Quantum Field Theory-II Problem Set n.

1
UZH and ETH, FS-2015 Prof. G. Isidori
Assistants: A. Greljio, S. Lionetti, D. Marzocca Due: 27-02-2015
http://www.physik.uzh.ch/∼marzocca/QFT2/

1 Gaussian Integrals
Let Z Z +∞
−xT Ax
G(A) = Πi dxi e = dx1 dx2 . . . dxn e−xi Aij xj , (1)
−∞

where A is a real symmetric n × n matrix with positive eigenvalues. Show that

G(A) = π n/2 det(A)−1/2 . (2)

Solution Since A is real and symmetric, the finite-dimensional spectral theorem guarantees that
it can be diagonalized, i.e. there is a matrix R ∈ O(n) such that
RAR−1 = Λ, Λij = λi δij . (3)
−1 0
Thus we may change variables according to xi = Rij xj , to find

Πi dx0i
Z Z
−x0T RAR−1 x0 2
P
G(A) = e = Πi dxi e− j λj xj , (4)
| det(R)|
where we used
R∈O(n)
xT = (R−1 x0 )T = x0T (R−1 )T = x0T R, (5)
together with | det(R)| = 1, always because of the orthogonality
√ of R. Assuming all λi > 0 as the
problem states, we may change variables once more to yi = λi xi and use Fubini’s theorem to
find Z Z r
dyi − Pj yj2 1 −yi2 π
G(A) = Πi √ e = Πi √ dyi e = Πi = π n/2 det(A)−1/2 . (6)
λi λi λ i

The last equality follows from

Πi λi = det(Λ) = det(RAR−1 ) = det(R) det(A) det(R−1 ) = 1 × det(A) × 1 = det(A). (7)

Similarly, considering n complex variables zi and the Hermitian positive-definite matrix C, show
that Z
†
Πi dzi dzi∗ e−z Cz = π n det(C)−1 . (8)

Solution The proof proceeds along the lines of the first one. Since C is hermitian again the
(complex) finite-dimensional spectral theorem guarantees that it may be diagonalized by means
of a unitary matrix U ,
U CU −1 = Λ, Λij = λi δij . (9)

1
Therefore changing variables to z 0 = U −1 z we find 1
Z Z Z 
∗ −z † Cz 0 0∗ −z 0† Λz 0 0 0∗ −λi zi0∗ zi0
Πi dzi dzi e = Πi dzi dzi e = Πi d(zi , zi ) e , (11)

where once more the jacobian of the transformation is 1 because U ∈ U(n). Now assuming that
C has positive eigenvalues, i.e. λj > 0, the integrals are convergent. We now evaluate each of the
independent integrals above by considering the measure of the complex plane in terms of the real
and imaginary coordinates (zi0 = xi + iyi ) and using the fact that d(zi0 , zi0∗ ) = d(xi , yi ):
Z Z  Z 
∗ −z † Cz 0 0∗ −λi zi0∗ zi0 −λi (x2i +yi2 ) π
Πi dzi dzi e = Πi d(zi , zi ) e = Πi d(xi , yi ) e = Πi (12)
R2 λi
The chain of equalities (7) is only slightly modified because the U matrices are unitary,
Πi λi = det(Λ) = det(U CU −1 ) = det(U ) det(C) det(U −1 ) = eiα det(C)e−iα = det(C), (13)
therefore we conclude Z
† Cz
Πi dzi dzi∗ e−z = π n det(C)−1 . (14)

Finally, prove that, for real variables,

Z
T T 1 T −1
Πi dxi e−x Ax+ω x = π n/2 e 4 ω A ω det(A)−1/2 (15)

Solution With the same notation as in the first part of the exercise, again we change variables
to x0 obtaining now
Z Z Z
−xT Ax+ω T x 0 −x0T Λx0 +ω T R−1 x0 02 0 0
P
Πi dxi e = Πi dxi e = Πi dx0i e− j [λj xj −ωj xj ] , (16)

where ωj0 = Rjk ωk . With this notation it is easy to complete the square by switching to
!
0
1 ω j
x0j = p yj + p , (17)
λj 2 λj
which yields

1 ωj02
Z Z  X 
−xT Ax+ω T x dyi 2
Πi dxi e = Πi √ exp − yj −
λi j
4 λj
1 ωj02 Y 1
 X 02 
ωj
X  Z
−yi2 1
= exp √ dyi e = exp π n/2 det(A)−1/2 . (18)
j
4 λj i
λ i 4 j
λj

