MATHEMATICAL IDEAS AND NOTIONS OF QUANTUM FIELD THEORY

3. Feynman calculus
3.1. Wicks theorem. Let V be a real vector space of dimension d with volume element dx. Let S(x)
be a smooth function on a box B V which attains a minimum at x = c Interior(B), and g be any
smooth function on B. In the last section we have shown that the function

d/2 S(c)/
e
g(x)eS(x)/ dx
I(
) =

B

admits an asymptotic power series expansion in
:

I(
) = A0 + A1
+ + Am
m +

(8)

Our main question now will be: how to compute the coecients Ai ?
It turns out that although the problem of computing I(
) is transcendental, the problem of computing
the coecients Ai is in fact purely algebraic, and involves only dierentiation of the functions S and g
at the point c. Indeed, recalling the proof of equation 8 (which we gave in the 1-dimensional case), we
see that the calculation of Ai reduces to calculation of integrals of the form

P (x)eB(x,x)/2 dx,
V

where P is a polynomial and B is a positive denite bilinear form (in fact, B(v, u) = (v u S)(c)). But
such integrals can be exactly evaluated. Namely, it is sucient to consider the case when P is a product
of linear functions, in which case the answer is given by the following elementary formula, known to
physicists as Wicks theorem.
For a positive integer k, consider the set {1, . . . , 2k}. By a pairing on this set we will mean its
partition into k disjoint two-element subsets (pairs). A pairing can be visualized by drawing 2k points
and connecting two points with an edge if they belong to the same pair (see Fig. 1). This will give k
edges, which are not connected to each other.

Figure 1. Pairings of the set {1, 2, 3, 4}

Let us denote the set of pairings on {1, . . . , 2k} by k . It is clear that |k | = 2(2n)!
n n! . For any k ,
we can think of as a permutation of {1, . . . , 2k}, such that 2 = 1 and has no xed points. Namely,
maps any element i to the second element (i) of the pair containing i.
Theorem 3.1. Let B 1 denote the inverse form on V , and let 1 , . . . , m V . Then, if m is even,
we have

(2)d/2
1 (x) . . . m (x)eB(x,x)/2 dx =
B 1 (i , (i) )
det B
V
i{1,...,m}/
m/2

If m is odd, the integral is zero.

MATHEMATICAL IDEAS AND NOTIONS OF QUANTUM FIELD THEORY

Proof. If m is odd, the statement is obvious, because the integrand is an odd function. So consider the
even case. Since both sides of the equation are symmetric polylinear forms in 1 , . . . , m , it suces to
prove the result when 1 = = m = . Further, it is clear that the formula to be proved is stable
under linear changes of variable (check it!), so we can choose a coordinate system in such a way that
B(x, x) = x21 + + xd2 , and (x) = x1 . Therefore, it is sucient to assume that d = 1, and (x) = x.
In this case, the theorem says that

2
(2k)!
x2k ex /2 dx = (2)1/2 k ,
2 k!

which is easily obtained from the denition of the Gamma function by change of variable y = x2 /2.
Examples.

(2)d/2 1
B (1 , 2 ).
1 (x)2 (x)eB(x,x)/2 dx =
det B

1 (x)2 (x)3 (x)4 (x)eB(x,x)/2 dx =

V

d/2

(2)

(B 1 (1 , 2 )B 1 (3 , 4 ) + B 1 (1 , 3 )B 1 (2 , 4 ) + B 1 (1 , 4 )B 1 (2 , 3 )).

det B
Wicks theorem shows that the problem of computing Ai is of combinatorial nature. In fact, the
central role in this computation is played by certain nite graphs, which are called Feynman diagrams.
They are the main subject of the remainder of this section.
3.2. Feynmans diagrams and Feynmans theorem. We come back to the problem of computing
the coecients Ai . Since each particular Ai depends only on a nite number of derivatives of g at
c, it suces to assume that g is a polynomial, or, more specically, a product of linear functions:
g = 1 . . . N , i V . Thus, it suces to be able to compute the series expansion of the integral

< 1 . . . N >:=
d/2 eS(c)/

1 (x) . . . N (x)eS(x)/ dx
B

Without loss of generality we may assume that c = 0, and S(c) = 0. Then the (asymptotic) Taylor

expansion of S at c is S(x) = B(x,x)

