Adam Mi kiewi z University
Pozna«, Poland
Ph.D. Thesis
Latti e QCD with
hirally invariant fermions
Krzysztof Ci hy
Supervisors
Dr hab. Piotr Tom zak, Prof. UAM
Adam Mi kiewi z University
Quantum Physi s Division
and
Dr. habil. Karl Jansen
NIC, DESY Zeuthen
Introdu tion 7
1 Theoreti al prin iples
of Latti e QCD 11
1.1 The QCD Lagrangian. . . 11
1.2 Dis retizing gaugeelds . . . 20
1.3 Dis retizing fermions . . . 22
1.3.1 Naive dis retization . . . 22
1.3.2 Wilson fermions . . . 23
1.3.3 Wilson twistedmass fermions . . . 26
1.4 Chiral symmetry onthe latti e . . . 29
1.4.1 Ginsparg-Wilson relation . . . 29
1.4.2 Overlapfermions . . . 30
1.4.3 Other kindsof hiral fermions . . . 32
1.4.3.1 Domain wall fermions . . . 32
1.4.3.2 Creutz fermions. . . 33
1.4.4 Topology onthe latti e . . . 34
1.5 Observables in Latti e QCD . . . 35
2 Tree-level s aling test 41 2.1 Fermion propagators . . . 41
2.1.1 Overlapfermions . . . 41
2.1.2 Wilson twistedmass fermions . . . 42
2.1.3 Creutz fermions . . . 43
2.2 Observables . . . 43
2.3 Test setup . . . 45
2.4 Comparison of overlap,twistedmass and Creutz fermions . . . 45
2.5 Mat hing twisted mass and overlap fermions . . . 52
2.5.1 Unmat hed quarkmasses. . . 54
3.1 SimulatingQCD . . . 61
3.1.1 General idea . . . 61
3.1.2 Hybrid MonteCarlo . . . 62
3.2 Computationof the overlap operator . . . 64
3.3 Redu ing the ondition numberof
A
†
A
. . . 663.3.1 Eigenvalue deation . . . 66
3.3.2 HYP smearingof gaugeelds . . . 68
3.4 Inverting the Dira operator . . . 69
3.4.1 Sto hasti sour es . . . 70
3.4.2 The SUMR solver . . . 72
4 Investigations of the ontinuum limit s aling properties of the mixed a tion setup 75 4.1 Mixed a tionapproa h . . . 75
4.2 S aling test light sea quark mass . . . 78
4.2.1 Simulationparameters . . . 79
4.2.2 Lo ality . . . 80
4.2.3 Mat hing the pion mass . . . 83
4.2.4 Pion de ay onstant s alingtest . . . 85
4.3 Chiralzero modes and their ontributionto mesoni orrelators 88 4.3.1 Chiral zero modes . . . 88
4.3.2 The ontributionofthezeromodestomesoni orrelators 90 4.3.3 Comparison of orrelation fun tions . . . 93
4.4 The roleof the zeromodes smallvolume, lightseaquark mass 96 4.5 The role of the zero modes nite volume ee ts analysis. . . 101
4.5.1 Simulationparameters . . . 101
4.5.2 Mat hing the pion mass PP orrelator . . . 101
4.5.3 Pion de ay onstant PP orrelator . . . 102
4.6 The role of the zero modes smallvolume, heavier sea quark mass . . . 104
4.6.1 Motivation and simulation setup. . . 104
4.6.2 Pion de ay onstant s alingtest . . . 106
4.7 The role of the zero modes on lusion . . . 109
4.8 Expli it subtra tion of zero modes . . . 111
4.8.1 Subtra tion pro edure . . . 111
4.8.2 Ee ts of expli it zero modes subtra tion . . . 112
5.1 Unitarity violations . . . 121
5.1.1 Motivation. . . 121
5.1.2 Small volumeanalysis . . . 124
5.2 Light baryonmasses . . . 128
5.3 Topologi al harge and sus eptibility . . . 131
Con lusions and prospe ts 135 A knowledgements 139 A Wilson gauge a tion 141 B Tree-level s aling test 143 B.1 Overlap fermions . . . 143
B.2 Creutz fermions . . . 145
B.3 Correlation fun tions . . . 147
C Improvements of the HMC algorithm 149 D Tree-level test of zero modes subtra tion 151 D.1 Analyti al formula . . . 152
D.2 GWC ode point sour es . . . 153
D.3 GWC ode sto hasti sour es . . . 155
The strongfor eplaysafundamentaland ru ialroleinnature. Itis
respon-sible for the formation of all hadrons, whi h an be lassied into mesons
and baryons. Examples of the former are the pion and the
ρ
meson and of the latter the proton and the neutron, whi h in turn form the nu lei ofall atoms. The theory of the strong intera tion is believed to be Quantum
ChromoDynami s (QCD). It postulatesthat allhadrons are not elementary
themselves but they have an inner stru ture and are built from onstituent
parti les. S attering experiments revealed that the onstituents are
point-like obje ts and we now have a large amount of eviden e that they an be
identied with quarks, whi h are spin-1/2 fermions and whose intera tion
is mediated by spin-1 bosons known as gluons. In order to understand the
intera tion amongquarksandgluonsand omprehendhowit an leadtothe
formation of hadrons anew quantum number, alled the olour harge, had
to beintrodu ed. However, allhadrons observed in experiment donot arry
this olour harge,butare olourless. Thismeansthatthequarksandgluons
an not beisolatedand donot existasfreeparti lesthey are onnedinto
olour-neutral omposite hadrons. This fundamental onnement property
of QCD resultsfrom the fa tthat at largedistan es (oratlow energies) the
QCD oupling onstant determining the intera tion strength between the
quarks and the gluons islarge. However, we know fromperturbation theory
analyses of QCDthat atsmalldistan es (orat highenergies) the QCD
ou-pling onstantbe omessmallandthe quarksbehaveasalmostfreeparti les.
This property of QCD is alled asymptoti freedom and has been tested by
onfronting experimental results with perturbative QCD predi tions. It is
one ofthe mostamazing hara teristi sof QCDthatitshoulddes ribeboth
phenomena, onnement and asymptoti freedom, simultaneously. Clearly,
inorder totest this theoreti alexpe tation, amethodisneeded whereQCD
an be evaluated both in the perturbative regime at small distan es and in
the non-perturbative regime at large distan es, where we enter the world of
the observed hadrons.
late the hadron spe trum and many stru tural properties of hadrons, like
form fa tors or parton distribution fun tions), non-perturbative methods
have to be employed. The only method whi h fullls the above riterion
and allows for pre ise quantitative predi tions is Latti e QCD (LQCD). It
onsists indis retizing spa e-time and formulatingQCD ona 4-dimensional
Eu lidean spa e-timegrid witha latti espa ing
a
. In thisway,the theoryis fullyregularizedandmathemati allywelldened, whi hledtomanyon ep-tualandtheoreti aldevelopmentsinourunderstandingofQCD.Ontheother
hand,byusingFeynman'spathintegralformulationofquantumeld theory,
LQCD anbeinterpreted asa kindof astatisti alme hani alsystem whi h
allows an evaluation with numeri al methods. LQCD was rst proposed in
a seminal paper by Wilson in 1974 [1℄ and shortly after Creutz indeed
per-formedsu hnumeri alsimulationsusingMarkov hainMonteCarlomethods
[2℄. IthastobesaidthatovermanyyearsLQCDsimulationswereperformed
in unphysi al setups with mu h too heavy and even innite quark masses.
