This action might not be possible to undo. Are you sure you want to continue?
Notes on (Semi)Advanced Quantum Mechanics:
The Path Integral Approach to Quantum Mechanics
Matthias Blau
Institut de Physique, Universit´e de Neuchˆ atel, Rue Breguet 1
CH2000 Neuchˆ atel, Switzerland
Contents
0 Preface 3
1 The Evolution Kernel 5
1.1 Review: The timeevolution Operator . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2 Deﬁnition of the Propagator / Kernel . . . . . . . . . . . . . . . . . . . . . . . . 5
1.3 Basic Properties of the Kernel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.4 Example: The Free Particle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2 Towards the Path Integral Representation of the Kernel 9
2.1 From the Kernel to the ShortTime Kernel and Back . . . . . . . . . . . . . . . . 9
2.2 The Trotter Product Formula and the Dirac ShortTime Kernel . . . . . . . . . . 9
2.3 The Path Integral Representation of the Kernel . . . . . . . . . . . . . . . . . . . 11
2.4 The Path Integral Representation of the Partition Function . . . . . . . . . . . . 14
2.5 Back from the Path Integral to the Schr¨ odinger Equation . . . . . . . . . . . . . 15
2.6 Convolution, Correlation Functions and TimeOrdered Products . . . . . . . . . 17
2.7 Comments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.8 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1
3 Gaussian Path Integrals and Determinants 22
3.1 Preliminary Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.2 The Free Particle and the Normalisation Constant N . . . . . . . . . . . . . . . . 23
3.3 The Free Particle: Fluctuation Determinant and Mode Expansion . . . . . . . . 25
3.4 The Harmonic Oscillator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.5 Gaussian Path Integrals and Determinants: the VVPM and GY Formulae . . . . 28
3.6 Some Comments on the Regularisation of Determinants . . . . . . . . . . . . . . 32
3.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4 Generating Functionals and Perturbative Expansions 37
4.1 The Generating Functional Z[J] . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.2 Green’s Functions and the Generating Functional for Quadratic Theories . . . . 39
4.3 Perturbative Expansion and Generating Functionals . . . . . . . . . . . . . . . . 41
4.4 The Stationary Phase Approximation for Oscillatory Integrals . . . . . . . . . . . 42
4.5 The SemiClassical Approximation . . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.6 Scattering Theory and the Path Integral . . . . . . . . . . . . . . . . . . . . . . . 47
4.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
A Gaussian and Fresnel Integrals 49
A.1 Basic 1dimensional Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
A.2 Related Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
A.3 Gaussian Integrals and Determinants . . . . . . . . . . . . . . . . . . . . . . . . . 51
B Inﬁnite Product Identities 53
C Equivalence of the VVPM and GY Formulae for the Fluctuation Determinant 54
2
0 Preface
These are notes for part of a course on advanced quantum mechanics given to 4th
year physics students. The only prerequisites, however, are a basic knowledge of the
Schr¨ odinger and Heisenberg pictures of standard quantum mechanics (as well as the will
ingness to occasionally and momentarily suspend disbelief). Thus the material could
easily be presented at an earlier stage. I covered the material in ﬁve 3“hour” lec
tures (1 “hour” = 45 minutes) and this time constraint (there are other topics that I
wanted to cover as well in the course) dictated the level of detail (or lack thereof) of
the presentation.
One of the aims of these lectures was to set the stage for a future course on quantum ﬁeld
theory. To a certain extent this motivated the choice of topics covered in these notes
(e.g. generating functionals are discussed, while concrete applications of path integrals
to nontrivial quantum mechanics problems are not).
These notes do not include an introductory section on motivations, history, etc.  such
things are best done orally anyway. My own point of view is that the path intgral
approach to quantum theories is simultaneously more intuitive, more fundamental, and
more ﬂexible than the standard operator  state description, but I do not intend to get
into an argument about this. Objectively, the strongest points in favour of the path
integral appoach are that
• unlike the usual Hamiltonian approach the path integral approach provides a man
ifestly Lorentz covariant quantisation of classical Lorentz invariant ﬁeld theories;
• it is perfectly adapted to perturbative expansions and the derivation of the Feyn
man rules of a quantum ﬁeld theory;
• it allows in a rather straightforward manner to calculate certain nonperturbative
contributions (i.e. eﬀects that cannot be seen to any order in perturbation theory)
to Smatrix elements.
The motivation for writing these notes was that I found the typical treatment of quantum
mechanics path integrals in a quantum ﬁeld theory text to be too brief to be digestible
(there are some exceptions), while monographs on path integrals are usually far too
detailed to allow one to get anywhere in a reasonable amount of time.
I have not provided any referenes to either the original or secondary literature since
most of the material covered in these notes is completely standard and can be found in
many places (the exception perhaps being the GelfandYaglom formula for ﬂuctuation
3
determinants for which some references to the secondary literature are given in section
3.5).
No attempt at mathematical rigour (not even the pretense of an attempt) is made in
these notes.
Updated corrected or expanded versions of these notes will be available at
http://www.unine.ch/phys/string/Lecturenotes.html
If you ﬁnd any mistakes, or if you have any other comments on these notes, complaints,
(constructive) criticism, or also if you just happen to ﬁnd them useful, please let me
know.
4
1 The Evolution Kernel
1.1 Review: The timeevolution Operator
The dynamical information about quantum mechanics is contained in the matrix ele
ments of the timeevolution operator U(t
f
, t
i
). For a timeindependent Hamiltonian
ˆ
H
one has
U(t
f
, t
i
) = e
−(i/¯ h)(t
f
−t
i
)
ˆ
H
(1.1)
whereas the general expression for a timedependent Hamiltonian involves the time
ordered exponential
U(t
f
, t
i
) = T
_
e
−(i/¯ h)
_
t
i
t
f
dt
ˆ
H(t
)
_
. (1.2)
It satisﬁes the evolution equation
i¯ h
∂
∂t
f
U(t
f
, t
i
) =
ˆ
H(t
f
)U(t
f
, t
i
) , (1.3)
with initial condition
U(t
f
= t
i
, t
i
) = I . (1.4)
As a consequence,
Ψ(t
f
) = U(t
f
, t
i
)ψ(t
i
) (1.5)
satisﬁes the timedependent Schr¨ odinger equation
i¯ h
∂
∂t
f
Ψ(t
f
) =
ˆ
H(t
f
)Ψ(t
f
) (1.6)
with initial condition
Ψ(t
i
) = ψ(t
i
) . (1.7)
Another keyproperty of the timeevolution operator is
U(t
f
, t
i
) = U(t
f
, t)U(t, t
i
) t
i
< t < t
f
. (1.8)
In words: evolution from time t
i
to t
f
is the same as evolving from t
i
to t, followed by
evolution from t to t
f
.
1.2 Definition of the Propagator / Kernel
Our main interest will be in matrix elements of U in the position representation. We
will denote these by (the letter K stands for “kernel”)
K(x
f
, x
i
; t
f
−t
i
) =< x
f
U(t
f
, t
i
)x
i
> . (1.9)
5
This can also be interpreted as the transition amplitude
K(x
f
, x
i
; t
f
−t
i
) =< x
f
, t
f
x
i
, t
i
> , (1.10)
where (in the timeindependent case)
x
i
, t
i
>= e
(i/¯ h)t
i
ˆ
H
x
i
> (1.11)
etc. Note that this is not the Schr¨ odinger timeevolution of the state x
i
>  this would
have the opposite sign in the exponent (this is related to the fact that U(t
f
, t
i
) satisﬁes
the Schr¨ odinger equation with respect to t
f
, not t
i
). Rather, this state is characterised
by the fact that it is the eigenstate of the Heisenberg picture operator ˆ x
H
(t) at t = t
i
,
ˆ x
H
(t)x, t >= e
(i/¯ h)t
ˆ
H
ˆ xe
−(i/¯ h)t
ˆ
H
e
(i/¯ h)t
ˆ
H
x >= xx, t > . (1.12)
Another way of saying this, or introducing the states x, t >, is the following: In general
the operators ˆ x
H
(t) and ˆ x
H
(t
) do not commute for t = t
. Hence they cannot be
simultaneously diagonalised. For any given t, however, one can choose a basis in which
ˆ x
H
(t) is diagonal. This is the basis {x, t >}.
Matrix elements of the evolution operator between other states, say the transition am
plitude between an initial state ψ
i
> and a ﬁnal state ψ
f
>, are determined by the
kernel through
< ψ
f
U(t
f
, t
i
)ψ
i
> =
_
dx
f
_
dx
i
< ψ
f
x
f
>< x
f
U(t
f
, t
i
)x
i
>< x
i
ψ
i
>
=
_
dx
f
_
dx
i
ψ
∗
f
(x
f
)ψ
i
(x
i
)K(x
f
, x
i
; t
f
−t
i
) . (1.13)
1.3 Basic Properties of the Kernel
Here are some of the basic properties of the Kernel:
1. With properly normalised position eigenstates, one has
lim
t
f
→t
i
< x
f
, t
f
x
i
, t
i
>=< x
f
x
i
>= δ(x
f
−x
i
) . (1.14)
2. The kernel satisﬁes the Schr¨ odinger equation with respect to (t
f
, x
f
), i.e.
[i¯ h∂
t
f
−
ˆ
H(x
f
, p
f
= (¯ h/i)∂
x
f
, t
f
)] < x
f
, t
f
x
i
, t
i
>= 0 . (1.15)
More generally, for any ket φ > the wave function
Ψ
φ
(x, t) =< x, tφ > (1.16)
6
is a solution of the Schr¨ odinger equation, with
Ψ
φ
(x, t) =
_
dx
i
< x, tx
i
, t
i
>< x
i
, t
i
φ >
=
_
dx
i
K(x, x
i
, t −t
i
)Ψ
φ
(x
i
, t
i
) . (1.17)
3. Since we can restrict our attention to evolution forwards in time, one frequently
also considers the causal propagator or retarded propagator
K
r
(x
f
, x
i
; t
f
−t
i
) = Θ(t
f
−t
i
)K(x
f
, x
i
; t
f
−t
i
) (1.18)
where Θ is the Heavyside step function, Θ(x) = 1 for x > 0 and Θ(x) = 0 for
x < 0. It follows from the above two properties of the kernel, and Θ
(x) = δ(x),
that the retarded propagator satisﬁes
[i¯ h∂
t
f
−
ˆ
H(x
f
, p
f
= (¯ h/i)∂
x
f
, t
f
)]K
r
(x
f
, x
i
; t
f
−t
i
) = i¯ hδ(t
f
−t
i
)δ(x
f
−x
i
) . (1.19)
Thus the retarded propagator is a Green’s function for the Schr¨ odinger equation.
4. A key property of the kernel is the convolution property
K(x
f
, x
i
; t
f
−t
i
) =
_
+∞
−∞
dx K(x
f
, x; t
f
−t)K(x, x
i
; t −t
i
) (1.20)
(for t arbitrary subject to the condition t
f
> t > t
i
). This follows from the
property (1.8) of the timeevolution operator by inserting a complete state of
states in the form
I =
_
∞
−∞
dx x >< x . (1.21)
One then ﬁnds
< x
f
, t
f
x
i
, t
i
> = < x
f
U(t
f
, t)U(t, t
i
)x
i
>
=
_
∞
−∞
dx < x
f
U(t
f
, t)x >< xU(t, t
i
)x
i
>
=
_
∞
−∞
dx < x
f
, t
f
x, t >< x, tx
i
, t
i
> , (1.22)
which is the claimed result.
5. Finally, we can also expand the kernel in terms of energy eigenstates
ψ
n
(x) =< xn > (1.23)
as
K(x
f
, x
i
; t
f
−t
i
) = < x
f
e
−(i/¯ h)(t
f
−t
i
)
ˆ
H
x
i
>
=
n
< x
f
e
−(i/¯ h)(t
f
−t
i
)
ˆ
H
n >< nx
i
>
=
n
e
−(i/¯ h)(t
f
−t
i
)E
n
ψ
n
(x
f
)ψ
∗
n
(x
i
) . (1.24)
7
1.4 Example: The Free Particle
For the free particle, with Hamiltonian H = p
2
/2m, it is straightforward to determine
the kernel K = K
0
and verify explicitly all the above properties. By inserting a complete
set of momentum eigenstates p >,
< xp >=
1
√
2π¯ h
e
(i/¯ h)px
(1.25)
(i.e. by Fourier transform), one ﬁnds
K
0
(x
f
, x
i
; t
f
−t
i
) = < x
f
e
−(i/¯ h)(t
f
−t
i
)ˆ p
2
/2m
x
i
>
=
_
+∞
−∞
dp < x
f
p > e
−(i/¯ h)(t
f
−t
i
)p
2
/2m
< px
i
>
=
1
2π¯ h
_
+∞
−∞
dp e
(i/¯ h)[p(x
f
−x
i
) −(t
f
−t
i
)p
2
/2m]
. (1.26)
Using the Fresnel integral formulae from Appendix A, one thus ﬁnds
K
0
(x
f
, x
i
; t
f
−t
i
) =
¸
m
2πi¯ h(t
f
−t
i
)
e
i
¯ h
m
2
(x
f
−x
i
)
2
(t
f
−t
i
)
(1.27)
Note that the exponent has the interpretation as the “classical action”, i.e. as the action
S
0
of the free particle evaluated on the classical path x
c
(t) satisfying the free equations
of motion and the boundary conditions x
c
(t
f,i
) = x
f,i
,
S
0
[x
c
] =
m
2
(x
f
−x
i
)
2
(t
f
−t
i
)
. (1.28)
This at this stage rather mysterious fact has a very natural explanation from the path
integral point of view.
1.5 Exercises
1. Verify the results (1.27) and (1.28) for the evolution kernel of the free particle.
2. Verify that (1.27) is normalised as in (1.14), and satisﬁes the convolution property
(1.20) and the free particle Schr¨ odinger equation (1.15).
3. Prove the basic integral identity (A.26),
_
+∞
−∞
d
d
x e
−A
ab
x
a
x
b
/2
=
_
det
A
2π
_
−1/2
. (1.29)
for Gaussian integrals (A
ab
is a real symmetric positive matrix), and calculate the
integral
_
+∞
−∞
d
d
x e
−A
ab
x
a
x
b
/2 +J
a
x
a
(1.30)
by completing the square (remember that A
ab
is invertible).
8
2 Towards the Path Integral Representation of the Kernel
2.1 From the Kernel to the ShortTime Kernel and Back
In general, it is a diﬃcult (if not impossible) task to ﬁnd a closed form expression for the
kernel. However, the convolution property allows us to reduce the determination of the
ﬁnitetime kernel to that of the shorttime (or even inﬁnitesimal time) kernel K(x, y; )
and, as we will see later on, this allows us to make some progress.
First of all we note that we can write (once again in the timeindependent case, but the
argument works in general)
< x
f
e
−(i/¯ h)(t
f
−t
i
)
ˆ
H
x
i
>=< x
f