+ r3 Br (x,x,...,x)
, where Br = dr f (0). Therefore, regarding
2
r!

the left hand side as a power series in
, and making a change of variable x x/
(like in the last
section), we get

B(x,x) P
r/21 Br (x,...,x)
r!
1 (x) . . . N (x)e 2 r3
dx.
< 1 . . . N >=
N/2
V

(This is an identity of expansions in
, as we ignored the rapidly decaying error which comes from
replacing the box by the whole space).
The theorem below, due to Feynman, gives the value of this integral in terms of Feynman diagrams.
This theorem is easy to prove but is central in quantum eld theory, and will be one of the main
theorems of this course. Before formulating this theorem, let us introduce some notation.
Let G3 (N ) be the set of isomorphism classes of graphs with N 1-valent external vertices, labeled
by 1, . . . , N , and a nite number of unlabeled internal vertices, of any valency 3. Note that here
and below graphs are allowed to have multiple edges between two vertices, and loops from a vertex to
itself (see Fig. 2).
For each graph G3 (N ), we dene the Feynman amplitude of as follows.
1. Put the covector j at the j-th external vertex.
2. Put the tensor Bm at each m-valent internal vertex.
3. Take the contraction of the tensors along edges of , using the bilinear form B 1 . This will
produce a number, called the amplitude of and denoted F (1 , . . . , N ).
Remark. If is not connected, then F is dened to be the product of numbers obtained from the
connected components. Also, the amplitude of the empty diagram is dened to be 1.

MATHEMATICAL IDEAS AND NOTIONS OF QUANTUM FIELD THEORY

Theorem 3.2. (Feynman) One has

(2)d/2
< 1 . . . N >=
det B

G3 (N )

b()
F (1 , . . . , N ),
|Aut()|

where b() is is the number of edges minus the number of internal vertices of .
(here by an automorphism of we mean a permutation of vertices AND edges which preserves
the graph structure, see Fig. 3; thus there can exist nontrivial automorphisms which act trivially on
vertices).
Remark 1. Note that this sum is innite, but
-adically convergent.
Remark 2. We note that Theorem 3.2 is a generalization of Wicks theorem: the latter is obtained
if S(x) = B(x, x)/2. Indeed, in this case graphs which give nonzero amplitudes do not have internal
vertices, and thus reduce to graphs corresponding to pairings .
Let us now make some comments about the terminology. In quantum eld theory, the function
< 1 . . . N > is called the N-point correlation function, and graphs are called Feynman diagrams.
The form B 1 which is put on the edges is called the propagator. The cubic and higher terms Bm /m! in
the expansion of the function S are called interaction terms, since such terms (in the action functional)
describe interaction between particles. The situation in which S is quadratic (i.e., there is no interaction)
is called a free theory; i.e. for the free theory the correlation functions are determined by Wicks formula.
Remark 3. Sometimes it is convenient to consider normalized correlation functions < 1 . . . N >norm
= < 1 . . . N > / < > (where < > denotes the integral without insertions). Feynmans theorem

0
N =0

2
1

N =0

3
1

N =1

N =2

4
1

2
N =2

Figure 2. Elements of G3 (N )

Figure 3. An automorphism of a graph

10

MATHEMATICAL IDEAS AND NOTIONS OF QUANTUM FIELD THEORY

implies that they are given by the formula

< 1 . . . N >norm =

G
(N )
3

b()
F (1 , . . . , N ),
|Aut()|

where G3 (N ) is the set of all graphs in G3 (N ) which have no components without external vertices.
3.3. Another version of Feynmans theorem. Before proving Theorem 3.2, we would like to slightly
modify and generalize it. Namely, in quantum eld theory it is often useful to consider an interacting
theory as a deformation of a free theory. This means that S(x) = B(x, x)/2 + S(x), where S(x) is the

perturbation S(x) = m0 gm Bm (x, x, . . . , x)/m!, where gm are (formal) parameters. Consider the
partition function

