Instantons in Particle Physics

Yuhsin Tsai

Institute for High Energy Phenomenology,

Newman Laboratory of Elementary Particle Physics,
Cornell University, Ithaca, NY 14853, USA
E-mail: yt237@cornell.edu

Abstract
In this report we review the basic idea of instanton, especially its connection to the
vacuum structure of the non-abelian gauge theory. The solution of the QCD U(1)-
problem is described, and the baryon and lepton number violation using instanton
effect is discussed.
1 Introduction
After Yang and Mills’s first paper of the isotopic gauge invariance [1], people’s understanding
of the non-abelian gauge theory has changed dramatically in this fifty five years. Not only in
discovering the gauge interactions and field representations in nature, but also have some idea
about how the vacuum looks like. As we will review in this paper, instantons play a central rule
in generating the vacuum structure of the nonabelian gauge theory. This is why instantons are
widely used in different aspects of particle physics.
This report focus on the basic property of instantons and their importance in the SM physics,
such as solving the U(1)-problem and giving the baryon and lepton number violation. We will
very closely follow the discussion of Coleman [2]. In order to use the newer notations and some
topological properties of the gauge theory, we also consult Ryder [3] and Terning [4]. To give a big
picture without focusing too much on the algebra, we put detailed derivations in the Appendix.
As you can easily see, this report is ‘about’ fifteen pages besides the Appendices.
The structure of this review is as follows: In Sec. 2, we define instantons in a 1 + 1D quantum
mechanical system as the barrier penetration between two potential wells. After learning how to
use instantons to calculate the tunneling amplitude, in Sec 3, we move to the non-abelian gauge
theory and looking for its vacuum structure. As we will see, instantons in this case also generate
barrier penetration between different vacua. Using the idea, we identify and solve the QCD U(1)-
problem in Sec. 4. In Sec. 5, we use the tool developed in the U(1) problem to generate the baryon
and lepton number violation. A short conclusion is given in the end.

2 Instanton and the tunneling amplitude

Instanton describes the tunneling between different vacua. The vacua can be as trivial as the
minimum of scalar potential or can be as subtle as the gauge field configuration on the boundary.
As a warm up, let us begin from the tunneling between potential wells in 1+1D quantum mechanics
in the semiclassical limit (small
).
Different from the standard quantum mechanical way of solving the differential equation, here
we calculate the penetration amplitude using the path integral with an Euclidean action. Why
Euclidean action? This is because when using the Euclidean action SE , e−SE gives the tunneling
amplitude. To see this, let us write the tunneling amplitude between points a and b with energy
E and potential barrier V in the WKB approximation

 
1 b 1
exp − [2m(V − E)] 2 dx . (2.1)

a
When E > V , the integral becomes
    
pdx = pẋdt = (H + L)dt = (E + L)dt = L dt. (2.2)

In the last equality we shift the energy to zero. As we can see, the WKB factor in this case gives
the usual action. When E < V , the way to get the same action result is to change V → −V .
For the equation of motion mẍ = −∂V /∂x, this is identical to changing the time t → it. That is,
when using the Euclidean action, we recover the penetration amplitude directly.

1
 Figure 1: The 1D double wells. The graph is copied from [2].

The goal of this section is to calculate the correlation function of a particle with unit mass
moving between two vacua in Fig. 1(a),

−HT /

xf |e |xi  = N [dx]e−S/
, xi , xf = ±a, (2.3)

where S is the Euclidean action and N is a normalization factor. The action in this case is
 T /2 
2 
1 dx
S= dt +V . (2.4)
−T /2 2 dt

Since the potential becomes negative in the Euclidean action, the tunneling between x = ±a
becomes the oscillation between the two points in Fig. 1(b). Let us focus on an interesting case
first, with the particle being static at x = −a when t = −T /2 and then moves to x = a at t = T /2.
The equation of motion in this case is the one with vanishing E,
1
dx/dt = (2V ) 2 . (2.5)

Equivalently,  x
1
t = t1 + dx
(2V )− 2 , (2.6)
0
where t1 is the time when x = 0. The solution is sketched in Fig. 2. For large t, x approaches a,
and eq. (2.5) can approximated by
dx/dt = ω(a − x). (2.7)
Thus we have
(a − x) ∝ e−ωt . (2.8)
This means the solution is a well-localized objects, having a size on the order of 1/ω. This object
is called instanton. When having a solution going from a to −a, we call it anti-instanton. The
instanton action can be written as
  a
1
2
S0 = dt(dx/dt) = dx(2V ) 2 . (2.9)
−a

2
 Figure 2: The one instanton solution for barrier penetration. [2]

This goes back to the WKB factor in eq. (2.1) as we expect when using the Eucliden action.
Go back to the corelation function in eq. (2.3). If the particle starts from −a and goes back
to −a, we have
∆tn ∆tn−1 ∆t1
−a|e−H(−a)
|−a−a|Ins(tn )|aa|e−H(a) |aa|Ins(tn−1 )|−a...a|Ins(t1 )|−a−a|e−H(−a)
|−a.

(2.10)
Here ∆tn ≡ tn −tn−1 . Ins(t) means the one-instanton-effect centered at t that changes the particle
position between ±a. There are even numbers of instantons in this case. If the process is −a → a,
there will be odd number of instantons. As we discuss before, instantons are well time-localized.
This means even if it changes the position, the particle stays at ±a in most of the time. Since the
potential at ±a are symmetric, the Lagrangian is almost a constant through the time evolution
(besides the short ‘instanton appearace’). We then be able to use this ‘dilute-gas approximation’
to get
∆tn ∆tn−1 ∆t1
−a|e−H(−a)
| − aa|e−H(a)
|a...−a|e−H(−a)
| − aK n = −a|e−HT /
| − a × K n . (2.11)

where K denotes the correction from the short intanton appearence. Define ω 2 ≡ V

(±a). In
Appendix. A, we calculate
 ω  12
−a|e−HT /
| − a = e−ω∆t . (2.12)
π

One can use functional integral to show [2]
 1
 det(−∂t2 + ω 2)  2

K = (S0 /2π
) 

1
2  , (2.13)
det (−∂t2 + V

) 

where det
indicates that the zero eigenvalue is removed.