_
e
−(i/¯ h)
(t
f
−t
i
)
N
ˆ
H
_
N
x
i
> . (2.1)
We think of this as dividing the timeinterval [t
i
, t
f
] into N equal timeintervals [t
k
, t
k+1
]
of length ,
= t
k+1
−t
k
= (t
f
−t
i
)/N . (2.2)
Here k = 0, . . . , N − 1 and we identify t
f
= t
N
and t
i
= t
0
. We can now insert N − 1
resolutions of unity at times t
k
, k = 1, . . . , N−1 into the above expression for the kernel
to ﬁnd
K(x
f
, x
i
; t
f
−t
i
) = [
N−1
k=1
_
∞
−∞
dx
k
][
N−1
k=0
K(x
k+1
, x
k
; = t
k+1
−t
k
)] , (2.3)
where x
f
= x
N
and x
i
= x
0
.
Now this expression holds for any N. But for ﬁnite N, the kernel is still diﬃcult to
calculate. As we will see below, things simplify in the limit →0, equivalently N →∞.
In this limit, the kernel
K(x
f
, x
i
; t
f
−t
i
) = lim
N→∞
[
N−1
k=1
_
∞
−∞
dx
k
][
N−1
k=0
K(x
k+1
, x
k
; = (t
f
−t
i
)/N)] (2.4)
is determined by the shorttime kernel K(x
k+1
, x
k
; ) as →0.
2.2 The Trotter Product Formula and the Dirac ShortTime Kernel
Formally, it is not diﬃcult to see that in the limit N → ∞ we only need to know
this shorttime kernel K(x
k+1
, x
k
; ) to order . If everything in sight were commuting
(instead of operators), the argument for this would be the following: one writes
e
x
=
_
e
x/N
_
N
= (1 +x/N +O(1/N
2
))
N
(2.5)
9
and compares with the formula
e
x
= lim
N→∞
(1 +x/N)
N
(2.6)
to conclude that in the limit N →∞ the subleading O(N
−2
) terms in (2.5) can indeed
be dropped.
1
Of course, to establish an analogous result for (unbounded) operators
requires some functional analysis.
We will now assume that the Hamiltonian is of the standard form
H = T(p) +V (x) =
p
2
2m
+V (x) . (2.7)
We are thus interested in determining the kernel
K(x
k+1
, x
k
; ) =< x
k+1
e
−(i/¯ h)(
ˆ
T +
ˆ
V )
x
k
> . (2.8)
Provided that we can justify dropping terms of O(
2
), things simplify quite a bit. Indeed,
even though
ˆ
T and
ˆ
V are noncommuting operators,
ˆ
T and
ˆ
V commute up to order
2
, because their commutator is
2
[
ˆ
T,
ˆ
V ]. Thus, using the BakerCampbellHausdorﬀ
formula one has
e
−(i/¯ h)(
ˆ
T +
ˆ
V )
= e
−(i/¯ h)
ˆ
T
e
−(i/¯ h)
ˆ
V
+O(
2
) . (2.9)
To justify dropping these O(
2
) commutator terms, however, one needs some control
over the operator [
ˆ
T,
ˆ
V ] which should not become too singular.
Concretely, what one needs is the validity of the Trotter product formula
e
ˆ
A+
ˆ
B
=
_
e
ˆ
A/N +
ˆ
B/N
_
N
?
= lim
N→∞
_
e
ˆ
A/N
e
ˆ
B/N
_
N
(2.10)
for
ˆ
A =
ˆ
T and
ˆ
B =
ˆ
V .
This identity is not diﬃcult to prove for bounded operators. The case of interest,
unbounded operators, is trickier. The identity holds, for instance, on the common
domain of A and B, provided that both are selfadjoint operators that are bounded
from below. Once again, we will gloss over these functional analysis complications and
proceed with the assumption that it is legitimate to drop the commutator terms (while
keeping in mind that this assumption is not valid e.g. for the Coulomb potential!).
1
The identity (2.6) can be proved directly, e.g. via a binomial expansion or by showing that
d
dx
lim
N→∞
(1 +x/N)
N
= lim
N→∞
(1 +x/N)
N
.
10
With the above assumptions, to order we can write the shorttime kernel as
K(x
k+1
, x
k
; ) =< x
k+1
e
−(i/¯ h)
ˆ
T
e
−(i/¯ h)
ˆ
V
x
k
> . (2.11)
To diagonalise the operator
ˆ
T we introduce a complete set of momentum eigenstates
p
k
> with
< xp
k
>=
1
√
2π¯ h
e
(i/¯ h)p
k
x
(2.12)
and calculate
K(x
k+1
, x
k
; ) =
_
+∞
−∞
dp
k
< x
k+1
e
−(i/¯ h)
ˆ
T
p
k
>< p
k
e
−(i/¯ h)
ˆ
V
x
k
>
=
_
+∞
−∞
dp
k
< x
k+1
p
k
>< p
k
x
k
> e
−(i/¯ h)(
p
2
k
2m
+V (x
k
))
=
1
2π¯ h
_
+∞
−∞
dp
k
e
(i/¯ h)[p
k
(x
k+1
−x
k
)
−H(p
k
, x
k
)]
(2.13)
This is a lovely result, ﬁrst obtained by Dirac in 1933. In the exponential, where we once
had operators, we now encounter the Legendre transform of the classical Hamiltonian,
i.e. the Lagrangian! Indeed, if we identify
x
k+1
−x
k
=
x
k+1
−x
k
t
k+1
−t
k
→
dx
k
dt
, (2.14)
as a discretised timederivative, the exponent takes the classical form p
k
˙ x
k
−H(p
k
, x
k
).
We can implement the Legendre transformation explicitly by performing the Gaussian
(Fresnel) integral over p
k
(see Appendix A) to ﬁnd
K(x
k+1
, x
k
; ) =
1
2π¯ h
_
+∞
−∞
dp
k
e
(i/¯ h)[p
k
(x
k+1
−x
k
) −(
p
2
k
2m
+V (x
k
))]
=
_
m
2πi¯ h
e
(i/¯ h)L(x
k
, ˙ x
k
)
(2.15)
where the Lagrangian L is
L(x
k
, ˙ x
k
) =
m(x
k+1
−x
k
)
2
2
2
−V (x
k
) →
m˙ x
2
k
2
−V (x
k
) . (2.16)
2.3 The Path Integral Representation of the Kernel
Having obtained the above explicit formula for the shorttime kernel in terms of the
Lagrangian, we can no go back to (2.4) to obtain an expression for the ﬁnitetime
kernel. We can use either the phase space expression (2.13) or the conﬁguration space
expression (2.15). This iteration of Dirac’s result is due to Feynman (1942).
11
In this way we arrive at
K(x
f
, x
i
; t
f
−t
i
) = lim
N→∞
[
N−1
k=1
_
∞
−∞
dx
k
][
N−1
k=0
_
∞
−∞
dp
k
2π¯ h
]e
(i/¯ h)
N−1
k=0
[p
k
x
k+1
−x
k
−H(p
k
, x
k
)]
= lim
N→∞
_
m
2πi¯ h
_
N/2
[
N−1
k=1
_
∞
−∞
dx
k
]e
(i/¯ h)
N−1
k=0
L(x
k
, ˙ x
k
)
(2.17)
Note that in the phase space representation of the kernel there is always one more
momentum than position integral. This is a consequence of the fact that each short
time propagator contains one momentum integral whereas the position integrals are
inserted between the shorttime propagators and the two endpoints x
i
and x
f
are not
integrated over.
In the →0 or N →∞ limit, we interpret the points x
k
as deﬁning a continuous (but
almost certainly nowhere diﬀerentiable  see below) curve, any two successive points
being joined by a straight line. That is, we think of them as deﬁning a curve x(t) with
endpoints
x(t
i
) = x
i
x(t
f
) = x
f
(2.18)
and
x
k
= x(t
i
+k) . (2.19)
Likewise, we think of the p
k
as deﬁnig a curve p(t) in momentum space such that
p
0
= p(t
i
) p
k
= p(t
i
+k) . (2.20)
With this interpretation the exponents in the integrand of the kernel can be written as
lim
N→∞
N−1
k=0
[p
k
x
k+1
−x
k
−H(p
k
, x
k
)] =
_
t
f
t
i
dt [p(t) ˙ x(t) −H(p(t), x(t))]
lim
N→∞
N−1
k=0
L(x
k
, ˙ x
k
) =
_
t
f
t
i
dt L(x(t), ˙ x(t)) = S[x(t); t
f
, t
i
] (2.21)
In the same spirit, we formally now write the integrals as integrals over paths, intro
ducing the notation
lim
N→∞
[
N−1
k=1
_
∞
−∞
dx
k
] =
_
x(t
f
)=x
f
x(t
i
)=x
i
D[x(t)]
lim
N→∞
[
N−1
k=0
_
∞
−∞
dp
k
2π¯ h
] =
_
D[p(t)/2π¯ h] . (2.22)
With this notation, we can now write the kernel as a path integral ,
K(x
f
, x
i
; t
f
−t
i
) =
_
x(t
f
)=x
f
x(t
i
)=x
i
D[x(t)]
_
D[p(t)/2π¯ h]e
(i/¯ h)
_
t
f
t
i
dt [p(t) ˙ x(t) −H(p(t), x(t))]
= N
_
x(t
f
)=x
f
x(t
i
)=x
i
D[x(t)]e
(i/¯ h)S[x(t)]
. (2.23)
12
This is an integral over all paths x(t) with the speciﬁed boundary conditions and, in
the ﬁrst version, all paths p(t) with t ∈ [t
i
, t
f
].
Some remarks:
1. The ﬁrst line, a phase space path integral, is valid for general Hamiltonians
H(p, x) = T(p) + V (x) (possibly timedependent). The measure appears to be
an inﬁnitedimensional analogue of the canonical Liouville measure. The latter
is of course invariant under canonical transformations. This should however not
lead one to believe that the path integral representation of the kernel also enjoys
this invariance. Indeed, this cannot possibly be true since it is well known that
under a canonical transformation any Hamiltonian can be mapped to zero (the
HamiltonJacobi transformation) and hence into any other Hamiltonian, while the
kernel depends nontrivially on the Hamiltonian.
2. To pass to the second line, a conﬁguration space path integral, we used the explicit
quadratic form T = p
2
/2m to perform the Gaussian integral over p. The derivation
of the path integral for a Hamiltonian with a velocity dependent potential, such
as the magnetic interaction (p − A)
2
/2m, is more subtle (the discretised version
requires a “midpoint rule” for the Hamiltonian) and will not be discussed here.
3. As mentioned above, as limits of piecewise linear continuous paths these paths are
continuous but not diﬀerentiable. Indeed diﬀerentiability would require existence
of a ﬁnite limit of (x
k+1
− x
k
)/ as → 0. But x
k+1
and x
k
are independent
variables, and hence there is no reason for the diﬀerence x
k+1
−x
k
to go to zero as
→ 0. Hence the paths entgering the above sum/integral are typically nowhere
diﬀerentiable. Evidently, then, things like ˙ x(t) require some (perhaps stochastic
or probabilistic) interpretation, but we will not open this Pandora’s box.
4. Finally, N is formally an inﬁnite normalisation constant,
N = lim
N→∞
_
mN
2πi¯ h(t
f
−t
i
)
_
N/2
(2.24)
(we have replaced by (t
f
− t
i
)/N) whose sole purpose in life is to make the
combined expression N times the integral well deﬁned, ﬁnite and equal to the left
hand side (provided that all our functional analysis assumptions are satisﬁed).
We could have absorbed much of the prefactor into the deﬁnition of the measure by
writing (2.17) as
K(x
f
, x
i
; t
f
−t
i
) =
_
m
2πi¯ h
_
1/2
lim
N→∞
[
N−1
k=1
_
∞
−∞
_
m
2πi¯ h
_
1/2
dx
k
]e
(i/¯ h)
N−1
k=0
L(x
k
, ˙ x
k
)
(2.25)
13
and deﬁning
_
x(t
f
)=x
f
x(t
i
)=x
i
˜
D[x(t)] = lim
N→∞
[
N−1
k=1
_
∞
−∞
_
m
2πi¯ h
_
1/2
dx
k
] . (2.26)
Then the kernel is
K(x
f
, x
i
; t
f
−t
i
) =
_
m
2πi¯ h
_
1/2
_
x(t
f
)=x
f
x(t
i
)=x
i
˜
D[x(t)]e
(i/¯ h)S[x(t)]
. (2.27)
Another useful normalisation of the measure is, as we will see in section 3, such that
N
_
x(t
f
)=x
f
x(t
i
)=x
i
D[x(t)]e
(i/¯ h)S[x(t)]
=
¸
m
2πi¯ h(t
f
−t
i
)
_
x(t
f
)=x
f
x(t
i
)=x
i
ˆ
D[x(t)]e
(i/¯ h)S[x(t)]
.
(2.28)
We will obtain a more informative expresion for N, and hence the relation between D[x]
and
ˆ
D[x], involving the determinant of a diﬀerential operator, in the next section.
2.4 The Path Integral Representation of the Partition Function
In analogy with the quantum statistical (or thermal) partition function
Z(β) = Tr e
−β
ˆ
H
(2.29)
one deﬁnes the quantum mechanical partition function, or the partition function for
short, as the trace of the time evolution operator,
Z(t
f
, t
i
) = Tr U(t
f
, t
i
) . (2.30)
In the timeindependent case, this becomes
Z(t
f
, t
i
) = Z(t
f
−t
i
) = Tr e
−(i/¯ h)(t
f
−t
i
)
ˆ
H
, (2.31)
and thus the quantum mechanical and thermal partition function are formally related
by continuation of the time interval (t
f
−t
i
) to the imaginary value
t
f
−t
i
= −i¯ hβ . (2.32)
Evaluating the trace in a basis of energy eingenstates n >, one ﬁnds
Z(t
f
−t
i
) =
n
e
−(i/¯ h)(t
f
−t
i
)E
n
. (2.33)
On the other hand, evaluting the trace in a basis of position eigenstates x >, one
obtains
Z(t
f
, t
i
) =
_
+∞
−∞
dx < xU(t
f
, t
i
)x >=
_
+∞
−∞
dx K(x, x; t
f
−t
i
) . (2.34)
14
This means that in the discretised expression (2.17) for the kernel there are now an
equal number of momentum and position integrals. In the continuum version (2.23),
setting x
f
= x
i
= x means that one is integrating not over all paths but over all closed
paths (loops) at x, and the integration over x means that one is integrating over all
closed loops,
Z(t
f
, t
i
) = N
_
+∞
−∞
dx
_
x(t
f
)=x
x(t
i
)=x
D[x(t)]e
(i/¯ h)S[x(t)]
= N
_
x(t
f
)=x(t
i
)
D[x(t)]e
(i/¯ h)S[x(t)]
. (2.35)
Since we now have an equal number of x and pintegrals, in terms of the measure
˜
D[x]
(2.26) the partition function reads
Z(t
f
, t
i
) =
_
x(t
f
)=x(t
i
)
˜
D[x(t)]e
(i/¯ h)S[x(t)]
. (2.36)
Comparison with (2.33) shows that, if we are able to calculate this path integral over
all closed loops we should (in the timeindependent case only, of course) be able to read
oﬀ the energy spectrum.
The relation between statistical mechanics and quantum mechanics at imaginary time
is rather deep. In particular, for quantum ﬁeld theory this implies a relation between
ﬁnite temperature quantum ﬁeld theory in Minkowski space and quantum ﬁeld theory in
Euclidean space with one compact (Euclidean “time”) direction. This has farreaching
consequences (none of which will be explored here).
2.5 Back from the Path Integral to the Schr¨ odinger Equation
Since inside the path integral one is dealing with classical functions and functionals
rather than with operators, fairly simple “classical” manipulations of the path integral
can lead to nontrivial quantum mechanical identities.
As an example, consider the “trivial” statement that the path integral is invariant under
an overall shift
x(t) →x(t) +y(t) y(t
i
) = y(t
f
) = 0 (2.37)
of the integration variable,
_
x(t
f
)=x
f
x(t
i
)=x
i
D[x(t)] e
(i/¯ h)S[x(t)]
=
_
x(t
f
)=x
f
x(t
i
)=x
i
D[x(t)] e
(i/¯ h)S[x(t) +y(t)]
(2.38)
which follows from the (presumed) translation invariance of the measure.
15
Inﬁnitesimally, with y(t) = δx(t) and
S[x(t) +δx(t)] = S[x(t)] +δS[x(t)] , (2.39)
this statement reduces to
_
x(t
f
)=x
f
x(t
i
)=x
i
D[x(t)] δS[x(t)] e
(i/¯ h)S[x(t)]
= 0 . (2.40)
Sometimes this statement is paraphrased as “the path integral of a total derivative is
zero”,
_
x(t
f
)=x
f
x(t
i
)=x
i
D[x(t)]
δ
δx(t)
e
(i/¯ h)S[x(t)]
= 0 . (2.41)
Consequences derived from such identities are known as “SchwingerDyson equations”
in the quantum ﬁeld theory context.
Since the variation of the action with δx(t) vanishing at the endpoints gives the Euler
Lagrange equations,
δS[x(t)] =
_
t
f
t
i
dt
_
∂L
∂x
−
d
dt
∂L
∂ ˙ x
_
δx(t) , (2.42)
and since (2.40) holds for any such δx(t), we deduce that the classical equation of motion
are valid inside the path integral (!),
_
x(t
f
)=x
f
x(t
i
)=x
i
D[x(t)]
_
∂L
∂x
−
d
dt
∂L
∂ ˙ x
_
e
(i/¯ h)S[x(t)]
= 0 . (2.43)
This is the path integral version of Ehrenfest’s theorem.
Since this was so easy, let us generalise this a bit and consider variations of the path
which ﬁx x(t
i
) but change x(t
f
),
δx(t
i
) = 0 δx(t
f
) = 0 . (2.44)
This means that we are calculating
δ < x
f
, t
f
x
i
, t
i
>= δx(t
f
)
∂
∂x
f
< x
f
, t
f
x
i
, t
i
> . (2.45)
In this case, the variation of the action is
δS[x(t)] =
_
t
f
t
i
dt
_
∂L
∂x
−
d
dt
∂L
∂ ˙ x
_
δx(t) +
∂L
∂ ˙ x
δx
t
f
=
_
t
f
t
i
dt
_
∂L
∂x
−
d
dt
∂L
∂ ˙ x
_
δx(t) +p(t
f
)δx(t
f
) , (2.46)
16
where p(t
f
) ≡ p
f
is the canonical momentum at t = t
f
. This leads to the standard
HamiltonJacobi relation
p
f
=
∂S[x
c
]
∂x
f
. (2.47)
We deduce that
∂
∂x
f
< x
f
, t
f
x
i
, t
i
>=
i
¯ h
_
x(t
f
)=x
f
x(t
i
)=x
i
D[x(t)] p(t
f
) e
(i/¯ h)S[x(t)]
, (2.48)
which is nothing other than the familiar statement that in the position representation
one has
ˆ p
f
=
¯ h
i
∂
∂x
f
. (2.49)
In the same way, by considering a variation of t
f
, and using
L(x
f
, ˙ x
f
) =
d
dt
f
S[x(t)] =
∂S
∂t
f
+p
f
˙ x
f
(2.50)
(modulo the EulerLagrange equations), and therefore
∂S[x
c
]
∂t
f
= −H(x
f
, p
f
) , (2.51)
one obtains a path integral of the statement (1.15) that the kernel satisﬁes the Schr¨ odinger
equation (Exercise),
_
i¯ h
∂
∂t
f
−
ˆ
H(ˆ x
f
, ˆ p
f
=
¯ h
i
∂
∂x
f
)
_
_
x(t
f
)=x
f
x(t
i
)=x
i
D[x(t)] e
(i/¯ h)S[x(t)]
= 0 . (2.52)
We had initially derived the path integral from the standard description, and now
we have employed path integral manipulations to pass back from the path integral to
the Schr¨ odinger representation. This essentially completes the (formal) proof of the
equivalence of the path integral description of quantum mechanics and the standard
Schr¨ odinger representation. One piece of the dictionary that is still missing is how to
translate matrix elements other than the basic transition amplitude < x
f
, t
f
x
i
, t
i
>
into the path integral language. This will be the subject of the next subsection.
2.6 Convolution, Correlation Functions and TimeOrdered Products
As we saw in section 1 and above, a crucial property of the kernel is the convolution
property (1.20). How is this encoded in the path integral representation (2.23)
K(x
f
, x
i
; t
f
−t
i
) = N
_
x(t
f
)=x
f
x(t
i
)=x
i
D[x(t)]e
(i/¯ h)S[x(t); t
f
, t
i
]
(2.53)
17
of the kernel? We need to consider seperately the integrand and the measure. As far as
the integrand is concerned, the action S[x(t); t
f
, t
i
] obviously satisﬁes
S[x(t); t
f
, t
i
] = S[x(t); t
f
, t
0
] +S[x(t); t
0
, t
i
] (2.54)
for any t
f
> t
0
> t
i
since
_
t
f
t
i
(. . .) =
_
t
f
t
0
(. . .) +
_
t
0
t
i
(. . .) . (2.55)
And for the path integral measure we have
_
x(t
f
)=x
f
x(t
i
)=x
i
D[x(t)] =
_
+∞
−∞
dx
0
_
x(t
f
)=x
f
x(t
0
)=x
0
D[x(t)]
_
x(t
0
)=x
0
x(t
i
)=x
i
D[x(t)] (2.56)
In words: we can perform the integral over all paths from x
i
to x
f
, by considering
all paths from x
i
to x
0
and x
0
to x
f
for some ﬁxed x
0
= x(t
0
) and then integrating
over x
0
. Taken together, the statements about the action and the measure imply
2
the
convolution property (1.20).
So far we have only discussed the transition amplitude < x
f
, t
f
x
i
, t
i
>, but it is also
possible to represent matrix elements of operators as path integrals. The most natu
ral operators to consider in the present context are products of Heisenberg operators
ˆ x
H
(t
1
)ˆ x
H
(t
2
) . . ..
We begin with a single operator ˆ x
H
(t
0
) with t
f
> t
0
> t
i
and, to simplify the notation,
we will from now on drop the subscript H on Heisenberg operators. By elementary
manipulations we ﬁnd
< x
f
, t
f
ˆ x(t
0
)x
i
, t
i
> =
_
+∞
−∞
dx(t
0
) < x
f
, t
f
ˆ x(t
0
)x(t
0
), t
0
>< x(t
0
), t
0
x
i
, t
i
>
=
_
+∞
−∞
dx(t
0
) < x
f
, t
f
x(t
0
), t
0
>< x(t
0
), t
0
x
i
, t
i
> x(t
0
) .
Turning this into a statement about path integrals, using the convolution property, we
therefore conclude that
< x
f
, t
f
ˆ x(t
0
)x
i
, t
i
>=
_
x(t
f
)=x
f
x(t
i
)=x
i
D[x(t)] x(t
0
) e
(i/¯ h)S[x(t); t
f
, t
i
]
. (2.57)
Thus the insertion of an operator in the usual prescription corresponds to the insertion
of a classical function in the path integral formulation. While this is very charming, and
in line with the replacement of the Hamiltonian operator by the Lagrange function and
2
At least as long as one pretends that N = 1 or that N has somehow been incorporated into the
deﬁnition of a suitably regularised path integral. This is the recommended attitude at the present level
of rigour (better: nonrigour), and one that we will adopt henceforth.
18
its action, it also immediately raises a puzzle. Namely, since x(t
1
) and x(t
2
) commute,
does the insertion of x(t
1
)x(t
2
) into the path integral calculate the matrix elements of
ˆ x(t
1
)ˆ x(t
2
) or ˆ x(t
2
)ˆ x(t
1
) or . . . ? To answer this question, we reverse the above calcu
lation, but this time with the insertion of x(t
1
)x(t
2
). In order to use the convolution
property of the kernel or path integral, we need to distinguish the two cases t
2
> t
1
and
t
2
< t
1
. Then one ﬁnds
_
x(t
f
)=x
f
x(t
i
)=x
i
D[x(t)] x(t
1
)x(t
2
) e
(i/¯ h)S[x(t); t
f
, t
i
]
=
_
x(t
f
)=x
f
x(t
i
)=x
i
D[x(t)] x(t
2
)x(t
1
) e
(i/¯ h)S[x(t); t
f
, t
i
]
=
_
< x
f
, t
f
ˆ x(t
2
)ˆ x(t
1
)x
i
, t
i
> t
2
> t
1
< x
f
, t
f
ˆ x(t
1
)ˆ x(t
2
)x
i
, t
i
> t
2
< t
1
(2.58)
Using the timeordering operator we can summarise these results as
_
x(t
f
)=x
f
x(t
i
)=x
i
D[x(t)] x(t
1
)x(t
2
) e
(i/¯ h)S[x(t); t
f
, t
i
]
=< x
f
, t
f
T (ˆ x(t
1
)ˆ x(t
2
))x
i
, t
i
> .
(2.59)
This immediately generalises to
_
x(t
f
)=x
f
x(t
i
)=x
i
D[x(t)] x(t
1
) . . . x(t
n
) e
(i/¯ h)S[x(t); t
f
, t
i
]
=< x
f
, t
f
T (ˆ x(t
1
) . . . ˆ x(t
n
))x
i
, t
i
> .
(2.60)
Thus the path integral always evaluates matrix elements of timeordered products of
operators. This explains why the ordering inside the path integral is irrelevant and how
the ordering ambiguity is resolved on the operator side.
As a ﬁnal variation of this theme, we consider the path integral representation of tran
sition amplitudes between states other than the position eigenstates x, t >. To obtain
these, we simply use the formula (1.13),
< ψ
f
U(t
f
, t
i
)ψ
i
>=
_
dx
f
_
dx
i
ψ
∗
f
(x
f
)ψ
i
(x
i
)K(x
f
, x
i
; t
f
−t
i
) , (2.61)
and the path integral expression for the kernel K(x
f
, x
i
; t
f
−t
i
).
2.7 Comments
We have thus passed from a formulation of quantum mechanics based on the Hamilto
nian (and operators and Hilbert spaces) to a Lagrangian description in which there are
only commuting objects, no operators. In this framework, the quantum nature, which
in the usual Hamiltonian approach is reﬂected in the noncommutativity of operators,
19
arises because one is instructed to consider not only classical paths, i.e. extrema of the
action (solutions to the classical equations of motion) but all possible paths, weighted
by the exponential of (i/¯ h) times the action.
This provides an extremely intuitive picture of quantum mechanics in which one is
never led to ask questions like ”which path did the electron take?”. One also sees very
explicitly how classical mechanics emerges in the limit ¯ h → 0. In that case, the path
integral will be dominated by contributions form the extrema of the action (by the usual
stationary phase approximation), i.e. precisely by those paths that are solutions to the
classical equations of motion (see the discussion in section 4.6).
It is important to note that at this point the path integral notation that we have
introduced is largely symbolic. It is a shorthand notation for the N →∞ limit (2.17).
Provided that all our assumptions are satisﬁed, this is just another (albeit complicated
looking) way of writing the propagator.
However, the signiﬁcance of introducing this symbolic notation should not be under
estimated. Indeed, if one always had to calculate path integrals as the limit of an
inﬁnite number of integrals, then path integrals might be conceptually interesting but
that approach would hardly be an eﬃcient calculational tool. One might like to draw an
analogy here with Riemann integrals, deﬁned as the limits of an inﬁnite sum. In practice,
of course, one does not calculate integrals that way. Rather, one can use that deﬁnition
to establish certain basic properties of the resulting inﬁnite sum, symbolically denoted
by the continuum sum (integral)
_
, and then one determines deﬁnite and indeﬁnite
integrals directly, without resorting to the discretised description. Of course, in this
process one may be glossing over several mathematical subtleties (which will ultimately
lead to the development of measure theory, the Lebesgue integral etc.), but this does
not mean that one cannot reliably calculate simple integrals without knowing about
these things.
The attitude regarding path integrals we will adopt in the following will be similar in
spirit. We will deduce some properties of path integrals from their “discretised” version
and then try to pass as quickly as possible to continuum integrals which will allow us to
perform path integrals via one “functional integration” instead of an inﬁnite number of
ordinary integrations. Once again, this will be sweeping many important mathematical
subtleties under the rug (not the least of which is “does something like what you have
called D[x] exist at all?”), but that does not mean that we cannot trust the results that
we have obtained.
3
3
By the way, the answer to that questions is ”no”, but that is irrelevant  it is simply not the right
question to ask. A more pertinent question might be “can one make sense of
_
D[x] exp(i/¯ h)S[x] as
something like a measure (or linear functional)?”.
20
For some historical comments on “eﬀorts to give an improved mathematical meaning to
Feynman’s path integral formulation of quantum mechanics”, see e.g.
J.R. Klauder, The Feynman Path Integral: An Historical Slice, quantph/0303034.
2.8 Exercises
1. Using the BakerCampbellHausdorﬀ formula, determine the operator
ˆ
X, deﬁned
by
e
−(i/¯ h)(
ˆ
T +
ˆ
V )
= e
−(i/¯ h)
ˆ
T
e
−(i/¯ h)
ˆ
V
e
−(/¯ h)
2
ˆ
X
, (2.62)
to order .
2. Generalise the derivation of the path integral to systems with d > 1 degrees of
freedom. Assume that the Hamiltonian has the standard form
H =
p
2
2m
+V (x) , (2.63)
where x = (x
1
, . . . , x
d
) etc.
3. Spell out the proof of (2.52) (the path integral satisﬁes the Schr¨ odinger equation)
in detail.
21
3 Gaussian Path Integrals and Determinants
3.1 Preliminary Remarks
The path integral expression for the propagator we have obtained is
K(x
f
, x
i
; t
f
−t
i
) = N
_
x(t
f
)=x
f
x(t
i
)=x
i
D[x(t)]e
(i/¯ h)S[x(t)]
. (3.1)
We thus now need to make sense of, and develop rules for evaluating, such path integrals.
For a general system described by an action S[x] an exact evaluation of the path integral
is certainly too much to hope for. Indeed, even in the ﬁnitedimensional case integrals
of exponentials of elementary functions can typically be evaluated in closed form only in
the purely quadratic (Gaussian, Fresnel) case, whereas more general integrals are then
evaluated ‘perturbatively’ in terms of a generating function as in (A.21).
For path integrals, the situation is quite analogous. Typically, the path integrals that
can be calculated in closed form are purely quadratic (Gaussian, Fresnel) integrals, i.e.
actions of the general timedependent harmonic oscillator type
S[x] =
m
2
_
t
f
t
i
( ˙ x(t)
2
−ω(t)
2
x(t)
2
) , (3.2)
perhaps with the addition of a ‘source’ term,
S[x] →S[x] +
_
j(t)x(t) . (3.3)
Then the strategy to deal with more general path integrals, corresponding e.g. to an
action of the form
S[x] =
_
t
f
t
i
(
m
2
˙ x(t)
2
−V (x(t)) , (3.4)
is to reduce it to an expansion about a quadratic action. In practice this is achieved
in one of two ways. Either the potential is of the harmonic oscillator form plus a
perturbation, V (x) = V
0
(x) + λW(x), and one deﬁnes the path integral via a power
series expansion in λ, a perturbative expansion. Or one deﬁnes the path integral by an
expansion around a classical solution x
c
(t) of the equations of motion m¨ x = −V
(x).
To quadratic order in the ‘quantum ﬂuctuations’ around the classical solution one then
ﬁnds the action (3.2) with ω(t)
2
=
1
m
V
(x
c
(t)). This turns out to lead to an expansion
of the path integral in a power series in ¯ h, a semiclassical expansion.
In either case, the Gaussian path integral can be evaluated in reasonably closed form
and the complete path integral is then deﬁned in terms of the generating functional
associated with this quadratic action. In this section 3 we will deal exclusively with
Gaussian integrals. The evaluation of more general integrals in terms of generating
functionals will then be one of the subjects of section 4.
22
3.2 The Free Particle and the Normalisation Constant N
We are now ready to tackle our ﬁrst path integral. For obvious reasons we will consider
the simplest dynamical system, namely the free particle, with Lagrangian
L
0
(x(t), ˙ x(t)) =
m
2
˙ x(t)
2
. (3.5)
We thus need to calculate
K
0
(x
f
, x
i
; t
f
−t
i
) = N
_
x(t
f
)=x
f
x(t
i
)=x
i
D[x(t)]e
(i/¯ h)S
0
[x(t)]
, (3.6)
and we know that we should ﬁnd the result (1.27),
K
0
(x
f
, x
i
; t
f
−t
i
) =
¸
m
2πi¯ h(t
f
−t
i
)
e
(i/¯ h)
m
2
(x
f
−x
i
)
2
(t
f
−t
i
)
(3.7)
Since in this case we already know the result, we can use this calculation to determine
the overall normalisation of the path integral from the continuum point of view (since
N is universal: it depends only on m and (t
f
−t
i
) and not on the potential V (x)).
As described above, the general strategy is to expand the paths around a solution
to the classical equations of motion. Here the starting action is already quadratic,
but this expansion will have the added beneﬁt of eliminating the boundary conditions
x(t
i,f
) = x
i,f
from the path integral. We thus split any path satisfying these boundary
conditions into the sum of the classical path x
c
(t),
x
c
(t) = x
i
+
x
f
−x
i
t
f
−t
i
(t −t
i
) , (3.8)
and a quantum ﬂuctuation y(t),
x(t) = x
c
(t) +y(t) , (3.9)
with y(t) satisfying the homogeneous boundary conditions
y(t
i
) = y(t
f
) = 0 . (3.10)
Plugging this into the action, one ﬁnds
S
0
[x
c
+y] = S
0
[x
c
] +
m
2
_
t
f
t
i
˙ y(t)
2
(3.11)
where S
0
[x
c
] is the classical action, already given in (1.28),
S
0
[x
c
] =
m
2
(x
f
−x
i
)
2
(t
f
−t
i
)
. (3.12)
23
There is no linear term in y(t) because we are expanding around a critical point of the
action S
0
[x(t)], and there are no higher than quadratic terms because the free particle
action itself is quadratic.
In the path integral, instead of integrating over all paths x(t) with the speciﬁed boundary
conditions we now integrate over all paths y(t) with the boundary conditions (3.10). We
thus ﬁnd that the path integral expression for the kernel becomes
K
0
(x
f
, x
i
; t
f
−t
i
) = e
(i/¯ h)S
0
[x
c
]
N
_
y(t
i
)=y(t
f
)=0
D[y(t)]e
i
¯ h
m
2
_
t
f
t
i
˙ y(t)
2
(3.13)
We thus automatically produce the classical action in the exponent, as in (3.7).
To get a handle on the path integral over y(t), we integrate by parts in the action to
obtain
_
y(t
i
)=y(t
f
)=0
D[y(t)]e
i
¯ h
m
2
_
t
f
t
i
˙ y(t)
2
=
_
y(t
i
)=y(t
f
)=0
D[y(t)]e
i
¯ h
m
2
_
t
f
t
i
y(t)(−∂
2
t
)y(t)
(3.14)
Comparing with the fundamental Fresnel integral formula (A.27),
_
+∞
−∞
d
d
x e
iA
ab
x
a
x
b
/2
=
_
det
A
2πi
_
−1/2
, (3.15)
we deduce that what this path integral formally calculates is the determinant of the
diﬀerential operator (−∂
2
t
),
_
y(t
i
)=y(t
f
)=0
D[y(t)]e
i
¯ h
m
2
_
t
f
t
i
y(t)(−∂
2
t
)y(t)
=
_
Det
m
2πi¯ h
[−∂
2
t
]
_
−1/2
. (3.16)
I denote the (functional) determinant of an operator by Det to distinguish it from
a standard ﬁnitedimensional determinant. We will discuss this determinant in more
detail momentarily. First of all, we note that comparison of (3.7) and (3.13) shows that
the inﬁnite normalisation constant N is related to this determinant by
N =
¸
m
2πi¯ h(t
f
−t
i
)
_
Det
m
2πi¯ h
[−∂
2
t
]
_
+1/2
. (3.17)
We also see that the normalised path integral measure
ˆ
D[y(t)] introduced in (2.28) is
such that it normalises the free particle Gaussian ﬂuctuation integral to unity,
_
y(t
i
)=y(t
f
)=0
ˆ
D[y(t)]e
i
¯ h
m
2
_
t
f
t
i
y(t)(−∂
2
t
)y(t)
= 1 . (3.18)
24
3.3 The Free Particle: Fluctuation Determinant and Mode Expansion
Now let us return to the deﬁnition of the determinant of a diﬀerential operator. As in
the ﬁnitedimensional case, this determinant is formally deﬁned as the product of the
eigenvalues. That this is indeed what the path integral gives rise to can be seen more
explicitly by expanding y(t) in normalised eigenmodes y
n
(t) of the operator (−∂
2
t
),
y(t) =
n
c
n
y
n
(t) , (3.19)
with
_
−∂
2
t
_
y
n
(t) = λ
n
y
n
(t)
_
t
f
t
i
y
n
(t)y
m
(t) = δ
m,n
y
n
(t
i
) = y
n
(t
f
) = 0 . (3.20)
In terms of this decomposition, the action becomes
m
2
_
t
f
t
i
y(t)(−∂
2
t
)y(t) =
m
2
n
λ
n
c
2
n
. (3.21)
Thus the expansion of y(t) in terms of eigenmodes is tantamount to diagonalising the
operator (this statement is obviously true more generally). Then the path integral over
all y(t) becomes an integral over the c
n
,
_
y(t
i
)=y(t
f
)=0
D[y(t)] =
n
__
+∞
−∞
dc
n
_
, (3.22)
and thus the path integral reduces to an inﬁnite product of ﬁnitedimensional Gaussian
integrals, with the result
_
y(t
i
)=y(t
f
)=0
D[y(t)]e
i
¯ h
m
2
_
t
f
t
i
y(t)(−∂
2
t
)y(t)
=
n
__
+∞
−∞
dc
n
_
e
i
¯ h
m
2
n
λ
n
c
2
n
=
_
n
m
2πi¯ h
λ
n
_
−1/2
. (3.23)
It is this inﬁnite product which we have deﬁned as the determinant.
It will be useful for later to know explicitly the eigenvalues λ
n
. The properly normalised
eigenfunctions are
y
n
(t) =
¸
2
t
f
−t
i
sin nπ
t −t
i
t
f
−t
i
. (3.24)
Since y
−n
(t) = −y
n
(t), the linearly independent solutions are y
n
(t) with n ∈ N and the
corresponding eigenvalues are
λ
n
=
n
2
π
2
(t
f
−t
i
)
2
. (3.25)
25
We thus have
Det[−∂
2
t
] =
∞
n=1
n
2
π
2
(t
f
−t
i
)
2
(3.26)
This is clearly inﬁnite, and thus Det
−1/2
[−∂
2
t
] is zero, but this is compensated by the
inﬁnite normalisation constant in precisely such a way that one obtains the ﬁnite result
(3.17).
N
_
Det
m
2πi¯ h
[−∂
2
t
]
_
−1/2
=
¸
m
2πi¯ h(t
f
−t
i
)
(3.27)
For more comments on determinants and regularised determinants see sections 3.5 and
3.6.
3.4 The Harmonic Oscillator
We have now accumulated all the techniques we need to tackle a more interesting ex
ample, namely the timeindependent harmonic oscillator, with action
S[x] =
m
2
_
t
f
t
i
( ˙ x(t)
2
−ω
2
0
x(t)
2
) . (3.28)
Following the same strategy as for the free particle, we decompose the path into a
classical path x
c
(t) and the ﬂuctuation y(t), determine the classical action and the (still
quadratic) action for y(t), and then perform the Gaussian integral over y(t).
The classical path and action are (see Exercise 1)
x
c
(t) = x
i
sin ω
0
(t
f
−t)
sin ω
0
(t
f
−t
i
)
+x
f
sin ω
0
(t −t
i
)
sin ω
0
(t
f
−t
i
)
S[x
c
] =
m
2
ω
0
sinω
0
(t
f
−t
i
)
_
(x
2
i
+x
2
f
) cos ω
0
(t
f
−t
i
) −2x
i
x
f
_
. (3.29)
Expanding the action arund x
c
(t), one ﬁnds
S[x
c
+y] = S[x
c
] +
m
2
_
t
f
t
i
y(t)(−∂
2
t
−ω
2
0
)y(t) (3.30)
Thus the path integral we need to calculate is
K(x
f
, x
i
; t
f
−t
i
) = e
(i/¯ h)S[x
c
]
N
_
y(t
i
)=y(t
f
)=0
D[y(t)]e
i
¯ h
m
2
_
t
f
t
i
y(t)(−∂
2
t
−ω
2
0
)y(t)
(3.31)
This is once again a straightforward Gaussian (Fresnel) integral, and thus one ﬁnds,
using the result (3.17),
K(x
f
, x
i
; t
f
−t
i
) =
¸
m
2πi¯ h(t
f
−t
i
)
¸
¸
¸
_
Det
m
2πi¯ h
[−∂
2
t
]
Det
m
2πi¯ h
[−∂
2
t
−ω
2
0
]
e
(i/¯ h)S[x
c
]
=
¸
m
2πi¯ h(t
f
−t
i
)
¸
Det[−∂
2
t
]
Det[−∂
2
t
−ω
2
0
]
e
(i/¯ h)S[x
c
]
(3.32)
26
Since in writing the above we have taken into account the normalisation factor, the
result should be well deﬁned and ﬁnite. This is indeed the case. To calculate this ratio
of determinants, we observe ﬁrst of all that if the eigenvalues of the operator (−∂
2
t
) are
λ
n
, the eigenvalues µ
n
of the operator (−∂
2
t
−ω
2
0
) are µ
n
= λ
n
−ω
2
0
. Thus the ratio of
determinants is
¸
Det[−∂
2
t
]
Det[−∂
2
t
−ω
2
0
]
=
_
n
λ
n
λ
n
−ω
2
0
_
1/2
=
_
n
(1 −
ω
2
0
λ
n
)
_
−1/2
. (3.33)
Even though this may not be obvious from this expression, the result is actually an
elementary function. First of all, using the explicit expression for the λ
n
, one has
∞
n=1
(1 −
ω
2
0
λ
n
) =
∞
n=1
(1 −
ω
2
0
(t
f
−t
i
)
2
n
2
π
2
) . (3.34)
The function
f(x) =
∞
n=1
(1 −
x
2
n
2
π
2
) (3.35)
is an even function of x with f(0) = 1 and simple zeros at x = ±nπ. This shows (see
also Appendix B) that this is an inﬁnite product representation of
f(x) =
sin x
x
. (3.36)
Therefore, the ﬁnal result for the ratio of determinants is
¸
Det[−∂
2
t
]
Det[−∂
2
t
−ω
2
0
]
=
¸
ω
0
(t
f
−t
i
)
sin ω
0
(t
f
−t
i
)
, (3.37)
and our ﬁnal compact result for the propagator of the harmonic oscillator is
K(x
f
, x
i
; t
f
−t
i
) =
¸
m
2πi¯ h(t
f
−t
i
)
¸
ω
0
(t
f
−t
i
)
sin ω
0
(t
f
−t
i
)
e
(i/¯ h)S[x
c
]
=
¸
mω
0
2πi¯ hsin ω
0
(t
f
−t
i
)
e
(i/¯ h)S[x
c
]
(3.38)
with
S[x
c
] =
m
2
ω
0
sin ω
0
(t
f
−t
i
)
_
(x
2
i
+x
2
f
) cos ω
0
(t
f
−t
i
) −2x
i
x
f
_
. (3.39)
It is easy to see that this result reduces to that for the free particle in the limit ω
0
→0.
27
Given the above result for the kernel, we also immediately obtain an expression for the
partition function by setting x
i
= x
f
= x and integrating over x. This is a Fresnel
integral, and the result is
Z(t
f
−t
i
) =
1
2i sin
ω
0
(t
f
−t
i
)
2
. (3.40)
This can be expanded as
Z(t
f
−t
i
) =
∞
n=0
e
−(i/¯ h)(t
f
−t
i
)E
n
, (3.41)
where (cf. (2.33))
E
n
= ¯ hω
0
(n +
1
2
) (3.42)
are the harmonic oscillator energy levels. (Exercise)
3.5 Gaussian Path Integrals and Determinants: the VVPM and GY For
mulae
Frequently one encounters timedependent harmonic oscillators with Hamiltonian
H(t) =
1
2m
p
2
+
mω(t)
2
2
x
2
(3.43)
and the quadratic action
S[x] =
m
2
_
t
f
t
i
( ˙ x(t)
2
−ω(t)
2
x(t)
2
) . (3.44)
For instance, expanding the general action
S[x] =
m
2
_
t
f
t
i
( ˙ x(t)
2
−
2
m
V (x(t)) (3.45)
to second order around a classical solution x
c
(t) of the equations of motion m¨ x =
−V
(x), one ﬁnds the action (3.44) with
ω(t)
2
=
1
m
V
(x
c
(t)) . (3.46)
The resulting path integral is still Gaussian, and exactly the same strategy as above can
be used to show that the path integral result for the kernel of the evolution operator is
(cf. (3.32))
< x
f