However, inthelastfewyearsatremendousprogresshas beena hievedwhen
new algorithmi developments provided a breakthrough in the performan e
oftheusedsimulationalgorithms. Atthesametime,thein reasing omputer
powermadeitpossibletosimulateonlargelatti eswithne latti espa ings
and pion masses approa hing the physi al pion mass. Latti e QCD
om-putations still require very large omputer resour es, parti ularly for fully
dynami al simulations, but its prospe ts are steadily improvingwith a new
generation of super omputers in the PetaFlop range. The algorithmi and
omputer improvementswere alsoa ompaniedby on eptualdevelopments
su hasonesleadingtoafasterapproa htothe ontinuumlimit(
a → 0
)and the formulationof non-perturbativerenormalizations hemes.Anotherimportantaspe t of QCDis hiral symmetry,i.e. the invarian e
of the theory under the ex hange of massless left- and right-handed quarks.
Itisa ontinuoussymmetryandwebelievethatitisspontaneouslybroken in
nature, thusgiving rise to the appearan e of Goldstone bosons. In QCDwe
identify these Goldstone bosons with the pions, whosemass is mu hsmaller
than the mass of any other observed hadron. Assuming su h spontaneous
breaking of hiral symmetry inQCD, manyphenomenologi alinvestigations
an beperformed tointerpretexperimental data,the most notableof whi h
is hiral perturbationtheory.
Inprin iple, LQCDshouldbe abletodedu e the phenomenon of
sponta-neous hiral symmetry from the QCD Lagrangian itself and one would not
have to rely on assumptions. However, for many years, it seemed
impos-sible to preserve hiral symmetry on the latti e. Only in the late 1990s it
that a latti e fermion an be hiral, provided that we allow for a
latti e-modied version of hiral symmetry. This dis overy led to the introdu tion
of so- alled overlap fermions, a kind of latti e fermion whi h respe ts this
latti e modied hiral symmetry. Overlap fermions have many appealing
properties, but are mu h more omputationallydemanding than other
pop-ular fermion dis retizations, su h as Wilson fermions, modi ations thereof
or staggered fermions. This makes the use of overlap fermions still a
hal-lenge,espe iallyindynami alsimulations. Assu h,alternativestodynami al
overlap fermions are being looked for to keep hiral symmetry. The goal is
is to prot from the good hiral properties of overlap fermions, but at the
same time avoid the high omputational ost of generating dynami al
over-lapgaugeeld ongurations. One su happroa his alledmixed a tionand
it onsists in using overlap fermions only as valen e quarks and for the sea
se tor a heaperfermion dis retizationis used.
The aimof this thesis is toinvestigatethe mixed a tionsetup of overlap
valen e fermions and Wilson twisted mass sea quarks. One may suspe t
that usingdierentlatti efermionformulationsinthesea andinthevalen e
se tor leads tounphysi al ee ts. And, aswe willshowin this thesis,this is
indeedthe ase. Aswewilldemonstrate,inordertohaveasafe simulation,
where su hee ts an be avoided,a areful tuningof the physi alsetup has
to be performed. It is one of the main goals of this thesis to spe ify the
regime of parameter values (su h as the latti e volume and the pion mass)
that allows toperformsu hsafesimulations. Knowingthese parameterswill
then allow to address physi al questions and ompute physi al observables
without being ae ted by possible unphysi al ee ts. Therefore, providing
theparametersforsafesimulationsopens thewayforfuturesimulationswith
hirallyinvariantoverlapfermionsinthevalen ese torto omputeimportant
physi al quantities.
The outlineof the thesis is the following.
In Chapter 1, we review the theoreti al prin iples of Latti e QCD. We
startbyintrodu ingthe ontinuumQCDLagrangian anddis ussingits
sym-metries,parti ularlythe hiralsymmetry. Next,weshowhowthe ontinuum
theory is dis retized and we introdu e dierent fermion dis retizations,
in- luding the hirally-symmetri overlap formalism. We also shortly dis uss
the ways of extra tingphysi al observables from asimulation.
Chapter 2 presents the results of a latti e spa ing s aling test of
dif-ferent fermion dis retizations at tree-level of perturbation theory. For this
we use overlap, twisted mass and Creutz fermions. We also investigate the
ee ts of mat hing of twisted mass and overlap fermions, whi h is relevant
of QCD simulations. We review the HMC algorithm and the te hniques
used to ee tively deal with overlap fermions, in parti ular the method of
omputation of the overlap Dira operator, ways of redu ing the ondition
numberof its kernel and the use of sto hasti sour es.
The main results of the thesis are reported in Chapter 4. First, the
motivation and the general idea of a mixed a tionsimulationare dis ussed.
Then,a ontinuumlimits alingtestofthepionde ay onstantisperformed.
This test motivates the analysis of the role of hiral zero modes of the
over-lap operator. We show that this is a very important ee t in the ase of a
hirally-symmetri valen eandnon- hirally-symmetri seaquarks
dis retiza-tion. This hapter on ludes with the aforementioned range of parameter
values that are ne essary fora simulationsafe against these ee ts.
In Chapter 5 we dis uss some further results, in luding the unitarity
violationspresentinthemixeda tionsetup,lightbaryonmasses omputation
Theoreti al prin iples
of Latti e QCD
1.1 The QCD Lagrangian
Quantum ChromoDynami s (QCD) is a gauge theory of strong nu lear
in-tera tions between the onstituents of hadrons. The hadrons are a lass of
parti les in luding baryons (e.g. the nu leon) and mesons (e.g. the pion).
The theory is based on the prin iple of lo al gauge invarian e with a
non-Abelian SU(3) gauge group [3, 4℄. The fundamental degrees of freedom of
the theory are quarks and gluons. The Lagrangian density of QCD an be
written as:
L
QCD
= L
quark
+ L
gluon
+ L
int
,
(1.1)where
L
quark
is the purely fermioni (quark) part,L
gluon
the purely bosoni (gluon) part andL
int
the intera tion part that ouples quarksand gluons.Let us now onsider the dierent parts that onstitute the QCD
La-grangian. The quark term is 1 :
L
quark
=
N
f
X
f =1
¯
ψ
f
(x)(iγ
µ
∂
µ
− m
f
)ψ
f
(x),
(1.2)where
N
f
is the number of avours 2,
ψ
f
(x)
is the quark (spinor) eld or-responding to avourf
andm
f
is thef
-avour bare quark mass and the1
Throughoutthethesis,weemploytheEinsteinsummation onventionforDira indi es
(denoted byGreekletters)andSU(3)-groupgeneratorindi es(denotedbyLatinletters).
2
The Standard Model in orporates 6 avours of quarks (up, down, strange, harm,
bottom, top). However, investigating the low-energy properties of QCD with Latti e
{γ
µ
, γ
ν
} = 2η
µν
,
(1.3) whereη
µν
=
diag
(1, −1, −1, −1)
isthe metri tensor. The gluon part reads:L
gluon
= −
1
4
F
a
µν
(x)F
a
µν
(x),
(1.4) whereF
a
µν
(x)
is the eld strength tensor, whi h is relatedto the gluon eld omponentsA
a
µ
(x)
:F
µν
a
(x) = ∂
µ
A
a
ν
(x) − ∂
ν
A
a
µ
(x) − gf
abc
A
b
µ
(x)A
c
ν
(x),
(1.5)where
g
isthe bare oupling onstantandf
abc
are the stru ture onstantsof
SU(3), satisfying the ommutation relations:
[t
a
, t
b
] = if
abc
t
c
,
(1.6)where
t
a
are the generators of the group SU(3).
The purely bosoni part of the Lagrangian is invariant with respe t to
the lo al gauge transformation. If we want the fermioni part to obey the
lo algauge symmetry aswell, we haveto introdu e a term that ouples the
fermions and bosons, i.e. des ribes the intera tion between them. This is
the basi buildingprin ipleof alllo algaugetheories. It wasrst dis overed
in the ase of the ele tromagneti intera tion, where a term that ouples
ele trons and photons is ne essary to guarantee the lo al gauge invarian e.