Z =
d/2

eS(x)/ dx

only positive powers of gi but arbitrary powers of
; however,

as a series in gi and
(this series involves

the coecient of a given monomial i gini is a nite sum, and hence contains only nitely many powers
of
).
Let n = (n0 , n1 , . . .) be a sequence of nonnegative integers, almost all zero. Let G(n) denote the set
of isomorphism classes of graphs with n0 0-valent vertices, n1 1-valent vertices, n2 2-valent vertices, etc.
(thus, now we are considering graphs without external vertices).
Theorem 3.3. One has

(2)d/2 ni

b()
F ,
Z=
gi
|Aut()|
det B n
i
G(n)

where F is the amplitude dened as before, and b() is the number of edges minus the number of
vertices of .
We will prove Theorem 3.3 in the next subsection. Meanwhile, let us show that Theorem 3.2 is
in fact a special case of Theorem 3.3. Indeed, because of symmetry of the correlation functions with
respect to 1 , . . . , N , it is sucient to consider the case 1 = = N = . In this case, denote the
correlation function < N > (expectation value of N ). Clearly, to compute < N > for all N , it is

N
< N > tN ! . But this expectation value is
sucient to compute the generating function < et >:=
exactly the one given by Theorem 3.3 for gi = 1, i 3, g0 = g2 = 0, g1 =
t, B1 = , B0 = 0, B2 = 0.
Thus, Theorem 3.3 implies Theorem 3.2 (note that the factor N ! in the denominator is accounted for
by the fact that in Theorem 3.3 we consider unlabeled, rather than labeled, 1-valent vertices convince
yourself of this!).
3.4. Proof
of Feynmans theorem. Now we will prove Theorem 3.3. Let us make a change of variable
y = x/
. Expanding the exponential in a Taylor series, we obtain

Z=
Zn ,
n

where
Zn =

eB(y,y)/2

gini
(
i/21 Bi (y, y, . . . , y))ni dy
(i!)ni ni !

Writing Bi as a sum of products of linear functions, and using Wicks theorem, we nd that the value
of the integral for each n can be expressed combinatorially as follows.
1. Attach to each factor Bi a ower a vertex with i outgoing edges (see Fig. 4).
2. Consider the set T of ends of these outgoing edges (see Fig. 5), and for any pairing of this set,
consider the corresponding contraction of tensors Bi using the form B 1 . This will produce a number
F ().
3. The integral Zn is given by

(2)d/2 gini
i
(9)
Zn =

ni ( 2 1)
F
n
i
det B i (i!) ni !

MATHEMATICAL IDEAS AND NOTIONS OF QUANTUM FIELD THEORY

11

0-valent ower

1-valent ower

3-valent ower

Figure 4

Now, recall that pairings on a set can be visualized by drawing its elements as points and connecting
them with edges. If we do this with the set T , all ends of outgoing edges will become connected with
each other in some way, i.e. we will obtain a certain (unoriented) graph = (see Fig. 6). Moreover,
it is easy to see that the number F () is nothing but the amplitude F .
It is clear that any graph with ni i-valent vertices for each i can be obtained in this way. However,
the same graph can be obtained many times, so if we want to collect the terms in the sum over , and
turn it into a sum over , we must nd the number of which yield a given .
For this purpose, we will consider the group G of permutations of T , which preserves owers
(i.e. endpoints of any two edges outgoing from the same ower end up again in the same ower). This
group involves
1) permutations of owers with a given valency;
2) permutation of the i edges inside each i-valent ower.

More
precisely, the group G is the semidirect product ( i Sni )  ( i Sini ). Note that |G| =

ni
i (i!) ni !, which is the product of the numbers in the denominator of the formula (9).

Figure 5. The set T for n = (0, 0, 0, 2, 1, 0, 0, . . .) (the set of white circles)

12

MATHEMATICAL IDEAS AND NOTIONS OF QUANTUM FIELD THEORY

r:

Figure 6. A pairing of T and the corresponding graph .

