson-Dira operator in
momentumspa e. In position spa e, this operator isgiven by:
D ˆ
Wilson= 1
2 γ µ (∇ ∗ µ + ∇ µ ) − ar∇ ∗ µ ∇ µ ,
(B.1)where we use the notation introdu ed in Chapter 1. The Eu lidean a tion
an bewritten as:
S = X
x,y
ψ(x)K ¯ xy ψ(y),
(B.2)with:
K xy = 1 2
X
µ
γ µ (δ x+ˆ µ,y − δ x−ˆ µ,y ) − r (δ x+ˆ µ,y + δ x−ˆ µ,y − 2δ x,y )
.
(B.3)Using the integral representation of the Krone ker delta:
δ x,y = Z π
−π
d 4 p
(2π) 4 e ip(x−y)a ,
(B.4)we obtain:
K xy = Z π
−π
d 4 p
(2π) 4 e ip(x−y)a
"
X
µ
1
2 γ µ e ipµa − e −ipµa +
(B.5)− r
2 e ipˆ µa + e −ipˆ µa − 2
!#
.
e ipˆ µa − e −ipˆ µa = 2i sin ap µ ,
(B.6)The expression inbra kets isthe Fouriertransform of
K mn
, i.e. theWilson-Dira operator inmomentum spa e
D ˆ
Wilson(p)
. Hen e, adding expli itlytheidentity matrix in Dira spa e, we have:
D ˆ
Wilson(p) = iγ µ ˚ p µ + ar
2 p ˆ 2
1,
(B.9)whi hisexpression (1.49)for theWilson-Dira operatorat
m = 0
orexpres-sion (1.45)for the naiveDira operator,ifthe Wilson parameter
r = 0
. Onean alsonoti ethattheinverseofthis formulaimmediatelygivesexpressions
for the fermion propagator (1.47)in the naive ase and (1.51)in the Wilson
ase.
We now set the Wilson parameter
r = 1
and then the denition of theoperator
A
for overlap fermions (1.75)(withs = 0
)yields:A = 1 − aD W (p) = 1 − iaγ µ ˚ p µ − a 2
2 p ˆ 2 .
(B.10)Sin e the massless overlap Dira operatoris given by:
D ˆ
ov(0) = 1
Using nowthe following rearrangements:
a 2 X
The nal form of the tree-level overlap Dira operator in momentum spa e
is:
One an show [55℄ that the free Dira operator for Creutz fermions an be
written inmomentum spa e as:
D Creutz (p) = i X
¯b = 1
where the onstants
C
,R
andS
are parameters hosen appropriately to ensurethe orre t ontinuumlimitofthe fermionpropagator. One anshowthat these values are:
C = 3/ √
with the fun tions
s
andc
given bys 1 (p) = [˚ p 1 + ˚ p 2 − ˚ p 3 − ˚ p 4 ] ,
(B.25)The freeDira operator forthe modi ation suggested by Bori i is:
D Borici (p) = i X
S Borici (p) = −i P
The starting point for the derivation of the expression for the pseudos alar
orrelation fun tion asa sum over momentais eq. (2.15):
orrelation fun tion asa sum over momentais eq. (2.15):