ˆ
Te
−(i/¯ h)
_
t
f
t
i
dt
ˆ
H(t)
x
i
>=
¸
m
2πi¯ h(t
f
−t
i
)
¸
Det[−∂
2
t
]
Det[−∂
2
t
−ω(t)
2
]
e
(i/¯ h)S[x
c
]
.
(3.47)
28
Here x
c
(t) now denotes the classical harmonic oscillator solution with the given bound
ary condition, x
c
(t
i
) = x
i
and x
c
(t
f
) = x
f
, S[x
c
] is the classical action, and the ﬂuctu
ation determinants are to be calculated for zero (Dirichlet) boundary conditions.
In order to evaluate the result for the propagator (3.47), one needs to determine the
classical action and the ratio of ﬂuctuation determinants. The former is rather straight
forward provided that one can ﬁnd the classical solution. An integration by parts shows
that the classical action can be calculated in terms of the boundary values of x
c
(t) and
˙ x
c
(t) at t = t
i
, t
f
,
S[x
c
] =
m
2
_
t
f
t
i
( ˙ x
c
(t)
2
−ω(t)
2
x
c
(t)
2
) =
m
2
[x
f
˙ x
c
(t
f
) −x
i
˙ x
c
(t
i
)] . (3.48)
The calculation of the ratio of determinants would be complicated if one tried to cal
culate these determinants directly, as we did in the timeindependent case. Fortunately
there are two elegant shortcuts to calculating this ratio of determinants which are ﬁnite
dimensional in nature and do not require the calculation of a fuctional determinant. I
will brieﬂy describe these below.
One can for instance use the useful and remarkable result that the ratio of determinants
can be calculated from the classical action via the Van Vleck  Pauli  Morette (VVPM)
formula
¸
m
2πi¯ h(t
f
−t
i
)
¸
Det[−∂
2
t
]
Det[−∂
2
t
−ω(t)
2
]
=
1
√
2πi¯ h
¸
−
∂
2
S[x
c
]
∂x
i
∂x
f
(3.49)
More generally, for a ddimensional quantum system, the 2nd derivative of the classical
action would be replaced by the ddimensional VVPM determinant
∂
2
S[x
c
]
∂x
i
∂x
f
→det
_
∂
2
S[x
c
]
∂x
µ
i
∂x
ν
f
_
.
This is a nontrivial but standard and wellknown result. Notice that, to evaluate the
ratio of quantum ﬂuctuations in this manner, one only needs to know the classical
action.
In the case of the harmonic oscillator with constant frequency, agreement between the
VVPM formula and the result we obtained in (3.38) can be immediately veriﬁed from
the classical action (3.39) which gives
−
∂
2
S[x
c
]
∂x
i
∂x
f
=
mω
0
sin ω
0
(t
f
−t
i
)
. (3.50)
Alternatively, instead of the VVPM result one can use the equally remarkable, but
apparently much less well known, GelfandYaglom (GY) formula that states that
Det[−∂
2
t
−ω(t)
2
]
Det[−∂
2
t
]
=
F
ω
(t
f
)
t
f
−t
i
(3.51)
29
where F
ω
(t) is the solution of the classical harmonic oscillator equation
(∂
2
t
+ω(t)
2
)F
ω
(t) = 0 (3.52)
with the initial conditions
F
ω
(t
i
) = 0
˙
F
ω
(t
i
) = 1 . (3.53)
Thus one has the simple result
¸
m
2πi¯ h(t
f
−t
i
)
¸
Det[−∂
2
t
]
Det[−∂
2
t
−ω(t)
2
]
=
¸
m
2πi¯ hF
ω
(t
f
)
(3.54)
It is quite remarkable that the ratio of ﬂuctuation determinants, which involves the
product over all eigenvalues of the operator −(∂
2
t
+ ω(t)
2
) (with zero boundary condi
tions), can be expressed in terms of the zero mode (solution with zero eigenvalue) of
the same operator with the GY boundary conditions (3.53).
Once again, as an example we consider the constant frequency harmonic oscillator. The
solution of the classical equations of motion satisfying the GY boundary conditions is
evidently
F
ω
0
(t) =
1
ω
0
sinω
0
(t −t
i
) . (3.55)
Thus
F
ω
0
(t
f
) =
1
ω
0
sinω
0
(t
f
−t
i
) (3.56)
and the GY formula predicts
Det[−∂
2
t
−ω
2
0
]
Det[−∂
2
t
]
=
sin ω
0
(t
f
−t
i
)
ω
0
(t
f
−t
i
)
, (3.57)
in perfect agreement with the result (3.37).
Some comments:
1. If f(t) is a solution of the harmonic oscillator equation with f(t
i
) = 0, the GY
solution is evidently F
ω
(t) = f(t)/
˙
f(t
i
) with
F
ω
(t
f
) =
f(t
f
)
˙
f(t
i
)
. (3.58)
If, on the other hand, f(t) is any solution of the harmonic oscillator equation with
f(t
i
) = 0, the GY solution can be constructed from it as
F
ω
(t) = f(t
i
)f(t)
_
t
t
i
dt
1
f(t
)
2
(3.59)
In particular, one has
F
ω
(t
f
) = f(t
i
)f(t
f
)
_
t
f
t
i
dt
1
f(t)
2
. (3.60)
30
2. The denominator t
f
− t
i
in the GY formula (3.51) can be interpreted as F
0
(t
f
),
where
F
0
(t) = t −t
i
(3.61)
solves the freeparticle (ω = 0) equations of motion with the GY boundary condi
tions, so that one can write (3.51) in the more suggestive form
Det[−∂
2
t
−ω(t)
2
]
Det[−∂
2
t
]
=
F
ω
(t
f
)
F
0
(t
f
)
. (3.62)
3. Comparison with the VVPM formula gives the relation
∂
2
S[x
c
]
∂x
i
∂x
f
= −
m
F
ω
(t
f
)
. (3.63)
With a bit of eﬀort this purely classical (in the sense of classical mechanics) formula
can be proved directly via HamiltonJacobi theory (see Appendix C), but this by
itself provides no insight into the reason for the validity of either the VVPM or
the GY formula.
A slick proof of the GY formula, in the form (3.62), has been given by S. Coleman in
his Erice lectures on The uses of instantons (Appendix A), reprinted in
S. Coleman, Aspects of Symmetry, Selected Erice Lectures, Cambridge University Press
(1985).
This proof is also reproduced in
L.S. Schulman, Techniques and Applications of Path Integration, Dover (2005).
It works roughly as follows (to make this argument more precise one should insert words
like “Fredholm operators” etc. in appropriate places):
Let F
ω,λ
(t) be the solution of the equation
(−∂
2
t
−ω(t)
2
)F
ω,λ
(t) = λF
ω,λ
(t) (3.64)
with the GY boundary conditions
F
ω,λ
(t
i
) = 0 ,
˙
F
ω,λ
(t
i
) = 1 . (3.65)
Thus F
ω,0
(t) is what we called F
ω
(t) above, and F
ω,λ
(t) has the property that F
ω,λ
(t
f
) =
0 iﬀ λ is an eigenvalue of the operator (−∂
2
t
− ω(t)
2
) with zero Dirichlet boundary
conditions (because then F
ω,λ
(t) is the corresponding eigenfunction).
The claim is now that
Det[−∂
2
t
−ω(t)
2
−λ]
Det[−∂
2
t
−λ]
=
F
ω,λ
(t
f
)
F
0,λ
(t
f
)
(3.66)
31
for all λ ∈ C. Here, Det is again deﬁned to be the product of all eigenvalues. In
particular, this implies the GY result (3.62) for λ = 0.
To establish this claim, one considers the left and right hand sides as functions of the
complex variable λ. The left hand side is a meromorphic function of λ with a simple
zero at each eigenvalue λ
n
of (−∂
2
t
−ω(t)
2
) (an eigenvalue λ
n
of (−∂
2
t
−ω(t)
2
) is a zero
eigenvalue of (−∂
2
t
− ω(t)
2
) − λ
n
)), and a simple pole at each eigenvalue λ
0,n
of (−∂
2
t
)
(for the same reason). By the remark above (3.66), exactly the same is true of the right
hand side. In particular, the ratio of the left and the right hand sides has no poles and
is therefore an analytic function of λ.
Moreover, provided that ω(t) is a bounded function of t, for λ suﬃciently large, λ →∞,
one can ignore ω(t), and hence both the left and the right hand side go to 1 in that limit
(everywhere except on the real positive line where one can ﬁnd large real eigenvalues).
Putting these two observations together, one concludes that the ratio of the two sides
is an analytic function of λ that goes to 1 in any direction except perhaps along the
positive real axis, and this implies that the ratio is equal to 1 identically, which concludes
the proof of the identity (3.66).
Another elegant continuum (i.e. nondiscretised) proof of the GY formula (3.51) can be
found in
H. Kleinert, A. Chervyakov, Simple Explicit Formulas for Gaussian Path Integrals with
TimeDependent Frequencies, Phys. Lett. A245 (1998) 345357; quantph/9803016.
Yet another proof can be assembled from sections 3.3, 3.5 and 34.2 of
J. ZinnJustin, Quantum Field Theory and Critical Phenomena (Oxford Science Publi
cations, 1989).
A proof along these lines is also outlined in Appendix A of
R. Rajaraman, Solitons and Instantons (North Holland 1982, 1987)
3.6 Some Comments on the Regularisation of Determinants
In the above we have repeatedly encountered determinants of inﬁnitedimensional oper
ators, and we have treated them in quite a formal and cavalier way. In the following, I
want to (very) brieﬂy indicate some ways to deﬁne these objects in a more satisfactory
manner.
There are certain inﬁnitedimensional operators for which the deﬁnition of a determinant
poses no real problem. For example, for most intents and purposes, trace class operators
K (i.e. operators for which the trace exists), and operators of the form I +K, where I is
32
the identity operator and K a trace class operator, behave like ﬁnitedimensional linear
operators (matrices). For invertible (n ×n)matrices M one can write the determinant
(with K = I −M) as
det M = det(I +K) = 1 + tr K + tr(∧
2
K) +. . . + tr(∧
n
K) . (3.67)
Here ∧
p
K (the p’th antisymmetric power of K) denotes the operator implementing
the induced action of K on antisymmetric pvectors. Analogously, one can deﬁne the
(Fredholm) determinant of an invertible operator I +K with K trace class by
Det
F
(I +K) = 1 + Tr K + Tr(∧
2
K) +. . . , (3.68)
since this series is absolutely convergent. However, most operators appearing in physics
are not of this form, and hence one needs to be more creative.
In section 3.4 we had seen that, even though Det(−∂
2
t
−ω
2
0
) (the product of the eigen
values) diverges, the ratio of this determinant and the (equally divergent) free particle
determinant gave us a ﬁnite result. More generally, the ratio of determinants that ap
pears in (3.47) can be interpreted as deﬁning a regularised functional determinant of
the operator (−∂
2
t
−ω(t)
2
), in the sense of
Det
reg
[−∂
2
t
−ω(t)
2
] :=
Det[−∂
2
t
−ω(t)
2
]
Det[−∂
2
t
]
(3.69)
It is this regularised determinant that the normalised path integral with measure
ˆ
D[x]
computes (cf. (2.28,3.18)),
_
y(t
i
)=y(t
f
)=0
ˆ
D[y(t)]e
i
¯ h
m
2
_
t
f
t
i
y(t)(−∂
2
t
−ω(t)
2
)y(t)
=
_
Det
reg
[−∂
2
t
−ω(t)
2
]
_
−1/2
.
(3.70)
However, usually in the physics literature one adopts a slighly diﬀerent attitude. Instead
of regularising explicitly by means of the free particle determinant (which is neverthe
less natural from the path integral point of view) one attempts to deﬁne meaningful
individual (instead of ratios of) regularised functional determinants in other ways, e.g.
via the socalled ζfunction or heat kernel regularisation.
For a suﬃciently reasonable (elliptic, selfadjoint, . . . ) operator A with eigenvalues λ
n
,
deﬁne the spectral ζfunction of A to be the function
ζ
A
(s) =
n
λ
−s
n
. (3.71)
This generalises the deﬁnition of the ordinary Riemann ζfunction ζ(s),
ζ(s) =
∞
n=1
n
−s
. (3.72)
33
To see the relation between the spectral ζfunction and the determinant, one diﬀerenti
ates once,
ζ
A
(s) = −
n
n
−s
log λ
n
, (3.73)
to conclude that
ζ
A
(0) = −
n
log λ
n
= −log
n
λ
n
. (3.74)
Formally, therefore, one has
n
λ
n
= e
−ζ
A
(0)
, (3.75)
but, at this point, the left hand side is as illdeﬁned as the right hand side because
ξ
A
(s), as it stands, will be convergent only for Re(s) suﬃciently large and positive (and
not at s = 0).
Likewise, the ordinary Riemann ζfunction, as it stands, converges only for Re(s) > 1.
However, in that case it is well known that ζ(s) can be analytically continued to a
meromorphic function of s in the entire complex splane, with a pole only at s = 1. In
particular, one then has cute results like
ζ(0) = −
1
2
(
∞
n=1
1 )
ζ(−1) = −
1
12
(
∞
n=1
n )
ζ(−2) = 0 (
∞
n=1
n
2
)
(3.76)
as well as (we will need this below)
ζ
(0) = −
1
2
log 2π . (3.77)
Analogously, under favourable circumstances the spectral ζfunction ζ
A
(s) can be ex
tended to a meromorphic function of s which is holomorphic at s = 0, and then (3.75)
can be used to deﬁne the ζfunction regularised determinant of A via
Det
ζ
A = e
−ζ
A
(0)
. (3.78)
To illustrate this method, let us go back to the calculation of the free particle determi
nant in section 3.3. There we had seen that the eigenvalues of the operator A = −∂
2
t
(with Dirichlet boundary conditions) are
λ
n
=
n
2
π
2
T
2
n = 1, 2, . . . (3.79)
where T = t
f
−t
i
. To deﬁne the determinant, we construct the spectral ζfunction
ζ
A
(s) =
∞
n=1
_
nπ
T
_
−2s
=
_
T
π
_
2s
ζ(2s) = e
2s log(T/π)
ζ(2s) (3.80)
34
and calculate
ζ
A
(0) = 2ζ(0) log(T/π) + 2ζ
(0) = −log(T/π) −log(2π) = −log(2T) . (3.81)
Thus we conclude that, with ζfunction regularisation,
Det
ζ
[−∂
2
t
] = 2T . (3.82)
Likewise, for the harmonic oscillator determinant of section 3.4, with
λ
n
=
n
2
π
2
T
2
−ω
2
0
n = 1, 2, . . . (3.83)
one ﬁnds
Det
ζ
[−∂
2
t
−ω
2
0
] = 2T
sin Tω
0
Tω
0
. (3.84)
In particular, we have
Det
ζ
[−∂
2
t
−ω
2
0
]
Det
ζ
[−∂
2
t
]
=
sin Tω
0
Tω
0
(3.85)
in perfect agreement with the previously obtained result (3.37),
¸
Det[−∂
2
t
]
Det[−∂
2
t
−ω
2
0
]
=
¸
ω
0
(t
f
−t
i
)
sin ω
0
(t
f
−t
i
)
. (3.86)
While this ζfunction regularised deﬁnition of the determinant captures most of the
essential properties of the standard determinant, some care is required in manipulating
these objects. For example, one important property of the standard determinant in
the ﬁnitedimensional case is its multiplicativity det(MN) = det(M) det(N). To what
extent an analogous identity
Det
ζ
(AB)
?
= Det
ζ
ADet
ζ
B (3.87)
holds in the inﬁnitedimensional case is discussed at length in (warning: not for the
faint of heart)
M. Kontsevich, S. Vishik, Determinants of elliptic pseudodiﬀerential operators (155 p.,
hepth/9404046).
3.7 Exercises
1. Verify (3.29). Instead of calculating the classical action by integration, try to
determine it in an intelligent way by proving ﬁrst that only boundary terms con
tribute to the classical action,
S[x
c
] =
m
2
[x
f
˙ x
c
(t
f
) −x
i
˙ x
c
(t
i
)] . (3.88)
35
2. Verify that the harmonic oscillator kernel (3.38)
• satisﬁes the harmonic oscillator Schr¨ odinger equation (1.15)
• and the initial condition (1.14),
• and reduces to the free particle propagator for ω
0
→0.
3. Verify the result (3.40) for the partition function of the harmonic oscillator, and
the expansion (3.41).
4. Fill in the missing steps in the proof of the identity (3.63) given in Appendix C.
In particular, verify (C.10).
5. Find an elementary proof of the VVPM formula
Det[−∂
2
t
]
Det[−∂
2
t
−ω(t)
2
]
= −
(t
f
−t
i
)
m
∂
2
S[x
c
]
∂x
i
∂x
f
(3.89)
or the GY formula
Det[−∂
2
t
]
Det[−∂
2
t
−ω(t)
2
]
=
F
0
(t
f
)
F
ω
(t
f
)
(3.90)
and publish it and/or tell me about it.
36
4 Generating Functionals and Perturbative Expansions
In this section we will study some other important properties of the path integral, in
particular the perturbative and semiclassical expansion of nonGaussian path integrals.
The treatment in this section is somewhat more cursory than in the previous sections 
the main intention is to give a ﬂavour of things to come (in a course on quantum ﬁeld
theory, say).
4.1 The Generating Functional Z[J]
The main objects of interest in quantum ﬁeld theory are vacuum expectation values
of timeordered products of ﬁeld operators. These matrix elements can be obtained
from a generating functional which, in turn, can be expressed as a path integral. This
motivates the following discussion of these concepts in the quantum mechanical context.
Before turning to the path integral, we introduce the generating functionals for the
correlation functions (npoint functions)
G
fi
(x(t
1
), . . . , x(t
n
)) = < x
f
, t
f
T (ˆ x(t
1
) . . . ˆ x(t
n
))x
i
, t
i
>
G(x(t
1
), . . . , x(t
n
)) = < 0T (ˆ x(t
1
) . . . ˆ x(t
n
))0 > . (4.1)
They are deﬁned as
Z
fi
[j] =
∞
n=0
(i/¯ h)
n
n!
_
t
f
t
i
dt
1
. . . dt
n
j(t
1
) . . . j(t
n
)G
fi
(x(t
1
), . . . , x(t
n
))
= < x
f
, t
f
T e
(i/¯ h)
_
t
f
t
i
dt j(t)ˆ x(t)
x
i
, t
i
> (4.2)
Z[j] =
∞
n=0
(i/¯ h)
n
n!
_
+∞
−∞
dt
1
. . . dt
n
j(t
1
) . . . j(t
n
)G(x(t
1
), . . . , x(t
n
))
= < 0T e
(i/¯ h)
_
+∞
−∞
dt j(t)ˆ x(t)
0 > . (4.3)
From these generating functionals, the individual correlation functions can evidently be
reconstructed by diﬀerentiation,
G
fi
(x(t
1
), . . . , x(t
n
)) =
_
¯ h
i
_
n
δ
n
Z
fi
[j]
δj(t
1
) . . . δj(t
n
)

j(t)=0
G(x(t
1
), . . . , x(t
n
)) =
_
¯ h
i
_
n
δ
n
Z[j]
δj(t
1
) . . . δj(t
n
)