The group G acts on the set of all pairings of T . Moreover, it acts transitively on the set P of
pairings of T which yield a given graph . Moreover, it is easy to see that the stabilizer of a given
pairing is Aut(). Thus, the number of pairings giving is

ni
i (i!) ni !
.
|Aut()|
Hence,

(i!)ni ni !
i
F .
|Aut()|

Finally, note that the exponent of
in equation (9) is i (i/2 1), which is the number of edges of
minus the number of vertices, i.e. b(). Substituting this into (9), we get the result.
Example. Let d = 1, V = R, gi = g, Bi = z i for all i (where z is a formal variable),
= 1. Then
we nd the asymptotic expansion

zx
x2
z 2k
1

,
e 2 +ge =
gn
|Aut()|
2

F =

n0

G(n,k)

where G(n, k) is the set of isomorphism classes of graphs with n vertices and k edges. Expanding the
left hand side, we get
2 2

ez n /2
z 2k
=
,
|Aut()|
n!
k G(n,k)

and hence

G(n,k)

n2k
1
= k
,
|Aut()|
2 k!n!

Exercise. Check this by direct combinatorics.

3.5. Sum over connected diagrams. Now we will show that the logarithm of the partition function
Z is also given by summation over diagrams, but with only connected diagrams taken into account.
This signicantly simplies the analysis of Z in the rst few orders of perturbation theory, since the
number of connected diagrams with a given number of vertices and edges is signicantly smaller than
the number of all diagrams.

MATHEMATICAL IDEAS AND NOTIONS OF QUANTUM FIELD THEORY

Theorem 3.4. Let Z0 =

(2)d/2
det(B) .

13

Then one has

ln(Z/Z0 ) =

gini

Gc (n)

b()
F ,
|Aut()|

where Gc (n) is the set of connected graphs in G(n).

Remark. We agree that the empty graph is not connected.
Proof. For any graphs 1 , 2 , let 1 2 stand for the disjoint union of 1 and 2 , and for any graph let
n denote the disjoint union of n copies of . Then every graph can be uniquely written as k11 . . . kl l ,
where j are connected non-isomorphic graphs. Moreover, it is clear that F1 2 = F1 F2 , b(1 2 ) =

b(1 ) + b(2 ), and |Aut(k11 . . . kl l )| = j (|(Aut(j )|kj kj !). Thus, exponentiating the equation of
Theorem 3.4, and using the above facts together with the Taylor series for the function ex , we arrive
at Theorem 3.3. As the Theorem 3.3 has been proved, so is Theorem 3.4

3.6. Loop expansion. It is very important to note that since summation in Theorem 3.4 is over
connected Feynman diagrams, the number b() is the number of loops in minus 1. In particular, the
lowest coecient in
is that of
1 , and it is the sum over all trees; the next coecient is of
0 , and
it is the sum over all diagrams with one loop (cycle); the next coecient to
is the sum over two-loop
diagrams, and so on. Therefore, physicists refer to the expansion of Theorem 3.4 as loop expansion.
Let us study the two most singular terms in this expansion (with respect to
), i.e. the terms given
by the sum over trees and 1-loop graphs.
Let x0 be the critical point of the function S. It exists and is unique, since gi are assumed to be
formal parameters. Let G(j) (n) denote the set of classes of graphs in Gc (n) with j loops. Let

F
.
(ln(Z/Z0 ))j =
gini
|Aut()|
(j)
n
i
G

(n)

Theorem 3.5.
(ln(Z/Z0 ))0 = S(x0 ),

(10)
and

(ln(Z/Z0 ))1 =

(11)

det(B)
1
.
ln
2 det S  (x0 )

Proof. First note that the statement is purely combinatorial. This means, in particular, that it is
sucient to check that the statement yields the correct asymptotic expansion of the right hand sides of
equations (10),(11). in the case when S is a polynomial with real coecients of the form B(x, x)/2 +
S(x)/
N
d/2
e
, where B is a suciently small box
i=0 gi Bi (x, x, . . . , x)/i!. To do this, let Z =

B
around 0. For suciently small gi , the function S has a unique global maximum point x0 in B, which
is nondegenerate. Thus, by the steepest descent formula, we have

where I(
) =

Z/Z0 = eS(x0 )/ I(
),

det(B)
det S  (x0 ) (1

+ a1
+ a2
2 + ) (asymptotically). Thus,

ln(Z/Z0 ) = S(x0 )
1 +

This implies the result.

det(B)
1
+ O(
).
ln
2 det S  (x0 )

Physicists call the expression (ln(Z/Z0 ))0 the classical (or tree) approximation to the quantum mechanical quantity
ln(Z/Z0 ), and the sum (ln(Z/Z0 ))0 +
(ln(Z/Z0 ))1 the one loop approximation.
Similarly one denes higher loop approximations. Note that the classical approximation is obtained by
nding the critical point and value of the classical action S(x), which in the mechanics and eld theory
situation corresponds to solving the classical equations of motion.