For the instanton part of the action, the total action of these widely separated objects is nS0 .
Also, we can parameterize the time integral as
 T /2  t1  tn−1
dt1 dt2 ... dtn = T n /n!. (2.14)
−T /2 −T /2 −T /2

3
Combining all the ingredients together, we have
 ω  12 (KT e−S0 /
)n
−HT /

−a|e | − a = e−ωT /2 , (2.15)
π
even n
n!

while for the −a → a case, we sum over odd ns. This gives the correlation function between two
vacua defined in eq. (2.3)
 ω  12 1
±a|e−HT /
| − a = e−ωT /2 exp(Ke−S0 /
T ) ∓ exp(−KeS0 /
T ) . (2.16)
π
2
For the ground state energy, let us define the energy eigenstates |n as

±a|e−HT /
| − a = e−En T /
±a|nn| − a. (2.17)

n

In large T limit, the correlation function only relates to the ground state energy . Comparing to
eq. (2.16), this gives
1
E± =
ω ±
Ke−S0 /
. (2.18)
2
That is, the instanton that corresponds to the barrier penetration breaks the degeneracy of the
two ground states. If we call these eigenstates |+ and |−, we have

1  ω  12
a|−−| − a = −a|++| − a = . (2.19)
2 π

To sum up, instantons are well time-localized objects describing the barrier penetration be-
tween two vacua. By using the Euclidean action and the dilute-gas approximation, we can calculate
the correlation function and the ground state energy.

3 The vacuum of the gauge theory

In this section, we review the role of instantons in describing the tunneling between vacua that
relate to different gauge field configurations. Limited by the pages, I will skip some detailed proof
and focus on the big picture of the relation between winding-number, |n-vacuum, |θ-vacuum and
instantons.

3.1 The winding number

The winding number is a topological quantity that denotes different mapping classes between a
group configuration space and a coordinate space. The simplest example is the mapping between
U(1) and a circle. Topologically, both of them are S 1 . The U(1) generator of the trivial mapping
is
h(0) (θ) = 1. (3.1)
We can have the identity mapping
h(1) (θ) = eiθ , (3.2)

4
or even more complicated mappings
h(ν) (θ) = eiνθ . (3.3)
These mappings can not be connected by a continuous deformation. They belong to different
‘homotopy classes’ of the mappings. We call the integer number ν as the winding number1 of the
mapping between S 1 and S 1 . The fancy convention is

π1 (S 1 ) = Z. (3.4)

Not only for abeliean groups, in the non-abelian case, we can also have homotopy classes betwen
specific mappings. For example, the mapping between SU(2) and the boundary of 4D Euclidean
spacetime (S 3 ) gives
π3 (S 3 ) = Z. (3.5)
One way of seeing this is that we can parametrize the SU(2) operator into (here x · σ ≡ xi σi )

ν
ν x4 + ix · σ
h (x) = √ , τ 2 = x24 + x2 . (3.6)
τ 2

i.e. with certain winding number ν, each point on the boundary corresponds to a group element.
An important formula of calculating the winding number from the operators is

1
ν= 2
d3 S n̂µ µνλσ (∂ν hh−1 , ∂λ hh−1 ∂σ hh−1 ). (3.7)
24π
One can show that this integral is topologically invariant.

3.2 The gauge field on the boundary

The question we are interested in is the non-perturbative gauge theory with the field strength
2
− 8π
part of the action looks like e g2 . This means, we want to study the semi-classical theories with
finite actions (large g but not too large).
For an SU(2) gauge theory with finite action (for conventions, see Appendix. B)

1
S = d4 x 2
(F, F ) − iψ̄Dµ γ µ ψ , (3.8)
4g
the gauge field on the boundary can only be in a pure gauge form
i
Aµ = − (∂µ h)h−1 + O(τ −2 ). (3.9)
g
Ignoring the O(τ −2 ) term that is suppressed at large radius, the gauge field on the boundary can
be written into the following form using eq. (3.6) (with ν = 1)
i −1
Ai = (xi − σi (σ · x + ix4 )), A4 = σ · x. (3.10)
gτ 2 gτ 2
1
For negative ν, we mean the mapping is ‘wided’ in the opposite direction.

5
When doing this, we map the gauge field configuration to the Euclidean boundary S 3 .
Besides the usual field strength term (F, F ) in the Lagrangian, there is another gauge invariant
term that should exist 
1
d4 x (F, F̃ ), F̃µν ≡ µνλσ F λσ . (3.11)
4
Doing some algebra, we can show that (F, F̃ ) can be written into a totally derivative term


2
(F, F̃ ) = ∂µ Kµ , Kµ = 2µνλδ Aν , ∂λ Aσ + Aλ Aσ . (3.12)
3

Using eq. (3.10), we have

2xµ
Kµ = , (3.13)
g2τ 4
and a nonvanishing volume integral
 
1 4 3 8π 2
d x(F, F̃ ) = K⊥ d S = 2 . (3.14)
4 g

One interesting thing is, the same integral can also be connected to eq. (3.7)
  

1 4 3 1 3 2
d x(F, F̃ ) = K⊥ d S = d S n̂µ µνλσ Aν , ∂λ Aσ + Aλ Aσ , (3.15)
4 2 3

1 3 −1 −1 −1 8π 2
= d S n̂  (∂
µ µνλσ ν hh , ∂λ hh ∂σ hh ) = ν. (3.16)
3g 2 g2

This means, the integral in eq. (3.14) is given by the nontrivial mapping between SU(2) and the
Euclidean S 3 with winding number ν = 1.