In the ase ofQCD, the sum
L
quark
+ L
gluon
is not invariant withrespe t to the lo al SU(3) transformation and the way to guarantee this invarian e isto introdu e the intera tion term
L
int
that ouples the quark eldsψ
and gluon eldsA
µ
:L
int
= g
N
f
X
f =1
¯
ψ
f
(x)γ
µ
A
µ
(x)ψ
f
(x),
(1.7)where the gluon eld
A
µ
is related toits omponentsin the followingway:A
µ
(x) = t
a
A
a
µ
(x).
(1.8)Conventionally,onewritestheterms
L
quark
andL
int
together,introdu ing the ovariant derivativeD
µ
:L
QCD
=
N
f
X
f =1
¯
ψ
f
(x)(iγ
µ
D
µ
− m
f
)ψ
f
(x) −
1
4
F
a
µν
(x)F
a
µν
(x).
(1.10)Letusalsodenethe ( lassi al)QCDa tion,whi his the integralof the
Lagrangian density overspa e-time:
S
QCD
=
Z
d
4
x L
QCD
.
(1.11)Anelegant(and relevantfrom the point of viewof Latti e QCD) way to
quantize a lassi al theory, like the one given by the lassi al QCD a tion
(1.11), is to use the Feynman path integral formalism [5℄. The expe tation
value of any observable
O
is given by:hOi =
Z
1
Z
D ¯
ψDψDA O[ψ, ¯
ψ, A] e
iS
QCD
[ψ, ¯
ψ,A]
,
(1.12)
with the partition fun tion:
Z =
Z
D ¯
ψDψDA e
iS
QCD
[ψ, ¯
ψ,A]
.
(1.13)
Itisworthtoemphasizethatalleldsinthepathintegralare lassi al. Su h
path integral an not beevaluated analyti ally (ex ept for few spe ial ases
mu h simpler than QCD) and one has to swit h to approximate methods.
Formany theories, likeQuantumEle troDynami s (QED), averysu essful
method is perturbation theory. It onsists in expanding the path integral
with respe t to a small parameter (e.g. the ne stru ture onstant
α ≈
1/137.036
inQED)anddroppingtermsbeyondsomeorder. Forexample,the most re ent al ulation of the anomalous magneti moment of the ele tron(usually parametrized in terms of the so- alled
g
-fa tor) up to fourth-order inα
agrees with experiment up to 10 signi ant digits, making it one of the most pre isely veried predi tion of physi s the ele trong
-fa tor isg
e
= 2a
e
+2
,wherethetheoreti alvalue:a
th
e
= 1 159 652 182.79(7.71)×10
−12
andtheexperimentalone:
a
exp
e
= 1 159 652 180.73(0.28)×10
−12
[6℄. However, for perturbative methods to work, there has to be a small parameter withrespe t to whi h one expands the path integral. In the ase of QCD, the
oupling onstant of the olour intera tion depends on energy and one has
to onsider two regimes. For high energy or large momentum transfer, the
QCD oupling onstantissmallenoughforperturbativemethodstowork. In
by Gross, Politzer and Wil zek. However, in the ase of low energy orsmall
momentum transfer, this oupling onstant be omes of the order of unity
and perturbation theory is bound to fail the strong intera tions be ome
strongindeed. Quantitatively,theenergys alewhenithappens
Λ
strong
≈ 250
MeV, where the value is not pre isely dened and depends on the hosenobservable. Anyway, its approximate value implies that a vast number of
relevant phenomena in QCD, su h asthe onnement of quarks and gluons
into hadrons, happen in the non-perturbative regime. Thus, one needs
non-perturbativemethods,su hasLatti eQCD,whi histheonlyknownmethod
of extra ting quantitative predi tions about the low-energy regimeof QCD.
This approa h onsists in dis retizing the QCD path integral. In this way,
one obtainsafullyregularized andwell-dened theory, whi h an bestudied
numeri ally, but also analyti ally the dis retized version of QCD enabled
manyrelevant on eptualdevelopmentsandledtoimportantinsightintothe
nature of strong intera tions.
However, the os illating exponential
e
iS
QCD
[ψ, ¯
ψ,A]
renders the numeri al
evaluation of the QCD path integral unfeasible from the pra ti al point of
view. Fortunately, integrals like (1.12) are tra table, if one swit hes from
Minkowski spa e-time with metri tensor
η
µν
with signature e.g.
(+ − −−)
toEu lideanspa e-timewithsignature(++++)
. Thisisa hievedbyanalyti ontinuation (Wi k rotation of the time dire tion:t → −iτ
). In order that the Eu lidean formulation an be ontinued ba k to physi al (Minkowski)spa e,the Eu lidean orrelationfun tionshavetosatisfya ertain ondition,
alled the Osterwalder-S hrader ree tion positivity [7, 8℄. This ondition
ensures that the transition probabilities between gauge-invariant states are
non-negative and the quantum me hani al Hamiltonian has only real and
positiveeigenvalues [9℄.
The QCDLagrangian density inEu lidean spa e reads [10℄:
L
E
QCD
=
N
f
X
f =1
¯
ψ
f
(x)(γ
µ
E
D
µ
+ m
f
)ψ
f
(x) −
1
4
F
a
µν
(x)F
a
µν
(x)
(1.14)and the Eu lidean gammamatri es satisfy:
{γ
µ
, γ
ν
} = 2δ
µν
,
(1.15)where
δ
µν
=
diag
(1, 1, 1, 1)
isthe Eu lidean metri tensor. Theexpe tation value of any observableO
is then given by:hOi =
Z
1
E
Z
D ¯
ψDψDA O[ψ, ¯
ψ, A] e
−S
QCD
E
[ψ, ¯
ψ,A]
,
where
S
E
QCD
=
R d
4
xL
E
QCD
is the Eu lidean a tion and the Eu lidean parti-tion fun tion reads:Z
E
=
Z
D ¯
ψDψDA e
−S
QCD
E
[ψ, ¯
ψ,A]
.
(1.17)
The os illating exponential in (1.12) is repla ed by the well-behaved fa tor
e
−S
QCD
E
and thus the multi-dimensional integral (1.16) an beevaluated
nu-meri ally,atleastinprin iple,e.g. withMonteCarlomethods. Formally,the
quantum eld theory dened by the partition fun tion (1.17) an be
inter-preted as a statisti al me hani al system and the exponential
e
−S
E
QCD
plays
the role of aBoltzmann fa tor.
Fromnowon, wewillworkonlywiththe Eu lideanformulationof SU(3)
non-Abelian gauge theory (QCD) and hen e we drop the supers ript
E
and the subs riptQCD
thatremind us of it.Now,wewilldis ussafewimportantfeaturesof ontinuumQCDthatare
relevant from the point of view of further onsiderations, espe iallythe role
of hiral symmetry and spontaneous hiral symmetry breaking[10, 11,4℄.
Tobespe i , letusrestri tourselvestotwoavours ofquarks (
u
andd
quarks). The lassi alQCD Lagrangian an be rewritten as:L = ¯uγ
µ
D
µ
u + ¯
dγ
µ
D
µ
d + ¯
um
u
u + ¯
dm
d
d −
1
4
F
a
µν
F
a
µν
≡
≡ L
u
+ L
d
+ L
m
u
+ L
m
d
+ L
gluon
,
(1.18)where
u ≡ ψ
u
andd ≡ ψ
d
are the orresponding spinors and we have sep-arated the mass terms in the fermioni Lagrangian. We an de ompose thequark Lagrangian further by dening left-handed and right-handed quark
spinor elds:
q
R
≡ P
+
q,
q
L
≡ P
−
q,
q = u, d,
(1.19) where:P
±
=
1 ± γ
5
2
.
(1.20)Eq. (1.19) impliesfor the onjugate spinor elds:
¯
q
R
= ¯
qP
−
,
q
¯
L
= ¯
qP
+
.