14

MATHEMATICAL IDEAS AND NOTIONS OF QUANTUM FIELD THEORY

3.7. Nonlinear equations and trees. As we have noted, Theorem 3.5 does not involve integrals
and is purely combinatorial. Therefore, there should exist a purely combinatorial proof of this theorem.
Such a proof indeed exists. Here we will give a combinatorial proof of the rst statement of the Theorem
(formula (10)).
Consider the equation S  (x) = 0, dening the critical point x0 . This equation can be written as
x = (x), where

1 Bi (x, x, . . . , x, ?)/(i 1)!,

(x) :=
gi B
i1

1 : V V is the operator corresponding to the form B 1 .

where B
In the sense of power series norm, is a contracting mapping. Thus, x0 = limN N (x), for any
initial vector x V . In other words, we will obtain x0 if we keep substituting the series (x) into itself.
This leads to summation over trees (explain why!). More precisely, we get the following expression for
x0 :

F
,
( gini )
x0 =
|Aut()|
(0)
n
G

(n,1)

where G(0) (n, 1) is the set

of trees with one external vertex and ni internal vertices of degree i. Now,
since
S(x)
=
B(x,
x)/2
+
gi Bi (x, x, . . . , x)/i!, the expression S(x0 ) equals the sum of expressions
ni
F
( gi ) |Aut()| over all trees (without external vertices). Indeed, the term B(x0 , x0 )/2 corresponds to
gluing two trees with external vertices (identifying the two external vertices, so that they disappear);
so it corresponds to summing over trees with a marked edge, i.e. counting each tree as many times as it
has edges. On the other hand, the term gi Bi (x0 , . . . , x0 )/i! corresponds to gluing
i trees with external
vertices together at the external vertices (making a tree with a marked vertex). So gi Bi (x0 , . . . , x0 )/i!
corresponds to summing over trees with a marked vertex, i.e. counting each trees as many times as it
has vertices. But the number of vertices of a tree exceeds the number of edges by 1. Thus, the dierence
S(x0 ) of the above two contributions corresponds to summing over trees, counting each exactly once.
This implies formula (10).
3.8. Counting trees and Cayleys theorem. In this section we will apply Theorem 3.5 to tree
counting problems, in particular will prove a classical theorem due to Cayley that the number of
labeled trees with n vertices is nn2 .
We consider essentially the same example as we considered above: d = 1, Bi = 1, gi = g. Thus, we
2
have S(x) = x2 gex . By Theorem 3.5, we have

1
= S(x0 ),
gn
|Aut()|
n0

T (n)

where T (n) is the set of isomorphism classes of trees with n vertices, and x0 is the root of the equation
S  (x) = 0, i.e. x = gex .
In other words, let f (z) be the function inverse to xex near x = 0. Then we have x0 = f (g). Thus,
let us nd the Taylor expansion of f . This is given by the following classical result.
Proposition 3.6. One has
f (g) =
Proof. Let f (g) =

nn2
gn.
(n 1)!

n1

n1

an g . Then

f (g)
1
x
1
an =
dg =
d(xex ) =
2i
g n+1
2i
(xex )n+1

1
1x
nn1
nn2
nn2

.
=
enx n dx =
2i
x
(n 1)! (n 2)!
(n 1)!

MATHEMATICAL IDEAS AND NOTIONS OF QUANTUM FIELD THEORY

15

Now we nd
S(x0 ) = f (g)2 /2 + gef (g) .
Thus
(d/dg)S(x0 ) = f (g)f  (g) + gef (g) f  (g) + ef (g) = ef (g) =
This means that

S(x0 ) =
0

This shows that

T (n)

f (g)
.
g

nn2
f (a)
da =
gn.
a
n!
n1

nn2
1
=
|Aut()|
n!