1
The notation dz dz ∗ ≡ d(z, z ∗ ) to indicate an integral over the complex numbers might be not familiar to everybody. It
represents the Lebesgue measure on the complex plane C, which is isomorphic to R2 , therefore:
Z Z Z Z Z
f (z)dzdz ∗ ≡ f (z)d(z, z ∗ ) = f (x + iy)d(x, y) = f (x + iy)dxdy. (10)
C C R2 R R

2
At this stage we only need to reconstruct the matrix in the exponent, which is accomplished by
X ωj02 X δij 0
= ωi0 ω = ω 0T Λ−1 ω 0 = ω T RΛ−1 R−1 ω 0 = ω T (RΛR−1 )−1 ω 0 = ω T A−1 ω 0 . (19)
j
λj i,j
λj j

2 Forced Harmonic Oscillator

Using the path-integral formulation of quantum mechanics, the transition amplitude of a one-
dimensional harmonic oscillator in presence of a driving force F (t) is
Z R tF 1 1 2 2
hQtF |QtI iF = Dq ei tI dt[ 2 q̇− 2 ω q +F (t)q(t)] ,
0
(20)

where QtI and Q0tF are the boundary conditions of q (for simplicity we set ~ = 1).
I. Show that
F̃ (E)F̃ (−E)
− 2i
R
dE
hQ0∞ |Q−∞ iF = hQ0∞ |Q−∞ iF =0 e E 2 −ω 2 +i , (21)
where Z +∞
dt
F̃ (E) = √ e−iEt F (t) . (22)
−∞ 2π
Suggestion: use a Fourier transformation (t ↔ E) in the exponential term in Eq. (20) and
introduce an i term to make it convergent.
Solution We would like to start by introducing in the action
Z tF 
1 tF  2
 Z
1 2 1 2 2
dt q̇ − ω 2 q 2 + F q + qF ,

S[q] = dt q̇ − ω q + F q = (23)
tI 2 2 2 tI

the antitransform
Z +∞
dE
f (t) = √ eiEt f˜(E) , (24)
−∞ 2π
for the functions q = q(t) and F = F (t), in order to turn time derivatives into iE factors. Before
doing this, however, we note that equations (22) and (24) are strictly speaking valid only when the
integration range of t is the whole real axis (for [tI , tF ] we would need a Fourier series instead). This
is ultimately what we want, since the problem asks us to compute hQ0∞ |Q−∞ i. Unfortunately there
is an important issue: if the integral in eq. (23) diverges in the limit tI,F → ∓∞, the exponential
of iS may oscillate indefinitely and be undefined in that limit.
One way to circumvent this is to add an i term (with  > 0) in the lagrangian, making the
exponential acquire a negative real part that quenches it when S → ∞. Although different
choices are possible, there is one that has a well-known physical meaning: sending ω 2 → ω 2 − i
corresponds to damping the forced harmonic oscillator.

3
Keeping this replacement for later, we may finally proceed to plug eq. (24) into the action
dE dE 0 i(E+E 0 )t h
Z Z
1 0 0 2 0 0 0
i
S= dt e (iE)(iE )q̃(E)q̃(E ) − ω q̃(E)q̃(E ) + F̃ (E)q̃(E ) + q̃(E)F̃ (E ) .
2 2π
(25)

Taking the limit tI,F → ∓∞ extends the time integral to the set (−∞, +∞) which allows us to
write
Z
0
dt ei(E+E )t = 2πδ(E + E 0 ). (26)

The integral over E 0 can be then performed free of charge

Z
1 h i
S= dE (E 2 − ω 2 )q̃(E)q̃(−E) + F̃ (E)q̃(−E) + q̃(E)F̃ (−E) . (27)
2
F̃ (E)
Shifting q̃(E) → q̃(E) − gives
E 2 −ω 2

Z " #
1 F̃ (E) F̃ (−E)
S= dE (E 2 − ω 2 )q̃(E)q̃(−E) − , (28)
2 E 2 − ω2

without changing the integration measure Dq since

Z
F̃ (E)
q(t) → q(t) − dE 2 , (29)
E − ω2
is a simple shift. Now the second term in the action is independent of the function q and may be
therefore carried out of the path integral, yielding
Z Z
F̃ (E)F̃ (−E)
− 2i dE
R
i 2 2
R
0
hQ∞ |Q−∞ iF = Dq e iS[Q]
=e 2
E −ω 2
Dq e 2 dE(E −ω )q̃(E)q̃(−E) =
F̃ (E)F̃ (−E)
− 2i
R
dE
=e E 2 −ω 2 hQ0∞ |Q−∞ iF =0 , (30)

which remembering that we still need to substitute ω 2 → ω 2 − i gives the desired result.

