A Exam Presentation:

Instantons and the U(1) problem

Christian Spethmann

May 1st , 2007

Outline
1 Instantons In Particle Mechanics

Euclidean path integrals

Double Well
Periodic Potential
2 Topology Of Non-Abelian Gauge Theories

Instantons and the Winding Number

n and theta-Vacuum States
3 The Chiral U(1) Problem and its Solution

The Anomalous Axial Current

The Gauge-Variant Conserved Current
The U(1) Problem
Chiral Ward identities in QCD
Quark Zero Modes and Index Theorem
Chiral U(1) is spontaneously broken
The chiral Goldstone boson in one-flavor QCD
QCD with two massless quarks

Euclidean Transition Amplitudes

Consider a spinless particle of unit mass, moving in one dimension,
with a Euclidean Hamiltonian
H=

p2
+ V (x).
2

We want to calculate the probability for the particle to move from an

initial position xi to a final position xf during an imaginary time interval
T . This probability is given by
Z
HT /~
hxf |e
|xi i = N [dx] eS/~ ,
using the usual path integral approach.

Recovering the Ground State Energy

Here S is the Euclidean action
Z

T /2

S=
T /2

"  
#
2
1 dx
dt
+V
2 dt

and [dx] means integration over all functions that satisfy the boundary
conditions x(T /2) = xi and x(T /2) = xf .
The transition amplitude can be expanded in energy eigenstates,
which gives
X
hxf |eHT /~ |xi i =
eEn T /~ hxf |nihn|xi i
n

For large T the lowest energy eigenstate will dominate the sum
The real-world energy of the ground state can be determined from
this Euclidean amplitude.

Integrating over Orthogonal Functions

To evaluate the path integral, we assume that the action has a
stationary point x (t),
S
d2 x
= 2 + V 0 (x ) = 0.
x
dt
A general function x(t) that satisfies the boundary conditions can
then be expressed by
X
x(t) = x (t) +
cn xn (t),
n

where the xn (t) are a set of real orthonormal functions that vanish at
the boundaries and are eigenfunctions of the second variational
derivative of S at x ,

d2 xn
+ V 00 (x ) xn = n xn .
dt 2

Functional Determinant
The integral measure can then be defined by
Y
[dx] =
(2~)1/2 dcn ,
n

and the action can be expanded in a Taylor series, so that the

functional integral becomes
Z
Y 1/2
N [dx]eS(x)/~ = NeS(x)/~
n
[1 + O(~)]
n

= Ne

S(x )/~



1/2
det t2 + V 00 (x )
[1 + O(~)]

including terms up to quadratic order in the cn s. The result is just one

over the square root of the functional determinant, as expected for
bosonic degrees of freedom.

Instanton in a Double Well

We now assume that the potential is a double well, with minima at a
and V 00 (a) = 2 , and we want to calculate
ha|eHT /~ | ai and ha|eHT /~ | ai
The first step is to find the stationary point of the action. This
equation describes a particle moving in an inverted potential V (x).
From this it is clear that there is a solution with zero energy, so that
dx/dt = (2V )1/2 ,
or

Z
t = t1 +
0

This is the so-called instanton.

dx 0 (2V )1/2 .

Dilute Gas Approximation

The action integral over the instanton solution is


Z
S0 =

dt

 Z
 2 Z a
1
dx
dx (2V )1/2 .
(dx/dt)2 + V = dt
=
2
dt
a

For large |t|, the particle is exponentially close to the minimum:

(a x) et ,
so a chain of isolated instantons and anti-instantons gives an
approximate solution to the problem.
Let us now try to find the functional determinant corresponding to
such a solution. In this approximation, the action is just the sum of n
instanton actions, or nS0 , since the action at the minima is zero.

Evaluating the Functional Determinant

To find the determinant we consider the following points:
In the interval between the instantons, the particle is very close

to the minima, so V 00 = 2 . If the particle would always be there,

the determinant would be

1/2  1/2 T /2
e
.
N det(t2 + 2 )
=
~
Let the correction to this result from a single instanton or

anti-instanton be K . The determinant then gets an additional

factor of K n .
The locations of the instantons are arbitrary, so we have to

integrate over the center positions. This gives a factor of

Z

T /2

t1

dt1
T /2

tn1

dt2 . . .
T /2

T /2

dtn = T n /n!.

Double Well Transition Amplitude

If we start at a and end at +a (or vice versa), we need an odd

number of instantons. If we get back to the same place where we

started, we need an even number of instantons. We therefore
have
ha|eHT /~ | ai =

 1/2
X (KeS0 /~ T )n
eT /2
~
n!
even n

and the corresponding sum over odd ns for ha|eHT /~ | ai.