But each isomorphism class of unlabeled trees with n vertices has

Thus the latter formula implies

n!
|Aut()|

nonisomorphic labelings.

Corollary 3.7. (A. Cayley) The number of labeled trees with n vertices is nn2 .
3.9. Counting trees with conditions. In a similar way we can count labeled trees with conditions
on vertices. For example, let us compute the number of labeled trivalent trees with m vertices (i.e. trees
that have only 1-valent and 3-valent vertices). Clearly, m = 2k, otherwise there is no such trees. The
2
relevant action functional is S(x) = x2 g(x + x3 /6). Then the critical point x0 is obtained from the

1 12g2
equation g(x2 /2 + 1) x = 0, which yields x0 =
. Thus, the tree sum (ln(Z/Z0 ))0 equals
g
(ln(Z/Z0 ))0 = S(x0 ) =

1 (1 2g 2 )3/2
1.
3g 2

Expanding this in a Taylor series, we nd

(ln(Z/Z0 ))0 =

1 3 (2n 1) 2n+2
g
(n + 2)!
n=0

Hence, we get
(2k)!
.
Corollary 3.8. The number of trivalent labeled trees with m = 2k vertices is (2k 3)!! (k+1)!

3.10. Counting oriented trees. Feynman calculus can be used to count not only non-oriented, but
also oriented graphs. For example, suppose we want to count labeled oriented trees, whose vertices are
either sources or sinks (see Fig. 7). In this case, it is easy to see (check it!) that the relevant integration
problem is in two dimensions, with the action S = xy bex aey . So the critical point is found from
the equations
xey = a, yex = b.
-

5
s

Figure 7. A labeled oriented tree with 3 sources and 3 sinks.

16

MATHEMATICAL IDEAS AND NOTIONS OF QUANTUM FIELD THEORY

Like before, look for a solution (x, y) = (x0 , y0 ) in the form

dpq ap bq .
cpq ap bq , y = b +
x=a+
p1,q1

p1,q1

A calculation with residues similar to the one we did for unoriented trees yields

qx+py
1
1
q p1 pq1
e
x
cpq =
da

db
=
(1

xy)dx

dy
=
.
(2i)2
ap+1 bq+1
(2i)2
xp y q+1
(p 1)!q!
p1 q1

p
. Now, similarly to the unoriented case, we nd that aa S(x, y) = x,
Similarly, dpq = qp!(q1)!
bb S(x, y) = y, so
a
pq1 q p1
x
du = a + b +
ap b q
S(x, y) = b +
p!q!
0 u
p,q1

This implies that the number of labeled trees with p sources and q sinks ispq1 q p1 (p+q)!
p!q! . In particular,
if we specify which vertices are sources and which are sinks, the number of trees is pq1 q p1 .
3.11. 1-particle irreducible diagrams and the eective action. Let Z = ZS be the partition
function corresponding to the action S. In the previous subsections we have seen that the classical
(or tree) part (ln(ZS /Z0 ))0 of the quantity ln(ZS /Z0 ) is quite elementary to compute it is just
minus the critical value of the action S(x). Thus, if we could nd a new eective action Se (a
deformation of S) such that
(ln(ZSeff /Z0 ))0 = ln(ZS /Z0 )
(i.e. the classical answer for the eective action is the quantum answer for the original one), then we
can regard the quantum theory for the action S as solved. In other words, the problem of solving
the quantum theory attached to S (i.e. nding the corresponding integrals) essentially reduces to the
problem of computing the eective action Se .
We will now give a recipe of computing the eective action in terms of amplitudes of Feynman
diagrams.
Denition 3.9. An edge e of a connected graph is said to be a bridge, if the graph \ e is
disconnected. A connected graph without bridges is called 1-particle irreducible (1PI).
Remark. This is the physical terminology. The mathematical terminology is 2-connected.
To compute the eective action, we will need to consider graphs with external edges (but having at
least one internal vertex). Such a graph (with N external edges) will be called 1-particle irreducible if
so is the corresponding amputated graph (i.e. the graph obtained from by removal of the external
edges). In particular, a graph with one internal vertex is always 1-particle irreducible (see Fig. 8), while
a single edge graph without internal vertices is dened not to be 1-particle irreducible.
Let us denote by G1irr (n, N ) the set of isomorphism classes of 1-particle irreducible graphs which
N external edges and ni i-valent internal vertices for each i (where isomorphisms are not allowed to
move external edges).
Theorem 3.10. The eective action Se is given by the formula
B(x, x) Bi
Se (x) =