3.3 The n-vacuum and the θ vacuum

In fact, the nontrivial gauge field Aµ we use in the previous section is the instanton in 4D. As we
just see, one instanton effect changes the winding number of the system by one. To see this more
clearly, let us do the winding number integral again by using the boundaries as shown in Fig. 3.
  ∞ 
1 3
ν= 2
d S 4ijk (Āi , Āj Āk ) + dx4 d2 S n̂i iµνλ (Āµ , Āν Āλ ) . (3.17)
24π I−II −∞ III

Here Āµ ≡ (∂µ h)h−1 . To do the integral, we need to know the instanton solution inside the
boundary, which is2

τ2 −i
Aµ = 2 2
(∂µ h)h−1 . (3.18)
τ +ρ e
2
We can get this by solving Fµν = F̃µν and match the boundary condition. The equation comes from getting
the equality of the Schwartz inequality (F, F ) ≥ (F, F̃ ), which gives the lower bound of the action.

6
Figure 3: I and II are the hypersurfaces x4 → ∞ and x4 → −∞ and III is the hypersurface joining
them. The diagram is copied from [3]

The ρ denotes the ‘size’ of the instanton. As we can see, Aµ goes back to eq. (3.9) when τ → ∞.
Since the winding number is gauge invariant, it is convenient to choose a gauge such that A
4 = 0
so that the integral over the ‘cylinder III’ vanishes3 . The integral then becomes
 
1 3
ν= 2
d S 4ijk (Āi , Āj Āk ) − d3 S 4ijk (Āi , Āj Āk ) = νI − νII . (3.19)
24π I II

i.e. the instanton really changes the winding number of the system.
Another way of interpreting the result is that as time evolves from −∞ to ∞, a vacuum (with
homotopy class νII ) evolves into another vacuum (with homotopy class νI ). The instanton solution
then represents the transition between one vacuum class to another. The Yang-Mills vacuum is
therefore infinitely degenerate, consisting of an infinite number of homotopically non-equivalent
vacua. Since they are denoted by different winding numbers, we call these vacua the |n-vucua.
We can also do the Fourier transform of it
|θ = einθ |n. (3.20)
n

This gives the so called |θ-vacua.

3.4 The energy of a θ vacuum

Let us calculate some quantities of a θ vacuum that we can use for the U(1) problem. First, using
the dilute-gas approximation, the correlation function θ|e−HT |θ is contributed by n instantons
and n̄ anti-instantons
 −S0 n+n̄ (V T )n+n̄ i(n−n̄)θ
θ|e−HT |θ = Ke e . (3.21)
n,n̄
n!n̄!
3 ix·σ −2 x4 π
Such a gauge transform is A
µ = U Aµ A−1 − i(∂µ U )U −1 with U = exp[ (τ 2 +ρ 2 )1/2 θ], θ = tan [ (τ 2 +ρ2 )1/2 − 2 ].

7
Similar to the way we derive eq. (2.16), here e−S0 is the one instanton action, K denotes the
‘instanton appearance’ effect (similar to eq. (2.13)), [(V T )n+n̄ /(n!n̄!)] comes from the 4-volume
integral as in eq. (2.14), and ei(n−n̄)θ is the Fourier transform factor as in eq. (3.20). The sum can
be written as

θ|e−HT |θ = exp(KV T e−S0 eiθ ) exp(KV T e−S0 e−iθ ) = exp(2KV T e−S0 cos θ), (3.22)

which gives the energy density

E(θ)/V = −2Ke−S0 cos θ. (3.23)
After putting in the instanton action S0 = 8π 2 /g 2 and the correct K value (coming from the
calculation of the instanton degree of freedom, dimensional analysis and the RG running of the
coupling), the result becomes
 ∞
−8π 2 /g 2 −8 dρ 2
E(θ)/V = −A cos θ e g 5
(ρM)8π β1 , (3.24)
0 ρ
where g is the gauge coupling, M is the renormalization scale, β1 is a number that can be calculated
from RG, and A is a constant independent of ρ, g and M.
The next thing we want to calculate is the expectation value of (F, F̃ ). By the translational
invariance, 
1 32π 2
θ|(F (x), F̃ (x))|θ = d4 xθ|(F, F̃ )|θ = 2 θ|ν|θ, (3.25)
VT g VT
here we have used eq. (3.14). This gives

32π 2
[dA]νe−S eiνθ
 
32π 2 i d
θ|(F (x), F̃ (x))|θ =  =− 2 ln −S
[dA]e e iνθ
. (3.26)
2 −S iνθ g V T dθ
g V T [dA]e e

Since 
[dA]e−S eiνθ = θ|e−HT |θ, (3.27)

we have
64π 2 i −S0
θ|(F (x), F̃ (x))|θ = − Ke sin θ. (3.28)
g2

4 The U(1) problem

The most successful use of instantons in the SM is to solve the U(1) problem. In this section, we
define what is the U(1) problem first and then give a solution using the concept of instantons.