The result for the path integral is therefore
ha|eHT /~ | ai
i
 1/2
1h
eT /2 exp(KeS0 /~ T ) exp(KeS0 /~ T ) .
=
~
2

Energy Eigenstates in the Double Well

Comparing this to the expansion of the transition amplitude, we see
that there are two states |+i and |i with energies
E =

1
~ ~KeS0 /~ .
2

Now we have to calculate the value of K . We start by looking at the

equation for the functions xn . It has an eigenfunction of eigenvalue
zero,
1/2
x1 = S0 dx /dt.
The integration over the corresponding expansion coefficient c1 can
be rewritten as an integration over the time variable t1 that defines the
center of the instanton,
(2~)1/2 dc1 = (S0 /2~)1/2 dt1 .

Double Well Functional Determinant

The integration over this zero eigenvalue function has already been
done above. To get the correct value of the determinant K , we
therefore have to include a factor of (S0 /2~)1/2 and omit the explicit
integration over this zero eigenvalue function. The one-instanton
contribution to the matrix element becomes

1/2
ha|eHT | ai = N T (S0 /2~)1/2 eS0 /~ det 0 [t2 + V 00 (x )]
.
Comparing this to the one-instanton term in the transition amplitude
sum, we see that
1/2

K = (S0 /2~)



 det(t2 + 2 ) 1/2


 det0 ( 2 + V 00 (x ))  .
t

Instantons in a Periodic Potential

In a periodic potential with minima at integer positions, the calculation
is almost the same. The only difference is that now instantons and
anti-instantons dont have to alternate. The transition matrix element
from position j to position j+ is then given by
hj+ |eHT /~ |j i =

 1/2
eT /2
~
X

X
1  S0 /~ n+n
Ke
T
nnj+ +j ,
!
n!n

n=0 n=0

are the numbers of instantons and anti-instantons.

where n and n
This can be rewritten by using
Z
ab =
0

d i(ab)
e
.
2

Continuous Energy Eigenstates

We then find
hj+ |eHT /~ |j i =

 1/2
eT /2
~
Z 2
h
i
d
exp 2KT cos eS0 /~ .
ei(j j+ )
2
0

There is a continuum of eigenstates with energies given by

E() =

1
~ + 2~K cos eS0 /~ .
2

This result is very familiar from the energy bands encountered in solid
state periodic potentials.

Classical Vacua in Yang-Mills Theory

A pure gauge theory has the Euclidean action
Z
1
S=
d4 x(F , F )
4g 2
with the field strength tensor
F = A A + [A , A ]
where the gauge potentials are defined as matrices by
A = gAa T a .
We now want to find field configurations with finite action. The above
formulas imply that the field strength tensor has to be of O(1/r 3 ) in
four dimensional Euclidean space-time.

Homotopy Classes of Classical Vacua

This means that the gauge potentials have to be of the form
A = g g 1 + O(1/r 2 ),
where the first term is a gauge transform of zero.
This corresponds to a continuous mapping from the
three-dimensional hypersphere S 3 (at infinite distance in four
Euclidean dimensions) to the gauge group.
In general, it is not possible to eliminate this term by a gauge
transformation that is continuous over all four-space, because those
kinds of gauge transformations correspond to the mappings from S 3
to g that can be reached by continuously deforming from the trivial
mapping S 3 1. The pure gauge field configurations therefore fall
into gauge-invariant homotopy classes.

Definition of the Winding Number

Let us assume that the gauge group is SU(2). A general group

element can be parameterized by
g = a + ib
where a2 + |b|2 = 1. SU(2) is therefore homomorphic to to S 3 , and
we have to consider mappings from S 3 to S 3 .
All possible mappings are homotopic to a member of the family given
by

g () (x) = [(x4 + ix )/r] ,

where is called the winding number. The trivial mapping e.g.
corresponds to = 0, the identity mapping to = 1.

Winding Number as Topological

Charge
For the two-dimensional representation of SU(2), the winding number
can be calculated from the gauge configuration by
Z
1
=
d1 d2 d3 Tr ijk gi g 1 gj g 1 gk g 1 ,
24 2
where the i are angles that parameterize the 3-sphere. Using
Gausss theorem, it can be shown that the winding number can also
be calculated by the volume integral
Z
d4 x (F , F ) = 32 2 ,
where

1
F =  F
2
is the dual of the field strength tensor.

The BPST Solution

The BPST solution is the explicit form of the = 1 instanton, which

can be written as
A =

r2
g (1)1 (x) g (1) (x).
r 2 + 2

where
r=

4
X

xi2 .

i=1

As it should be, this solution is a continuous function that is zero at

the origin.