(1.21)Thus, the rst twoterms in Lagrangian (1.18)be ome:
L
u
+ L
d
= ¯
u
L
γ
µ
D
µ
u
L
+ ¯
u
R
γ
µ
D
µ
u
R
+ ¯
d
L
γ
µ
D
µ
d
L
+ ¯
d
R
γ
µ
D
µ
d
R
=
(1.22)=
u
¯
L
d
¯
L
γ
µ
D
µ
0
0
γ
µ
D
µ
u
L
d
L
+ ¯
u
R
d
¯
R
γ
µ
D
µ
0
0
γ
µ
D
µ
u
R
d
R
,
terms we obtain:
L
m
u
+ L
m
d
= m
u
(¯
u
L
u
R
+ ¯
u
R
u
L
) + m
d
( ¯
d
L
d
R
+ ¯
d
R
d
L
) =
(1.23)=
u
¯
L
d
¯
L
m
u
0
0 m
d
u
R
d
R
+ ¯
u
R
d
¯
R
m
u
0
0 m
d
u
L
d
L
,
i.e. the mass terms oupleelds of opposite hiralities.
Let us now onsider the massless terms
L
u
andL
d
in the Lagrangian. They are invariant with respe t to the following transformations,respe -tively:
u
L
d
L
→ L
u
d
L
L
,
u
R
d
R
→ R
u
d
R
R
,
(1.24)where
L
andR
areunitary2×2
matri es,i.e. elementsofthe(avour)group U(2). This means that the LagrangianL
u
+ L
d
is invariant with respe t to the group U(2)L
×
U(2)R
.Let us take a loser look at the possible forms of transformations. The
masslessquark Lagrangianisinvariantunder fourSU(2)
×
U(1)ve tor trans-formations:u
d
→ e
iαu
i
u
d
,
u ¯
¯
d
→ ¯u ¯
d e
−iαu
i
,
(1.25)wherethesubs ript
i = 0, 1, 2, 3
,u
0
istheidentitymatrixinavourspa eandu
i
(i = 1, 2, 3
) are avour SU(2) group generators. There are 4 onserved (ve tor) Noether urrentsj
µ
i
asso iated with these 4 transformations and hen e 4 onserved hargesQ
i
=
R d
3
xj
0
i
the baryon number (i = 0
) and the isospin (i = 1, 2, 3
).Inaddition,therearetransformationsinvolving
γ
5
, alled hiralrotations:u
d
→ e
iαγ
5
u
i
u
d
,
u ¯
¯
d
→ ¯u ¯
d e
iαγ
5
u
i
.
(1.26)Togetherwithtransformations(1.25),themasslessquarkLagrangian
L
u
+L
d
is invariant underthe symmetry groupSU(2)R
×
SU(2)L
×
U(1)V
×
U(1)A
.However, it an be shown that the fermion integration measure in the
quantized theory is not invariant under the transformation (1.26) for
i = 0
, whi hredu es the fullsymmetry to SU(2)R
×
SU(2)L
×
U(1)V
. This isthe so- alled axial anomaly and it has important onsequen es e.g. for the mesonspe trum the hiral avour singlet symmetry an not be broken
sponta-neously and hen e there isno Goldstone boson asso iatedwith spontaneous
breaking of this symmetry. This implies that the mass of the avour
sin-glet
η
′
opposed to the mass of the
η
meson, whi h is one of the pseudo-Goldstone bosons),but it isrelatedto topologi alu tuationsof the QCD va uumviathe Witten-Veneziano formula[12, 13℄:
f
2
π
2N
f
m
2
η
+ m
2
η
′
− 2m
2
K
= χ
top
,
(1.27)where
f
π
is the pion de ay onstant,m
x
the mass of thex
meson andχ
top
the topologi al sus eptibility, whi h willbe dened later.Letus now onsider the mass terms of the QCD Lagrangian
L
m
u
+ L
m
d
. They areinvariantwithrespe t tothe transformation(1.25)fori = 0
,sothe baryon number is onserved also in the massive theory. Fori = 1, 2, 3
the transformation(1.25)isasymmetryonlyif thequarkmassesare equalm
u
=
m
d
. Hen e, theisospinis onserved inthemassivetheory,butonlyfor mass-degenerate quarks. However, the mass termsL
m
u
+ L
m
d
are not invariant under hiralrotations(1.26),whi his ausedbythefa tthattheexponentialin (1.26) is the same for the spinor
(u d)
T
and the onjugate spinor
u ¯
¯
d
,
whi h is, in turn, due to the anti ommutation relation
{γ
µ
, γ
5
} = 0
. Thus, the symmetry ofthe quantum QCDLagrangianisbroken toSU(2)V
×
U(1)V
in the mass-degenerate ase and to U(1)V
×
U(1)V
ifm
u
6= m
d
.In the ase of arbitrary number
N
f
of quark avours, the analysis is easilygeneralized (the matri esu
i
are nowtheN
f
× N
f
identity matrix andN
2
f
− 1
generatorsofthe avour groupSU(N
f
))andthefullsymmetry ofthe quantized massless QCD Lagrangian is SU(N
f
)R
×
SU(N
f
)L
×
U(1)V
, whi h is redu ed to SU(N
f
)V
×
U(1)V
in the mass-degenerate ase and further to U(1)V
× . . . ×
U(1)V
(withN
f
fa tors U(1)V
) in the ase of dierent quark masses. Thus, in the latter ase, the only exa t symmetry is the baryonnumber onservation.
However,sin etheisospinsymmetryisonlyslightlybrokenforthelightest
twoquarks, itisoftentreatedasexa t 3
,while theheavierquarksare treated
separately. Moreover, sin e the up and down quarks are so light, ompared
to the heavier quarks (
m
u
≈ m
d
≈
a few MeV, whereas alreadym
s
≈ 100
MeV), the full symmetry of the massless Lagrangian withN
f
= 2
avours SU(2)R
×
SU(2)L
×
U(1)V
remains an important approximate symmetry and is the basis ofN
f
= 2
hiral perturbation theory (χ
PT ). At low energy, the quarks and gluons are onned into hadrons and hen e one an denean ee tive eld theory, in whi h the fundamental degrees of freedom are
not quarksand gluons,but lighthadrons. Two-avour
χ
PT wasformulated by Gasser and Leutwyler [14℄. The Lagrangian of this theory is onstru tedfrom elds des ribing the pions (
π
±
,
π
0
) in a way whi h is onsistent with
3
ganized in terms of expansion parameters
p/Λ
χ
andm
π
/Λ
χ
, wherep
is the momentum,m
π
thepion massandΛ
χ
= (4πf )
2
thetypi alhadroni s ale
≈
1 GeV,withf
the pion de ay onstantin the hirallimit. Thereare many appli ations ofχ
PT in the analysis of the low-energy regime of QCD, e.g. pion s attering experiments. Moreover, it is also essential in the analysis ofLatti eQCDdata, sin emost of ontemporaryLatti eQCDsimulationsare
performed at unphysi al values of the pion mass 4
hen e an extrapolation
to the physi al point (physi al pion mass) is ne essary and is performed by
tting
χ
PT formulas. What is more, even thoughthe strange quarkmass is mu h larger than the mass of the up and down quarks, it is still relativelysmall ompared to the typi al QCD s ale of
≈
1 GeV and the symmetry SU(3)R
×
SU(3)L
×
U(1)V
of the masslessN
f
= 3
Lagrangian is also an ap-proximate symmetry and forms the basis ofN
f
= 3
hiral perturbation the-ory,whi hisalsoof use inthe analysis oflow-energyQCDexperiments,e.g.in luding the kaons (also in kaon physi s from Latti e QCD). Three-avour
χ
PT was also introdu ed by Gasser and Leutwyler [15℄ as a generalization of the two-avour ase to in lude the strange quark. The three-avourLa-grangianin ludes,besidesthepionelds,alsootherlightpseudos alarmeson
elds (of the remaining pseudo-Goldstone bosons the kaons
K
±
,
K
0
,
K
¯
0
and the
η
meson). Quantitatively, the expli it breaking of hiral symmetry by the quark masses an be expressed by the ratiosm
2
π
/(4πf )
2
≈ 0.007
andm
2
K
/(4πf )
2
≈ 0.09
. In this sense, the expli itbreaking by the strangequark mass isroughlya10%ee t, whileforthe lightestquarksitisa<1% ee t.Obviously,itisnot possibletotreat the
N
f
= 4
symmetryasapproximately valid,sin ethe harmquarkisalreadyheavy(m
c
≈ 1.3
GeV)andthemesons ontaining itare mu hheavier than the s aleΛ
χ
.However, if hiralsymmetrywasbroken onlyexpli itly,wewouldobserve
degenerate multiplets of hadrons e.g. there should be s alar mesons with
massesverysimilartothepseudos alarones. Also,inthis aseoneshouldnot
expe t su h big dieren e between the masses of the pions and kaons. The
explanation of these phenomena an beprovided by anassumption that the
hiral symmetry of QCD is not only expli itly broken by the quark masses,
butalsospontaneouslybroken. Wespeakofspontaneoussymmetrybreaking
if asymmetrywhi hispresentatthe Lagrangianlevelisabsentinthe
phys-i al ground state 5
. If a ontinuous symmetry is broken spontaneously, then
4
Some ollaborationshavere entlystartedorarepreparingsimulationsatthephysi al
pion mass.