8
4.1 The Goldstone-boson of the U(1)A
In SM, QCD is the formal description of the strong force. The quarks including up and down
types in each of the three families interact with each other through changing the SU(3) gauge
fields. This gives successful prediction in experiments and has been well accepted. However, one
interesting feature of the strong interaction is that besides the gauge symmetry, we also have some
approximated global symmetry that gives corresponding selection rules to a very good level. For
example, when ignoring the u and d quark masses in the high energy scale (like GeV scale), the
U(2)L × U(2)R transform acting on the chiral spinors



uL uR
ψL = , ψR = , (4.1)
dL dR

is invariant in the kinetic term

L ψ̄L iDµ γ µ ψL + ψ̄L iDµ γ µ ψL . (4.2)

Because of the quark condensation, this global symmetry is spontaneously broken into

U(2)L × U(2)L → SU(2)D × U(1)B , (4.3)

where the SU(2)D gives the isospin symmetry and the U(1)B gives the baryon number conservation
in the strong dynamics. The three Goldstone bosons from SU(2)L × SU(2)R → SU(2)D becomes
the three light pions, which is a strong evidence that the spontaneous symmetry breaking really
happens.
Extending the same idea to the broken U(1)A , which has the transform
5
ψf → e−iαγ ψf , (4.4)

(Now ψf is written into the four component Dirac spinor with flavor f = 1, 2) and the associated
current
2
jµ5 = ψ̄f γµ γ5 ψf , (4.5)
f =1

the SSB (spontaneous symmetry breaking) of it should also give a Goldstone boson with the same
mass scale as pions. However, we do not see it. This gives the most naive version of the U(1)
problem, i.e., where is the Goldstone of the broken U(1)A ?
The reason why we say “naive” is that the SSB argument we just give is not quite correct.
In fact, the U(1)A is broken by anomaly in the perturbation theory. In the limit of N massless
quarks,
N
∂ µ jµ5 = (Fµν , F̃λσ ). (4.6)
32π 2
This means jµ5 is not a conserved current. It looks like we just solve the problem since U(1)A is
not a symmetry in the beginning. However, if we redifine the current into
N 2 λ σ N
Jµ5 ≡ jµ5 − µνλσ (Aν
, F λσ
− A A ) ≡ jµ
5
− Gµ , (4.7)
32π 2 3 32π 2

9
since
∂µ Gµ = (Fµν , F̃λσ ), (4.8)
Jµ5 becomes a gauge-variant but conserved current4 . Gauge-variant means it is not observable.
Nevertheless, because its charge 
Q5 = d3 xJ05 (4.9)

is conserved, this theory when realized in the Goldstone mode would demand the existence of a
pseudoscalar meson with mass m0 [6]5
m0 mπ . (4.10)
To see this we can use the standard current-algebra technique to obtain a Ward identity [7] 6
 
m20 − k 2        
2 2
m0 f0 = i 2
ik ν 4
d xe −ik·x
0 T (∂ Jµ (0)Jν (x)) 0 + d4 xe−ik·x 0 δ(x0 ) ∂ µ Jµ5 (0), J05 (x)
 µ 5 5  0 ,
m0

m0 − k 2
2    
= i 2
ik ν
d4 xe−ik·x 0 T (∂ µ Jµ5 (0)Jν5 (x)) 0 + m2π fπ2 , (4.11)
m0

where f0 is the isoscalar meson decay constant. In low energy limit k ν → 0, if there is no zero
mass pole in the first term on the right-hand side, we relate m20 f02 to a the second integral term
which is m2π fπ2 ,
m20 f02 = m2π fπ2 (4.12)
To go further, one can write Jµ5 into the sum of an SU(3) octet and a singlet. Since all the
pseudoscalar octet decay constants are equal, f0 fπ , we have the mass relation in eq. (4.10).
This brings the trouble back; we have to explain where this pseudo scalar goes.
It is pointed out by Kogut and Susskind [8] that one way to avoid the problem is for the
gauge-variant Jµ5 to be coupled to a massless ‘particle’, then the first integral in the right side
gets a pole and does not drop out when k ν → 0. There then be no constraint on m0 . Since Jµ5
is gauge-variant, this gauge-dependent massless particle does not couple to the physical quantity
and brings no further problems.
To identify this massless field, Kogut and Susskind study the Schwinger model - massless spinor
electrodynamics, in 1 + 1D in a covariant gauge. This model is solvable and has the ingredients
we want:

• a gauge-invariant axial current with an anomalous divergence.

• a gauge-variant but conserved axial current.

• a U(1)A breaking without Goldstone poles in the gauge-invariant Green’s functions.

4
The new added gauge-variant term can be seen as spurions [5]. That is the way we change an explicit SSB (of
U (1)A from anomaly) into a spontaneous one. The price of doing this is that we should have a Goldstone-like field
couples to the this gauge-variant term. This√ is what we want to show here.
5
A more precise result should be m0 ≤ 3mπ .
6
I do not know how to derive this, just to describe the problem is in a more quantitative way.

10
What Kogut and Susskind find is that after using boson fields to describe the 1D system (this is the
feature of the Schwinger model), two free massless fields φ+ and φ− , create quanta of positive and
negative norm. This gives the propagators carrying opposite signs. All gauge-invariant quantities
couple to the sum of these fields (φ+ + φ− ). This has zero propagator and is free of the new poles.
On the other hand, the gauge-variant quantities couples to ∂µ (φ+ − φ− ). The coupling to φ− then
carries an additional sign which compensates the sign in the propagator and thus gives the pole
we want. This setup is called the Goldstone dipole.
According to this, the missing degree of freedom coming from the SSB does not exist in the
gauge-invariant greens functions but are shown in the gauge-variant ones. We then be able to
formulate the U(1)-problem in a more precise way: Is the U(1)A in QCD spontaneously broken
via a Goldstone dipole?

4.2 QCD (baby version)

’t Hooft gives a brilliant solution of the U(1) problem when connecting this chiral symmetry to
instanton [9]. To get an idea about how this is done, let us solve the U(1)-problem in a baby
version QCD first. The basic steps are as follows
• Show the relation between the U(1)A transform and the shift of the θ-vacua.

• Show that the U(1)A is SSB when staying in a θ-vacuum.

• Look for the Goldstone-dipole in the gauge-variant Green’s functions.