,
2
i!
i0

where
BN (x, x, . . . , x) =

( gini )
n

where x V

G1irr (n,N )

b()+1
F (x , x , . . . , x ),
|Aut()|

is dened by x (y) := B(x, y)

Thus, Se = S +
S1 +
2 S2 + .. The expressions
j Sj are called the j-loop corrections to the eective
action.
This theorem allows physicists to worry only about 1-particle irreducible diagrams, and is the reason
why you will rarely see other diagrams in a QFT textbook. As before, it is very useful in doing low

MATHEMATICAL IDEAS AND NOTIONS OF QUANTUM FIELD THEORY

17

not a bridge

6

a bridge

1PI graph with

two external edges

non-1PI graph with

four external edges

Figure 8
order computations, since the number of 1-particle irreducible diagrams with a given number of loops
is much smaller than the number of connected diagrams with the same number of loops.
Proof. The proof is based on the following lemma from graph theory.
Lemma 3.11. Any connected graph can be uniquely represented as a tree, whose vertices are 1-particle
irreducible subgraphs (with external edges), and edges are the bridges of .
The lemma is obvious. Namely, let us remove all bridges from . Then will turn into a union of
1-particle irreducible graphs, which should be taken to be the vertices of the said tree.
The tree corresponding to the graph is called the skeleton of (see Fig. 9).

Graph:

Skeleton:

Figure 9. The skeleton of a graph.

It is easy to see that the lemma implies the theorem. Indeed, it implies that the sum over all
connected graphs occurring in the expression of ln(ZS /Z0 ) can be written as a sum over skeleton trees,
so that the contribution from each tree is (proportional to) the contraction of tensors Bi put in its

18

MATHEMATICAL IDEAS AND NOTIONS OF QUANTUM FIELD THEORY

vertices, and Bi is the (weighted) sum of amplitudes of all 1-particle irreducible graphs with i external
edges.

3.12. 1-particle irreducible graphs and the Legendre transform. Recall the notion of Legendre
transform.
Let f be a smooth function on a vector space Y , such that the map Y Y given by x df (x) is
a dieomorphism. Then one can dene the Legendre transform of f as follows. For p Y , let x0 (p)
be the critical point of the function (p, x) f (x) (i.e. the unique solution of the equation df (x) = p).
Then the Legendre transform of f is the function on Y dened by
L(f )(p) = (p, x0 ) f (x0 ).
It is easy to see that the dierential of L(f ) is also a dieomorphism Y Y (in fact, inverse to df (x)),
and that L2 (f ) = f .
Example. Let f (x) = ax2 /2. Then px f = px x2 /2 has a critical point at p = x/a, and the
critical value is p2 /2a. L(ax2 /2) = p2 /2a. Similarly, if f (x) = B(x, x)/2 where B is a nondegenerate
symmetric form, then L(f )(p) = B 1 (p, p)/2.
Now let us consider Theorem 3.10 in the situation of Theorem 3.2. Thus, S(x) = B(x, x)/2 + O(x3 ),
and we look at

JxS(x)
d/2
Z(J ) =

e dx
V

By Theorem 3.10, one has

ln(Z(J )/Z0 ) = Se (x0 , J ),
where the eective action Se (x, J ) given by summation over 1-particle irreducible graphs.
Now, we must have Se (x, J ) = J x + Se (x), since the only 1PI graph which contains 1-valent
internal vertices (corresponding to J ) is the graph with one edge, connecting an internal vertex with
an external one (so it yields the term J x, and other graphs contain no J -vertices). This shows that
ln(Z(J )/Z0 ) is the critical value of J x Se (x). Thus we have proved the following.
Proposition 3.12. We have
Se (x) = L(ln(Z(J )/Z0 )), ln(Z(J )/Z0 ) = L(Se (x)).
Physicists formulate this result as follows: the eective action is the Legendre transform of the
logarithm of the generating function for quantum correlators (and vice versa).