LOG#149. Path integral(IV).


Final round of this thread!

Approximate methods

We will evaluate the path integral (PI) using an approximation known as “saddle point”. It is a semiclassical approximation sometimes referred as the method of steepest descent. Moreover, it is based on the Lagrange method. Firstly, we take the integral

\displaystyle{I=\int_{-\infty}^\infty ds e^{-\lambda S(x)}}

where S(x) is any function with (likely) several local minima (maxima). The constant \lambda can be take both real or complex, in particular purely imaginary!

Suppose now that we are interested in the limit of this integral when \lambda\rightarrow\infty or equivalently (in the complex case) when \vert\lambda\vert\rightarrow\infty. The, the integral will be dominated by the minima of S. In that case, we can aproximate it by a series of gaussian integrals, one for every minimum of S. If x_i is such a minimum, then, in its neighbourhood, we obtain an expansion

S(x)\simeq S(x_i)+\dfrac{1}{2}(x-x_i)^2S''(x_i)

and we can write

\displaystyle{I(x)\simeq \sum_i I_i}


\displaystyle{\boxed{I_i=\int_{-\infty}^\infty dx\exp\left(S(x_i)+\dfrac{1}{2}(x-x_i)^2S''(x_i)\right)=e^{-\lambda S(x_i)}\sqrt{\dfrac{2\pi}{\lambda S''(x_i)}}}}

As a simple example, we consider again the harmonic oscillator perturbed by an anharmonic term quartic in q, i.e., we take an harmonic oscillator plus an interacting term with quartic power in the potential. The propagator reads, of any initial and final points (take the mass equal to one for simplicity):

\displaystyle{K=\int \mathcal{D}q\exp \left(i\int dt\left(\dfrac{1}{2}\dot{q}^2-\dfrac{1}{2}\omega^2q^2-\dfrac{\lambda}{4!}q^4\right)\right)}

Now, we define the PI for the harmonic oscillator with a source term K_0(J) (physically, it could describe the action of an external force) added to the lagrangian as follows:


This PI can be evaluated exactly. Now, if we choose F(q) such as

\displaystyle{F(q)=\exp\left(i\int dt\dfrac{\lambda}{4!}q^4\right)}

we get for the first order correction the propagator

\displaystyle{K=K_0(0)+\dfrac{i\lambda}{4!}\int dt\left(\dfrac{\delta}{\delta J(t)}\right)^4K_0(J)\vert_{J=0}}

In the last term above, we take four functional derivatives of K_0(J) and then we set J=0, and thus only the piece of K_0(J), quartic in J, is relevant. Fewer than four J’s will be killed (annihilated) by the derivatives, more than four will be killed when we set J=0. That is,

\displaystyle{K_0(J)=C\exp\left(\dfrac{1}{2}\int dt dt'J(t)G(t,t')J(t')\right)}

\displaystyle{K_0(J)=\mbox{irrelevant term}+\dfrac{C}{2}\dfrac{1}{2}\left(\dfrac{1}{2}\int dtdt'J(t)G(t,t')J(t')\right)^2=\dfrac{C}{8}\langle J_1G_{12}{J_2}\rangle\langle J_3G_{34}J_4\rangle}

where we have introduced the notation

\displaystyle{\langle J_1G_{12}J_2\rangle=\int dt_1dt_2J(t_1)G(t_1,t_2)J(t_2)}

Substituting the last compact expression into the actual PI, we obtain the formula

\displaystyle{K=C\left(1+i\dfrac{\lambda}{4!8}\int dt\left(\dfrac{\delta}{\delta J(t)}\right)^4\langle J_1G_{12}J_2\rangle\langle J_3G_{34}J_4\rangle+o(\lambda^2)\right)}

and now,working out a little bit more

\displaystyle{X=\left(\dfrac{\delta}{\delta J(t)}\right)^2\langle J_1G_{12}J_2\rangle=\left(\dfrac{\delta}{\delta J(t)}\right)^2\int dt_1dt_2J(t_1)G(t_1,t_2)J(t_2)}

then the first derivative can act either on J_1 or J_2. It provides a delta function in any of these two cases and it reduces the integral to

\displaystyle{X=\dfrac{\delta}{\delta J(t)}\int dt_1dt_2\left(\delta (t-t_1)G(t_1,t_2)J(t_2)+J(t_1)G(t_1,\tau)\delta (\tau-t_2)\right)=\dfrac{\delta}{\delta J(t)}\left(\int dt_2G(t,t_2)J(t_2)+\int dt_1J(t_1)G(t_1,t)\right)}