4.2.1 U(1)A and the θ-vacua

In this baby theory the gauge symmetry is SU(2), and there exists only a single isodoublet quark
with zero mass. The action looks like

1
S = d4 x (F, F ) − iψ̄Dµ γµ ψ . (4.13)
4g 2
There are θ-vacua in this case, with the same quantities derived in Sec. 3.4,
64π 2 i −S0
E(θ)/V = −2K cos θe−S0 , θ|(F, F̃ )|θ = − Ke sin θ. (4.14)
g2
The only difference for quarks is now the K-factor (see eq. (2.13)) contains

iD/ i(∂µ + Aµ )γµ

det = det , (4.15)
i∂/ i∂/
where Aµ is the field of an instanton. To calculate this, we need to find the eigenfields of ψ
under the iD/ operator. As we will show now, there exists zero eigenvalue modes when there is a
nontrivial winding number. This makes the determinant vanish, as does E(θ)/V and θ|(F, F̃ )|θ.
Decomposing the fermion into different eigenfunctions of iD,/

/ r = λr ψr .
iDψ (4.16)

11
 / is Hermitian, all λr ’s are real. Using γ 5 , we can get the other set of the eigenfields
Since iD
/ 5 ψr = −λr γ 5 ψr .
iDγ (4.17)
Thus non-vanishing eigenvalues always occur in pairs of opposite sign. The eigenfunctions of γ 5
/ (since γ52 = 1, we have χr = ±1)7
are the zero eigenfunctions of iD
γ5 ψr = χr ψr , (λr = 0). (4.18)

Let us denote the number of zero eigenfunctions by n± . As proved in Appendix. C.2, we have
n− − n+ = ν. (4.19)

That is, there is a zero eigenvalue in any gauge field of non-zero winding number. When there is
an instanton background centering at X and having size ρ, the zero mode eigenfunction is
− 32
ψ0 (x − X, ρ) = ρ ρ2 + (x − X)2 u, (4.20)

where u is a constant spinor. When having n widely separated instantons and anti-instantons,
there are n such eigenfunctions centered about each object.
The result of having zero modes is that, the E(θ) and the expectation value of (F, F̃ ) with
different θ are always zero. All the θ-vacua are degenerate. Where does this degeneracy come
from? As we show in Appendix.X, when applying a U(1)A transform (ψ → exp(−iαγ 5 )ψ) to the
θ-vacuum expectation value of quark mutilinears φ(i) ’s, we have
∂ ∂   (1)  
+2 θ φ (x1 )... θ = 0. (4.21)
∂α ∂θ
This means the U(1)A transform rotates one θ-vacuum into another. When staying in one |θ,
U(1)A is spontaneously broken, and the θ-vacua are the many vacua that appear when a symmetry
suffers SSB. This explains where the degeneracy comes from - it comes from the U(1)A invariance.

4.2.2 The SSB of U(1)A

In fact, it might be too early to say that the SSB really occurs. If the ∂α and ∂θ for every
Green’s function just vanish independently, the θ-vacua would have nothing to do with the U(1)A
transform. To show that the vacuum really ‘feels’ the U(1)A transform, let us calculate the Green’s
function of a fermion condensate ψ̄ψ that breaks U(1)A .

[dA][dψ][dψ̄]e−S eiνθ σ± (x)
θ |σ± (x)| θ =  , (4.22)
−S iνθ
[dA][dψ][dψ̄]e e

where
1
σ± = ψ̄(1 ± γ 5 )ψ. (4.23)
2
7 / 5 ψr = iDψ
To see this, when having χr = 1, iDγ / r = λr ψr = λr γ 5 ψr , which contradicts eq. (4.17) if λr is
non-zero.

12
Since under U(1)A
1   
σ± → σ± + δσ± = ψ̄ − iψ̄γ 5 δα 1 ± γ 5 ψ − iγ 5 ψδα , (4.24)
2
The σ± are eigenfunctions of the U(1)A transform
∂σ± /∂α = δσ± /δα = ∓2iσ± . (4.25)
If eq. (4.22) is non-zero, the vacuum expectation value feels the U(1)A transform, and the SSB
really happens.
To calculate the Green’s function, we use the dilute-gas approximation to multiply all the
instanton and anti-instons together like what we do in Sec. 3.4. The similar factors such as
 ∞
−8 dρ
K = 2g 5
f (ρM) (4.26)
0 ρ

still exists. For the functional integral of fermions, if there exists one instanton, the winding
number ν = 1 gives one zero mode and the integral containing det(iD) / vanishes. The only non-
zero case is when having ψ̄0 ψ0 in the integral, such that the zero mode fermions in the action
contract with it and the determinant of the rest of the fermions exists. This case is exactly the
same as the one in eq. (4.22). For the σ− in the numerator, the Fermi integral under one instanton
background is8
1 † 

ψ0 (x − X, ρ)(1 − γ 5 )ψ0 (x − X, ρ) λr = ψ0† (x − X, ρ)ψ0 (x − X, ρ) det(iD).

/ (4.27)
2 λ =0
r

Here det denotes a determinant with vanishing eigenvalues removed. Since the determinant term
does not depend on the instanton location X, the integral over X is trivial

d4 Xψ0† ψ0 = 1. (4.28)
2
− 8π
For [dA], the instanton action gives a factor e g2 , and there is also a factor eiθ for ν = 1. In the
denomenator, since there is no fermions in the integral that contract the zero modes, there can be
no instantons and anti-instantons. Thus we only have det(i∂). /
Putting all the ingredient together, the Green’s function becomes
 ∞
2
− 8π2 ∓iθ −8 dρ det
(iD)
/
θ |σ± (x)| θ = e g e g 2 5
f (ρM) . (4.29)
0 ρ det(i∂)/
/ should have 1/length, the det
which has
Using dimensional analysis, since the eigenvalues of iD
one less eigenvalue gives
det
(iD)
/
= ρ × h(ρM), (4.30)
/
det(i∂)
where h is an function of a quantity. The expectation value carries dimension 1/(length)3 as
expected. We can go further by using the RG running of the coupling to figure out f and h, but
the more important result for us now is that eq. (4.22) is non-zero and the SSB of U(1)A do occur.
8
Here we use the fact that one instanton has one zero eigenfunction with γ 5 ψ0 = −ψ0 and none with γ 5 ψ0 = ψ0 .
The anti-instanton is reversed.