II. Transform the exponential term in Eq. (21) into a time integral,
Z Z
i F̃ (E)F̃ (−E) i
− dE 2 →− dtdt0 F (t)D(t − t0 )F (t0 ) (31)
2 E − ω 2 + i 2
and discuss its physical meaning.

4
Solution Simply plugging in the definition of Fourier transforms gives

dt dt0 −iE(t−t0 ) F (t)F (t0 )

Z Z
i F̃ (E)F̃ (−E) i
− dE 2 = − dE e =
2 E − ω 2 + i 2 2π E 2 − ω 2 + i
0
dE e−iE(t−t )
Z Z Z
i 0 0 i
=− dt dt F (t)F (t ) ≡− dt dt0 F (t)F (t0 )D(t − t0 ). (32)
2 2π E 2 − ω 2 + i 2
In order to understand the physical meaning of this result, we note that

− 2i
R
dt dt0 F (t)F (t0 )D(t−t0 ) hQ0∞ |Q−∞ iF
e = . (33)
hQ0∞ |Q−∞ i0
Since the position eigenstates form a complete set, the exponential on the left-hand side equals
the ratio of amplitudes with and without an external force for the transition from any state |ii at
t = −∞ to any state |fi at t = +∞:
Z
i 0 0 0
R
hf∞ |i−∞ iF = dQ dQ0 hf∞ |Q0∞ i hQ0∞ |Q−∞ iF hQ−∞ |ii = hf∞ |i−∞ i0 e− 2 dt dt F (t)F (t )D(t−t ) . (34)

III. Defining Z 2
h i
1 dq
( dτ ) + 12 ω 2 q 2 −F (τ )q(τ )
R
− dτ
WE [F ] = Dq e 2
, (35)

show that 1
dτ dτ 0 F (τ )DE (τ −τ 0 )F (τ 0 )
R
WE [F ] = WE [0] e 2 , (36)
and discuss how to recover D(t) from DE (τ ) via analytic continuation.
Solution The proof initially proceeds as in point I, with the important difference that here we
have to trust our action integral to converge. Using the Fourier transforms
Z Z
˜ dτ −iEτ dE
f (E) = √ e f (τ ), f (τ ) = √ eiEτ f˜(E), (37)
2π 2π
we find
Z   
1 dq 2 2 2
SE [q] = dτ + ω q − F q − qF
2 dτ
dE dE 0 i(E+E 0 )τ h
Z
1 i
= dτ e (iE)(iE 0 )q̃(E)q̃(E 0 ) + ω 2 q̃(E)q̃(E 0 ) − F̃ (E)q̃(E 0 ) − q̃(E)F̃ (E 0 )
2 2π
Z
1 h i
= dE (E 2 + ω 2 )q̃(E)q̃(−E) − F̃ (E)q̃(−E) − q̃(E)F̃ (−E) . (38)
2
F̃
We can once more change variables in the path integral by shifting q̃ → q̃ + E 2 +ω 2 to get

Z
1
R F̃ (E)F̃ (−E) 1 2 2 1 0 0 0
dE
R R
WE [F ] = e 2 E 2 +ω 2
Dq e− 2 dE(E +ω )q̃(E)q̃(−E) = e 2 dτ dτ F (τ )F (τ )DE (τ −τ ) WE [0] , (39)

5
where we defined
dE e−iEτ
Z
1
DE (τ ) ≡ . (40)
2 2π E 2 + ω 2
This last integral can be interpreted as a contour integral along the real axis E ∈ R of a complex
function
e−iEτ
f (E) = , (41)
E 2 + ω2
which has poles in E = ±iω. We are therefore allowed to deform the integration path in all the
analytic domain of this function, and in particular to rotate it of ±(π/2 − ) so that it lies (almost)
along the imaginary axis. It is possible to show that, in the limit  → 0, this is equivalent to
shifting the poles a bit to the left and to the right by sending ω 2 → ω 2 ± i, according to the
direction of the rotation. Replacing E = ±iE, τ = ∓it then gives

e−iEt
Z
i dE
DE (∓it) = − . (42)
2 2π E 2 − ω 2 ∓ i
On the other hand, we see that carrying out the replacement τ = ∓it in the action yields

−SE [q] = ∓iS[q] , (43)

thus choosing the lower sign in all these equations gives us back the original path integral together
with the causal propagator D(t) = DE (it). The opposite rotation changes the sign in front of S[q],
which is equivalent to flipping the sign of time, and flips the i prescription in the propagator,
making it anti-causal.