5
A learexampleis provided byferromagnets. EventhoughtheHamiltonian ofsu h
system is invariantwith respe t to a simultaneous ip of allspins, in an experimentall
interpreted asthe wouldbe-Goldstonebosons of hiralsymmetrybreaking,
wheretheprex wouldbe-referstothefa tthattheyare notmassless,but
have asmallmass ( ompared tothe masses of other hadrons) that isdue to
(small) expli it breaking of hiral symmetry by the quark masses.
Also, spontaneous breaking of hiral symmetry an be observed in the
mass dieren e of parti les that are hiral partners and should have the
same mass,if hiralsymmetry wasexa t. Sin e hiralsymmetry isexpli itly
broken bythe quark masses,the experimental massvalues of hiral partners
should not be equal, but they should be lose to ea h other, be ause the
masses of the lightquarks are so small. This is not observed. Forexample,
the ve tor mesons
ρ
anda
1
have massesequalto,respe tively, 770 and 1260 MeV,whi hisamu hlargerdieren ethanone wouldexpe tfromthesmallexpli itbreakingof hiralsymmetry[16℄. Anotherexampleisthenu leonand
itsnegative-paritypartner,usuallydenotedby
N
∗
[11,17℄. Theexperimental
value of the nu leon mass is
m
N
≈ 940
MeV, whilem
N
∗
≈ 1535
MeV. Spontaneous hiralsymmetrybreakingissignalledbyanon-zerovalueofthe hiral ondensate
h0|¯uu|0i
,where|0i
is theva uumstate. This quantity emerges in hiral perturbation theory as an important low-energy onstantB
0
:B
0
= −f
−2
h0|¯uu|0i,
(1.28) where the tree-level pion de ay onstantf
is another low-energy onstant. A well-known relationthat involves the hiral ondensate is the Gell-Mann,Oakes, Renner (GMOR) relation[18℄:
f
2
m
2
π
= −(m
u
+ m
d
)h0|¯uu|0i,
(1.29)whi h an be derived in
χP T
. As su h, it is desirable to assess the value of the hiral ondensate from experiment thus the value ofB
0
would be known. It has been argued that the best estimate an be obtained fromthe low-energypion-pions attering [19, 20℄. However, the al ulation of the
ondensatefromempiri aldatarequiressomemodelassumptions,i.e. one in
fa t has to assumethat spontaneous hiral symmetry breaking takes pla e.
Therefore,animportant he k would beto al ulatethe ondensate
non-perturbativelyfromrstprin iples,withoutanyadditionalassumptions. One
su hway isprovidedby Latti eQCD.Indeed, Latti eQCDsimulations
on-rmthatitisnon-zeroatzerotemperature(areviewofresultsonthistopi is
provided e.g. in. [21℄). However, thereexists atemperaturewherethe hiral
ondensate vanishes,thus signalling hiral symmetry restoration. Moreover,
ment temperature, i.e. the temperature at whi h the quark-gluon plasma
formsand quarksand gluons areno longer onned intohadrons. Up tothe
presentday,thisissuehasnotbeenresolved ompletely,butitisastronghint
that Latti eQCD al ulations point to the fa t that both temperatures are
equal, up tostatisti alerror. This strongly suggests that spontaneous hiral
symmetry breaking isrelated to onnement and onrms that
understand-ing hiral symmetry and spontaneous hiral symmetry breaking is essential
to fully omprehend QCD. However, mu h more pre ise results are needed
to unambiguously resolve this question. In Latti e QCD investigations of
these phenomena it is therefore essential to take hiral symmetry properly
into a ount,i.e. fermionswith good hiralpropertieshave tobeused. This
is one of the motivations for employing overlap fermions, whi h will be the
main subje t of this thesis.
1.2 Dis retizing gauge elds
In this se tionand the next one, we showhow QCD an be formulated ina
non-perturbative way on a Eu lidean 4-dimensional hyper ubi latti e with
latti e spa ing denoted by
a
[22℄.The basi relationshipbetween the ontinuum and latti eformulationof
gauge elds is given by the followingequation:
U(x, x + aˆ
µ) = e
igaA
µ
(x)
,
(1.30)
where
U(x, x + aˆ
µ)
represents the gauge eld on the latti e (itis a variable dened onthe link onne tingsitesx
andx + aˆ
µ
, whereµ
ˆ
is the unitve tor in theµ
-dire tion) andA
µ
(x)
isthe ontinuum gauge eld. This expression also implies that the link variables are SU(3) matri es, sin e it involves thegenerators of SU(3), a ording to eq. (1.8).
We nowdis uss the simplest gauge eld latti ea tion, alled the Wilson
a tion [1℄, and show that in the ontinuum limit itis equivalent tothe
on-tinuum gauge a tion. It is worth to emphasize that the hoi e of the latti e
a tion is non-unique. In prin iple, any latti e a tion an be used, provided
that it has the orre t ontinuum limit. The Wilson a tionreads:
S
Wilson[U] =
β
3
X
x
X
1≤µ<ν≤4
(1 −
ReTrU
P
(x, µ, ν)) ,
(1.31)where
U
P
is alled the plaquette variableand is dened as:U
P
(x, µ, ν) ≡ U(x, x + aˆµ)U(x + aˆµ, x + aˆµ + aˆν)
(1.32)To simplify notation, one usually denes
U(x, x + aˆ
µ) ≡ U
x,µ
andU(x, x −
aˆ
µ) ≡ U
x−aˆ
†
µ,µ
. Theshort utnotationfortheplaquettevariableis:U
P
(x, µ, ν)
≡ U
x,µν
, whereµν
identies the plane of the plaquette. In this way, the pla-quette an be writtenas:U
x,µν
= U
x,µ
U
x+aˆ
µ,ν
U
x+aˆ
†
ν,µ
U
x,ν
†
.
(1.33)Thegaugetransformationonthe latti eisasso iatedwithmultipli ation
of the fermion and gluon elds by a site-dependent SU(3) matrix
G(x)
. For the link matri esit an bewritten as:U
x,µ
→ U
x,µ
′
= G(x)U
x,µ
G(x + aˆ
µ)
†
.