13
4.2.3 The Goldstone dipole
Now we know spontaneous symmetry breaking occurs. Are there Goldstone bosons? Let us look
for them in
θ |σ+ (x)σ− (0)| θ , (4.31)
which is the propagator of the fermion-condensate that breaks the symmetry. If there is a Gold-
stone pole couple in the gauge invariant quantity, it must be in this propagator9.
The calculation of this is similar to the one we just did. The only difference now is since there
are two σ’s, we can have either no instantons or one instanton and one anti-instanton that get rid
of the zero modes. The first case gives the usual one-loop perturbation theory expression, having
a two-quark cut, but no Goldstone pole. The second case just gives the product θ|σ+ |θθ|σ− |θ.
This also has no Goldstone pole. We can also check other gauge-invariant quantities, and the
Goldstone never shows up.
Following Kogut and Susskind’s idea, let us check if the gauge-variant current Jµ5 couples to
the Goldstone dipole. For example,
1
θ|Jµ5 (x)σ− (0)|θ = θ|jµ5 (x)σ− (0)|θ + θ|Gµ (x)σ− (0)|θ. (4.32)
32π 2
The first term on the right side is gauge-invariant and non-conserved by the explicit breaking,
which has nothing to do with the Goldstone poles. However, the second term does. The way to
see this is that the integration of the totally derivative term
 
d x∂µ θ|Gµ σ− (0)|θ = d4 xθ|(F, F̃ )σ− (0)|θ = 32π 2 θ|σ− (0)|θ = 0
4
(4.33)

(here we have used eqs. (3.14) and (4.29)) is non-vanishing. This means the Green’s function
connecting to x → ∞ is non-zero. Since only a massless field can propagate like this, there must
be a Goldstone dipole couple to this.
As a reminder, the nonvanishing integral requires a nontrivial winding number in θ|σ− (0)|θ,
i.e. the instanton background is vital to the argument. On the other hand, since there is no
instanton configuration for θ|Jµ5 Jλ5 |θ, we have ν = 0, and there is no poles in it. To summarize, in
the dilute-gas approximation, the SU(2) gauge theory with one massless fermion-doublet contains:

• SSB of U(1)A .

• no Goldstone poles in gauge-invariant Green’s functions.

• no Goldston dipoles in the propagator of gauge-variant conserved current.

• a Goldstone dipole in the Green’s function of a gauge-variant conserved current and a gauge-
invariant operator.

This gives the Goldstone dipole of Kogut and Susskind.

9
I think one way to see this is, the fermion-condensate is the ‘higgs’ in the dual picture that generates SSB. The
Green’s function should include the propagator of Goldstone-bosons if there exists any.

14
4.3 QCD (the real version)
Real QCD in the chiral U(2) × U(2) limit differs from the baby model in two respects. Firstly,
we have triplet quarks with gauge group SU(3) but not doublet quarks with SU(2). Second, we
have three massless quarks (u and d) rather than one.
For the symmetry part, there is a remarkable theorem due to Raoul Bott that states that any
continuous mapping of S 3 into G can be continuously deformed into a mapping into an SU(2)
subgroup of G. Thus, everything we used about the winding number is the same. The only thing
we have to change is the g −8 from the integration of the instanton phase space now becomes g −12 .
Changing the number of massless fermions effects more. As we show in the Appendix. C.2,
when having two massless fermions, the sum rule in eq. (4.19) changes to

n− − n+ = 2ν. (4.34)

This means ν = 1 gives two vanishing eigenvalues instead of one. We then have θ|σ− |θ vanishing
in the one instanton background. The way to show the SSB in this case is to consider the
expectation value of two σ’s10
1
ij kl ψ̄i (1 − γ 5 )ψk ψ̄j (1 − γ 5 )ψl = det[(ψ̄R ψL )f f
], (4.35)
2
where f f
are the flavor (u and d) indices that SU(2)L × SU(2)R acts on.
All in all, the result we find in the previous section still applies, U(1)A is broken spontaneously
in the instanton background, and the missing degree of freedom becomes the Goldstone dipole
couple to the gauge-variant but conserved axial current Jµ5 .

5 The baryon and the lepton number violation

One very important concept we developed when solving the U(1)-problem is that the instanton
changes the number of zero modes in eq. (4.34). This special sum rule is called the ’t Hooft
term [9]:11
nL − nR = nr 2C(r), (5.1)
r

This corresponds to the one-instant case, where nr is the number of fermions in the representation
r and C(r) is defined in eq. (B.4). When having an instanton effect, it creates the LH fermions,
while for an anti-instanton effect it creates the RH fermions.
Since the instanton effect also exists in the electroweak symmetry SU(2)×U(1), we can change
the number of fermions charged under SU(2). The fermions that have non-trivial C(r) in SU(2)
are the three (generation f ) lepton doublets (νLf , fL ) and three quark doublets (uf,r f,r
L , dL ). For each
flavor of the leptons, C(r) = 1/2 and nr = 1. For each flavor of the quarks, we have C(r) = 1/2
10
Instanton does not break chiral SU (2)L × SU (2)R , and the θ-vacuum is invariant under this. This means the
non-zero vacuum configuration has to be a chiral SU (2)L × SU (2)R singlet.
11
The origin of this comes from the non-conservation of the chiral current ∂µ jµ5 ∝ (F, F̃ ). When doing the volume
integral, the nonconservation of the LH and RH fermions is proportional to the winding number.