Definition of the |ni-Vacua

There exists an infinite family of classical field configurations |ni that

are analogous to the degenerate minima in the 1-D periodic potential
problem. Those states are pure gauge configurations in 3-space, with
the additional constraint that g(r ) has to approach the same constant
value as r . 3-space then becomes topologically equivalent to
S 3 and the same arguments as above can be used.
If a spatial field configuration is continuously deformed to a different
homotopy class, the space of pure gauge transformations has to be
left. Therefore an energy barrier exists and the situation is analogous
to the periodic potential.

Energy Eigenstates
We now consider a large 4-dimensional box of Euclidean spacetime.
Let the integral over all configurations inside the box with winding
number n be
Z
F (V , T , n) = N [dA] eS n ,
where we have used some gauge-fixing condition and set ~ = 1.
For large times T1 and T2 ,
X
F (V , T1 + T2 , n) =
F (V , T1 , n1 )F (V , T2 , n2 ),
n1 +n2 =n

because the winding number can be written as the integral over a

local density. This equation tells us that the |ni-vacua are not the
energy eigenstates, because the transition element is not a simple
exponential with a multiplicative composition law.

Definition of the |i-Vacua

Using a Fourier transform, it is possible to bring this expression into
the desired form:
Z
X
F (V , T , ) =
ein F (V , T , n) = N [dA] eS ei .
n

This can now be interpreted as the expectation value of eHT in an

energy eigenstate, called the |i-vacuum:
Z
HT
0
F (V , T , ) h|e
|i = N
[dA] eS ei .
This new vacuum state can be written as a sum over the classical
field configurations
X
|i
ein |ni.
n

h|eHT |i Transition Amplitude

Using the same methods as for the 1D particle case, let us now try to
calculate the energy density of this new vacuum state. In the dilute
gas approximation, the integral over all gauge configurations with the

same winding number is replaced by a sum over n instantons and n

anti-instantons, so that = n n. This gives

X
!)
h|eHT |i
(KeS0 )n+n (VT )n+n ei(nn) /(n!n

n,n



= exp 2KVTeS0 cos .
The energy density of a |i-vacuum is then given by
E()/V = 2K cos eS0 .

Action of a Yang-Mills Instanton

The action S0 of a single instanton can be found using the Schwartz
inequality:
2

4g S0 =

(F ) =

Z

Z
(F , F )

so that
S0


1/2 Z




(F , F )
 (F , F ) = 32 2 ||,
8 2
||.
g2

Equality holds for F = F .

The symmetry group of the solution has eight parameters: the size of
the instanton, the 4D location of its center, and three global
gauge
transformation parameters. From this we get a factor of (1/ ~)8 , or
equivalently 1/g 8 .

|i-Vacuum Energy-Density
Using this, the formula for the vacuum energy becomes
Z
2
2
d%
E()/V = cos e8 /g g 8
f (%M),
%5
0
where M is a mass-dimension parameter that defines the
renormalization scale. From RGE analysis it follows that observable
quantities can only depend on
1
1 log M + O(g 2 ),
g2
where 1 can be calculated from one-loop perturbation theory, and is
here given by 11/12 2 . Taking this into account, the energy density
becomes
Z


2
2
2
d%
E()/V = A cos e8 /g g 8
(%M)8 1 1 + O(g 2 ) .
5
%
0

Axial U(1) symmetry

We now include massless fermions in our gauge theory. The
Lagrangian in Minkowski space becomes
L=

X
1
(F , F ) +
f iD f
2
4g
f

where f is a flavor index and D = + A the usual covariant

derivative. This Lagrangian is symmetric under chiral rotations of the
fermion fields
X
f ei5 f

j5 =
f 5 f .
f

It is well known that this current is not conserved in the quantum

theory because of the Adler-Bell-Jackiw anomaly:
j5 =

Nf

(F , F, ).
32 2

Gauge-Variant Conserved Current

The anomalous term is proportional to the divergence of the current
G = 2 (A , F

2
A A ),
3

with

1
G = (F , F ) =  (F , F ).
2
It is therefore possible to define an anomaly free (but gauge
dependent) axial current
J5 = j5

Nf
G
16 2

J5 = 0.

The conserved charge corresponding to this current is given by

Z
5
Q = d3 x J05 .

Rotation of the |i-Vacuum

It can be shown that the effect of a gauge transformation in the n-th
homotopy class is to change the value of this charge by 2n.
We therefore get
Q 5 |ni = 2n|ni,
because the charge of the trivial vacuum A = 0 is obviously zero.
The axial charge is the generator of a U(1) symmetry group, and the
effect of a rotation with angle 0 on the |i-vacuum is therefore




1 0 5 X in
1 0 5
i Q
|i = exp
i Q
e |ni
exp
2
2
n
X
0
=
ei( +)n |ni
n

= | + 0 i.