(1.34)Thisformofthegaugetransformationimpliesthatthetra eoftheplaquette
(a tually, the tra e of any losed loopof link variables) is a gauge-invariant
quantity:
U
x,µν
→ U
x,µν
′
= G(x)U
x,µ
G(x + aˆ
µ)
†
G(x + aˆ
µ)U
x+aˆ
µ,ν
G(x + aˆ
µ + aˆ
ν)
†
× G(x + aˆµ + aˆν)U
x+aˆ
†
ν,µ
G(x + aˆ
ν)
†
G(x + aˆ
ν)U
x,ν
†
G(x)
†
=
= G(x)U
x,µ
U
x+aˆ
µ,ν
U
x+aˆ
†
ν,µ
U
x,ν
†
G(x)
†
,
(1.35)Tr
U
′
x,µν
=
TrG(x)U
x,µ
U
x+aˆ
µ,ν
U
†
x+aˆ
ν,µ
U
†
x,ν
G(x)
†
=
(1.36)=
TrU
x,µ
U
x+aˆ
µ,ν
U
†
x+aˆ
ν,µ
U
†
x,ν
=
TrU
x,µν
.
We will onsider the gauge transformationfor the fermionelds in the next
se tion.
InAppendix A,weshow thatthe Wilsongauge a tion anbewrittenas:
S
gauge
[U] = β
g
2
a
4
6
X
x
X
µ,ν
1
4
F
µν
(x)
2
+ O(a
2
)
.
(1.37)Comparing this expression with the ontinuum gauge a tion
R d
4
x
1
4
F
µν
(x)
2
,
we an immediatelysee that the ontinuumlimitof the dis retized a tionis
the ontinuum gauge a tion if we set:
β =
6
g
2
.
(1.38)The leading dis retization ee ts are
O(a
2
)
, sin e the fa tor
a
4
in front of
the sum omesjust from the dis retizationof the integral
R d
4
x → a
4
P
improved a tion, whi h helps to de rease the size of latti e dis retization
ee ts. Su h a tions have the same ontinuum limit, but this limit is
ap-proa hed faster. One ofthe rst improved a tions wasderived by Weisz [23℄
and it is usually referred to as tree-level Symanzik improved gauge a tion.
The form of this a tion is:
S
tlSym[U] =
β
3
X
x
b
0
X
µ,ν=1
1≤µ<ν
(1 −
ReTrU
x,µν
) + b
1
X
µ,ν=1
µ6=ν
1 −
ReTrU
re tx,µν
,
(1.39)where
b
0
,b
1
areparameters 6,
U
x,µν
isthe (denedabove)plaquettetermandU
re tx,µν
is the re tangle term:U
re tx,µν
= U
x,µ
U
x+aˆ
µ,µ
U
x+2aˆ
µ,ν
U
x+aˆ
†
ν+aˆ
µ,µ
U
x+aˆ
†
ν,µ
U
x,ν
†
.
(1.40) Ifb
1
= 0
,this a tionbe omesthe Wilson a tion.1.3 Dis retizing fermions
1.3.1 Naive dis retization
Letusstartwithadis retizationofone-avour ontinuumfreefermiona tion
in Eu lidean spa e, given by:
S
free quark=
Z
d
4
x ¯
ψ(x)Dψ(x),
(1.41)where
D = γ
µ
∂
µ
+ m
is the Dira operator andm
is the quark mass. The dis retization pro edure is not unique and we show here one of the hoi esfor the latti e derivative[25℄:
ˆ
∂
µ
ψ(x) =
1
2a
(ψ(x + aˆ
µ) − ψ(x − aˆµ)) .
(1.42) This an alsobe writtenas:ˆ
∂
µ
ψ(x) =
1
2a
(ψ(x + aˆ
µ) − ψ(x) + ψ(x) − ψ(x − aˆµ)) ≡
1
2
∇
µ
+ ∇
∗
µ
ψ(x),
(1.43) 6The omputationsrelevantforfurtherpartofthisworkusedgaugeeld ongurations
generatedbytheEuropeanTwistedMassCollaboration(ETMC),whousedthisa tionin
where we have dened the forward latti e derivative
∇
µ
and the ba kward latti e derivative∇
∗
µ
. We also dis retize the spa e-time integral (R d
4
x →
a
4
P
x
),thus arrivingat:ˆ
S
free quark= a
4
X
x
X
µ
¯
ψ(x)(γ
µ
∂
ˆ
µ
+ m)ψ(x),
(1.44)where the hat denotes latti e quantities.
ByFourier-transforming the latti e Dira operator (whi h is
onvention-ally alled the naive operator, sin e it orresponds to the simplest possible
dis retization)
D
ˆ
naive
= γ
µ
ˆ
∂
µ
+m
,one anobtaintheexpression forthe Dira operator in momentum spa e:ˆ
D
naive(p) = i˚
p
µ
γ
µ
+ m
1
,
(1.45)where we havedened:
˚
p
µ
≡
1
a
sin(ap
µ
)
(1.46)for later onvenien e and 1 isthe unit matrix inDira spa e.
The tree-level fermion propagator in momentum spa e is given by the
inverse of the Dira operator (1.45) and thusequals:
ˆ
D
−1
naive(p) =
−i˚
p
µ
γ
µ
+ m
1P
µ
˚
p
2
µ
+ m
2
.
(1.47)Letus onsiderthe aseofmasslessfermions. One aneasilyobservethatthis
expression hasthe right ontinuumlimit
−ip
µ
γ
µ
/p
2
. However, italsoimplies
that the number of fermionsis doubled forea h spa e-timedimension, sin e
the poles of the fermion propagator are lo ated not onlyat zero momentum
(
ap
µ
= (0, 0, 0, 0)
), whi h orresponds to the single fermion given by the ontinuum Dira operator, but also whenever any momentum omponentequals
π/a
. Thus, in4-dimensionalspa e-time, wehave2
4
= 16
fermions,of
whi h15areunphysi alandare alleddoublers. Thisistheso- alledfermion
doublingproblem.
1.3.2 Wilson fermions
The rst way to over ome the doubling problem onsists in treating
dier-entlythephysi alpoleandtheunphysi alonesandwasintrodu edbyWilson
[26℄, who suggested the following formof the latti eDira operator:
ˆ
D
Wilson=
1
2
γ
µ
(∇
∗
µ
+ ∇
µ
) − ar∇
∗
µ
∇
µ
+ m,
(1.48)where
r
is the Wilson parameter. The se ond-derivative term is now alled the Wilson term. In momentumspa e, this operatorreads:ˆ
D
Wilson(p) = i˚
p
µ
γ
µ
+
ar
2
p
ˆ
2
µ
1+ m
1,
(1.49)where we havedened:
ˆ
p
µ
≡
2
a
sin
ap
µ
2
(1.50)and the tree-level fermion propagatoris:
ˆ
D
−1
Wilson(p) =
−i˚
p
µ
γ
µ
+ (
ar
2
P
µ
p
ˆ
2
µ
+ m)
1P
µ
˚
p
2
µ
+ (
ar
2
P
µ
p
ˆ
2
µ
+ m)
2
.
(1.51)The physi al pole at
ap
µ
= (0, 0, 0, 0)
gets no ontributionfrom the Wilson term, but the unphysi al ones a quire an additionalmass, whi h ispropor-tional to
a
−1
and hen e be omeinnitely heavy inthe ontinuum limitand
de ouple.
However, the pri e one has to pay for removing the doublers is twofold.