15
and nr = 3 for three colors. When having one anti-instanton, since only the LH fermions are
effected by the SU(2), the LH anti-fermions are generated and satisfy [9]
∆eL +∆νLe = ∆µL +∆νLµ = ∆τL +∆νLτ = −1, ∆uL +∆dL = ∆cL +∆sL = ∆tL +∆bL = −3.
(5.2)
From this, we can have processes like

u
u e+
d
ν̄µ
c
s
s τ+

t t b
Through the CKM rotation, this generates the process like
p + pc + nc → e+ + τ + + ν̄µ , (5.3)
where “c” means the u and d come from the CKM rotation. The cross section of this process is
2 2
suppressed by the instanton action e−16π /gEW like the one in the amplitude eq. (4.29). In this
case, it is
2 2 2 2 −2
e−16π /gEW = e−16π /e sin θW 10−262 . (5.4)
This gives a deuteron lifetime of the order of 10218 yr. Does this mean the instanton-type baryon
and the lepton number violation can never happen? In fact, all the discussion we just have are
2 2
in the zero temperature case. The free energy scale comparing to the factor e−16π /gEW is about
the EW symmetry breaking scale. In the early universe, before the temperature is lowered down
to the EWSB scale, we can get ‘over’ the free-energy barriers (between different θ-vacua) instead
of tunneling through them. This different vacua changing effect is called the ‘sphaleron effect’12 ,
which is important for baryogenesis models such as leptogenesis.

6 Conclusion
We have seen that the penetration between different vacua can be described using instantons.
The vacuum structure of a non-abelian gauge theory is related to different homotopy classes of
the mapping between the configuration space and the Euclidean spacetime. The different vacua
in this case are also connected by instantons.
Besides the solution of the U(1)-problem and a tool for the baryon and lepton violation de-
scribed in this report, the idea of instantons are heavily used in many different aspects of the field
theory, such as the strong CP problem, the NSVZ β function and the gaugino mass in SUSY,
the model of composite fields, and the tunneling mechanism in the inflation models. The physics
hidden in the vacuum structure keeps bringing us new surprise.
12
A good introduction is in [10].

16
 Figure 4

Acknowledgements
Thanks Csaba for giving me this instanton question. I finally make my mind to read Coleman’s
book. Thanks Yang and Yong-Hui for useful discussions, and Flip for introducing me useful review
articles. Finally, even though you do not know me at all, Mr. Coleman and Mr. ’t Hooft, you are
amazing.

A Euclidean functional integrals

In this appendix we calculate the correlation function 0|e−HT /
|0 which gives eq. (2.12) usgin
Euclidean functional integral. The action we have is
 T /2 
2 
1 dx
S= dt +V , (A.1)
−T /2 2 dt

with V given in Fig. ??. Write x(t) into the classical route x̄(t) (withδS/δx̄ = 0) plus the
eigenfunctions of the second variational derivative xn (t) (with δ 2 S/δx̄2 = n λn xn ), we have

x(t) = x̄(t) + cn xn (t), (A.2)

n

and  T /2  1
dtxn (t)xm (t) = δnm , [dx] = (2π
)− 2 dcn . (A.3)
−T /2 n
Use this to calculate the correlation function,
 −1 1
xf |e−HT /
|xi  = Ne−S(x̄)/
λn 2 = Ne−S(x̄)/
[det(−∂t2 + V

(x̄))]− 2 . (A.4)
n

Here we ignore the term with higher power of
when doing the perturbation. When the particle
stays at the vacuum in the classical limit, x̄ = 0. The correlation function becomes
1
xf |e−HT /
|xi  = N[det(−∂t2 + ω 2 )]− 2 . (A.5)

17
Here we define ω 2 ≡ V

(0). One can show that for large T ,
 ω  12
− 12
N[det(−∂t2


+ V (x̄))] = e−ωT /2 , (A.6)
π

this gives the vacuum correlation function eq. (2.12).

B Conventions for the gauge theory

In this appendix we establish notational conventions for the gauge theory.

B.1 Lie algebra

For two matrices T a amd T b in a representation of a Lie group, we can always choose them such
that T r(T a T b ) ∝ δ ab . We define the Cartan inner product as

(T a , T b ) = δ ab . (B.1)

For the SU(2) adjoint representations T a = −iσ a /2, we have

(T a , T b ) = −2T r(T a T b ). (B.2)

For the SU(2) fundamental representations φT1 = √i (1, 0) and φT2 = √i (0, 1), we have
2 2

1
(φa , φb ) = − T r(φa φb ). (B.3)
2
For two SU(n) representations ta and tb ,

T r(T a T b ) = −C(r)δ ab , (B.4)

where r denotes the representations. For the fundamental rep, C(n) = 1/2. For the adjoint rep,
C(adj) = n.

B.2 Gauge fields

We define the gauge fields Aµ as
Aµ = gAaµ T a , (B.5)
where g is the gauge coupling. The field strength tensor is defined by

Fµν = ∂µ Aν − ∂ν Aµ + [Aµ , Aν ]. (B.6)

Pure gauge field theory is defined by the Euclidean action

 
1 4 1
S = 2 d x(Fµν , Fµν ) ≡ 2 (F, F ). (B.7)
4g 4g

18
B.3 Gauge transform
A gauge transformation is a function, h(x), from Euclidean space into the gauge group, G

h(x) = exp λa (x)T a , (B.8)

where λs are arbitrary functions. Under such a transform

Aµ → hAµ h−1 + h∂µ h−1 , Fµν → gFµν g −1. (B.9)

If Fµν vanishes,
Aµ = h∂µ h−1 . (B.10)

C Some tools for the U (1) problem

Here we derive some tools used in the U(1) problem.