j(t)=0
. (4.4)
For Z
fi
[j] we can easily deduce a path intgral representation. Using (2.60) and the
deﬁnition (4.2), one ﬁnds
Z
fi
[j] =
_
x(t
f
)=x
f
x(t
i
)=x
i
D[x(t)] e
(i/¯ h)S[x(t); j(t); t
f
, t
i
]
, (4.5)
37
where
S[x(t); j(t); t
f
, t
i
] = S[x(t); t
f
, t
i
] +
_
t
f
t
i
dt j(t)x(t) (4.6)
is the action with a source term (or the action with a coupling of x(t) to the current
j(t)). Thus the generating functional Z
fi
[j] is the path integral for the action with a
source term.
Our aim is now to ﬁnd a similar path integral representation for Z[j]. For that we need
to project from the states x, t > to the ground state 0 >. To that end we ﬁrst expand
the state x, t > in a basis of eigenstates n > of the Hamiltonian,
x, t >= e
(i/¯ h)t
ˆ
H
x >=
n
e
(i/¯ h)t
ˆ
H
n >< nx >=
n
e
(i/¯ h)tE
n
ψ
∗
n
(x)n > . (4.7)
Thus for the correlation functions (2.60) of timeordered products of operators one ﬁnds
G
fi
(x(t
1
), . . . , x(t
n
)) =
m,n
e
−(i/¯ h)t
f
E
n
+ (i/¯ h)t
i
E
m
×
ψ
n
(x
f
)ψ
∗
m
(x
i
) < nT (ˆ x(t
1
) . . . ˆ x(t
n
))m > . (4.8)
To accomplish the projection onto the vacuum expectation value, we now take the limit
t
f,i
→ ±∞. This can be understood in a number of related ways. They all amount
to the statement that, in the sense of distributions, exp(−itE) → 0 for t → ∞, the
dominant contribution in that limit coming from the smallest possible value of E, i.e.
from the ground state. Explicitly, one can for instance replace E
n
→ (1 − i)E
n
for
a small positive and then take the limit → 0 at the end. Alternatively, one can
“analytically continue” to imaginary time, take the limit there, and then continue back
to real time. In whichever way one proceeds, one can conclude that
G(x(t
1
), . . . , x(t
n
)) = < 0T (ˆ x(t
1
) . . . ˆ x(t
n
))0 > (4.9)
= lim
t
f,i
→±∞
e
(i/¯ h)(t
f
−t
i
)E
0
ψ
0
(x
f
)ψ
∗
0
(x
i
)
_
x(t
f
)=x
f
x(t
i
)=x
i
D[x(t)] x(t
1
) . . . x(t
n
) e
(i/¯ h)S[x(t); t
f
, t
i
]
.
As the left hand side is independent of the boundary conditions x
f,i
imposed at t →±∞,
so is the right hand side. Passing now to the generating functional Z[j] (4.3), we can
once again rewrite the inﬁnite sum as an exponential in the path integral to deduce
(suppressing the dependence on the boundary conditions) that
Z[j] ∼
_
D[x(t)] e
(i/¯ h)S[x(t); j(t); t
f,i
= ±∞]
, (4.10)
where
S[x(t); j(t), t
f,i
= ±∞] =
_
+∞
−∞
dt [L(x(t), ˙ x(t)) +j(t)x(t)] . (4.11)
38
The proportionality factor (normalisation constant) is ﬁxed by
Z[j = 0] =< 00 >= 1 . (4.12)
We therefore obtain the ﬁnal result
Z[j] =
_
D[x(t)] e
(i/¯ h)S[x(t); j(t); t
f,i
= ±∞]
_
D[x(t)] e
(i/¯ h)S[x(t); t
f,i
= ±∞]
. (4.13)
The right hand side is independent of the boundary conditions imposed at t = ±∞
provided that one chooses the same boundary conditions in the numerator and the
denominator.
4.2 Green’s Functions and the Generating Functional for Quadratic
Theories
As we will reduce the calculation of a general path integral and its generating functional
to that of a quadratic theory, in this section we will determine explicitly the generating
functional for the latter.
For ﬁnitedimensional Fresnel integrals one has (see e.g. Exercise 1.5.3 and equation
(A.13))
_
d
d
x e
iA
ab
x
a
x
b
/2 +ij
a
x
a
=
_
det
A
2πi
_
−1/2
e
−iG
ab
j
a
j
b
/2
, (4.14)
where G
ab
is the inverse matrix (“Green’s function”) to A
ab
, G
ab
A
bc
= δ
a
c
. Thus the
“generating function” is
z
0
[j] :=
_
d
d
x e
iA
ab
x
a
x
b
/2 +ij
a
x
a
_
d
d
x e
iA
ab
x
a
x
b
/2
= e
−iG
ab
j
a
j
b
/2
. (4.15)
It follows from this that the “2point function”
< x
c
x
d
>:=
_
d
d
x x
c
x
d
e
iA
ab
x
a
x
b
/2
_
d
d
x e
iA
ab
x
a
x
b
/2
(4.16)
is
< x
c
x
d
> =
_
1
i
∂
∂j
c
1
i
∂
∂j
d
z
0
[j]
_
j=0
=
_
1
i
∂
∂j
c
_
−G
ad
j
a
z
0
[j]
_
_
j=0
= iG
cd
. (4.17)
39
Thus the “2point function” is the “Green’s function” of the Fresnel integral.
Higher moments (npoint functions) can be calculated in a similar way. For n odd they
are manifestly zero. For the 4point function one ﬁnds (Exercise)
< x
a
x
b
x
c
x
d
>=< x
a
x
b
>< x
c
x
d
> + < x
a
x
c
>< x
b
x
d
> + < x
a
x
d
>< x
b
x
c
> (4.18)
etc. The general result, expressing the 2npoint functions as a sum over all possible
pairings P(x
1
, . . . , x
2n
),
< x
1
. . . x
2n
>=
P(x
1
,...,x
2n
)
< x
i
1
x
i
2
> . . . < x
i
2n−1
x
i
2n
> (4.19)
(there are (2n − 1)!! terms) is also easily deduced from the generating function z
0
[j].
In the quantum ﬁeld theory context, this result is known as Wick’s Theorem and, even
though a simple result, is of enormous practical signiﬁcance in perturbative calculations.
We now consider the analagous question for harmonic oscillator path integrals. In this
case, one ﬁnds, in precise analogy with the ﬁnitedimensional case,
Z
0
[j] = e
−(i/m¯ h)
_
+∞
−∞
dt
_
+∞
−∞
dt
j(t)G
0
(t, t
)j(t
)/2
, (4.20)
where G(t, t
) is a Green’s function (the inverse) of the operator −∂
2
t
−ω(t)
2
,
(−∂
2
t
−ω(t)
2
)G
0
(t, t
) = δ(t −t
) . (4.21)
For the ﬁnitetime path integral, the Green’s function that appears here would have
been determined by the Dirichlet (zero) boundary conditions at t
i,f
to be the Green’s
function with
G
0
(t
f
, t
) = G
0
(t, t
i
) = 0 . (4.22)
In the present case (inﬁnite time interval), the relevant Green’s function is implicitly
determined by an i prescription. In particular, for the timeindependent harmonic
oscillator with constant frequency ω
0
one ﬁnds
G
0
(t, t
) =
1
2iω
0
e
−iω
0
t −t

. (4.23)
Once we know the generating functional, we can use it to calculate npoint functions.
In particular, for the twopoint function one has
< 0T (ˆ x(t)ˆ x(t
))0 > =
_
¯ h
i
δ
δj(t)
¯ h
i
δ
δj(t
)
Z
0
[j]
_
j=0
=
i¯ h
m
G
0
(t, t
) . (4.24)
That the timeordered product < 0T (ˆ x(t)ˆ x(t
))0 > is a Green’s function can of course
also be veriﬁed directly in the standard operator formulation of quantum mechanics
(Exercise). Likewise, higher npoint functions can be expressed in terms of products of
2point functions (Wick’s theorem again).
40
4.3 Perturbative Expansion and Generating Functionals
As we have seen, the generating functional Z[j] encodes all the information about the
npoint functions G(x
1
, . . . , x
n
). However, this is only useful if supplemented by a
prescription for how to calculate Z[j]. Since in practice the only path integrals that
we can do explicitly are Gaussian path integrals and their close relatives, the question
arises how to reduce the evaluation of Z[j] to an evaluation of Gaussian integrals.
This is achieved via a perturbative expansion of the path integral around a quadratic
(Gaussian) action. The (assumed) small perturbation expansion parameter can be a
coupling constant λ, as in
ˆ
H =
ˆ
H
0
+λ
ˆ
W , (4.25)
with
ˆ
H
0
a harmonic oscillator Hamiltonian. In this case one is studying a path integral
counterpart of standard quantum mechanical perturbation theory.
Alternatively, the small parameter could be Planck’s constant ¯ h itself. In this case one
is interested in evaluating the path integral
_
D[x] e
(i/¯ h)S[x]
(4.26)
for ¯ h →0 as a power series in ¯ h by expanding the action around a classical solution x
c
(t).
This is a semiclassical expansion of the path integral, a counterpart of the standard
WKB approximation of quantum mechanics.
Both physically and technically (in one case one has a small parameter in front of a part
of the action, the perturbation, in the other a large parameter 1/¯ h in front of the entire
action) these two expansions appear to be quite distinct. Calculationally, however,
they are rather similar, since in both cases the path integral can be reduced to a series
expansion in derivatives of the generating functional of the quadratic theory. This is
immediate for the perturbative λexpansion (which we will consider in this section)
but requires a minor bit of trickery for the ¯ hexpansion (hence the excursion into the
stationary phase approximation for ﬁnitedimensional integrals in section 4.5).
For deﬁniteness, we will assume that the perturbation
ˆ
W arises from a velocityindependent
perturbation of the potential
V (x) = V
0
(x) +λW(x) , (4.27)
with V
0
(x) a harmonic oscillator potential. For more complicated perturbations one
would have to go back to the phase space path integral, introduce sources for both ˆ x(t)
and ˆ p(t), etc. Then the action takes the form
S[x] = S
0
[x] +λS
I
[x] = S
0
[x] −λ
_
dt W(x(t)) , (4.28)
41
where S
0
[x] is the free action, and S
I
[x] the perturbation or interaction term.
To calculate the path integral, one introduces a sourceterm for the free action S
0
[x],
S
0
[x, j] = S
0
[x] +
_
dt j(t)x(t); , (4.29)
and determines Z
fi,0
[j] or Z
0
[j]. Focussing on the latter, the vacuum generating func
tional Z[j] for the perturbed action can then be written in terms of that for the free
action as
Z[j] = Ne
(iλ/¯ h)S
I
[
¯ h
i
δ
δj(t)
]
Z
0
[j] . (4.30)
The normalisation constant, determined by the condition Z[j = 0] = 1, is
N
−1
=
_
e
(iλ/¯ h)S
I
[
¯ h
i
δ
δj(t)
]
Z
0
[j]
_
j=0
(4.31)
This result for Z[j] is manifestly a power series expansion in λ  and the fact that this
perturbative expansion is so straightforward to obtain in the path integral formalism is
one of the reasons that makes the path integral approach to quantum ﬁeld theory so
powerful.
Moreover, using the explicit expression (4.20) for the generating functional Z
0
[j] in terms
of the Green’s functions of the free theory, one sees that the generating functional Z[j],
and thus all the npoint functions of the perturbed theory, are expressed as a series
expansion in terms of the Green’s functions of the unperturbed theory. A graphical
representation of this expansion leads to the Feynman diagram expansion of quantum
mechanics (and quantum ﬁeld theory).
4.4 The Stationary Phase Approximation for Oscillatory Integrals
The integrals of interest in this section are oscillatory integrals of the kind
_
+∞
−∞
dx e
(i/¯ h)f(x)
. (4.32)
The basic tenet of the stationary phase approximation of such integrals is that for small
¯ h, ¯ h →0, the integrand oscillates so rapidly that the integral over any small xinterval
will give zero unless one is close to a critical point x
c
of f(x), f
(x
c
) = 0, for which to
ﬁrst order around x
c
there are no oscillations. This suggests that for ¯ h →0 the integral
is dominated by the contribution from the neighbourhood of some critical point(s) x
c
of f(x), and that therefore in this limit the dominant contribution to the integral can
be obtained by a Taylor expansion of f(x) around x = x
c
.
42
To set the stage for this discussion, we will ﬁrst reconsider the Gaussian and Fresnel
integrals of Appendix A from this point of view. There we had obtained the formula
(A.13),
_
+∞
−∞
dx e
iax
2
/2 +ijx
=
¸
2πi
a
e
−ij
2
/2a
(4.33)
by analytic continuation from the corresponding Gaussian integral and/or completing
the square. A (for the following) more instructive way of obtaining this result is to
consider the integral
_
+∞
−∞
dx e
(i/¯ h)q(x)
(4.34)
for some quadratic function of x (q(x) = ax
2
/2 + jx + c, say). Such a function has a
unique critical point x
c
(x
c
= −j/a), and since q(x) is quadratic and q
(x
c
) = 0 one can
write q(x) as
q(x) = q(x
c
) +
1
2
(x −x
c
)
2
q
(x
c
) . (4.35)
Thus the integral is
_
+∞
−∞
dx e
(i/¯ h)q(x)
= e
(i/¯ h)q(x
c
)
_
+∞
−∞
dy e
(i/¯ h)q
(x
c
)y
2
/2
= e
(i/¯ h)q(x
c
)
¸
2πi¯ h
q
(x
c
)
(4.36)
This reproduces and generalises (4.33).
It will be more convenient to move the factor
√
2πi¯ h to the left, so that the result is
1
√
2πi¯ h
_
+∞
−∞
dx e
(i/¯ h)q(x)
=
1
_
q
(x
c
)
e
(i/¯ h)q(x
c
)
. (4.37)
We see that a Fresnel integral is determined exactly by the contribution from the critical
point and the quadratic ﬂuctuations around it. Another way of saying this is that, as
we will see, for a Gaussian/Fresnel integral the stationary phase approximation is exact.
Now let us consider the oscillatory integral
I[f] =
1
√
2πi¯ h
_
+∞
−∞
dx e
(i/¯ h)f(x)
(4.38)
for a general function f(x). Let x
c
denote the critical point of f(x) with the smallest
absolute value, and assume that x
c
is isolated. Then we can expand f(x) around x
c
as
f(x) = f(x
c
) +
1
2
(x −x
c
)
2
f
(x
c
) +R(x −x
c
) , (4.39)
where the rest R(x −x
c
) is at least cubic in (x −x
c
). Thus the integral is
I[f] =
1
√
2πi¯ h
e
(i/¯ h)f(x
c
)
_
+∞
−∞
dx e
(i/¯ h)[f
(x
c
)(x −x
c
)
2
/2 +R(x −x
c
)]
. (4.40)
43
The stationary phase approximation (also known as the saddle point approximation) to
the integral amounts to ignoring the higher than quadratic terms encoded in R(x −x
c
)
and leads to the approximate result
1
√
2πi¯ h
_
+∞
−∞
dx e
(i/¯ h)f(x)
≈
1
_
f
(x
c
)
e
(i/¯ h)f(x
c
)
. (4.41)
To justify this approximation, one needs to show that the contributions due to the
remainder R(x −x
c
) are indeed subleading in ¯ h as ¯ h →0. We will establish this below.
Another cause for concern may be that, while we have (handwavingly) argued that the
dominant contribution to the oscillatory integral should arise from a small neighbour
hood of the critical point(s), in order to arrive at (4.41) we have taken the integral not
over a small neighbourhood of x
c
but, quite on the contrary, over (−∞, +∞).
To justify this, consider the contribution to the integral from an interval [a, b] without
critical points. In that interval, one can change the integration variable from x to f(x)
(this would not be allowed if f(x) had a critical point in the interval). Then it is easy
to see (e.g. by an integration by parts) that the integral
_
b
a
dx e
(i/¯ h)f(x)
= O(¯ h) . (4.42)
Thus regions without critical points contribute O(¯ h) terms to the integral. On the
other hand, (4.41) shows that, regardless of what the neglected terms do, there are
some contributions to the total integral which are of order O(¯ h
1/2
). Thus these must
be due to the contributions from (integrals over arbitrarily small neighbourhoods of)
the critical points. As these are dominant relative to the O(¯ h) contributions as ¯ h →0,
the diﬀerence betwen integrating over such a neighbourhood of the critical point and
integrating over all x is negligible in this limit.
To analyse the contributions due to R(x−x
c
) and to make the dependence of the various
terms on ¯ h more transparent, it is convenient to deﬁne the ﬂuctuation variable y not as
(x −x
c
), as was implicitly done in (4.36), but via
x = x
c
+
√
¯ hy . (4.43)
This has the eﬀect of making the Gaussian part of the integral independent of ¯ h,
(x −x
c
)
2
/2¯ h = y
2
/2 . (4.44)
Moreover, since R(x−x
c
) is at least cubic, (1/¯ h)R(
√
¯ hy) is now a power series in strictly
positive powers of
√
¯ h,
r(y) ≡ (1/¯ h)R(
√
¯ hy) = ¯ h
1/2
f
(3)
(x
c
)y
3
/3! + ¯ hf
(4)
(x
c
)y
4
/4! +. . . (4.45)
44
even (odd) powers of y appearing with integral (halfintegral) powers of ¯ h. All in all,
we thus have
1
√
2πi¯ h
_
+∞
−∞
dx e
(i/¯ h)f(x)
=
1
√
2πi
e
(i/¯ h)f(x
c
)
_
+∞
−∞
dy e
i[f
(x
c
)y
2
/2 +r(y)]
.
(4.46)
To obtain the higher order corrections to the stationary phase approximation, one can
expand exp ir(y). Remembering that only even powers of y in that expansion give a
nonzero contribution to the yintegral, one concludes that this expresses the integral
as a power series in (integral) powers of ¯ h,
1
√
2πi¯ h
_
+∞
−∞
dx e
(i/¯ h)f(x)
=
1
_
f
(x
c
)
e
(i/¯ h)f(x
c
)
(1 + ¯ h(. . .) + ¯ h
2
(. . .) +. . .) (4.47)
Alternatively, this series can be expressed in terms of generating functions as
1 + ¯ h(. . .) + ¯ h
2
(. . .) +. . . = e
ir(−i∂/∂j)
e
−ij
2
/2f
(x
c
)

j=0
. (4.48)
The stationary phase approximation can be used to calculate the integral with reason
able accuracy for very small ¯ h, ¯ h → 0. However, the series expansion should not be
expected to converge in general, and the series is only an asymptotic series. If contribu
tions from all critical points are included, it is under certain conditions possible to obtain
error estimates, but in practice applications of the stationary phase approximation are
usually restricted to (4.41).
4.5 The SemiClassical Approximation
It is now straightforward to formally apply this stationary phase approximation to
path integrals. The crucial point is that in the path integral formulation of quantum
mechanics the classical limit and the emergence of classical mechanics from quantum
mechanics are extremely transparent: in the limit ¯ h →0, the path integral is dominated
by paths that are critical points of the action, i.e. the classical solutions to the equations
of motion. This should be contrasted with the much more cumbersome eikonal or WKB
semiclassical approximation to the Schr¨ odinger equation.
Essentially all the work has already been done in sections 3.5 and 4.6 and we can be
brief about this here. Instead of the function f(x) we have an action S[x], and the
semiclasical approximation amounts to expanding S[x] around a critical point, i.e. a
classical solution x
c
(t) of the corresponding EulerLagrange equations. This is precisely
the procedure we had already advocated for how to deal with general nonGaussian
path integrals. We now see that this will lead to an expansion of the path integral in
45
powers of ¯ h, the stationary phase approximation to the path integral agreeing with the
harmonic oscillator path integral of section 3.5.
First of all, with
x(t) = x
c
(t) +δx(t) , (4.49)
we expand the action as
S[x] = S[x
c
] +
_
t
f
t
i
dt
δS
δx(t)

x(t)=xc(t)
δx(t)
+
1
2
_
t
f
t
i
dt
_
t
f
t
i
dt
δ
2
S
δx(t)δx(t
)

x(t)=xc(t)
δx(t)δx(t
) +. . .
= S[x
c
] +
1
2
_
t
f
t
i
dt
_
t
f
t
i
dt
δ
2
S
δx(t)δx(t
)

x(t)=xc(t)
δx(t)δx(t
) +. . . (4.50)
For an action of the standard form
S[x] =
m
2
_
t
f
t
i
( ˙ x(t)
2
−
1
m
V (x(t)) (4.51)
the quadratic term is
1
2
_
t
f
t
i
dt
_
t
f
t
i
dt
δ
2
S
δx(t)δx(t
)

x(t)=xc(t)
δx(t)δx(t
) =
m
2
_
t
f
t
i
(δ ˙ x(t)
2
−
1
m
V
(x
c
(t))δx(t)
2
)
(4.52)
As already noted in section 3.5, this is the action of a harmonic oscillator with time
dependent frequency
ω(t)
2
=
1
m
V
(x
c
(t)) . (4.53)
Thus the stationary phase or semiclassical approximation to the path integral is (cf.
(3.47))
_
x(t
f
)=x
f
x(t
i
)=x
i
D[x(t)] e
(i/¯ h)S[x(t); t
f
, t
i
]
≈
¸
m
2πi¯ h(t
f
−t
i
)
¸
Det[−∂
2
t
]
Det[−∂
2
t
−ω(t)
2
]
e
(i/¯ h)S[x
c
]
.
(4.54)
This can be evaluated using either the VVPM or the GY method.
The diﬀerence between (3.47)) and (4.54) is that in the former case one was dealing
with a quadratic action and the result was exact (i.e. the semiclassical approximation
is exact for the harmonic oscillator), whereas here this is really just the semiclassical
approximation. Note also that x
c
in (4.54) refers to a classical solution of the full
equations of motion m¨ x = −V
(x) whereas x
c
in (3.47) is of course a solution of the
harmonic oscillator equation.
If desired, higher order corrections in ¯ h to the semiclassical result can be calculated, as
in (4.48), by using the generating functional of the quadratic theory.
46
4.6 Scattering Theory and the Path Integral
Formal scattering theory can be succinctly described by the evolution operator U
I
(t
f
, t
i
)
in the interaction representation. Thus let the Hamiltonian be (in the simplest case of
twobody potential scattering)
H = H
0
+V (4.55)
where H
0
is the freeparticle Hamiltonian H
0
= p
2
/2m. Then U
I
(t
f
, t
i
) is given by
U
I
(t
f
, t
i
) = e
(i/¯ h)t
f
H
0
e
−(i/¯ h)(t
f
−t
i
)H
e
−(i/¯ h)t
i
H
0
(4.56)
The Møller operators
Ω
±
= lim
t→∓∞
U
I
(0, t) (4.57)
act on free particle states 
k
a
> (plane wave eigenstates of the free Hamiltonian H
0
)
< x
k
a
>=
1
(2π)
3/2
e
i
k
a
.x
, (4.58)
p
a
= ¯ h
k
a
, H
0

k
a
>= E
a

k
a
>, to produce the stationary scattering states ψ
±
a
(x),
Ω
±

k
a
>= ψ
±
a
> . (4.59)
The Smatrix elements S
ab
are the transition amplitudes among the asymptotic in and
out scattering states ψ
±
a
>,
S
ab
=< ψ
−
a
ψ
+
b
> , (4.60)
and therefore they are the matrix elements
S
ab
=<
k
a
S
k
b
> (4.61)
of the scattering operator
S = Ω
†
−
Ω
+
= lim
t
f,i
→±∞
U
I
(t
f
, t
i
) (4.62)
between free particle states.
Such matrix elements can readily be expressed in terms of the path integral. First of
all, we have
S
ab
= lim
t
f,i
→±∞
<
k
a
e
(i/¯ h)t
f
H
0
e
−(i/¯ h)(t
f
−t
i
)H
e
−(i/¯ h)t
i
H
0

k
b
>
= lim
t
f,i
→±∞
e
(i/¯ h)(t
f
E
a
−t
i
E
b
)
<
k
a
e
−(i/¯ h)(t
f
−t
i
)H

k
b
> . (4.63)
47
To pass from the momentum space matrix elements of the evolution operator for H to
the kernel (the position space matrix elements), we perform a double Fourier transform
(this is a special case of (2.61)),
<
k
a
e
−(i/¯ h)(t
f
−t
i
)H