First, the Wilson term leads to an
O(a)
leading ut-o dependen e in ob-servables, whi h makes it, from the point of view of pra ti al simulations,advantageous to introdu e further terms to the a tion, e.g. a twisted mass
term, whi hwillbedis ussed later,or ountertermswithin theframeworkof
the Symanzik improvement programme. The simplest way to obtain
O(a)
-improvement (the absen e ofO(a)
ut-o ee ts) is to add to the a tion a single term, alled the Sheikholeslami-Wohlert( lover) term [27℄.Se ond, the Wilson term, being a mass term, expli itly breaks hiral
symmetryeveninthe hirallimit
m = 0
,i.e. eveninthislimit{ ˆ
D
Wilson
, γ
5
} 6=
0
. Moreover, it has been proven by Nielsen and Ninomiya[28℄ that it is not possiblethatalatti eDira operatorD
ˆ
fulllsatthesametimethefollowing onditions7
:
1. lo ality i.e. the norm of the Dira operator
D
ˆ
de ays exponentially, as afun tion of the distan e between latti epoints,2. translational invarian e i.e. the Fouriertransform of the Dira
oper-ator exists and equals
D(p) = iγ
ˆ
µ
p
µ
+ O(ap
2
)
for
p ≪ π/a
,3. no fermion doublers i.e.
D(p)
ˆ
is invertible everywhere, ex ept forp
µ
= (0, 0, 0, 0)
, 7Original formulation of the Nielsen-Ninomiya theorem is in fa t dierent. Here we
present anequivalent formulation(given e.g. in [29, 30℄), whi h stresses the important
{ ˆ
D, γ
5
} = 0.
(1.52)Formanyyears,itseemedthatitwasnotpossibletohave hiral fermionson
the latti e without violating one of the other onditions. However, a great
progress has been made onthis topi when it wasrealized that(1.52) is not
the only possible form of latti e hiral symmetry. The impli ations of this
dis overy willbedis ussed inthe next se tion.
An important onsequen e of hiral symmetry breaking for the Wilson
a tion is that the quark mass
m
requires additive renormalization. Hen e, the massless ase does not orrespond tom = 0
, but tom = m
c
, wherem
c
is alled the riti alquark mass.The quarkmass isoften expressed with the so- alledhoppingparameter
κ
, dened as:κ =
1
8 + 2m
.
(1.53)Now, we dis uss how to add gauge elds to the Wilson fermion a tion.
It is believed that in the intera ting ase the doubler modes also de ouple.
However, there isno rigorousproof of it.
Undergaugetransformation,thefermioneldstransforminthefollowing
way:
ψ(x) → ψ
′
(x) = G(x)ψ(x),
ψ(x) → ¯
¯
ψ
′
(x) = ¯
ψ(x)G(x)
†
.
(1.54)For onvenien e, weremind here that the gauge elds transform as:
U
x,µ
→ U
x,µ
′
= G(x)U
x,µ
G(x + aˆ
µ)
†
.
(1.55)In this way, the fermion mass term is obviously gauge-invariant, but the
derivative terms, e.g.
ψ(x)γ
¯
µ
∇
µ
ψ(x) = ¯
ψ(x)γ
µ
(ψ(x + aˆ
µ) − ψ(x))
are not, sin e:¯
ψ(x)ψ(x + aˆ
µ) → ¯
ψ(x)G
†
(x)G(x + aˆ
µ)ψ(x + aˆ
µ).
(1.56) However, introdu ingthe ovariantderivative:ˆ
D
µ
ψ(x) =
1
2a
U
x,µ
ψ(x + aˆ
µ) − U
x−aˆ
†
µ,µ
ψ(x − aˆµ)
,
(1.57)one nds forthe derivativeterm (1.56):
¯
ψ(x)U
x,µ
ψ(x + aˆ
µ) → ¯
ψ(x)G
†
(x)G(x)U
x,µ
G(x + aˆ
µ)
†
G(x + aˆ
µ)ψ(x + aˆ
µ)
The gauge-invariantWilson-Dira operator an bewritten as:
ˆ
D
Wilson(m) =
1
2
γ
µ
(∇
∗
µ
+ ∇
µ
) − ar∇
∗
µ
∇
µ
+ m,
(1.59)whi his exa tlythe same formasineq. (1.48), but now
∇
µ
and∇
∗
µ
are the forward and the ba kward ovariantderivatives8 , dened by:
∇
µ
=
1
a
(U
x+aˆ
µ,µ
ψ(x + aˆ
µ) − ψ(x)) ,
(1.60)∇
∗
µ
=
1
a
ψ(x) − U
x−aˆ
†
µ,µ
ψ(x − aˆµ)
.
(1.61)1.3.3 Wilson twisted mass fermions
Intheremainderofthisse tion,wewilldis ussWilsontwistedmassfermions,
whi h are relevant from the point of view of further onsiderations.
Origi-nally, they were introdu ed to deal with the problem of unphysi ally small
eigenvalues (zero modes) of the Wilson-Dira operator [31℄, whi h is
an-other onsequen e of additive quark mass renormalization, whi h an bring
the renormalized quark mass to zero. In the quen hed approximation, the
ontribution of these modes is not balan ed by the fermioni determinant
and leads tolarge u tuations, whi hae t ensembleaverages in an
un on-trolled way. The gauge eld ongurations whi h ause this problem are
referred toas ex eptional ongurations. Thisis espe iallydangerousin the
ase ofsmallquark masses and makes the approa htowards the hiral limit
pra ti ally impossible with Wilson fermions. In dynami al simulationswith
Wilsonfermionstheproblemissuppressedbythefermioni determinant,but
it an still ause te hni al problems, su h as long auto orrelation times in
ertain observables, oming froma idental zero modes of the Wilson-Dira
operator. Moreover, it wasrealized that the twistedmass dis retization an
redu e the ee ts of expli it hiral symmetry breaking by the Wilson term
by suppressing the mixing problem of operators belonging to dierent
hi-ral representations. Finally,twisted mass a tion makes it possible toobtain
automati
O(a)
-improvement, by tuning just one parameter. This is an es-sentialadvantageoftwistedmassfermions,sin eotherimprovements hemesmake itne essary to ompute improvement oe ients for dierent
interpo-lating operators.
8
Wewillusethesamesymbols
∇
µ
and∇
∗
µ
forthenon- ovariantand ovariant deriva-tivesandthemeaningofthesesymbolswillbedeterminedfromthe ontext.rate quarks is given by:
ˆ
S
TM= a
4
X
x
¯
χ(x) ˆ
D
TMχ(x),
(1.62) with:ˆ
D
TM= ˆ
D
Wilson(m) + iµγ
5
τ
3
,
(1.63)where
µ
is an additionalmass parameter, alled the twisted mass,τ
3
is the third Paulimatrix inavour spa e andχ(x)
isthe quarkeld inthe twisted basis.The physi aland twistedbases are relatedby an axialtransformation:
ψ(x) → χ(x) = e
iωγ
5
τ
3
/2
ψ(x),
¯
ψ(x) → ¯
χ(x) = ¯
ψ(x)e
iωγ
5
τ
3
/2
,
(1.64)
where
ω
is alledthe twist angle. Thistransformationleavesthe formof the a tion invariant,only transforming the mass parameters a ording to:m → m cos(ω) + µ sin(ω),
(1.65)µ → −m sin(ω) + µ cos(ω).
(1.66) A spe ial ase of this transformation, referred to as maximal twist, isω =
π/2
, whi h orresponds to sending the bare quark massm
to 0 or, taking additivemass renormalizationintoa ount,toits riti alvaluem
c
. Conven-tionally, the value of the riti al bare quark mass is expressed in terms ofthe parameter
κ
c
,given by eq. (1.53). This istheonlyparameter thatneeds to be tuned to obtain automatiO(a)
-improvement. The tuning is usually done by employingone of two methods. First, one an just nd the riti albare quark mass by looking for a quark mass value that gives a vanishing
pion mass. Alternatively, one an also tune the so- alled untwisted PCAC
mass:
m
P CAC
=
P
~
x
h∂
0
A
a
0
(~x, t)P
a
(0)i
2
P
~
x
hP
a
(~x, t)P
a
(0)i
,
a = 1, 2
(1.67)to zero [24℄. The latter method seems towork very wellin pra ti al
simula-tions.