Symmetries of Massless QCD

Let us consider QCD with two quarks:
In the massless limit, the Lagrangian of this theory is symmetric

under U(2) U(2), because the left- and righthanded quark

fields can be independently rotated with arbitrary unitary
matrices.
The observed spectrum on the other hand has only

(approximate) U(1)B SU(2)V symmetry, where U(1)B

corresponds to baryon number and SU(2)V to the vector
subgroup (with identical rotation of left- and righthanded fields).
Since the axial SU(2)A and preudoscalar U(1)P symmetries are
not realized, this symmetry has to be spontaneously broken, and
one would expect to see four Goldstone bosons corresponding to
those generators.

Broken Chiral Symmetries

However, only three low-mass pseudoscalar bosons (the pions) are
observed, corresponding to the generators of SU(2)A . As Weinberg
showed,
the mass of the missing boson would have to be less than

3m , so that it can not be any other of the observed mesons. This

missing Goldstone boson is the U(1) problem.
Kogut and Susskind found the solution of this problem by looking at
the 1+1 dimensional Schwinger model. It has two massless fields +
and , one with the usual propagator, one with minus the usual
propagator. All gauge invariant operators couple to the sum of those
fields, so that the propagators cancel and no Goldstone pole is
observed. The gauge-variant, conserved current couples to the
gradient of the difference (+ ), so that a Greens function for
one gauge-variant current and a string of gauge-invariant fields has
the expected pole. Following Colemans analysis, I will now show that
this also happens in QCD.

Greens Function for Fermion Fields

The Euclidean action for massive fermions in a gauge field is given by
Z
D M).
S = i d4 x (
The Greens function for a set (r ) of local functions of those fields is
given by
R
S (1) (x1 ) . . . (m) (xm )
[d][d]e
R
,
h(1) (x1 ) . . . (m) (xm )iA =
S
[d][d]e
We now perform an axial rotation
ei5 ,

ei5 .

or (for infinitesimal )
= i5 ,

5 .
= i

Deriving the Ward Identity

The Lagrangian is invariant under this rotation, and we can derive the
chiral Ward identity
hj5 (y )(1) (x1 ) . . . (m) (xm )iA
5 (y )(1) (x1 ) . . . (m) (xm )iA
+ 2hM
+ 4 (y x1 )h(1) (x1 )/ . . . (m) (xm )iA
+ ...
+ 4 (y xm )h(1) (x1 ) . . . (m) (xm )/iA
=

iC
(F (y ), F (y )) h(1) (x1 ) . . . (m) (xm )iA ,
8 2

5 and the right-hand side would be zero without the

where j5 =
anomaly.

Integral Ward Identity Formula

Integrating this equation over y gives

Z
2

A
5 (y )(1) (x1 ) . . . (m) (xm ) + h(1) (x1 ) . . . (m) (xm )iA
d4 y M

= 4iCh(1) (x1 ) . . . (m) (xm )iA

where is the winding number of the gauge field. The constant C
depends on the representation of the fermions and is defined by
Tr T a T b = C ab .

Fermion Functional Determinant

In a gauge theory with one massless quark, the energy density of the
|i-vacua is still given by
E()/V = 2K cos eS0 ,
but the determinant K now contains a term of the form
 


iD
/
i( + A )
det
= det
.
i/
i
This factor is zero, because the Dirac field has zero modes with
vanishing eigenvalues of iD
/ . Let us assume that the spectrum is
discrete,
iD
/ r = r r .

Odd and Even Parity Eigenfunctions

The s are real and occur in pairs of opposite sign, because 5

anticommutes with . Eigenfunctions of vanishing eigenvalue can be
choosen to be also eigenfunctions of 5
5 r = r ,

(r = 0).

Let n be the number of eigenfunctions with positive/negative parity.

We then have the sum rule
n n+ = ,
so there are zero eigenvalues in any gauge field with nonvanishing
winding number.