k
b
>=
1
(2π)
3
_
dx
a
_
dx
b
e
i(
k
b
.x
b
−
k
a
.x
a
)
K(x
a
, x
b
, t
f
−t
i
) .
(4.64)
Representing, in the usual way, the kernel by the path integral, we conclude that the
Smatrix elements S
ab
are given by a Fourier transform of the path integral. Either per
turbative or semiclassical expansion techniques, as described earlier on in this section,
can now be employed to obtain series expansions for these Smatrix elements.
4.7 Exercises
1. Using the Heisenberg picture operator equations of motion or Ehrenfest’s theorem,
show that for a (possibly timedependent) harmonic oscillator the timeordered
2point function is a Green’s function,
(−∂
2
t
−ω(t)
2
) < 0T (ˆ x(t)ˆ x(t
))0 >=
i¯ h
m
δ(t −t
) . (4.65)
2. Using the generating functional Z
0
[j] (4.20), express the 4point function
< 0T (ˆ x(t
1
)ˆ x(t
2
))ˆ x(t
3
)ˆ x(t
4
))0 > (4.66)
in terms of sums of products of Green’s functions (2point functions).
3. Calculate the order ¯ h correction to the stationary phase approximation of the
integral
1
√
2πi¯ h
_
+∞
−∞
dx e
(i/¯ h)f(x)
=
1
_
f
(x
c
)
e
(i/¯ h)f(x
c
)
(1 + ¯ h(. . .) +. . .) (4.67)
Note that two diﬀerent terms contribute to the integral at this order.
48
A Gaussian and Fresnel Integrals
A.1 Basic 1dimensional Integrals
The basic Gaussian integral is
I
0
[α] =
_
+∞
−∞
dx e
−αx
2
/2
=
_
2π
α
α ∈ R, α > 0 . (A.1)
This result can for instance be established by the standard trick of squaring the integral
and passing to polar coordinates. The ﬁrst generalisation we will consider is the integral
I
1
[α, j] =
_
+∞
−∞
dx e
−αx
2
/2 +jx
j ∈ R . (A.2)
By completing the square and shifting the integration variable (using translation invari
ance of the measure) one ﬁnds
I
1
[α, j] =
_
+∞
−∞
dx e
−α(x −j/α)
2
/2 +j
2
/2α
=
_
2π
α
e
j
2
/2α
. (A.3)
Another proof of this identity is based on the trivial identity
_
+∞
−∞
dx
d
dx
e
−αx
2
/2 +jx
= 0 . (A.4)
Expanding this out, one ﬁnds
0 =
_
+∞
−∞
dx (−αx +j)e
−αx
2
/2 +jx
= −α
∂
∂j
I
1
[α, j] +jI
1
[α, j] . (A.5)
This diﬀerential equation for I
1
[α, j],
∂
∂j
I
1
[α, j] = (j/α)I
1
[α, j] , (A.6)
called a SchwingerDyson Equation in the quantum ﬁeld theory context, is evidently
solved by
I
1
[α, j] = ce
j
2
/2α
, (A.7)
and the normalisation (initial) condition I
1
[α, 0] = I
0
[α] then leads to
I
1
[α, j] = I
0
[α]e
j
2
/2α
, (A.8)
which agrees with the result (A.3).
49
The other generalisation that will interest us is the oscillatory Fresnel integral
J
0
[a] = I
0
[−ia] =
_
+∞
−∞
dx e
iax
2
/2
. (A.9)
It can be obtained by noting that the basic Gaussian integral is well deﬁned for α ∈ C
with Re(α) > 0, and that it continues to be well deﬁned for Re(α) → 0 provided that
Im(α) = 0. It follows that
J
0
[a] =
¸
2πi
a
=
_
2π
a
e
iπ/4
. (A.10)
A useful way to remember this result is to relate it to the integral I
0
with a phase
depending on the sign of a,
J
0
[a] = I
0
[a]e
isign(a)π/4
. (A.11)
By the same reasoning, for the integral
J
1
[a, j] = I
1
[−ia, ij] =
_
+∞
−∞
dx e
iax
2
/2 +ijx
(A.12)
one ﬁnds
J
1
[a, j] =
¸
2πi
a
e
−ij
2
/2a
. (A.13)
For an alternative proof of this identity see section 4.5.
As an application, we consider an integral of the kind we encountered in sections 1 and
2, namely
K =
1
2π¯ h
_
+∞
−∞
dp e
(i/¯ h)[py −p
2
/2m]
. (A.14)
By completing the square, shifting the integration variable and using the Fresnel integral
formula, or directly upon using the above result for J
1
[a, j], one ﬁnds
K =
1
2π¯ h
¸
2π¯ hm
i
e
(i/¯ h)(m/2)(y/)
2
=
_
m
2πi¯ h
e
i
¯ h
m
2
y
2
. (A.15)
A.2 Related Integrals
By symmetry considerations, one has
_
+∞
−∞
dx e
−αx
2
/2
x
2m+1
= 0 . (A.16)
50
The integrals of even powers x
2m
can be calculated by relating them to the integrals I
0
or I
1
. For example, one has
_
+∞
−∞
dx e
−αx
2
/2
x
2m
= (−2)
m
∂
m
∂α
m
_
+∞
−∞
dx e
−αx
2
/2
= (−2)
m
∂
m
∂α
m
I
0
[α] . (A.17)
This can be evaluated to give
_
+∞
−∞
dx e
−αx
2
/2
x
2m
= (−2)
m
√
2π
∂
m
∂α
m
α
−1/2
=
√
2π(2m−1)!!α
−(2m+1)/2
. (A.18)
Alternatively, one can write
_
+∞
−∞
dx e
−αx
2
/2
x
n
=
_
∂
n
∂j
n
_
+∞
−∞
dx e
−αx
2
/2 +jx
_
j=0
=
_
∂
n
∂j
n
I
1
[α, j]
_
j=0
. (A.19)
More generally, one has
_
+∞
−∞
dx e
−αx
2
/2
F(x) =
_
F
_
∂
∂j
_
I
1
[α, j]
_
j=0
(A.20)
and
_
+∞
−∞
dx e
−αx
2
/2 +F(x)
=
_
e
F
_
∂
∂j
_
I
1
[α, j]
_
j=0
(A.21)
for any function F(x). This explains why the central object of interest for integrals with
Gaussian weight is the generating function I
1
[α, j].
A.3 Gaussian Integrals and Determinants
The simplest 2dimensional Gaussian integral is
_
+∞
−∞
dx
_
+∞
−∞
dy e
−αx
2
/2 −βy
2
/2
=
2π
√
αβ
. (A.22)
This generalises to
_
+∞
−∞
dx
_
+∞
−∞
dy e
−(αx
2
+βy
2
+ 2γxy)/2
=
2π
_
αβ −γ
2
, (A.23)
as can be seen by completing the square.
51
In terms of the matrix
A =
_
α γ
γ β
_
(A.24)
the exponent of the integrand can be written as A
ab
x
a
x
b
with x
a
= (x, y) and the above
identity reads
_
+∞
−∞
d
2
x e
−A
ab
x
a
x
b
/2
=
(
√
2π)
2
√
det A
=
_
det
A
2π
_
−1/2
. (A.25)
This way of writing this result generalises to ddimensional integrals, d ≥ 2. Let A be
a symmetric, positive real (d ×d)matrix. Then one has
_
+∞
−∞
d
d
x e
−A
ab
x
a
x
b
/2
=
_
det
A
2π
_
−1/2
. (A.26)
Likewise, for the oscillatory version of this integral one has
_
+∞
−∞
d
d
x e
iA
ab
x
a
x
b
/2
=
_
det
A
2πi
_
−1/2
. (A.27)
Thus the task of calculating Gaussian integrals and their various relatives and descen
dants has been reduced to the purely algebraic task of calculating determinants of
matrices.
The proof of the fundamental identities (A.26,A.27) is remarkably simple and will be
left as an exercise. The strategy is to ﬁrst prove them for A diagonal (this is trivial) and
to then use the fact that any real symmetric A can be diagonalised by an orthogonal
transformation, combined with the rotation invariance of the measure d
d
x.
52
B Infinite Product Identities
A useful identity, ubiquitous in determinant and index calculations, is the following
inﬁnite product representation for (sin x)/x,
sin x
x
=
∞
n=1
_
1 −
x
2
n
2
π
2
_
(B.1)
The validity of this formula is plausible as the right hand side has exactly the same zeros
and identical behaviour as x →0 as the left hand side. It can be proved rigorously in a
variety of ways. The easiest (and closest in spirit to the rough argument in the previous
sentence) is to extend (sin x)/x to a holomorphic function (sin z)/z in the complex plane
and factorise using the MittagLeﬄer pole expansion  see e.g. section 7 of
G. Arfken and H. Weber, Mathematical Methods for Physicists, Elsevier (2005).
The single identity above can be used to generate an inﬁnite number of other identities,
in the spirit of Euler. The ﬁrst nontrivial one of these results from comparing the
quadratic term in the expansion of the left hand side,
sin x
x
=
x −x
3
/6 +. . .
x
= 1 −
x
2
6
+. . . (B.2)
with the quadratic term arising from the right hand side,
∞
n=1
_
1 −
x
2
n
2
π
2
_
= 1 −
∞
n=1
x
2
n
2
π
2
+. . . (B.3)
which provides a cute proof of the famous identity
∞
n=1
1
n
2
=
π
2
6
. (B.4)
There are a host of other occasionally useful inﬁnite product identities (none of which,
however, will be used in these notes). For example, closely related inﬁnite product
representations of other trigonomotric and hyperbolic functions are
sinh x
x
=
∞
n=1
_
1 +
x
2
n
2
π
2
_
(B.5)
cos x =
∞
n=1
_
1 −
x
2
(n −1/2)
2
π
2
_
(B.6)
cosh x =
∞
n=1
_
1 +
x
2
(n −1/2)
2
π
2
_
(B.7)
53
There is also the more mysterious identity
sin x
x
=
∞
n=1
cos
_
x
2
n
_
. (B.8)
As another example, Euler’s identity for the Riemann ζfunction
ζ(s) =
∞
n=1
1
n
s
(B.9)
is
ζ(s) =
∞
n=1
1
1 −p
−s
n
, (B.10)
where p
n
= 2, 3, 7, 11, . . . is the sequence of prime numbers. This identity provides the
cornerstone of the relation between number theory and complex analysis.
C Equivalence of the VVPM and GY Formulae for the Fluctua
tion Determinant
The article of Kleinert and Chervyakov cited in section 3.5 contains a nice proof of the
classical identity (3.63) expressing the equivalence of the VVPM (3.49) and Gelfand
Yanglom (3.51) results for the ratio of ﬂuctuation determinants.
Consider the classical solution x
c
(t) with x
c
(t
f,i
) = x
f,i
. This solution can equally well
be regarded as a function of the initial position x
i
and velocity ˙ x
i
,
x
c
= x
c
(t, x
i
, ˙ x
i
) . (C.1)
Writing this as a linear combination of two linearly independent solutions f
1
(t) and
f
2
(t) of the oscillator equation,
x
c
(t, x
i
, ˙ x
i
) = x
i
f
1
(t) + ˙ x
i
f
2
(t) (C.2)
and imposing the conditions x
c
(t
i
) = x
i
and ˙ x
c
(t
i
) = ˙ x
i
, one ﬁnds
x
c
(t
i
) = x
i
⇒ f
1
(t
i
) = 1 f
2
(t
i
) = 0
˙ x
c
(t
i
) = ˙ x
i
⇒
˙
f
1
(t
i
) = 0
˙
f
2
(t
i
) = 1 . (C.3)
This shows that f
2
(t) = F
ω
(t) is the GY solution,
F
ω
(t) =
∂x
c
(t, x
i
, ˙ x
i
)
∂ ˙ x
i
. (C.4)
At this point, one can either use directly the HamiltonJacobi relation
∂S[x
c
]
∂x
i
= −p
i
= −m˙ x
i
, (C.5)
54
to deduce (3.63). Or one can evaluate explicitly the classical action in terms of F
ω
(t)
and f
1
(t) ≡ G
ω
(t) (the “dual” GY solution, which plays a role analogous to the GY
solution when one considers periodic or antiperiodic boundary conditions instead of
zero boundary conditions). Thus one has
x
c
(t) = x
i
G
ω
(t) + ˙ x
i
F
ω
(t) . (C.6)
and therefore, in particular,
x
f
= x
i
G
ω
(t
f
) + ˙ x
i
F
ω
(t
f
)
˙ x
f
= x
i
˙
G
ω
(t
f
) + ˙ x
i
˙
F
ω
(t
f
) . (C.7)
Since only boundary terms contribute to the classical action, it is simply given by (3.48)
S[x
c
] =
m
2
[x
f
˙ x
c
(t
f
) −x
i
˙ x
c
(t
i
)] . (C.8)
Using (C.7) to eliminate ˙ x
i,f
in favour of x
i,f
, and using the fact that the Wronskian of
F
ω
(t) and G
ω
(t) is tindependent,
(F
ω
˙
G
ω
−
˙
F
ω
G
ω
)(t
i
) = −1 = (F
ω
˙
G
ω
−
˙
F
ω
G
ω
)(t
f
) , (C.9)
one ﬁnds that the classical action, as a function of x
i
and x
f
, is
S[x
c
] =
m
2F
ω
(t
f
)
[
˙
F
ω
(t
f
)x
2
f
+G
ω
(t
f
)x
2
i
−2x
i
x
f
] . (C.10)
It follows that
∂
2
S[x
c
]
∂x
i
∂x
f
= −
m
F
ω
(t
f
)
, (C.11)
as we set out to show.
55
3 Gaussian Path Integrals and Determinants 3.1 3.2 3.3 3.4 3.5 3.6 3.7 Preliminary Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The Free Particle and the Normalisation Constant N . . . . . . . . . . . . . . . . The Free Particle: Fluctuation Determinant and Mode Expansion . . . . . . . . The Harmonic Oscillator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Gaussian Path Integrals and Determinants: the VVPM and GY Formulae . . . . Some Comments on the Regularisation of Determinants . . . . . . . . . . . . . . Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
22 22 23 25 26 28 32 35 37 37 39 41 42 45 47 48 49 49 50 51 53
4 Generating Functionals and Perturbative Expansions 4.1 4.2 4.3 4.4 4.5 4.6 4.7 The Generating Functional Z[J] . . . . . . . . . . . . . . . . . . . . . . . . . . . Green’s Functions and the Generating Functional for Quadratic Theories . . . . Perturbative Expansion and Generating Functionals . . . . . . . . . . . . . . . . The Stationary Phase Approximation for Oscillatory Integrals . . . . . . . . . . . The SemiClassical Approximation . . . . . . . . . . . . . . . . . . . . . . . . . . Scattering Theory and the Path Integral . . . . . . . . . . . . . . . . . . . . . . . Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
A Gaussian and Fresnel Integrals A.1 Basic 1dimensional Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A.2 Related Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A.3 Gaussian Integrals and Determinants . . . . . . . . . . . . . . . . . . . . . . . . . B Inﬁnite Product Identities
C Equivalence of the VVPM and GY Formulae for the Fluctuation Determinant 54
2
0
Preface
These are notes for part of a course on advanced quantum mechanics given to 4th year physics students. The only prerequisites, however, are a basic knowledge of the Schr¨dinger and Heisenberg pictures of standard quantum mechanics (as well as the willo ingness to occasionally and momentarily suspend disbelief). Thus the material could easily be presented at an earlier stage. I covered the material in ﬁve 3“hour” lectures (1 “hour” = 45 minutes) and this time constraint (there are other topics that I wanted to cover as well in the course) dictated the level of detail (or lack thereof) of the presentation. One of the aims of these lectures was to set the stage for a future course on quantum ﬁeld theory. To a certain extent this motivated the choice of topics covered in these notes (e.g. generating functionals are discussed, while concrete applications of path integrals to nontrivial quantum mechanics problems are not). These notes do not include an introductory section on motivations, history, etc.  such things are best done orally anyway. My own point of view is that the path intgral approach to quantum theories is simultaneously more intuitive, more fundamental, and more ﬂexible than the standard operator  state description, but I do not intend to get into an argument about this. Objectively, the strongest points in favour of the path integral appoach are that • unlike the usual Hamiltonian approach the path integral approach provides a manifestly Lorentz covariant quantisation of classical Lorentz invariant ﬁeld theories; • it is perfectly adapted to perturbative expansions and the derivation of the Feynman rules of a quantum ﬁeld theory; • it allows in a rather straightforward manner to calculate certain nonperturbative contributions (i.e. eﬀects that cannot be seen to any order in perturbation theory) to Smatrix elements. The motivation for writing these notes was that I found the typical treatment of quantum mechanics path integrals in a quantum ﬁeld theory text to be too brief to be digestible (there are some exceptions), while monographs on path integrals are usually far too detailed to allow one to get anywhere in a reasonable amount of time. I have not provided any referenes to either the original or secondary literature since most of the material covered in these notes is completely standard and can be found in many places (the exception perhaps being the GelfandYaglom formula for ﬂuctuation
3
determinants for which some references to the secondary literature are given in section 3.5). No attempt at mathematical rigour (not even the pretense of an attempt) is made in these notes. Updated corrected or expanded versions of these notes will be available at http://www.unine.ch/phys/string/Lecturenotes.html If you ﬁnd any mistakes, or if you have any other comments on these notes, complaints, (constructive) criticism, or also if you just happen to ﬁnd them useful, please let me know.
4
1
1.1
The Evolution Kernel
Review: The timeevolution Operator
The dynamical information about quantum mechanics is contained in the matrix eleˆ ments of the timeevolution operator U (t f , ti ). For a timeindependent Hamiltonian H one has ˆ h U (tf , ti ) = e −(i/¯ )(tf − ti )H (1.1) whereas the general expression for a timedependent Hamiltonian involves the timeordered exponential U (tf , ti ) = T It satisﬁes the evolution equation i¯ h with initial condition U (tf = ti , ti ) = I . As a consequence, Ψ(tf ) = U (tf , ti )ψ(ti ) satisﬁes the timedependent Schr¨dinger equation o i¯ h with initial condition Ψ(ti ) = ψ(ti ) . Another keyproperty of the timeevolution operator is U (tf , ti ) = U (tf , t)U (t, ti ) ti < t < t f . (1.8) (1.7) ∂ ˆ Ψ(tf ) = H(tf )Ψ(tf ) ∂tf (1.6) (1.5) (1.4) ∂ ˆ U (tf , ti ) = H(tf )U (tf , ti ) , ∂tf (1.3) e −(i/¯ ) h
ti tf
ˆ dt H(t )
.
(1.2)
In words: evolution from time ti to tf is the same as evolving from ti to t, followed by evolution from t to tf . 1.2 Definition of the Propagator / Kernel
Our main interest will be in matrix elements of U in the position representation. We will denote these by (the letter K stands for “kernel”) K(xf , xi ; tf − ti ) =< xf U (tf , ti )xi > . 5 (1.9)
12) Another way of saying this. t > . ˆ h ˆˆ h ˆ h ˆ xH (t)x. ti >=< xf xi >= δ(xf − xi ) . tf )] < xf . ti > . xi . (1. ti >= e (i/¯ )ti H xi > (1. tf xi . Rather. tf − ti ) =< xf .14) 2. The kernel satisﬁes the Schr¨dinger equation with respect to (t f . ˆ (1.e. where (in the timeindependent case) h ˆ xi . this state is characterised o by the fact that it is the eigenstate of the Heisenberg picture operator x H (t) at t = ti .10) etc. For any given t. o ˆ h [i¯ ∂tf − H(xf . t >}. one can choose a basis in which xH (t) is diagonal.16) (1.this would o have the opposite sign in the exponent (this is related to the fact that U (t f . tφ > 6 (1.13) Basic Properties of the Kernel Here are some of the basic properties of the Kernel: 1. not ti ). one has tf →ti lim < xf . Note that this is not the Schr¨dinger timeevolution of the state x i > . ti ) satisﬁes the Schr¨dinger equation with respect to t f . pf = (¯ /i)∂xf . ˆ Matrix elements of the evolution operator between other states. (1. t >.3 dxf dxf dxi < ψf xf >< xf U (tf . ti )ψi > = = 1. tf xi . are determined by the kernel through < ψf U (tf .This can also be interpreted as the transition amplitude K(xf . ti >= 0 . h More generally. With properly normalised position eigenstates. say the transition amplitude between an initial state ψ i > and a ﬁnal state ψf >.11) (1. ti )xi >< xi ψi > ∗ dxi ψf (xf )ψi (xi )K(xf . or introducing the states x. xi . t) =< x. Hence they cannot be ˆ ˆ simultaneously diagonalised. for any ket φ > the wave function Ψφ (x. is the following: In general the operators xH (t) and xH (t ) do not commute for t = t . tf − ti ) . xf ). tf xi . t >= e (i/¯ )tH xe −(i/¯ )tH e (i/¯ )tH x >= xx.15) . This is the basis {x. however. i.
xi . It follows from the above two properties of the kernel. tf −ti ) = i¯ δ(tf −ti )δ(xf −xi ) . A key property of the kernel is the convolution property +∞ K(xf .18) where Θ is the Heavyside step function. (1. one frequently also considers the causal propagator or retarded propagator Kr (xf . that the retarded propagator satisﬁes ˆ [i¯ ∂tf −H(xf . tf xi . t)x >< xU (t. ti > . xi . xi . tf − t)K(x.23) = n h ∗ e −(i/¯ )(tf − ti )En ψn (xf )ψn (xi ) .8) of the timeevolution operator by inserting a complete state of states in the form ∞ I= dx x >< x . ti >< xi . (1. ti φ > dxi K(x. t − ti ) (1. ti )xi > ∞ = −∞ ∞ dx < xf U (tf . ti )xi > dx < xf . xi . xi . x. ti > = < xf U (tf . ti ) . This follows from the property (1. t − ti )Ψφ (xi .24) 7 . (1. pf = (¯ /i)∂xf .20) (for t arbitrary subject to the condition t f > t > ti ). t)U (t. with o Ψφ (x.17) 3. (1.19) h h h Thus the retarded propagator is a Green’s function for the Schr¨dinger equation. xi . tf − ti ) (1. tf − ti ) = < xf e −(i/¯ )(tf − ti )H xi > ˆ h = < xf e −(i/¯ )(tf − ti )H n >< nxi > n (1. o 4. t >< x. we can also expand the kernel in terms of energy eigenstates ψn (x) =< xn > as ˆ h K(xf . t) = = dxi < x. Since we can restrict our attention to evolution forwards in time.21) −∞ One then ﬁnds < xf .is a solution of the Schr¨dinger equation.22) = −∞ which is the claimed result. 5. (1. Finally. tf − ti ) = −∞ dx K(xf . xi . txi . Θ(x) = 1 for x > 0 and Θ(x) = 0 for x < 0. txi . tf x. tf )]Kr (xf . tf − ti ) = Θ(tf − ti )K(xf . and Θ (x) = δ(x).
by Fourier transform).26). and satisﬁes the convolution property o (1. with Hamiltonian H = p 2 /2m. one thus ﬁnds = K0 (xf . Prove the basic integral identity (A. it is straightforward to determine the kernel K = K0 and verify explicitly all the above properties.27) Note that the exponent has the interpretation as the “classical action”.20) and the free particle Schr¨dinger equation (1. +∞ −∞ a b A dd x e −Aab x x /2 = det 2π −1/2 . xi .26) 2π¯ −∞ h Using the Fresnel integral formulae from Appendix A.28) for the evolution kernel of the free particle.27) and (1. Verify the results (1. 2 (tf − ti ) (1. 1 h < xp >= √ e (i/¯ )px (1. S0 [xc ] = m (xf − xi )2 . xi .15). By inserting a complete set of momentum eigenstates p >. 3. (1.29) for Gaussian integrals (Aab is a real symmetric positive matrix).e.27) is normalised as in (1. 1. and calculate the integral +∞ a b a dd x e −Aab x x /2 + Ja x (1.14). i. Verify that (1.30) −∞ by completing the square (remember that A ab is invertible). tf − ti ) = i m m ¯ eh 2 2πi¯ (tf − ti ) h (xf −xi )2 (tf −ti ) (1. one ﬁnds h p2 K0 (xf . as the action S0 of the free particle evaluated on the classical path x c (t) satisfying the free equations of motion and the boundary conditions x c (tf. 8 .25) 2π¯ h (i.4 Example: The Free Particle For the free particle. tf − ti ) = < xf e −(i/¯ )(tf − ti )ˆ /2m xi > +∞ = −∞ 2 h dp < xf p > e −(i/¯ )(tf − ti )p /2m < pxi > +∞ 2 1 h dp e (i/¯ )[p(xf − xi ) − (tf − ti )p /2m] .1.e.i ) = xf. 2.5 Exercises 1.28) This at this stage rather mysterious fact has a very natural explanation from the path integral point of view. (1.i .
. this allows us to make some progress.2) Here k = 0. . y.1 Towards the Path Integral Representation of the Kernel From the Kernel to the ShortTime Kernel and Back In general. things simplify in the limit → 0. (2. . First of all we note that we can write (once again in the timeindependent case. it is a diﬃcult (if not impossible) task to ﬁnd a closed form expression for the kernel. ) and. tf ] into N equal timeintervals [tk . As we will see below. We can now insert N − 1 resolutions of unity at times tk . equivalently N → ∞. dxk ][ k=0 k=1 K(xk+1 .2 2. . tf − ti ) = [ where xf = xN and xi = x0 . xk . = tk+1 − tk )] . N −1 into the above expression for the kernel to ﬁnd N −1 ∞ −∞ N −1 K(xf .5) . xk . it is not diﬃcult to see that in the limit N → ∞ we only need to know this shorttime kernel K(xk+1 . But for ﬁnite N . k = 1. .3) Now this expression holds for any N . tf − ti ) = lim [ N →∞ dxk ][ k=0 k=1 K(xk+1 . ) to order . the argument for this would be the following: one writes e x = e x/N N = (1 + x/N + O(1/N 2 ))N 9 (2. (2. However. . (2.1) We think of this as dividing the timeinterval [t i .4) is determined by the shorttime kernel K(x k+1 . N − 1 and we identify tf = tN and ti = t0 . xk . the convolution property allows us to reduce the determination of the ﬁnitetime kernel to that of the shorttime (or even inﬁnitesimal time) kernel K(x. but the argument works in general) ˆ h h < xf e −(i/¯ )(tf − ti )H xi >=< xf  e −(i/¯ ) (tf −ti ) ˆ N H N xi > . = tk+1 − tk = (tf − ti )/N . xi . as we will see later on. ) as 2. the kernel is still diﬃcult to calculate. tk+1 ] of length .2 The Trotter Product Formula and the Dirac ShortTime Kernel Formally. the kernel N −1 ∞ −∞ N −1 K(xf . . If everything in sight were commuting (instead of operators). xi . (2. . xk . In this limit. = (tf − ti )/N )] → 0.
and compares with the formula e x = lim (1 + x/N )N N →∞ (2.g. is trickier. for instance. e. for the Coulomb potential!). Thus. The identity holds.6) to conclude that in the limit N → ∞ the subleading O(N −2 ) terms in (2. to establish an analogous result for (unbounded) operators requires some functional analysis. 2m (2. we will gloss over these functional analysis complications and proceed with the assumption that it is legitimate to drop the commutator terms (while keeping in mind that this assumption is not valid e.g. via a binomial expansion or by showing that d lim (1 + x/N )N = lim (1 + x/N )N N →∞ dx N →∞ .9) To justify dropping these O( 2 ) commutator terms. 10 . The case of interest. xk .5) can indeed be dropped. V ] which should not become too singular. We will now assume that the Hamiltonian is of the standard form H = T (p) + V (x) = p2 + V (x) .10) The identity (2.1 Of course. 1 N = lim ? N →∞ ˆ ˆ e A/N e B/N N (2. Once again. on the common domain of A and B.7) We are thus interested in determining the kernel h ˆ ˆ K(xk+1 . using the BakerCampbellHausdorﬀ ˆ ˆ formula one has h ˆ ˆ h ˆ h ˆ e −(i/¯ ) (T + V ) = e −(i/¯ ) T e −(i/¯ ) V + O( 2 ) . one needs some control ˆ ˆ over the operator [T . things simplify quite a bit. ) =< xk+1 e −(i/¯ ) (T + V ) xk > . unbounded operators. (2. ˆ ˆ ˆ ˆ even though T and V are noncommuting operators.8) Provided that we can justify dropping terms of O( 2 ). Indeed. because their commutator is 2 [T . (2. provided that both are selfadjoint operators that are bounded from below. V ]. however. Concretely. This identity is not diﬃcult to prove for bounded operators.6) can be proved directly. T and V commute up to order 2 . what one needs is the validity of the Trotter product formula ˆ ˆ ˆ ˆ e A + B = e A/N + B/N ˆ ˆ ˆ ˆ for A = T and B = V .
xk ). xk ) = ˙ 2.With the above assumptions.14) as a discretised timederivative. ) = = where the Lagrangian L is L(xk .12) < xpk >= √ 2π¯ h and calculate +∞ K(xk+1 . to order we can write the shorttime kernel as (2. ﬁrst obtained by Dirac in 1933.3 mx2 ˙k m(xk+1 − xk )2 − V (xk ) → − V (xk ) .15).11) h ˆ h ˆ K(xk+1 . xk )] (2.4) to obtain an expression for the ﬁnitetime kernel. In the exponential. ˙ We can implement the Legendre transformation explicitly by performing the Gaussian (Fresnel) integral over pk (see Appendix A) to ﬁnd K(xk+1 . tk+1 − tk dt (2. where we once had operators. This iteration of Dirac’s result is due to Feynman (1942). xk . xk . if we identify xk+1 − xk = xk+1 − xk dxk → .13) or the conﬁguration space expression (2.e. We can use either the phase space expression (2. i. ) = = −∞ +∞ −∞ h ˆ h ˆ dpk < xk+1 e −(i/¯ ) T pk >< pk e −(i/¯ ) V xk > dpk +∞ −∞ k h < xk+1 pk >< pk xk > e −(i/¯ ) ( 2m + V (xk )) p2 = 1 2π¯ h h dpk e (i/¯ ) [pk (xk+1 −xk ) − H(pk . ˆ To diagonalise the operator T we introduce a complete set of momentum eigenstates pk > with 1 h e (i/¯ )pk x (2.15) The Path Integral Representation of the Kernel Having obtained the above explicit formula for the shorttime kernel in terms of the Lagrangian. xk ) ˙ e h 2πi¯ h (2. ) =< xk+1 e −(i/¯ ) T e −(i/¯ ) V xk > . xk . 11 .16) 1 2π¯ h +∞ −∞ k h dpk e (i/¯ )[pk (xk+1 − xk ) − ( 2m + V (xk ))] p2 m (i/¯ ) L(xk . the Lagrangian! Indeed. we can no go back to (2. the exponent takes the classical form p k xk − H(pk . 2 2 2 (2.13) This is a lovely result. we now encounter the Legendre transform of the classical Hamiltonian.