C.1 Chiral Ward identity

We want to study a theory of fermions interacting with c-number gauge fields,

S = −i d4 xψ̄(D / − M)ψ. (C.1)

For the Green’s function with m local multilinear functions of ψ’s, the Green’s function are defined
by 
 (1) A [dψ][dψ̄]e−S φ(1) (x1 )...φm (xm )
φ (x1 )...φm (xm ) =  . (C.2)
−S
[dψ][dψ̄]e

The superscript A is to remind us that we are working in an external gauge field. When having
a chiral transform with exp(−iγ5 δα)

δψ = −iγ5 ψδα, δ ψ̄ = −iψ̄γ5 δα, (C.3)

we have the Ward identity in Schwinger-Dyson equation as

 A  A
∂ µ jµ5 (y)φ(1) (x1 )...φm (xm ) + ψ̄Mγ5 ψ(y)φ(1) (x1 )...φm (xm )
 A
+δ (4) (y − x1 ) ∂φ(1) (x1 )/∂α...φm (xm )
 A
+... + δ (4) (y − xm ) φ(1) (x1 )...∂φm (xm )/∂α
iC  A
= − 2 (F (y), F̃ (y)) φ(1) (x1 )...φm (xm ) . (C.4)
8π
Where the last term comes from the anomaly, C is defined by the

T r(T aT b ) = −Cδ ab . (C.5)

19
For N SU(n) fundamental fields, C = N/2.
Integrating out the 4-space in eq. (C.4), the first term only has boundary contribution and
vanishes since there is no massless fields can give a non-vanishing surface term. On the right we
can use 
d4 y(F, F̃ ) = 32π 2 ν. (C.6)
We then have
 A
4 (1) m ∂  (1) A  A
2 d y ψ̄Mγ5 ψ(y)φ (x1 )...φ (xm ) + φ (x1 )...φm (xm ) = −4iCν φ(1) (x1 )...φm (xm ) .
∂α
(C.7)
We will use this later.

C.2 The sum rule

Now let us prove eq. (4.19). Using eq. (C.7), when there is no φ’s, we have
 
 A 2 [dψ][dψ̄]e −S
d4 y ψ̄Mγ5 ψ
− 2Niν = 2 d4 y ψ̄Mγ5 ψ(y) =  . (C.8)
−S
[dψ][dψ̄]e

/ − M), and using the orthogonal relation for the

Writing the fermion into eigenfunctions of i(D
zero and non-zero modes
 
4 †
d yψr γ5 ψr = 0, with λr = 0, d4 yψs† γ5 ψs = χs , with λs = 0, (C.9)

we can separate the numerator into the zero-mode part with ψs and the non-zero-mode part with
ψr   

−Sr=s ψ̄s (−iMs )ψs
2 [dψr ][dψ̄r ]e [dψs ][dψ̄s ]e s d4 y ψ̄s Ms γ5 ψs . (C.10)
s
The pseudoscalar
 integral only exists for zero-modes. The first integral gives the usual determinant
term r=s (λr − iM). Using eq. (4.18) and ψ̄s (−iMs )ψs = ψ̄s (−iMs )γ 5 γ 5 ψs = chis ψ̄s (−iMs )γ 5 ψs ,
we can write the last two integrals into
 
  
χs ψ̄s (−iMs γ 5 )ψs 4 iMs ∂ 5
[dψs ][dψ̄s ]e s d y ψ̄s Ms γ5 ψs = [dψs ][dψ̄s ]e s χs ψ̄s (−iMs γ )ψs ,
s s
χs ∂Ms
  
iMs ∂  iMs ∂  
= [dψs ][dψ̄s ]e s ψ̄s (−iMs )ψs = (−iMs ) = i χ−1
s (−iMs ).
s
χs ∂Ms s
χs ∂Ms s s
(C.11)
Combine the denominator term together, the green function becomes
 
 
2i χ−1
s (−iM s ) (λr − iM)
s s r=s
− 2Niν =   = 2i χ−1
s = 2i(n+ − n− ). (C.12)
(−iMs ) (λr − iM) s
s r=s

20
For one quak only, we have N = 1. This gives eq. (4.19). In the real QCD case, N = 2. This
gives eq. (4.34).

C.3 U (1)A and the θ-vacua transform

In this section we want to proof eq. (4.21). Define the denominator-free Green’s function

φ (x1 )... ≡ [dψ][dψ̄]e−S φ(1) ....
(1) A
(C.13)

We can get the chiral Ward identity in the same way as deriving eq. (C.7). The only difference
now is that M = 0 for massless quarks, and C = 12 . This gives
∂
+ 2iν φ1 (x1 )...A = 0. (C.14)
∂α
Fourier transforms the n-vacua into θ-vacua, for a given |θ, the Green’s function of the baby
QCD can be written as

[dA]e−Sg eiνθ φ(1) (x1 )...A
θ|φ1 (x1 )...|θ =  , (C.15)
−Sg iνθ A
[dA]e e 1

where Sg is the gauge-field part of the action. By eq. (C.14), we have eq. (4.21)
∂
+ 2iν θ|φ1 (x1 )...|θA = 0. (C.16)
∂α

References
[1] C. N. Yang and R. L. Mills, Phys. Rev. 96, 191 (1954).
[2] Sidney Coleman, “Aspects of symmetry”, Cambride University press.
[3] Lewis H.Ryder, “Quantum field theory”, Cambride University press.
[4] John Terning, “Modern supersymmetry”, Oxford science publications.
[5] G. ’t Hooft, Phys. Rept. 142, 357 (1986).
[6] S. Weinberg, Phys. Rev. D 11, 3583 (1975).
[7] Ta-Pei Cheng and Ling-Fong Li, “Gauge theory of elementary particle physics,” Clarendon
Press. Oxford.
[8] J. B. Kogut and L. Susskind, Phys. Rev. D 11, 3594 (1975).
[9] G. ’t Hooft, Phys. Rev. Lett. 37, 8 (1976); Phys. Rev. D 14, 3432 (1976) D 18, 2199 (1978).
[10] Edward W. Kolb and Michael S. Turner, “The early universe”, Westview press.

21