Proof of the Index Theorem

To prove this, we take the integrated Ward identity for a massive
quark and no s,
Z
2i = 2

5
d y m

A

R
R
5
S d4 y m
2 [d][d]e
R
=
.
S

[d][d]e

The eigenfunctions of i(D

/ m) are the same as before,
i(D
/ m)r = (r im)r ,
so that the functional integral becomes
P R
Q
2m r d4 y r 5 r s6=r (s im)
Q
2i =
r (r im)
Z
X
= 2m
d4 r 5 r (r im)1 .

|i-Vacua are Degenerate

Using the fact that eigenfunctions of a Hermitian operator with

different eigenvalues are orthogonal,
Z
Z

4
d y r 5 r = 0 (r 6= 0),
d4 y r 5 r = 1 (r = 0)
we see that only the zero modes contribute, and the integral becomes
2i = 2i(n+ n )

This is the above sum rule, so we now know that every vacuum with
nonvanishing winding number has at least one zero eigenvalue. The
|i-vacua have therefore the same energy.

Chiral Rotations of |i
Let us define denominator-free Greens functions by
Z
(1)
A
S (1) (x1 ) . . . ;
hh (x1 ) . . .ii = [d][d]e
they obey the Ward identity



+ 2i hh(1) (x1 ) . . .iiA = 0

The Greens functions of one-flavor QCD are given by

R
[dA]eSg ei hh(1) (x1 ) . . .iiA
R
,
h|(1) (x1 ) . . . |i =
[dA]eSg ei hh1iiA
so that

+2

h|(1) (x1 ) . . . |i = 0.

Vacuum Expectation Value of

This proves that chiral U(1) rotations, parameterized by , rotate the
|i-vacuum. The energy density of the vacuum is an invariant quantity
under U(1).
To show that chiral symmetry is spontaneously broken, it suffices to
find a Greens function that has a non-zero derivative with respect to
. We choose to calculate
R
S ei (x)
[dA][d][d]e
R
h| (x)|i =
S ei
[dA][d][d]e
5 ) are chiral eigenfields, so that
where = 12 (1
/ = 2i .

Evaluating h| (x)|i
The calculation is similar to the one for the pure gauge theory. One
important difference, however, is that there is now a fermion
determinant that has n vanishing eigenvalues, if there are n
instantons and anti-instantons. The integral can therefore only give a
non-zero value if there are at least 2n Dirac fields in the Greens
function.
The path integral in the denominator has zero Dirac fields, so the only
contribution comes from the classical vacuum A = 0, and the result
is a product of a Bose and a Fermi determinant. The numerator for
needs to have one instanton and no anti-instantons. The Fermi
integral gives
Y
1
0 (1 5 )0
r = 0 0 det 0 (iD
/)
2
r 6=0

since the zero eigenvalue function has odd parity in this case.

Proof for Spontaneous Symmetry

Breaking
Including all factors from the pure gauge case, and doing the
corresponding calculation for + , we get
Z
2
2
det 0 (iD
/)
d%
f (%M)
.
h| (x)|i = e8 /g ei g 8 2
5
%
det(i
/
)
0
Since one eigenvalue has been removed from the numerator, the
ratio of determinants has to be of the form
det 0 (iD
/)
= % h(%M),
det(i/ )
by dimensional analysis, where % is the size of the instanton and h is
some unknown function. So we now know that spontaneous
symmetry breakdown occurs.

One Flavor QCD Model

If there is a chiral Goldstone boson, the corresponding pole should
appear in
h|+ (x) (0)|i.
The only contributions to this are from the zero instanton field and
from the one instanton, one anti-instanton field configuration. The
former is the one-loop perturbative expression, the latter gives the
product
h|+ |ih| |i
which has no Goldstone pole either. As in the Schwinger model, the
pole only appears in gauge-variant Greens functions like
h|J5 (x) (0)|i = h|j5 (x) (0)|i +

i
h|G (x) (0)|i.
16 2

Goldstone Pole in One Flavor QCD

The second term has a Goldstone pole if and only if
Z
d4 x h|G (x) (0)|i =
6 0.
But we know that

d4 x G = 32 2

and that only = 1 field configurations contribute, so that

Z
d4 h|G (x) (0)|i = 32 2 h| |i =
6 0.
So a Goldstone pole appears in the Greens function of one
gauge-variant current and one gauge-invariant operator. However, if
we calculate the propagator of the gauge-variant current h|J J |i,
we again find no pole, because only the classical vacuum with = 0
contributes.

Goldstone Pole in Two Flavor QCD

For the real world case of two massless quarks, the sum rule now
becomes
n n+ = 2.
This means that iD
/ in a one instanton field now has two vanishing
eigenvalues, and all quark bilinears will have zero expectation values.
To prove that chiral U(1) is spontaneously broken, we can instead
calculate quadrilinears like
1 (1 5 )1 2 (1 5 )2 .
Another difference is that the |i-vacuum rotates twice as much under
axial rotations, as necessary to preserve SU(2) SU(2) symmetry:



h|1 (x1 ) . . . |i = 0,
+4