ti ] ˙ In the same spirit. h With this notation. This is a consequence of the fact that each shorttime propagator contains one momentum integral whereas the position integrals are inserted between the shorttime propagators and the two endpoints x i and xf are not integrated over. tf − ti ) = = N →∞ lim [ k=1 dxk ][ k=0 N/2 N −1 dpk (i/¯ ) ]e h 2π¯ h xk+1 −xk N −1 k=0 [pk − H(pk .In this way we arrive at N −1 ∞ −∞ N −1 ∞ −∞ ∞ −∞ K(xf . In the → 0 or N → ∞ limit. xk ) = ˙ k=0 dt L(x(t).see below) curve. xk )] (2.18) and xk = x(ti + k ) . x(t))] ˙ (2.23) = N h D[x(t)]e (i/¯ )S[x(t)] .17) N →∞ lim m 2πi¯ h [ k=1 h dxk ]e (i/¯ ) N −1 k=0 L(xk .20) With this interpretation the exponents in the integrand of the kernel can be written as lim [pk k=0 N −1 xk+1 − xk − H(pk . we can now write the kernel as a path integral. we think of the pk as deﬁnig a curve p(t) in momentum space such that p0 = p(ti ) N −1 N →∞ pk = p(ti + k ) . introducing the notation N −1 N →∞ ∞ −∞ ∞ −∞ lim [ k=1 N −1 k=0 dxk ] = x(tf )=xf x(ti )=xi D[x(t)] (2. xi . tf . xk ) ˙ Note that in the phase space representation of the kernel there is always one more momentum than position integral. we think of them as deﬁning a curve x(t) with endpoints x(ti ) = xi x(tf ) = xf (2. 12 . we formally now write the integrals as integrals over paths. xk )] = tf ti tf ti dt [p(t)x(t) − H(p(t). x(t))] ˙ (2. K(xf . (2. any two successive points being joined by a straight line.21) N →∞ lim L(xk . That is. x(t)) = S[x(t). we interpret the points x k as deﬁning a continuous (but almost certainly nowhere diﬀerentiable . xi . (2.22) N →∞ lim [ dpk ] = 2π¯ h D[p(t)/2π¯ ] . tf − ti ) = x(tf )=xf x(ti )=xi x(tf )=xf x(ti )=xi D[x(t)] h D[p(t)/2π¯ ]e (i/¯ ) h tf ti dt [p(t)x(t) − H(p(t).19) Likewise.
3. Hence the paths entgering the above sum/integral are typically nowhere diﬀerentiable.This is an integral over all paths x(t) with the speciﬁed boundary conditions and. This should however not lead one to believe that the path integral representation of the kernel also enjoys this invariance. Indeed diﬀerentiability would require existence of a ﬁnite limit of (xk+1 − xk )/ as → 0. To pass to the second line. The ﬁrst line. The derivation of the path integral for a Hamiltonian with a velocity dependent potential.24) (we have replaced by (tf − ti )/N ) whose sole purpose in life is to make the combined expression N times the integral well deﬁned. But xk+1 and xk are independent variables.25) 13 . is more subtle (the discretised version requires a “midpoint rule” for the Hamiltonian) and will not be discussed here. this cannot possibly be true since it is well known that under a canonical transformation any Hamiltonian can be mapped to zero (the HamiltonJacobi transformation) and hence into any other Hamiltonian. while the kernel depends nontrivially on the Hamiltonian. Evidently. Some remarks: 1. as limits of piecewise linear continuous paths these paths are continuous but not diﬀerentiable. then. all paths p(t) with t ∈ [t i . tf ]. The measure appears to be an inﬁnitedimensional analogue of the canonical Liouville measure. We could have absorbed much of the prefactor into the deﬁnition of the measure by writing (2. N = lim N →∞ mN 2πi¯ (tf − ti ) h N/2 (2. Finally. but we will not open this Pandora’s box. things like x(t) require some (perhaps stochastic ˙ or probabilistic) interpretation. in the ﬁrst version. Indeed. is valid for general Hamiltonians H(p. and hence there is no reason for the diﬀerence x k+1 − xk to go to zero as → 0. 4. a phase space path integral. a conﬁguration space path integral. xk ) ˙ (2.17) as K(xf . ﬁnite and equal to the left hand side (provided that all our functional analysis assumptions are satisﬁed). N is formally an inﬁnite normalisation constant. x) = T (p) + V (x) (possibly timedependent). xi . we used the explicit quadratic form T = p2 /2m to perform the Gaussian integral over p. tf − ti ) = m 2πi¯ h 1/2 N →∞ N −1 ∞ −∞ lim [ k=1 m 2πi¯ h 1/2 h dxk ]e (i/¯ ) N −1 k=0 L(xk . As mentioned above. such as the magnetic interaction (p − A)2 /2m. The latter is of course invariant under canonical transformations. 2.
xi .33) On the other hand. ti ) = −∞ dx < xU (tf . (2. ti ) = Z(tf − ti ) = Tr e −(i/¯ )(tf − ti )H . such that N x(tf )=xf x(ti )=xi h D[x(t)]e (i/¯ )S[x(t)] = (2. h Evaluating the trace in a basis of energy eingenstates n >.32) h e −(i/¯ )(tf − ti )En . ti )x >= 14 −∞ dx K(x. (2. or the partition function for short. and hence the relation between D[x] ˆ and D[x]. In the timeindependent case. (2.31) (2. 2. one obtains +∞ +∞ Z(tf . this becomes ˆ h Z(tf . ti ) = Tr U (tf . as we will see in section 3. one ﬁnds Z(tf − ti ) = n (2. (2.and deﬁning x(tf )=xf x(ti )=xi ˜ D[x(t)] = lim [ N →∞ N −1 k=1 ∞ −∞ m 2πi¯ h 1/2 dxk ] .29) one deﬁnes the quantum mechanical partition function.30) and thus the quantum mechanical and thermal partition function are formally related by continuation of the time interval (t f − ti ) to the imaginary value tf − ti = −i¯ β .4 The Path Integral Representation of the Partition Function m 2πi¯ (tf − ti ) h x(tf )=xf x(ti )=xi h ˆ D[x(t)]e (i/¯ )S[x(t)] . tf − ti ) = m 2πi¯ h 1/2 x(tf )=xf x(ti )=xi h ˜ D[x(t)]e (i/¯ )S[x(t)] . involving the determinant of a diﬀerential operator. x. (2.27) Another useful normalisation of the measure is. ti ) . evaluting the trace in a basis of position eigenstates x >.28) We will obtain a more informative expresion for N .34) . tf − ti ) . In analogy with the quantum statistical (or thermal) partition function ˆ Z(β) = Tr e −β H (2. Z(tf . as the trace of the time evolution operator. in the next section.26) Then the kernel is K(xf .
x(tf )=x(ti ) ˜ Since we now have an equal number of x. ti ) = h ˜ D[x(t)]e (i/¯ )S[x(t)] . and the integration over x means that one is integrating over all closed loops.5 ¨ Back from the Path Integral to the Schr odinger Equation Since inside the path integral one is dealing with classical functions and functionals rather than with operators.33) shows that. x(tf )=x(ti ) (2. in terms of the measure D[x] (2. This has farreaching consequences (none of which will be explored here).This means that in the discretised expression (2. if we are able to calculate this path integral over all closed loops we should (in the timeindependent case only.26) the partition function reads Z(tf . As an example. 15 . ti ) = N = N dx −∞ x(tf )=x x(ti )=x h D[x(t)]e (i/¯ )S[x(t)] (2.38) which follows from the (presumed) translation invariance of the measure. In particular. fairly simple “classical” manipulations of the path integral can lead to nontrivial quantum mechanical identities.17) for the kernel there are now an equal number of momentum and position integrals. of course) be able to read oﬀ the energy spectrum. setting xf = xi = x means that one is integrating not over all paths but over all closed paths (loops) at x. 2.and pintegrals. +∞ Z(tf .35) h D[x(t)]e (i/¯ )S[x(t)] . for quantum ﬁeld theory this implies a relation between ﬁnite temperature quantum ﬁeld theory in Minkowski space and quantum ﬁeld theory in Euclidean space with one compact (Euclidean “time”) direction. The relation between statistical mechanics and quantum mechanics at imaginary time is rather deep.23).37) of the integration variable. x(tf )=xf x(ti )=xi x(tf )=xf x(ti )=xi h D[x(t)] e (i/¯ )S[x(t)] = h D[x(t)] e (i/¯ )S[x(t) + y(t)] (2. In the continuum version (2. consider the “trivial” statement that the path integral is invariant under an overall shift x(t) → x(t) + y(t) y(ti ) = y(tf ) = 0 (2.36) Comparison with (2.
39) h D[x(t)] δS[x(t)] e (i/¯ )S[x(t)] = 0 . x(tf )=xf δ h e (i/¯ )S[x(t)] = 0 . with y(t) = δx(t) and S[x(t) + δx(t)] = S[x(t)] + δS[x(t)] .44) ∂ < xf . − ∂x dt ∂ x ˙ dt (2. ti >= δx(tf ) In this case. the variation of the action is δS[x(t)] = = tf ti tf ti δx(tf ) = 0 . tf xi . δx(ti ) = 0 This means that we are calculating δ < xf . x(tf )=xf x(ti )=xi D[x(t)] d ∂L ∂L h e (i/¯ )S[x(t)] = 0 . (2.41) D[x(t)] δx(t) x(ti )=xi Consequences derived from such identities are known as “SchwingerDyson equations” in the quantum ﬁeld theory context.43) This is the path integral version of Ehrenfest’s theorem.40) holds for any such δx(t).45) ∂L tf ∂L d ∂L δx(t) + − δx ∂x dt ∂ x ˙ ∂x ˙ ∂L d ∂L dt δx(t) + p(tf )δx(tf ) . Since this was so easy. Since the variation of the action with δx(t) vanishing at the endpoints gives the EulerLagrange equations. (2. tf xi . δS[x(t)] = tf ti dt ∂L d ∂L − δx(t) . − ∂x dt ∂ x ˙ (2. let us generalise this a bit and consider variations of the path which ﬁx x(ti ) but change x(tf ). ti > .46) 16 . this statement reduces to x(tf )=xf x(ti )=xi (2. ∂x dt ∂ x ˙ (2. ∂xf (2.42) and since (2. we deduce that the classical equation of motion are valid inside the path integral (!).40) Sometimes this statement is paraphrased as “the path integral of a total derivative is zero”. (2.Inﬁnitesimally.
and therefore ∂S[xc ] = −H(xf .20). by considering a variation of t f . ∂tf (2. (2. One piece of the dictionary that is still missing is how to o translate matrix elements other than the basic transition amplitude < x f . ti >= ∂xf h ¯ x(tf )=xf x(ti )=xi h D[x(t)] p(tf ) e (i/¯ )S[x(t)] . (2. This essentially completes the (formal) proof of the o equivalence of the path integral description of quantum mechanics and the standard Schr¨dinger representation. xf ) = ˙ d ∂S S[x(t)] = + pf xf ˙ dtf ∂tf (2.23) K(xf .6 Convolution. pf = ) ∂tf i ∂xf x(tf )=xf x(ti )=xi h D[x(t)] e (i/¯ )S[x(t)] = 0 .51) one obtains a path integral of the statement (1. (2. ti ] (2. This leads to the standard HamiltonJacobi relation ∂S[xc ] pf = .47) ∂xf We deduce that ∂ i < xf . This will be the subject of the next subsection. ti > into the path integral language. How is this encoded in the path integral representation (2. and now we have employed path integral manipulations to pass back from the path integral to the Schr¨dinger representation.48) which is nothing other than the familiar statement that in the position representation one has h ∂ ¯ . tf xi . a crucial property of the kernel is the convolution property (1. tf − ti ) = N x(tf )=xf x(ti )=xi h D[x(t)]e (i/¯ )S[x(t). tf xi . Correlation Functions and TimeOrdered Products As we saw in section 1 and above. xi .49) pf = ˆ i ∂xf In the same way. tf . pf ) .where p(tf ) ≡ pf is the canonical momentum at t = tf .50) (modulo the EulerLagrange equations). and using L(xf . i¯ h ∂ h ∂ ¯ ˆ x ˆ − H(ˆf . 2. (2.52) We had initially derived the path integral from the standard description.53) 17 .15) that the kernel satisﬁes the Schr¨dinger o equation (Exercise).
The most natural operators to consider in the present context are products of Heisenberg operators xH (t1 )ˆH (t2 ) . tf . ˆ we will from now on drop the subscript H on Heisenberg operators. 2 18 .57) Thus the insertion of an operator in the usual prescription corresponds to the insertion of a classical function in the path integral formulation.54) (. to simplify the notation. By elementary manipulations we ﬁnd +∞ < xf . ti >= x x(tf )=xf x(ti )=xi h D[x(t)] x(t0 ) e (i/¯ )S[x(t). . ti > x(t0 ) . by considering all paths from xi to x0 and x0 to xf for some ﬁxed x0 = x(t0 ) and then integrating over x0 . This is the recommended attitude at the present level of rigour (better: nonrigour).) = tf t0 t0 (. Taken together. and in line with the replacement of the Hamiltonian operator by the Lagrange function and At least as long as one pretends that N = 1 or that N has somehow been incorporated into the deﬁnition of a suitably regularised path integral. ˆ x We begin with a single operator xH (t0 ) with tf > t0 > ti and. the action S[x(t). ti ] for any tf > t0 > ti since tf ti (2. the statements about the action and the measure imply 2 the convolution property (1. tf . tf ˆ(t0 )xi . but it is also possible to represent matrix elements of operators as path integrals. t0 xi . t0 xi . ti ] obviously satisﬁes S[x(t). t0 ] + S[x(t). ti ] = S[x(t). . . ti ] .) . t0 >< x(t0 ). (2. ti > x dx(t0 ) < xf . tf x(t0 ). t0 >< x(t0 ). As far as the integrand is concerned.20).56) In words: we can perform the integral over all paths from x i to xf . tf xi . Turning this into a statement about path integrals.) + ti (. While this is very charming. we therefore conclude that < xf . . So far we have only discussed the transition amplitude < x f . tf ˆ(t0 )xi . tf . ti >. . t f . tf ˆ(t0 )x(t0 ). ti > = x = −∞ +∞ −∞ dx(t0 ) < xf . and one that we will adopt henceforth. . (2. . using the convolution property. t0 . .of the kernel? We need to consider seperately the integrand and the measure.55) And for the path integral measure we have x(tf )=xf x(ti )=xi +∞ D[x(t)] = −∞ dx0 x(tf )=xf x(t0 )=x0 x(t0 )=x0 D[x(t)] x(ti )=xi D[x(t)] (2..
does the insertion of x(t1 )x(t2 ) into the path integral calculate the matrix elements of x(t1 )ˆ(t2 ) or x(t2 )ˆ(t1 ) or . 2. tf . As a ﬁnal variation of this theme. x(tn ) e (i/¯ )S[x(t).59) This immediately generalises to x(tf )=xf x(ti )=xi h D[x(t)] x(t1 ) . 19 . tf ˆ(t1 )ˆ(t2 )xi . tf T (ˆ(t1 ) . . tf ˆ(t2 )ˆ(t1 )xi . xi . we need to distinguish the two cases t 2 > t1 and t2 < t1 . since x(t 1 ) and x(t2 ) commute. we consider the path integral representation of transition amplitudes between states other than the position eigenstates x. ti ] t2 > t1 t2 < t1 (2. tf . we reverse the above calcuˆ x ˆ x lation. ti ] h D[x(t)] x(t2 )x(t1 ) e (i/¯ )S[x(t).60) Thus the path integral always evaluates matrix elements of timeordered products of operators. ti > . x(tn ))xi .its action. we simply use the formula (1. . ti ] =< xf . . x ˆ (2. no operators. tf . This explains why the ordering inside the path integral is irrelevant and how the ordering ambiguity is resolved on the operator side.13). tf − ti ) . xi .61) and the path integral expression for the kernel K(x f . which in the usual Hamiltonian approach is reﬂected in the noncommutativity of operators. (2.58) = = x(tf )=xf x(ti )=xi < xf . Namely. In order to use the convolution property of the kernel or path integral. ti > . To obtain these. x x (2. but this time with the insertion of x(t 1 )x(t2 ). it also immediately raises a puzzle. tf T (ˆ(t1 )ˆ(t2 ))xi . ti ] =< xf . In this framework. < ψf U (tf .7 Comments We have thus passed from a formulation of quantum mechanics based on the Hamiltonian (and operators and Hilbert spaces) to a Lagrangian description in which there are only commuting objects. ? To answer this question. . ti )ψi >= dxf ∗ dxi ψf (xf )ψi (xi )K(xf . . t >. ti > x x < xf . tf . . the quantum nature. tf − ti ). ti > x x Using the timeordering operator we can summarise these results as x(tf )=xf x(ti )=xi h D[x(t)] x(t1 )x(t2 ) e (i/¯ )S[x(t). Then one ﬁnds x(tf )=xf x(ti )=xi h D[x(t)] x(t1 )x(t2 ) e (i/¯ )S[x(t).
the Lebesgue integral etc. i. In that case. h This provides an extremely intuitive picture of quantum mechanics in which one is never led to ask questions like ”which path did the electron take?”. the answer to that questions is ”no”. In practice. precisely by those paths that are solutions to the classical equations of motion (see the discussion in section 4.6). the signiﬁcance of introducing this symbolic notation should not be underestimated. of course. The attitude regarding path integrals we will adopt in the following will be similar in spirit. However. One might like to draw an analogy here with Riemann integrals. this will be sweeping many important mathematical subtleties under the rug (not the least of which is “does something like what you have called D[x] exist at all?”). Once again. A more pertinent question might be “can one make sense of D[x] exp(i/¯ )S[x] as h something like a measure (or linear functional)?”.it is simply not the right question to ask. and then one determines deﬁnite and indeﬁnite integrals directly. symbolically denoted by the continuum sum (integral) . i.17). but this does not mean that one cannot reliably calculate simple integrals without knowing about these things. 3 20 . one can use that deﬁnition to establish certain basic properties of the resulting inﬁnite sum. We will deduce some properties of path integrals from their “discretised” version and then try to pass as quickly as possible to continuum integrals which will allow us to perform path integrals via one “functional integration” instead of an inﬁnite number of ordinary integrations. but that does not mean that we cannot trust the results that we have obtained. the path ¯ integral will be dominated by contributions form the extrema of the action (by the usual stationary phase approximation). then path integrals might be conceptually interesting but that approach would hardly be an eﬃcient calculational tool. without resorting to the discretised description. but that is irrelevant .3 By the way. One also sees very explicitly how classical mechanics emerges in the limit h → 0. Of course. extrema of the action (solutions to the classical equations of motion) but all possible paths.).arises because one is instructed to consider not only classical paths. weighted by the exponential of (i/¯ ) times the action. Provided that all our assumptions are satisﬁed.e. deﬁned as the limits of an inﬁnite sum. Indeed. in this process one may be glossing over several mathematical subtleties (which will ultimately lead to the development of measure theory.e. one does not calculate integrals that way. this is just another (albeit complicated looking) way of writing the propagator. Rather. if one always had to calculate path integrals as the limit of an inﬁnite number of integrals. It is a shorthand notation for the N → ∞ limit (2. It is important to note that at this point the path integral notation that we have introduced is largely symbolic.
g. 2. determine the operator X. see e.52) (the path integral satisﬁes the Schr¨dinger equation) o in detail. Using the BakerCampbellHausdorﬀ formula.63) 21 . Generalise the derivation of the path integral to systems with d > 1 degrees of freedom. 2m (2. Assume that the Hamiltonian has the standard form H= where x = (x1 .8 Exercises ˆ 1. 2. deﬁned by h ˆ ˆ h ˆ h ˆ h 2ˆ e −(i/¯ ) (T + V ) = e −(i/¯ ) T e −(i/¯ ) V e −( /¯ ) X . .R. (2. Klauder. quantph/0303034. J. xd ) etc. p2 + V (x) . . . Spell out the proof of (2. The Feynman Path Integral: An Historical Slice.62) to order . 3. .For some historical comments on “eﬀorts to give an improved mathematical meaning to Feynman’s path integral formulation of quantum mechanics”.
Indeed. In this section 3 we will deal exclusively with Gaussian integrals. and develop rules for evaluating. a perturbative expansion. such path integrals. (3. actions of the general timedependent harmonic oscillator type S[x] = m 2 tf ti (x(t)2 − ω(t)2 x(t)2 ) .1 Gaussian Path Integrals and Determinants Preliminary Remarks The path integral expression for the propagator we have obtained is K(xf . even in the ﬁnitedimensional case integrals of exponentials of elementary functions can typically be evaluated in closed form only in the purely quadratic (Gaussian. x To quadratic order in the ‘quantum ﬂuctuations’ around the classical solution one then 1 ﬁnds the action (3. 22 .4) 2 ti is to reduce it to an expansion about a quadratic action. i.2) with ω(t)2 = m V (xc (t)).2) perhaps with the addition of a ‘source’ term.e. ¯ In either case. The evaluation of more general integrals in terms of generating functionals will then be one of the subjects of section 4.21). For path integrals. Either the potential is of the harmonic oscillator form plus a perturbation.1) We thus now need to make sense of. xi . In practice this is achieved in one of two ways. Typically.g. S[x] = ˙ (3. the path integrals that can be calculated in closed form are purely quadratic (Gaussian.3) Then the strategy to deal with more general path integrals. tf − ti ) = N x(tf )=xf x(ti )=xi h D[x(t)]e (i/¯ )S[x(t)] . This turns out to lead to an expansion of the path integral in a power series in h. whereas more general integrals are then evaluated ‘perturbatively’ in terms of a generating function as in (A. the Gaussian path integral can be evaluated in reasonably closed form and the complete path integral is then deﬁned in terms of the generating functional associated with this quadratic action. ˙ (3. V (x) = V0 (x) + λW (x). to an action of the form tf m ( x(t)2 − V (x(t)) . Fresnel) integrals. a semiclassical expansion. (3. Fresnel) case. the situation is quite analogous. corresponding e. and one deﬁnes the path integral via a power series expansion in λ. Or one deﬁnes the path integral by an expansion around a classical solution x c (t) of the equations of motion m¨ = −V (x). S[x] → S[x] + j(t)x(t) .3 3. For a general system described by an action S[x] an exact evaluation of the path integral is certainly too much to hope for.
5) h D[x(t)]e (i/¯ )S0 [x(t)] .3. with Lagrangian L0 (x(t). tf − t i (3. S0 [xc ] = m (xf − xi )2 . Here the starting action is already quadratic. we can use this calculation to determine the overall normalisation of the path integral from the continuum point of view (since N is universal: it depends only on m and (t f − ti ) and not on the potential V (x)).2 The Free Particle and the Normalisation Constant N We are now ready to tackle our ﬁrst path integral. xc (t) = xi + and a quantum ﬂuctuation y(t). but this expansion will have the added beneﬁt of eliminating the boundary conditions x(ti. one ﬁnds S0 [xc + y] = S0 [xc ] + m 2 tf ti xf − x i (t − ti ) . Plugging this into the action.27). xi .12) . As described above. We thus split any path satisfying these boundary conditions into the sum of the classical path x c (t). x(t)) = ˙ We thus need to calculate K0 (xf . For obvious reasons we will consider the simplest dynamical system. ˙ 2 (3.f ) = xi.8) (3.6) and we know that we should ﬁnd the result (1. tf − ti ) = (i/¯ ) m h 2 m e 2πi¯ (tf − ti ) h (xf −xi )2 (tf −ti ) (3.11) where S0 [xc ] is the classical action.7) Since in this case we already know the result.9) (3.28). already given in (1. xi . namely the free particle. (3.f from the path integral.10) y(t)2 ˙ (3. tf − ti ) = N x(tf )=xf x(ti )=xi m x(t)2 . with y(t) satisfying the homogeneous boundary conditions y(ti ) = y(tf ) = 0 . x(t) = xc (t) + y(t) . K0 (xf . 2 (tf − ti ) 23 (3. the general strategy is to expand the paths around a solution to the classical equations of motion.
We will discuss this determinant in more detail momentarily.28) is such that it normalises the free particle Gaussian ﬂuctuation integral to unity. as in (3.18) 24 . (3.13) We thus automatically produce the classical action in the exponent. instead of integrating over all paths x(t) with the speciﬁed boundary conditions we now integrate over all paths y(t) with the boundary conditions (3. (3. tf − ti ) = e (i/¯ )S0 [xc ] N ¯ D[y(t)]e h y(ti )=y(tf )=0 i m tf 2 ti y(t)2 ˙ (3. (3. we note that comparison of (3.17) ˆ We also see that the normalised path integral measure D[y(t)] introduced in (2.7) and (3. xi .16) I denote the (functional) determinant of an operator by Det to distinguish it from a standard ﬁnitedimensional determinant. We thus ﬁnd that the path integral expression for the kernel becomes h K0 (xf .7).15) we deduce that what this path integral formally calculates is the determinant of the 2 diﬀerential operator (−∂t ).27).14) Comparing with the fundamental Fresnel integral formula (A.There is no linear term in y(t) because we are expanding around a critical point of the action S0 [x(t)]. To get a handle on the path integral over y(t).13) shows that the inﬁnite normalisation constant N is related to this determinant by N = m m 2 [−∂t ] Det 2πi¯ (tf − ti ) h 2πi¯ h +1/2 . First of all. and there are no higher than quadratic terms because the free particle action itself is quadratic. (3. In the path integral.10). ¯ D[y(t)]e h y(ti )=y(tf )=0 i m tf 2 ti 2 y(t)(−∂t )y(t) = Det m [−∂ 2 ] t 2πi¯ h −1/2 . ˆ ¯ D[y(t)]e h y(ti )=y(tf )=0 i m tf 2 ti 2 y(t)(−∂t )y(t) = 1 . +∞ −∞ a b A dd x e iAab x x /2 = det 2πi −1/2 . we integrate by parts in the action to obtain ¯ D[y(t)]e h y(ti )=y(tf )=0 i m tf 2 ti y(t)2 = ˙ y(ti )=y(tf )=0 ¯ D[y(t)]e h i m tf 2 ti 2 y(t)(−∂t )y(t) (3.