Thus, one an writethe maximallytwistedmass(MTM) QCDa tion as:
ˆ
S
MTM= a
4
X
x
¯
χ(x) ˆ
D
MTMχ(x),
(1.68) with:ˆ
D
MTM= ˆ
D
Wilson(m
c
) + iµγ
5
τ
3
.
(1.69)r
χ
0
m
PS
= 0.614
r
χ
0
m
PS
= 0.900
r
χ
0
m
PS
= 1.100
r
χ
0
f
PS
(a/r
χ
0
)
2
0.06
0.04
0.02
0
0.42
0.38
0.34
0.30
0.26
= 0.045
= 0.090
r
χ
0
µ
R
= 0.130
(r
χ
0
m
PS
)
2
(a/r
χ
0
)
2
0.06
0.04
0.02
0
1.4
1.0
0.6
0.2
Figure 1.1: Continuum limits aling inxed nitevolume for
r
0
f
P S
at xed values ofr
0
m
P S
(a) and for(r
0
m
P S
)
2
at xed values of renormalized quark
mass
r
0
µ
R
(b). In (b) data atβ = 4.2
((a/r
0
)
2
= 0.0144
) are not in luded,
due tothe missing value of the renormalizationfa tor
Z
P
. Sour e: [33℄.The spe ial meaningof the maximally twisted ase is that it guarantees an
automati
O(a)
-improvement,whi hwasproven in[32℄. Thismeansthat all terms ofO(a)
in the Symanzik expansion of parity even operators (whi h givee.g. the hadronmasses)are absent. This observation makesthe twistedmass formulation (at maximal twist) very useful from the point of view of
pra ti alsimulations.
Anexample of
O(a)
-improvement isprovided by ETMC simulations[33℄ and isdepi ted in Fig. 1.1. The left plot shows the ontinuum limits alingofthe pseudos alarde ay onstant(inxed volume)atxedreferen evalues
of the pseudos alar mass
r
0
m
P S
. Four latti e spa ings are in luded, but the linear t does not in lude the data at the largest latti e spa ing. Therightplotshows the s alingofthe pseudos alarmass(againinxed volume)
at xed values of the renormalized quark mass
r
0
µ
R
. Here, the data for only three latti e spa ings are presented (all of them in luded in the t),sin e it was impossible to in lude the points at the nest latti e spa ing
(
(a/r
0
)
2
= 0.0144
), due tothe missing value of the renormalizationfa tor of
the pseudos alar urrent
Z
P
. Bothplotsshowthatthe leading ut-o ee ts are indeedO(a
2
)
and their overall magnitude israther small.
However, one should mention here that the twisted mass term violates
parity and the isospin symmetry. This ee t e.g. makes the masses of the
harged and neutral pions dierent from ea h other 9
and in fa t this mass
9
Both parity and isospinbreaking are
O(a
2
)
ee ts and hen e they vanish in
the ontinuum limit.
1.4 Chiral symmetry on the latti e
In this se tion we dis uss the great breakthrough asso iated with the fa t
that itwas realized that there is an alternative viewon hiral symmetry on
the latti e, i.e. that the Nielsen-Ninomiya theorem an be over ome in an
elegantway.
1.4.1 Ginsparg-Wilson relation
In 1982 (i.e. only one year after establishing the Nielsen-Ninomiya
theo-rem), Ginsparg and Wilson, basing on renormalization group
transforma-tions,showed [34℄thataremnantof hiralsymmetryispresentonthelatti e
without the doublermodes, if the orresponding Dira operator
D
ˆ
obeys an equation now alledthe Ginsparg-Wilsonrelation:γ
5
D + ˆ
ˆ
Dγ
5
= a ˆ
Dγ
5
D.
ˆ
(1.70)It is a modi ation of the anti ommutation relation(1.52) and the term on
the right-handside vanishesinthe ontinuum limithen e, inthislimitthe
standard hiral symmetry relation(1.52) is regained.
However, for many years it has not been realized that the
Ginsparg-Wilson relation allows one to dene hiral symmetry also on the latti e,
i.e. at non-vanishinglatti e spa ing. It lasted until around 1997 before the
Ginsparg-Wilson relation was redis overed. First, P. Hasenfratz realized
that a kind of latti e fermions alled the xed point fermions satises this
relation [35, 36℄ and shortly afterwards a similar observation was made by
Neuberger regarding the overlap formalism [37, 38℄. Moreover, Lüs her [29℄
found that the Ginsparg-Wilsonrelationleads toa non-standardrealization
oflatti e hiralsymmetry. Thefermiona tionisinvariantunderthefollowing
hiral rotation:
ψ → e
iθγ
5
“
1−
a ˆ
D
2
”
ψ, ¯
ψ → ¯
ψe
iθγ
5
“
1−
a ˆ
D
2
”
.
(1.71)Inthe ontinuumlimitthistransformationis(1.26)with
u
i
= 1
. Toprovethe invarian e of the massless a tion with respe t to the above transformation,intera tionsandtheexpli itviolationofisospinsymmetrybydierentupanddownquark
masses. InLatti eQCDwithmass-degeneratequarksthe hargedandneutralpionmasses
the left- and right-handed proje tors of fermion elds [11℄ with a modied
γ
5
-matrixγ
ˆ
5
= γ
5
(1 − a ˆ
D)
:ˆ
P
±
=
1 ± ˆγ
5
2
.
(1.72)Thusdened proje torshavethe sameproperties asthe standard ontinuum
proje tors andhen e one an de ompose the fermion part ofthe Lagrangian
into left- and right-handed massless parts and a symmetry breaking mass
term that mixes the left- and right-handed omponents. Taking the
proper-ties ofthe latti eproje tors
P
ˆ
±
intoa ount,onends thatthe masstermis of the formm ¯
ψ
1 −
a ˆ
D
2
ψ
, whi h meansthat themassive Ginsparg-Wilson Dira operatorD(m)
ˆ
that orresponds tothe massless operatorD
ˆ
reads:ˆ
D(m) = ˆ
D + m
1 −
a ˆ
D
2
!
=
1 −
am
2
ˆ
D + m.
(1.73)Sin etheGinsparg-Wilsonrelationisanon-standardrealizationof hiral
symmetry,the onditionsoftheNielsen-Ninomiyatheoremdonotapply and
one anhave hiralsymmetry withoutthe doublers,whi hwas onsidered to
beimpossible for many years.
Moreover, itwasalsoshownbyHasenfratz,Laliena,Niedermayer[39℄and
in adierentway by Lüs her [29℄ thatthe Ginsparg-Wilsonrelationimplies
thatthe axialanomalyis orre tlyreprodu ed. Thea tionisinvariantunder
the transformation (1.71), but the fermioni measure
D ¯
ψDψ
is not its Ja obianJ
is non-trivial:J = exp[−2iθ
Tr(γ
5
(1 − a ˆ
D/2))]
and it an also be expressed asJ = exp[−2iθQ
top
]
, whereQ
top
is the topologi al harge, to be dis ussed later. This issue was further elu idated by Fujikawa [40℄, whostudied the ontinuum limitof this Ja obian.
Furthermore, a onsequen e of the Ginsparg-Wilson relation is that
fer-mionsare prote tedfromadditivemass renormalizationandmixingbetween
four-fermion operators in dierent hiral representations (Hasenfratz [36℄)
and there an beno
O(a)
latti e artefa ts (Niedermayer [30℄).Inthenextse tionwedis ussoneofthesolutionsofthe Ginsparg-Wilson
equation, dening the so- alled overlap operator.
1.4.2 Overlap fermions
As wehave already stated, for many years it has not been realizedthat the