The properly normalised eigenfunctions are 2 t − ti yn (t) = sin nπ . (3. As in the ﬁnitedimensional case. this determinant is formally deﬁned as the product of the eigenvalues. (3. the action becomes m 2 tf ti 2 y(t)(−∂t )y(t) = m 2 λn c2 .25) (tf − ti )2 25 .19) with 2 −∂t yn (t) = λn yn (t) tf ti yn (t)ym (t) = δm. It will be useful for later to know explicitly the eigenvalues λ n . with the result ¯ D[y(t)]e h y(ti )=y(tf )=0 i m tf 2 ti 2 y(t)(−∂t )y(t) = n +∞ −∞ ¯ dcn e h −1/2 i m 2 2 n λn cn = n m λn 2πi¯ h .23) It is this inﬁnite product which we have deﬁned as the determinant.22) and thus the path integral reduces to an inﬁnite product of ﬁnitedimensional Gaussian integrals. (3. y(t) = n cn yn (t) . the linearly independent solutions are y n (t) with n ∈ N and the corresponding eigenvalues are n2 π 2 λn = .20) yn (ti ) = yn (tf ) = 0 . That this is indeed what the path integral gives rise to can be seen more 2 explicitly by expanding y(t) in normalised eigenmodes y n (t) of the operator (−∂t ). +∞ D[y(t)] = y(ti )=y(tf )=0 n −∞ dcn . (3. (3. Then the path integral over all y(t) becomes an integral over the c n .n (3. n n (3.3 The Free Particle: Fluctuation Determinant and Mode Expansion Now let us return to the deﬁnition of the determinant of a diﬀerential operator.3.24) tf − t i tf − t i Since y−n (t) = −yn (t).21) Thus the expansion of y(t) in terms of eigenmodes is tantamount to diagonalising the operator (this statement is obviously true more generally). In terms of this decomposition.
The classical path and action are (see Exercise 1) xc (t) = xi S[xc ] = sin ω0 (tf − t) sin ω0 (t − ti ) + xf sin ω0 (tf − ti ) sin ω0 (tf − ti ) m ω0 (x2 + x2 ) cos ω0 (tf − ti ) − 2xi xf i f 2 sin ω0 (tf − ti ) m 2 tf ti 2 2 y(t)(−∂t − ω0 )y(t) . and thus Det −1/2 [−∂t ] is zero.We thus have 2 Det[−∂t ] = For more comments on determinants and regularised determinants see sections 3. one ﬁnds S[xc + y] = S[xc ] + (3. we decompose the path into a classical path xc (t) and the ﬂuctuation y(t).26) We have now accumulated all the techniques we need to tackle a more interesting example. (3. using the result (3.30) Thus the path integral we need to calculate is h K(xf . with action S[x] = m 2 tf ti 2 (x(t)2 − ω0 x(t)2 ) .28) Following the same strategy as for the free particle.27) N Det = 2πi¯ h 2πi¯ (tf − ti ) h n2 π 2 (t − ti )2 n=1 f ∞ (3.31) This is once again a straightforward Gaussian (Fresnel) integral.4 The Harmonic Oscillator 2 This is clearly inﬁnite. tf − ti ) = e (i/¯ )S[xc ] N ¯ D[y(t)]e h y(ti )=y(tf )=0 i m tf 2 ti 2 2 y(t)(−∂t − ω0 )y(t) (3.5 and 3. 3. −1/2 m m 2 [−∂t ] (3. and then perform the Gaussian integral over y(t). but this is compensated by the inﬁnite normalisation constant in precisely such a way that one obtains the ﬁnite result (3.6.17).32) . tf − ti ) = = m 2πi¯ (tf − ti ) h m 2πi¯ (tf − ti ) h 26 m 2 Det 2πi¯ [−∂t ] (i/¯ )S[xc ] h h m 2 e 2 Det 2πi¯ [−∂t − ω0 ] h 2 Det[−∂t ] (i/¯ )S[xc ] h 2 2 e Det[−∂t − ω0 ] (3. xi .29) Expanding the action arund xc (t).17). namely the timeindependent harmonic oscillator. K(xf . ˙ (3. determine the classical action and the (still quadratic) action for y(t). xi . and thus one ﬁnds.
33) Even though this may not be obvious from this expression. This shows (see also Appendix B) that this is an inﬁnite product representation of f (x) = sin x .38) mω0 h e (i/¯ )S[xc ] 2πi¯ sin ω0 (tf − ti ) h . x (3. Thus the ratio of determinants is 2 Det[−∂t ] 2 2 Det[−∂t − ω0 ] = n λn 2 λn − ω 0 ω2 (1 − 0 ) λn 1/2 −1/2 = n . λn n π n=1 n=1 ∞ (3. m ω0 (x2 + x2 ) cos ω0 (tf − ti ) − 2xi xf i f 2 sin ω0 (tf − ti ) (3. To calculate this ratio 2 of determinants. the ﬁnal result for the ratio of determinants is 2 Det[−∂t ] 2 2 = Det[−∂t − ω0 ] ω0 (tf − ti ) .39) 27 . First of all.34) The function f (x) = ∞ n=1 (1 − x2 ) n2 π 2 (3. the result should be well deﬁned and ﬁnite. the eigenvalues µn of the operator (−∂t − ω0 ) are µn = λn − ω0 . It is easy to see that this result reduces to that for the free particle in the limit ω 0 → 0.37) and our ﬁnal compact result for the propagator of the harmonic oscillator is K(xf . we observe ﬁrst of all that if the eigenvalues of the operator (−∂ t ) are 2 2 2 λn .35) is an even function of x with f (0) = 1 and simple zeros at x = ±nπ. tf − ti ) = = with S[xc ] = m 2πi¯ (tf − ti ) h ω0 (tf − ti ) (i/¯ )S[xc ] e h sin ω0 (tf − ti ) (3. using the explicit expression for the λ n .36) Therefore. the result is actually an elementary function. sin ω0 (tf − ti ) (3. xi . This is indeed the case. one has ∞ 2 ω0 ω 2 (tf − ti )2 (1 − )= (1 − 0 2 2 ) . (3.Since in writing the above we have taken into account the normalisation factor.
(3. (2.46) The resulting path integral is still Gaussian. and exactly the same strategy as above can be used to show that the path integral result for the kernel of the evolution operator is (cf.5 Gaussian Path Integrals and Determinants: the VVPM and GY Formulae Frequently one encounters timedependent harmonic oscillators with Hamiltonian H(t) = and the quadratic action S[x] = m 2 tf ti 1 2 mω(t)2 2 p + x 2m 2 (3. m (3. This is a Fresnel integral. h 2 − ω(t)2 ] e Det[−∂t (3. ˙ (3. (Exercise) 3.41) n=0 1 En = hω0 (n + 2 ) ¯ (3. we also immediately obtain an expression for the partition function by setting xi = xf = x and integrating over x.32)) h ˆ < xf T e −(i/¯ ) tf ti ˆ dt H(t) x >= i m 2πi¯ (tf − ti ) h 28 2 Det[−∂t ] (i/¯ )S[xc ] . expanding the general action S[x] = m 2 tf ti (x(t)2 − ˙ 2 V (x(t)) m (3.33)) h e −(i/¯ )(tf − ti )En . (3.Given the above result for the kernel.43) (x(t)2 − ω(t)2 x(t)2 ) .40) Z(tf − ti ) = where (cf. and the result is Z(tf − ti ) = This can be expanded as ∞ 1 2i sin ω0 (tf −ti ) 2 . (3.47) . one ﬁnds the action (3.44) with ω(t)2 = 1 V (xc (t)) .45) to second order around a classical solution x c (t) of the equations of motion m¨ = x −V (x).44) For instance.42) are the harmonic oscillator energy levels.
In order to evaluate the result for the propagator (3.Morette (VVPM) formula 2 m Det[−∂t ] 1 ∂ 2 S[xc ] (3. GelfandYaglom (GY) formula that states that 2 Det[−∂t − ω(t)2 ] Fω (tf ) = 2] tf − t i Det[−∂t (3. ˙ ˙ 2 (3. for a ddimensional quantum system. to evaluate the ratio of quantum ﬂuctuations in this manner. tf . The former is rather straightforward provided that one can ﬁnd the classical solution.47). but apparently much less well known.39) which gives − mω0 ∂ 2 S[xc ] = .49) =√ − 2 2πi¯ (tf − ti ) Det[−∂t − ω(t)2 ] h ∂xi ∂xf 2πi¯ h More generally. one only needs to know the classical action.51) 29 . xc (ti ) = xi and xc (tf ) = xf . as we did in the timeindependent case. In the case of the harmonic oscillator with constant frequency. ˙ S[xc ] = m 2 tf ti (xc (t)2 − ω(t)2 xc (t)2 ) = ˙ m [xf xc (tf ) − xi xc (ti )] . I will brieﬂy describe these below.Here xc (t) now denotes the classical harmonic oscillator solution with the given boundary condition. Notice that. This is a nontrivial but standard and wellknown result. the 2nd derivative of the classical action would be replaced by the ddimensional VVPM determinant ∂ 2 S[xc ] ∂ 2 S[xc ] → det ∂xi ∂xf ∂xµ ∂xν i f . ∂xi ∂xf sin ω0 (tf − ti ) (3. agreement between the VVPM formula and the result we obtained in (3.Pauli .50) Alternatively. Fortunately there are two elegant shortcuts to calculating this ratio of determinants which are ﬁnitedimensional in nature and do not require the calculation of a fuctional determinant.38) can be immediately veriﬁed from the classical action (3. one needs to determine the classical action and the ratio of ﬂuctuation determinants. One can for instance use the useful and remarkable result that the ratio of determinants can be calculated from the classical action via the Van Vleck . and the ﬂuctuation determinants are to be calculated for zero (Dirichlet) boundary conditions.48) The calculation of the ratio of determinants would be complicated if one tried to calculate these determinants directly. An integration by parts shows that the classical action can be calculated in terms of the boundary values of x c (t) and xc (t) at t = ti . S[xc ] is the classical action. instead of the VVPM result one can use the equally remarkable.
60) 30 .52) with the initial conditions Fω (ti ) = 0 Thus one has the simple result m 2πi¯ (tf − ti ) h 2 Det[−∂t ] = 2 Det[−∂t − ω(t)2 ] ˙ Fω (ti ) = 1 . can be expressed in terms of the zero mode (solution with zero eigenvalue) of the same operator with the GY boundary conditions (3. f (t)2 (3. as an example we consider the constant frequency harmonic oscillator. (3.37). the GY solution is evidently Fω (t) = f (t)/f˙(ti ) with Fω (tf ) = f (tf ) .58) If.59) tf ti dt (3. Some comments: 1. the GY solution can be constructed from it as t Fω (t) = f (ti )f (t) In particular. on the other hand.53) m 2πi¯ Fω (tf ) h (3.54) It is quite remarkable that the ratio of ﬂuctuation determinants. ˙ f(ti ) (3. which involves the 2 product over all eigenvalues of the operator −(∂ t + ω(t)2 ) (with zero boundary conditions). The solution of the classical equations of motion satisfying the GY boundary conditions is evidently 1 Fω0 (t) = sin ω0 (t − ti ) .57) in perfect agreement with the result (3. (3.56) ω0 and the GY formula predicts 2 2 Det[−∂t − ω0 ] sin ω0 (tf − ti ) .where Fω (t) is the solution of the classical harmonic oscillator equation 2 (∂t + ω(t)2 )Fω (t) = 0 (3.55) ω0 Thus 1 Fω0 (tf ) = sin ω0 (tf − ti ) (3. f (t) is any solution of the harmonic oscillator equation with f (ti ) = 0. = 2] ω0 (tf − ti ) Det[−∂t (3. Once again. one has Fω (tf ) = f (ti )f (tf ) dt ti 1 f (t )2 1 . If f (t) is a solution of the harmonic oscillator equation with f (t i ) = 0.53).
61) solves the freeparticle (ω = 0) equations of motion with the GY boundary conditions.65) Thus Fω.λ (t) (3.51) in the more suggestive form 2 Fω (tf ) Det[−∂t − ω(t)2 ] = .λ (ti ) = 1 . Comparison with the VVPM formula gives the relation m ∂ 2 S[xc ] =− . Techniques and Applications of Path Integration. Cambridge University Press (1985). reprinted in S.62). The denominator tf − ti in the GY formula (3.2. so that one can write (3. in appropriate places): Let Fω.66) 31 .λ (tf ) = 2 0 iﬀ λ is an eigenvalue of the operator (−∂ t − ω(t)2 ) with zero Dirichlet boundary conditions (because then Fω. Selected Erice Lectures.λ (ti ) = 0 .λ (t) = λFω. It works roughly as follows (to make this argument more precise one should insert words like “Fredholm operators” etc. in the form (3.λ (tf ) = 2 − λ] F0. Coleman. Aspects of Symmetry.64) with the GY boundary conditions Fω.λ (t) is the corresponding eigenfunction). but this by itself provides no insight into the reason for the validity of either the VVPM or the GY formula. ∂xi ∂xf Fω (tf ) (3. The claim is now that 2 Det[−∂t − ω(t)2 − λ] Fω. 2] Det[−∂t F0 (tf ) (3.51) can be interpreted as F 0 (tf ). This proof is also reproduced in L.S.λ (t) has the property that Fω. (3. has been given by S. Coleman in his Erice lectures on The uses of instantons (Appendix A). Schulman. and Fω.λ (t) be the solution of the equation 2 (−∂t − ω(t)2 )Fω. Dover (2005). where F0 (t) = t − ti (3.λ (tf ) Det[−∂t (3.62) 3. A slick proof of the GY formula. ˙ Fω.63) With a bit of eﬀort this purely classical (in the sense of classical mechanics) formula can be proved directly via HamiltonJacobi theory (see Appendix C).0 (t) is what we called Fω (t) above.
For example. To establish this claim. operators for which the trace exists). for λ suﬃciently large. Here. Kleinert. where I is 32 . Moreover. A proof along these lines is also outlined in Appendix A of R.62) for λ = 0. this implies the GY result (3. Det is again deﬁned to be the product of all eigenvalues. exactly the same is true of the right hand side. A245 (1998) 345357. 1987) 3. and we have treated them in quite a formal and cavalier way. In particular.n of (−∂t ) (for the same reason). Another elegant continuum (i.5 and 34. Rajaraman. Lett. provided that ω(t) is a bounded function of t. 3. and operators of the form I + K. which concludes the proof of the identity (3. nondiscretised) proof of the GY formula (3.e. There are certain inﬁnitedimensional operators for which the deﬁnition of a determinant poses no real problem. In particular. and a simple pole at each eigenvalue λ 0. Solitons and Instantons (North Holland 1982.3. one can ignore ω(t). and this implies that the ratio is equal to 1 identically. for most intents and purposes. In the following. By the remark above (3. one concludes that the ratio of the two sides is an analytic function of λ that goes to 1 in any direction except perhaps along the positive real axis.66). trace class operators K (i. 1989). Yet another proof can be assembled from sections 3. Simple Explicit Formulas for Gaussian Path Integrals with TimeDependent Frequencies. Phys.2 of J. Quantum Field Theory and Critical Phenomena (Oxford Science Publications.e. one considers the left and right hand sides as functions of the complex variable λ. The left hand side is a meromorphic function of λ with a simple 2 2 zero at each eigenvalue λn of (−∂t − ω(t)2 ) (an eigenvalue λn of (−∂t − ω(t)2 ) is a zero 2 2 eigenvalue of (−∂t − ω(t)2 ) − λn )).for all λ ∈ C.66). Putting these two observations together. quantph/9803016. I want to (very) brieﬂy indicate some ways to deﬁne these objects in a more satisfactory manner. A. ZinnJustin.6 Some Comments on the Regularisation of Determinants In the above we have repeatedly encountered determinants of inﬁnitedimensional operators. the ratio of the left and the right hand sides has no poles and is therefore an analytic function of λ. λ → ∞. and hence both the left and the right hand side go to 1 in that limit (everywhere except on the real positive line where one can ﬁnd large real eigenvalues).51) can be found in H. Chervyakov.
+ tr(∧n K) .69) ˆ It is this regularised determinant that the normalised path integral with measure D[x] computes (cf. . one can deﬁne the (Fredholm) determinant of an invertible operator I + K with K trace class by DetF (I + K) = 1 + Tr K + Tr(∧2 K) + . even though Det(−∂ t − ω0 ) (the product of the eigenvalues) diverges. ) operator A with eigenvalues λ n . (2. .18)). selfadjoint.47) can be interpreted as deﬁning a regularised functional determinant of 2 the operator (−∂t − ω(t)2 ).72) . most operators appearing in physics are not of this form. usually in the physics literature one adopts a slighly diﬀerent attitude.g. Analogously. the ratio of this determinant and the (equally divergent) free particle determinant gave us a ﬁnite result. ∞ ζ(s) = n=1 n−s . .28. (3. deﬁne the spectral ζfunction of A to be the function ζA (s) = n λ−s . (3. via the socalled ζfunction or heat kernel regularisation. 33 (3. . . in the sense of 2 Detreg [−∂t − ω(t)2 ] := 2 Det[−∂t − ω(t)2 ] 2 Det[−∂t ] (3.3.4 we had seen that. and hence one needs to be more creative.70) However. .67) Here ∧p K (the p’th antisymmetric power of K) denotes the operator implementing the induced action of K on antisymmetric pvectors. 2 2 In section 3. More generally.71) This generalises the deﬁnition of the ordinary Riemann ζfunction ζ(s). Instead of regularising explicitly by means of the free particle determinant (which is nevertheless natural from the path integral point of view) one attempts to deﬁne meaningful individual (instead of ratios of) regularised functional determinants in other ways.68) since this series is absolutely convergent. . ˆ ¯ D[y(t)]e h y(ti )=y(tf )=0 i m tf 2 ti 2 y(t)(−∂t − ω(t)2 )y(t) = Det [−∂ 2 − ω(t)2 ] reg t −1/2 . .the identity operator and K a trace class operator. the ratio of determinants that appears in (3. behave like ﬁnitedimensional linear operators (matrices). For a suﬃciently reasonable (elliptic. For invertible (n × n)matrices M one can write the determinant (with K = I − M ) as det M = det(I + K) = 1 + tr K + tr(∧2 K) + . However. n (3. e. (3.
74) λn = e −ζA (0) .75) but.73) ζA (s) = − n to conclude that ζA (0) = − Formally. . Likewise. one has n n log λn = − log λn . we construct the spectral ζfunction ∞ ζA (s) = n=1 nπ T −2s = T π 2s ζ(2s) = e 2s log(T /π) ζ(2s) (3. with a pole only at s = 1. therefore. To deﬁne the determinant.78) To illustrate this method. one diﬀerentiates once. .76) as well as (we will need this below) 1 ζ (0) = − 2 log 2π . let us go back to the calculation of the free particle determi2 nant in section 3. the left hand side is as illdeﬁned as the right hand side because ξA (s). the ordinary Riemann ζfunction. . (3. under favourable circumstances the spectral ζfunction ζ A (s) can be extended to a meromorphic function of s which is holomorphic at s = 0. However. will be convergent only for Re(s) suﬃciently large and positive (and not at s = 0). n−s log λn . (3. In particular. There we had seen that the eigenvalues of the operator A = −∂ t (with Dirichlet boundary conditions) are λn = n2 π 2 T2 n = 1.80) 34 . (3. converges only for Re(s) > 1.79) where T = tf − ti . 2.75) can be used to deﬁne the ζfunction regularised determinant of A via Detζ A = e −ζA (0) . one then has cute results like 1 ζ(0) = − 2 1 ζ(−1) = − 12 ζ(−2) = 0 ( ( ( ∞ n=1 ∞ n=1 ∞ n=1 1 ) n ) n2 ) (3. at this point.To see the relation between the spectral ζfunction and the determinant. (3. in that case it is well known that ζ(s) can be analytically continued to a meromorphic function of s in the entire complex splane. as it stands. as it stands. n (3.77) Analogously.3. (3. and then (3.
37).87) holds in the inﬁnitedimensional case is discussed at length in (warning: not for the faint of heart) M.. 2 Detζ [−∂t ] = 2T .29). ˙ ˙ 2 35 (3. Instead of calculating the classical action by integration. .88) . (3. we have 2 2 sin T ω0 Detζ [−∂t − ω0 ] = 2] T ω0 Detζ [−∂t (3.82) Likewise. with ζfunction regularisation.85) in perfect agreement with the previously obtained result (3. sin T ω0 .83) (3. Thus we conclude that. Verify (3. . S[xc ] = m [xf xc (tf ) − xi xc (ti )] . Determinants of elliptic pseudodiﬀerential operators (155 p.7 Exercises 1. S.84) In particular. one important property of the standard determinant in the ﬁnitedimensional case is its multiplicativity det(M N ) = det(M ) det(N ). T ω0 (3. try to determine it in an intelligent way by proving ﬁrst that only boundary terms contribute to the classical action. Kontsevich.4. .and calculate ζA (0) = 2ζ(0) log(T /π) + 2ζ (0) = − log(T /π) − log(2π) = − log(2T ) .86) While this ζfunction regularised deﬁnition of the determinant captures most of the essential properties of the standard determinant. sin ω0 (tf − ti ) (3. 3. some care is required in manipulating these objects. for the harmonic oscillator determinant of section 3. 2 Det[−∂t ] 2 − ω2] = Det[−∂t 0 ω0 (tf − ti ) . hepth/9404046). with λn = one ﬁnds 2 2 Detζ [−∂t − ω0 ] = 2T n2 π 2 2 − ω0 T2 n = 1. To what extent an analogous identity Detζ (AB) = Detζ A Detζ B ? (3. 2.81) (3. Vishik. For example.
Verify the result (3.38) • satisﬁes the harmonic oscillator Schr¨dinger equation (1.14).15) o • and the initial condition (1. Fill in the missing steps in the proof of the identity (3.41).90) and publish it and/or tell me about it. In particular. 3.89) or the GY formula 2 F0 (tf ) Det[−∂t ] 2 − ω(t)2 ] = F (t ) Det[−∂t ω f (3. 4. 36 .63) given in Appendix C. verify (C. Verify that the harmonic oscillator kernel (3.10). 5.2. • and reduces to the free particle propagator for ω 0 → 0. and the expansion (3. Find an elementary proof of the VVPM formula 2 (tf − ti ) ∂ 2 S[xc ] Det[−∂t ] =− 2 Det[−∂t − ω(t)2 ] m ∂xi ∂xf (3.40) for the partition function of the harmonic oscillator.
. This motivates the following discussion of these concepts in the quantum mechanical context. . . . in turn. . . δj(tn ) j(t)=0 δ n Z[j]  . x(tn ))0 > . . . can be expressed as a path integral. x(tn )) = G(x(t1 ). δj(tn ) j(t)=0 (4. . . say). j(tn )G(x(t1 ). . tf T e (i/¯ ) Z[j] = n=0 (i/¯ )n h n! +∞ dt j(t)ˆ(t) x . x(tn )) = < xf . . ti ] . x(tn ))xi . x(tn )) j(t)ˆ(t) 0 > . j(tn )Gf i (x(t1 ). Gf i (x(t1 ). . . . x(tn )) = h ¯ i h ¯ i n n δ n Zf i [j]  δj(t1 ) . . tf T (ˆ(t1 ) . δj(t1 ) . x (4. x(tn )) = < 0T (ˆ(t1 ) . . . These matrix elements can be obtained from a generating functional which. . . . . . . . ti > x ˆ G(x(t1 ). . in particular the perturbative and semiclassical expansion of nonGaussian path integrals. . 37 (4. Before turning to the path integral.1 The Generating Functional Z[J] The main objects of interest in quantum ﬁeld theory are vacuum expectation values of timeordered products of ﬁeld operators.1) dt1 . . we introduce the generating functionals for the correlation functions (npoint functions) Gf i (x(t1 ).3) h = < 0T e (i/¯ ) +∞ −∞ dt From these generating functionals.4 Generating Functionals and Perturbative Expansions In this section we will study some other important properties of the path integral. t > x i i (4.60) and the deﬁnition (4. dtn j(t1 ) . . j(t). The treatment in this section is somewhat more cursory than in the previous sections the main intention is to give a ﬂavour of things to come (in a course on quantum ﬁeld theory. . . . . x(tn )) tf ti h = < xf . . .5) . . . one ﬁnds Zf i [j] = x(tf )=xf x(ti )=xi h D[x(t)] e (i/¯ )S[x(t). Using (2. .2) −∞ dt1 . . 4.4) For Zf i [j] we can easily deduce a path intgral representation. x ˆ They are deﬁned as Zf i [j] = (i/¯ )n h n! n=0 ∞ ∞ tf ti (4.2). the individual correlation functions can evidently be reconstructed by diﬀerentiation. dtn j(t1 ) . tf . .
x(t)) + j(t)x(t)] . t >= e (i/¯ )tH x >= n h ˆ e (i/¯ )tH n >< nx >= h ∗ e (i/¯ )tEn ψn (x)n > . h ˆ x.10) dt [L(x(t). For that we need to project from the states x. . so is the right hand side. tf . .i →±∞ ψ0 (xf )ψ0 (xi ) x(ti )=xi As the left hand side is independent of the boundary conditions x f. Explicitly. t > to the ground state 0 >. Alternatively. one can conclude that G(x(t1 ). ∗ tf. j(t).7) n Thus for the correlation functions (2. Thus the generating functional Z f i [j] is the path integral for the action with a source term. ti ] + tf ti dt j(t)x(t) (4. in the sense of distributions.i = ±∞] = h D[x(t)] e (i/¯ )S[x(t). . x(tn ) e (i/¯ )S[x(t). To accomplish the projection onto the vacuum expectation value. .3). . ˙ −∞ (4. Passing now to the generating functional Z[j] (4. one can for instance replace E n → (1 − i )En for a small positive and then take the limit → 0 at the end.i → ±∞. we now take the limit tf. exp(−itE) → 0 for t → ∞. x(tn ))0 > x ˆ (4. x(tn )) = < 0T (ˆ(t1 ) . . (4. tf . take the limit there. They all amount to the statement that. .n h h e −(i/¯ )tf En + (i/¯ )ti Em × (4. To that end we ﬁrst expand the state x. .8) ∗ x ˆ ψn (xf )ψm (xi ) < nT (ˆ(t1 ) . we can once again rewrite the inﬁnite sum as an exponential in the path integral to deduce (suppressing the dependence on the boundary conditions) that Z[j] ∼ where S[x(t).11) 38 . ti ] = S[x(t). x(tn ))m > . from the ground state.where S[x(t). t > in a basis of eigenstates n > of the Hamiltonian. tf.i imposed at t → ±∞.9) h e (i/¯ )(tf − ti )E0 x(tf )=xf h = lim D[x(t)] x(t1 ) . ti ] . . and then continue back to real time. . In whichever way one proceeds. tf. . j(t). the dominant contribution in that limit coming from the smallest possible value of E. x(tn )) = m. one can “analytically continue” to imaginary time.60) of timeordered products of operators one ﬁnds Gf i (x(t1 ). tf . .e. i. . This can be understood in a number of related ways. Our aim is now to ﬁnd a similar path integral representation for Z[j].6) is the action with a source term (or the action with a coupling of x(t) to the current j(t)). +∞ (4. . j(t).i = ±∞] .
16) j=0 = iG . (4. We therefore obtain the ﬁnal result Z[j] = h D[x(t)] e (i/¯ )S[x(t). Gab Abc = δ a . j(t).The proportionality factor (normalisation constant) is ﬁxed by Z[j = 0] =< 00 >= 1 . h D[x(t)] e (i/¯ )S[x(t). in this section we will determine explicitly the generating functional for the latter.5. 4.12) The right hand side is independent of the boundary conditions imposed at t = ±∞ provided that one chooses the same boundary conditions in the numerator and the denominator. tf. tf.14) e 2πi where Gab is the inverse matrix (“Green’s function”) to A ab .i = ±∞] (4. Thus the c “generating function” is z0 [j] := = a b a dd x e iAab x x /2 + ija x a b dd x e iAab x x /2 ab e −iG ja jb /2 .15) It follows from this that the “2point function” < xc xd >:= is < x c xd > = = 1 ∂ 1 ∂ z0 [j] i ∂jc i ∂jd j=0 1 ∂ −Gad ja z0 [j] i ∂jc cd a b dd x xc xd e iAab x x /2 a b dd x e iAab x x /2 (4.13)) a b a A −1/2 −iGab ja jb /2 dd x e iAab x x /2 + ija x = det . Exercise 1. (4.17) 39 . (4.i = ±∞] .2 Green’s Functions and the Generating Functional for Quadratic Theories As we will reduce the calculation of a general path integral and its generating functional to that of a quadratic theory.g.3 and equation (A. For ﬁnitedimensional Fresnel integrals one has (see e.13) (4.
Thus the “2point function” is the “Green’s function” of the Fresnel integral. In the quantum ﬁeld theory context. . . (4. In particular.. (4. In this case. (4. < x1 . for the timeindependent harmonic oscillator with constant frequency ω 0 one ﬁnds G0 (t.. For the 4point function one ﬁnds (Exercise) < xa xb xc xd >=< xa xb >< xc xd > + < xa xc >< xb xd > + < xa xd >< xb xc > (4. one ﬁnds. . x2n ). in precise analogy with the ﬁnitedimensional case. this result is known as Wick’s Theorem and. t )j(t )/2 . t ) = δ(t − t ) . higher npoint functions can be expressed in terms of products of 2point functions (Wick’s theorem again). 2iω0 (4.20) 2 where G(t. . ti ) = 0 .. < xi2n−1 xi2n > (4. The general result. h Z0 [j] = e −(i/m¯ ) +∞ +∞ −∞ dt −∞ dt j(t)G0 (t. even though a simple result. . x2n >= P (x1 . (4.23) Once we know the generating functional. we can use it to calculate npoint functions.f to be the Green’s function with G0 (tf . . the Green’s function that appears here would have been determined by the Dirichlet (zero) boundary conditions at t i. t ) . expressing the 2npoint functions as a sum over all possible pairings P (x1 . is of enormous practical signiﬁcance in perturbative calculations.18) etc. t ) = 1 −iω0 t − t  e . In particular.. for the twopoint function one has h δ h δ ¯ ¯ Z0 [j] i δj(t) i δj(t ) j=0 i¯ h = G0 (t. t ) = G0 (t. < 0T (ˆ(t)ˆ(t ))0 > = x x 40 .24) m That the timeordered product < 0T (ˆ(t)ˆ(t ))0 > is a Green’s function can of course x x also be veriﬁed directly in the standard operator formulation of quantum mechanics (Exercise). t ) is a Green’s function (the inverse) of the operator −∂ t − ω(t)2 .x2n ) < xi1 xi2 > . the relevant Green’s function is implicitly determined by an i prescription.21) For the ﬁnitetime path integral.19) (there are (2n − 1)!! terms) is also easily deduced from the generating function z 0 [j]. For n odd they are manifestly zero. . 2 (−∂t − ω(t)2 )G0 (t. Likewise. . We now consider the analagous question for harmonic oscillator path integrals.22) In the present case (inﬁnite time interval). Higher moments (npoint functions) can be calculated in a similar way.
(4. In this case one ¯ is interested in evaluating the path integral h D[x] e (i/¯ )S[x] (4.25) ˆ with H0 a harmonic oscillator Hamiltonian. ¯ ¯ This is a semiclassical expansion of the path integral.28) . . Calculationally. a counterpart of the standard WKB approximation of quantum mechanics. however. the perturbation. etc. . However. . ˆ For deﬁniteness. in the other a large parameter 1/¯ in front of the entire h action) these two expansions appear to be quite distinct.26) for h → 0 as a power series in h by expanding the action around a classical solution x c (t).3 Perturbative Expansion and Generating Functionals As we have seen. (4. This is achieved via a perturbative expansion of the path integral around a quadratic (Gaussian) action. the small parameter could be Planck’s constant h itself. In this case one is studying a path integral counterpart of standard quantum mechanical perturbation theory. as in ˆ ˆ ˆ H = H0 + λ W . the question arises how to reduce the evaluation of Z[j] to an evaluation of Gaussian integrals. xn ).5). This is immediate for the perturbative λexpansion (which we will consider in this section) but requires a minor bit of trickery for the hexpansion (hence the excursion into the ¯ stationary phase approximation for ﬁnitedimensional integrals in section 4. Then the action takes the form ˆ S[x] = S0 [x] + λSI [x] = S0 [x] − λ 41 dt W (x(t)) . (4.4. For more complicated perturbations one would have to go back to the phase space path integral. we will assume that the perturbation W arises from a velocityindependent perturbation of the potential V (x) = V0 (x) + λW (x) . since in both cases the path integral can be reduced to a series expansion in derivatives of the generating functional of the quadratic theory. . Since in practice the only path integrals that we can do explicitly are Gaussian path integrals and their close relatives. Alternatively. Both physically and technically (in one case one has a small parameter in front of a part of the action. they are rather similar.27) with V0 (x) a harmonic oscillator potential. The (assumed) small perturbation expansion parameter can be a coupling constant λ. the generating functional Z[j] encodes all the information about the npoint functions G(x1 . introduce sources for both x(t) ˆ and p(t). this is only useful if supplemented by a prescription for how to calculate Z[j].
determined by the condition Z[j = 0] = 1. (4. Moreover. Focussing on the latter.and the fact that this perturbative expansion is so straightforward to obtain in the path integral formalism is one of the reasons that makes the path integral approach to quantum ﬁeld theory so powerful. one sees that the generating functional Z[j]. S0 [x. for which to ﬁrst order around xc there are no oscillations.32) The basic tenet of the stationary phase approximation of such integrals is that for small h.29) and determines Zf i.31) This result for Z[j] is manifestly a power series expansion in λ . To calculate the path integral. the integrand oscillates so rapidly that the integral over any small xinterval ¯ ¯ will give zero unless one is close to a critical point x c of f (x). is N −1 = e ¯ δ (iλ/¯ )SI [ h δj(t) ] h i Z0 [j] j=0 (4. and thus all the npoint functions of the perturbed theory. f (xc ) = 0. are expressed as a series expansion in terms of the Green’s functions of the unperturbed theory. the vacuum generating functional Z[j] for the perturbed action can then be written in terms of that for the free action as ¯ δ (iλ/¯ )SI [ h δj(t) ] h i Z0 [j] . This suggests that for h → 0 the integral ¯ is dominated by the contribution from the neighbourhood of some critical point(s) x c of f (x). (4. (4. using the explicit expression (4. and that therefore in this limit the dominant contribution to the integral can be obtained by a Taylor expansion of f (x) around x = x c .30) Z[j] = N e The normalisation constant.20) for the generating functional Z 0 [j] in terms of the Green’s functions of the free theory. j] = S0 [x] + dt j(t)x(t). 4. 42 . and SI [x] the perturbation or interaction term.where S0 [x] is the free action. .0 [j] or Z0 [j]. h → 0. one introduces a sourceterm for the free action S 0 [x]. A graphical representation of this expansion leads to the Feynman diagram expansion of quantum mechanics (and quantum ﬁeld theory).4 The Stationary Phase Approximation for Oscillatory Integrals The integrals of interest in this section are oscillatory integrals of the kind +∞ −∞ h dx e (i/¯ )f (x) .
34) −∞ for some quadratic function of x (q(x) = ax 2 /2 + jx + c. +∞ 2 2πi −ij 2 /2a dx e iax /2 + ijx = e (4.37) h dx e (i/¯ )q(x) = We see that a Fresnel integral is determined exactly by the contribution from the critical point and the quadratic ﬂuctuations around it. as we will see. 2 where the rest R(x − xc ) is at least cubic in (x − xc ). Now let us consider the oscillatory integral 1 I[f ] = √ 2πi¯ h +∞ −∞ h dx e (i/¯ )f (x) (4. (4. A (for the following) more instructive way of obtaining this result is to consider the integral +∞ h dx e (i/¯ )q(x) (4. There we had obtained the formula (A. Another way of saying this is that. we will ﬁrst reconsider the Gaussian and Fresnel integrals of Appendix A from this point of view.33) a −∞ by analytic continuation from the corresponding Gaussian integral and/or completing the square. and since q(x) is quadratic and q (xc ) = 0 one can write q(x) as q(x) = q(xc ) + 1 (x − xc )2 q (xc ) . so that the result is h 1 h e (i/¯ )q(xc ) . It will be more convenient to move the factor 1 √ 2πi¯ h +∞ −∞ √ 2πi¯ to the left. (4.38) for a general function f (x). Then we can expand f (x) around x c as f (x) = f (xc ) + 1 (x − xc )2 f (xc ) + R(x − xc ) .39) (4.36) This reproduces and generalises (4. for a Gaussian/Fresnel integral the stationary phase approximation is exact. Let xc denote the critical point of f (x) with the smallest absolute value. say).13). and assume that xc is isolated.40) 43 . Such a function has a unique critical point xc (xc = −j/a).To set the stage for this discussion. Thus the integral is 1 h I[f ] = √ e (i/¯ )f (xc ) 2πi¯ h +∞ −∞ 2 h dx e (i/¯ )[f (xc )(x − xc ) /2 + R(x − xc )] .33). q (xc ) (4.35) 2 Thus the integral is +∞ −∞ h h dx e (i/¯ )q(x) = e (i/¯ )q(xc ) h = e (i/¯ )q(xc ) +∞ −∞ 2 h dy e (i/¯ )q (xc )y /2 2πi¯ h q (xc ) (4.
since R(x−xc ) is at least cubic.44) √ Moreover. f (xc ) (4.41) To justify this approximation. +∞).43) This has the eﬀect of making the Gaussian part of the integral independent of h.45) . We will establish this below. In that interval. As these are dominant relative to the O(¯ ) contributions as h → 0.g. but via x = xc + √ hy . h (4. ¯ √ r(y) ≡ (1/¯ )R( hy) = h1/2 f (3) (xc )y 3 /3! + hf (4) (xc )y 4 /4! + . there are some contributions to the total integral which are of order O(¯ 1/2 ). ¯ (4. over (−∞. quite on the contrary.42) Thus regions without critical points contribute O(¯ ) terms to the integral. one needs to show that the contributions due to the remainder R(x − xc ) are indeed subleading in h as h → 0. On the h other hand. .41) shows that. h ¯ the diﬀerence betwen integrating over such a neighbourhood of the critical point and integrating over all x is negligible in this limit. (4. consider the contribution to the integral from an interval [a. one can change the integration variable from x to f (x) (this would not be allowed if f (x) had a critical point in the interval). while we have (handwavingly) argued that the dominant contribution to the oscillatory integral should arise from a small neighbourhood of the critical point(s). Then it is easy to see (e. it is convenient to deﬁne the ﬂuctuation variable y not as ¯ (x − xc ). in order to arrive at (4.41) we have taken the integral not over a small neighbourhood of xc but. b] without critical points.The stationary phase approximation (also known as the saddle point approximation) to the integral amounts to ignoring the higher than quadratic terms encoded in R(x − x c ) and leads to the approximate result 1 √ 2πi¯ h +∞ −∞ h dx e (i/¯ )f (x) ≈ 1 h e (i/¯ )f (xc ) . regardless of what the neglected terms do. .36). h ¯ ¯ ¯ 44 (4. ¯ (x − xc )2 /2¯ = y 2 /2 . h (4. Thus these must h be due to the contributions from (integrals over arbitrarily small neighbourhoods of) the critical points. To analyse the contributions due to R(x−x c ) and to make the dependence of the various terms on h more transparent. To justify this. as was implicitly done in (4. (1/¯ )R( hy) is now a power series in strictly h ¯ √ positive powers of h. ¯ ¯ Another cause for concern may be that. by an integration by parts) that the integral b a h dx e (i/¯ )f (x) = O(¯ ) .
47) ¯ ¯ f (xc ) Alternatively. If contributions from all critical points are included. h → 0. o Essentially all the work has already been done in sections 3. This should be contrasted with the much more cumbersome eikonal or WKB semiclassical approximation to the Schr¨dinger equation. ¯ 1 √ 2πi¯ h +∞ −∞ h dx e (i/¯ )f (x) = 1 h e (i/¯ )f (xc ) (1 + h(. The crucial point is that in the path integral formulation of quantum mechanics the classical limit and the emergence of classical mechanics from quantum mechanics are extremely transparent: in the limit h → 0.5 The SemiClassical Approximation It is now straightforward to formally apply this stationary phase approximation to path integrals. and the series is only an asymptotic series. . one can expand exp ir(y).5 and 4.e. i. but in practice applications of the stationary phase approximation are usually restricted to (4.) (4. . . i. . .) + h2 (.even (odd) powers of y appearing with integral (halfintegral) powers of h. ¯ ¯ (4.6 and we can be brief about this here. . a classical solution xc (t) of the corresponding EulerLagrange equations. the series expansion should not be ¯ ¯ expected to converge in general.46) To obtain the higher order corrections to the stationary phase approximation. 4.48) The stationary phase approximation can be used to calculate the integral with reasonable accuracy for very small h. We now see that this will lead to an expansion of the path integral in 45 . one concludes that this expresses the integral as a power series in (integral) powers of h. this series can be expressed in terms of generating functions as 2 1 + h(. = e ir(−i∂/∂j) e −ij /2f (xc ) j=0 . and the semiclasical approximation amounts to expanding S[x] around a critical point. .) + h2 (. . . it is under certain conditions possible to obtain error estimates.) + . the path integral is dominated ¯ by paths that are critical points of the action. However. All in all. ¯ we thus have 1 √ 2πi¯ h +∞ −∞ 1 h h e (i/¯ )f (xc ) dx e (i/¯ )f (x) = √ 2πi +∞ −∞ 2 dy e i[f (xc )y /2 + r(y)] . Remembering that only even powers of y in that expansion give a nonzero contribution to the yintegral. . the classical solutions to the equations of motion. .) + .e. Instead of the function f (x) we have an action S[x].41). . This is precisely the procedure we had already advocated for how to deal with general nonGaussian path integrals. (4.
51) δ2 S m  δx(t)δx(t ) = δx(t)δx(t ) x(t)=xc (t) 2 tf h D[x(t)] e (i/¯ )S[x(t). higher order corrections in h to the semiclassical result can be calculated.powers of h. ti ] ≈ m 2πi¯ (tf − ti ) h The diﬀerence between (3. . as ¯ in (4. the stationary phase approximation to the path integral agreeing with the ¯ harmonic oscillator path integral of section 3. the semiclassical approximation is exact for the harmonic oscillator). Note also that xc in (4. First of all. with x(t) = xc (t) + δx(t) . 2 Det[−∂t − ω(t)2 ] x(ti )=xi (4. tf . (3. x(tf )=xf tf tf m 2 tf ti (x(t)2 − ˙ 1 V (x(t)) m (4.54) is that in the former case one was dealing with a quadratic action and the result was exact (i. If desired.5. whereas here this is really just the semiclassical approximation. + 2 ti δx(t)δx(t ) x(t)=xc (t) ti tf 1 tf δ2 S  δx(t)δx(t ) + .47)) 1 2 dt dt (δ x(t)2 − ˙ 2 Det[−∂t ] h e (i/¯ )S[xc ] . (4. = S[xc ] + dt dt 2 ti δx(t)δx(t ) x(t)=xc (t) ti dt tf (4.e.52) As already noted in section 3. 46 .49) (4.47)) and (4.48).5.54) This can be evaluated using either the VVPM or the GY method.47) is of course a solution of the x harmonic oscillator equation. we expand the action as S[x] = S[xc ] + δS  δx(t) δx(t) x(t)=xc (t) ti tf δ2 S 1 tf dt dt  δx(t)δx(t ) + . . . . by using the generating functional of the quadratic theory.50) For an action of the standard form S[x] = the quadratic term is 1 V (xc (t))δx(t)2 ) m ti ti ti (4. this is the action of a harmonic oscillator with timedependent frequency 1 ω(t)2 = V (xc (t)) .53) m Thus the stationary phase or semiclassical approximation to the path integral is (cf.54) refers to a classical solution of the full equations of motion m¨ = −V (x) whereas xc in (3.
First of all. we have Sab = = tf. Thus let the Hamiltonian be (in the simplest case of twobody potential scattering) H = H0 + V (4. t) t→ ∞ (4.6 Scattering Theory and the Path Integral Formal scattering theory can be succinctly described by the evolution operator U I (tf . (2π)3/2 (4.57) act on free particle states ka > (plane wave eigenstates of the free Hamiltonian H 0 ) < xka >= 1 e ika .i →±∞ (4. Then UI (tf . (4. to produce the stationary scattering states ψ a (x).63) lim 47 . ti ) in the interaction representation. ti ) (4.61) lim UI (tf . ¯ ± Ω± ka >= ψa > . Such matrix elements can readily be expressed in terms of the path integral.55) where H0 is the freeparticle Hamiltonian H 0 = p2 /2m. H0 ka >= Ea ka >.62) lim h h h < ka e (i/¯ )tf H0 e −(i/¯ )(tf − ti )H e −(i/¯ )ti H0 kb > h h e (i/¯ )(tf Ea − ti Eb ) < ka e −(i/¯ )(tf − ti )H kb > .i →±∞ tf.56) (4.60) and therefore they are the matrix elements Sab =< ka Skb > of the scattering operator S = Ω † Ω+ = − between free particle states.x . ti ) is given by h h h UI (tf . − + Sab =< ψa ψb > . (4.4.58) ± pa = hka . ti ) = e (i/¯ )tf H0 e −(i/¯ )(tf − ti )H e −(i/¯ )ti H0 The Møller operators Ω± = lim UI (0. (4.i →±∞ tf.59) The Smatrix elements Sab are the transition amplitudes among the asymptotic in and ± out scattering states ψa >.
in the usual way. show that for a (possibly timedependent) harmonic oscillator the timeordered 2point function is a Green’s function. .65) 2.7 Exercises 1.xa ) K(xa . 3.61)). m (4. the kernel by the path integral. xb . 4. express the 4point function < 0T (ˆ(t1 )ˆ(t2 ))ˆ(t3 )ˆ(t4 ))0 > x x x x in terms of sums of products of Green’s functions (2point functions). we perform a double Fourier transform (this is a special case of (2.) ¯ f (xc ) (4. 48 .64) Representing. (4. Using the generating functional Z 0 [j] (4.67) Note that two diﬀerent terms contribute to the integral at this order. tf − ti ) .xb − ka . we conclude that the Smatrix elements Sab are given by a Fourier transform of the path integral. . as described earlier on in this section. h < ka e −(i/¯ )(tf − ti )H kb >= 1 (2π)3 dxa dxb e i(kb . Calculate the order h correction to the stationary phase approximation of the ¯ integral 1 √ 2πi¯ h +∞ −∞ (4.) + . 2 x x (−∂t − ω(t)2 ) < 0T (ˆ(t)ˆ(t ))0 >= i¯ h δ(t − t ) . Either perturbative or semiclassical expansion techniques. . .20).66) h dx e (i/¯ )f (x) = 1 h e (i/¯ )f (xc ) (1 + h(.To pass from the momentum space matrix elements of the evolution operator for H to the kernel (the position space matrix elements). Using the Heisenberg picture operator equations of motion or Ehrenfest’s theorem. can now be employed to obtain series expansions for these Smatrix elements.
The ﬁrst generalisation we will consider is the integral +∞ I1 [α.7) and the normalisation (initial) condition I 1 [α. j] = ce j /2α . ∂j (A.3). α > 0 . (A. j]. ∂ I1 [α. α (A. j] = −∞ 2 dx e −αx /2 + jx j∈R . 49 .8) which agrees with the result (A. j] .3) Another proof of this identity is based on the trivial identity +∞ dx −∞ d −αx2 /2 + jx e =0 . j] = −∞ 2 2 dx e −α(x − j/α) /2 + j /2α = 2π j 2 /2α e .1) This result can for instance be established by the standard trick of squaring the integral and passing to polar coordinates. ∂j (A.2) By completing the square and shifting the integration variable (using translation invariance of the measure) one ﬁnds +∞ I1 [α.4) Expanding this out. j] . one ﬁnds +∞ 0 = −∞ 2 dx (−αx + j)e −αx /2 + jx = −α ∂ I1 [α.6) called a SchwingerDyson Equation in the quantum ﬁeld theory context. is evidently solved by 2 I1 [α. j] = I0 [α]e j /2α .A A. j] + jI1 [α. (A. (A.1 Gaussian and Fresnel Integrals Basic 1dimensional Integrals The basic Gaussian integral is +∞ I0 [α] = −∞ 2 dx e −αx /2 = 2π α α ∈ R. j] = (j/α)I1 [α.5) This diﬀerential equation for I1 [α. dx (A. 0] = I0 [α] then leads to 2 I1 [α. (A.
J0 [a] = I0 [a]e isign(a)π/4 .The other generalisation that will interest us is the oscillatory Fresnel integral +∞ J0 [a] = I0 [−ia] = 2 dx e iax /2 . ij] = one ﬁnds J1 [a. (A. a (A. or directly upon using the above result for J 1 [a. eh 2 2πi¯ h A.16) 50 .9) −∞ It can be obtained by noting that the basic Gaussian integral is well deﬁned for α ∈ C with Re(α) > 0. one has +∞ −∞ 2 dx e −αx /2 x2m+1 = 0 . namely +∞ 2 1 h K= dp e (i/¯ )[py − p /2m] . and that it continues to be well deﬁned for Re(α) → 0 provided that Im(α) = 0.11) J1 [a. As an application. (A.13) For an alternative proof of this identity see section 4. By the same reasoning. j] = I1 [−ia. we consider an integral of the kind we encountered in sections 1 and 2.2 Related Integrals By symmetry considerations.12) −∞ 2πi −ij 2 /2a e .15) i m y2 m ¯ . j] = 2 dx e iax /2 + ijx (A. for the integral +∞ (A. (A. a (A.14) 2π¯ −∞ h By completing the square.5. one ﬁnds K = = 1 2π¯ h 2π¯ m (i/¯ ) (m/2)(y/ )2 h e h i (A. shifting the integration variable and using the Fresnel integral formula. j]. It follows that J0 [a] = 2πi = a 2π iπ/4 e .10) A useful way to remember this result is to relate it to the integral I 0 with a phase depending on the sign of a.
j] j=0 (A. αβ − γ 2 (A.17) This can be evaluated to give +∞ −∞ √ 2 ∂m dx e −αx /2 x2m = (−2)m 2π m α−1/2 ∂α √ = 2π(2m − 1)!!α−(2m+1)/2 . j] j=0 (A. dy e −αx /2 − βy /2 = √ αβ (A.20) and +∞ −∞ 2 F dx e −αx /2 + F (x) = e I1 [α. j] . 51 .18) Alternatively.21) for any function F (x). For example. one can write +∞ −∞ 2 dx e −αx /2 xn = 2 ∂ n +∞ dx e −αx /2 + jx ∂j n −∞ ∂n I1 [α.19) 2 dx e −αx /2 F (x) = F ∂ ∂j ∂ ∂j I1 [α. one has +∞ −∞ +∞ 2 2 ∂m dx e −αx /2 dx e −αx /2 x2m = (−2)m m ∂α −∞ ∂m = (−2)m m I0 [α] . ∂α (A.3 Gaussian Integrals and Determinants The simplest 2dimensional Gaussian integral is +∞ +∞ dx −∞ −∞ 2 2 2π .22) This generalises to +∞ +∞ dx −∞ −∞ 2 2 dy e −(αx + βy + 2γxy)/2 = 2π . j]. A. ∂j n j=0 j=0 = More generally.23) as can be seen by completing the square. one has +∞ −∞ (A.The integrals of even powers x2m can be calculated by relating them to the integrals I 0 or I1 . This explains why the central object of interest for integrals with Gaussian weight is the generating function I 1 [α. (A.
Then one has +∞ −∞ a b A dd x e −Aab x x /2 = det 2π −1/2 . (A. (A.27) is remarkably simple and will be left as an exercise.A.24) the exponent of the integrand can be written as A ab xa xb with xa = (x. y) and the above identity reads √ 2 −1/2 +∞ 2 −Aab xa xb /2 = ( 2π) = det A √ d xe . (A. The strategy is to ﬁrst prove them for A diagonal (this is trivial) and to then use the fact that any real symmetric A can be diagonalised by an orthogonal transformation. d ≥ 2. 52 .27) Thus the task of calculating Gaussian integrals and their various relatives and descendants has been reduced to the purely algebraic task of calculating determinants of matrices. for the oscillatory version of this integral one has +∞ −∞ a b A dd x e iAab x x /2 = det 2πi −1/2 .25) 2π det A −∞ This way of writing this result generalises to ddimensional integrals. Let A be a symmetric.26. combined with the rotation invariance of the measure d d x. The proof of the fundamental identities (A. positive real (d × d)matrix.In terms of the matrix A= α γ γ β (A.26) Likewise.
Arfken and H.3) which provides a cute proof of the famous identity 1 π2 = . ∞ n=1 (B. however. n2 6 n=1 ∞ (B. . n2 π 2 n=1 ∞ (B. in the spirit of Euler. It can be proved rigorously in a variety of ways. sin x x − x3 /6 + .see e. section 7 of G.g. The easiest (and closest in spirit to the rough argument in the previous sentence) is to extend (sin x)/x to a holomorphic function (sin z)/z in the complex plane and factorise using the MittagLeﬄer pole expansion . closely related inﬁnite product representations of other trigonomotric and hyperbolic functions are sinh x x ∞ = n=1 ∞ n=1 ∞ 1+ 1− 1+ n=1 x2 n2 π 2 x2 (n − 1/2)2 π 2 x2 (n − 1/2)2 π 2 (B. The ﬁrst nontrivial one of these results from comparing the quadratic term in the expansion of the left hand side. Weber..1) The validity of this formula is plausible as the right hand side has exactly the same zeros and identical behaviour as x → 0 as the left hand side.. x2 = =1− + . For example.5) (B. The single identity above can be used to generate an inﬁnite number of other identities. Elsevier (2005)..7) cos x = cosh x = 53 .4) There are a host of other occasionally useful inﬁnite product identities (none of which. will be used in these notes). x x 6 with the quadratic term arising from the right hand side. .6) (B.B Infinite Product Identities A useful identity. ubiquitous in determinant and index calculations.. ∞ x2 sin x = 1− 2 2 x n π n=1 (B. is the following inﬁnite product representation for (sin x)/x. Mathematical Methods for Physicists.2) 1− x2 n2 π 2 =1− x2 + .
2) (C.51) results for the ratio of ﬂuctuation determinants. (C. ˙ (C. xi ) = xi f1 (t) + xi f2 (t) ˙ ˙ and imposing the conditions xc (ti ) = xi and xc (ti ) = xi . is the sequence of prime numbers.8) As another example. xi ) . Fω (t) = ∂xc (t. . n=1 1 − pn (B. xi .5) . C Equivalence of the VVPM and GY Formulae for the Fluctuation Determinant The article of Kleinert and Chervyakov cited in section 3. xi .3) At this point. (B.63) expressing the equivalence of the VVPM (3. This identity provides the cornerstone of the relation between number theory and complex analysis.1) Writing this as a linear combination of two linearly independent solutions f 1 (t) and f2 (t) of the oscillator equation. ˙ ∂xi 54 (C. ˙ xc = xc (t.10) where pn = 2. xi . xi ) ˙ .There is also the more mysterious identity ∞ x sin x = cos n x 2 n=1 .i . 7. .4) f2 (ti ) = 0 f˙2 (ti ) = 1 . Consider the classical solution xc (t) with xc (tf.49) and GelfandYanglom (3. 11.5 contains a nice proof of the classical identity (3. xc (t. 3.9) 1 −s . ∂ xi ˙ (C. . one can either use directly the HamiltonJacobi relation ∂S[xc ] = −pi = −mxi . one ﬁnds ˙ ˙ xc (ti ) = xi ⇒ f1 (ti ) = 1 xc (ti ) = xi ⇒ f˙1 (ti ) = 0 ˙ ˙ This shows that f2 (t) = Fω (t) is the GY solution. Euler’s identity for the Riemann ζfunction ζ(s) = is ζ(s) = ∞ 1 ns n=1 ∞ (B. This solution can equally well be regarded as a function of the initial position x i and velocity xi .i ) = xf.
˙ ˙ 2 (C.9) (C.63). it is simply given by (3.f in favour of xi.10) (C.7) Since only boundary terms contribute to the classical action. which plays a role analogous to the GY solution when one considers periodic or antiperiodic boundary conditions instead of zero boundary conditions).f . xf xf ˙ = xi Gω (tf ) + xi Fω (tf ) ˙ ˙ = xi Gω (tf ) + xi Fω (tf ) .to deduce (3. ∂xi ∂xf Fω (tf ) (C. in particular. 55 . Or one can evaluate explicitly the classical action in terms of F ω (t) and f1 (t) ≡ Gω (t) (the “dual” GY solution. is S[xc ] = It follows that m ˙ [Fω (tf )x2 + Gω (tf )x2 − 2xi xf ] . ˙ and therefore. f i 2Fω (tf ) m ∂ 2 S[xc ] =− . one ﬁnds that the classical action.48) S[xc ] = m [xf xc (tf ) − xi xc (ti )] .6) (C.7) to eliminate xi. ˙ ˙ (C.8) Using (C.11) as we set out to show. and using the fact that the Wronskian of ˙ Fω (t) and Gω (t) is tindependent. Thus one has xc (t) = xi Gω (t) + xi Fω (t) . as a function of x i and xf . ˙ ˙ ˙ ˙ (Fω Gω − Fω Gω )(ti ) = −1 = (Fω Gω − Fω Gω )(tf ) .
This action might not be possible to undo. Are you sure you want to continue?
We've moved you to where you read on your other device.
Get the full title to continue reading from where you left off, or restart the preview.