In fact, the remaining derivative acts similarly an dills the remaining integral, and the result simplifies


Thus, the functional derivatives above, as a generalization of this simple result, provide

\displaystyle{K=C\left(1+\dfrac{i\lambda}{8}\int dt G(t,t)^2\right)}

and again, we get


The time integral is just the time interval \Delta T=T_f-T_i, and the final result of this long calculation can be written at last as follows

\boxed{K(0,0;T_f,T_i)=C\left(1+\dfrac{i\Delta T\lambda}{32m^2\omega^2}+o(\lambda^2)\right)}

Please, note the dependence in \Delta T,\omega^2, m^2 in the propagator at first order in \lambda. It is a very different behaviour to the mere free particle or the simple harmonic oscillator!

QFT and field theory path integrals

It is very easy to generalize the PI to many, finite or even infinite, degrees of freedom. Field theory is indeed a “complex” or complicated system with infinite degrees of freedom. We turn particle positions into field functions:

q(t)\rightarrow \phi(X,t)=\phi(x)

and it constitutes a continuous (infinitely coordinated) system (N\rightarrow \infty). The analogue of the quantum mechanical propagator is the transition amplitude to go from one field configuration \phi(x) at t=0 to another \phi'(x') at t'=T:

\displaystyle{K(\phi'(x'),\phi(x);T,0)=\int\mathcal{D}\phi e^{S(\phi)}}

where S is the field action, for instance, for some scalar field we could choose arbitrarily. Take, e.g., the field theory action

\displaystyle{S(\phi)=\int d^4x\left(\dfrac{1}{2}\left(\partial_\mu\phi)^2-\dfrac{1}{2}m^2\phi^2\right)\right)}

The field theory PI captures ALL the possible field configurations obeying the stated initial and final boundary conditions.

In field theory we are also interested in the Green functions. Quantum Field Theory calculations require the computation of the next object (the correlation functions!):

\langle 0\vert T\phi(x_1)\phi(x_2)\cdots\phi(x_n)\vert 0\rangle

where the terms inside the vacuum expectation value are quantum operators, operator valued quantities of coordinates (and momenta). The vacuum expectation value (v.e.v.)= of any time-ordered product of Heisenberg field operators are very important quantities. In fact, this quantity IS known as the Green function or the correlation function. Motto to remember in QFT:

\mbox{Green functions=Correlation functions=v.e.v. of any time ordered product of Heisenberg field operators}

This lemma provides interesting connection and links, and it is a useful reminder or dictionary between different “worlds” that are, secretly, the same thing.

The order of the operators is such that the earliest field is written last (right-the most), the secon earliest second last, and so on. Example:

T\phi(x_1)\phi(x_2)=\begin{cases}\phi(x_1)\phi(x_2),\mbox{if}\;\; x_1^0>x_2^0\\ \phi(x_2)\phi(x_1),\mbox{if}\;\; x_1^ 0<x_2^0\end{cases}

Green functions are related to amplitudes for the physical processes sush as scattering and decay processes in particle physics as well. But this is a long story some day I will tell to you!

Examples from field theory

 Example 1. Free scalar field.

In the case of the free scalar field, the generating functional is given by

Z_0(J)=\dfrac{\langle 0\vert 0\rangle_J}{\langle 0\vert 0\rangle_{J=0}}

Both numerator and denominator can be expressed in terms of PI. The numerator reads

\displaystyle{N_0=\int\mathcal{D}q\exp\left(i\int d^4x\left(\dfrac{1}{2}(\partial_\mu\phi)^2-\dfrac{1}{2}m^2\phi^2+J\phi\right)\right)}

We put


where \phi_c is the classical configuration and \varphi are deviations or quantum fluctuations of the field configuration. Then, we integrate out over the deviations from the classical configurations. The action can be proved to be

\displaystyle{S(\phi_c+\varphi)=\int d^4x\left(\dfrac{1}{2}(\partial_\mu\phi_c)^2-\dfrac{1}{2}m^2\phi_c^2+J\phi_c\right)+\int d^4x\left(\dfrac{1}{2}(\partial_\mu\varphi)^2-\dfrac{1}{2}m^2\varphi^2\right)}

where we neglected the linear terms in \varphi since they are reduced to zero due to the equations of motion. Thus, we get

\displaystyle{N_0=C\exp\left(i\int d^4x\left(\dfrac{1}{2}(\partial_\mu\phi_c)^2-\dfrac{1}{2}m^2\phi_c^2+J\phi_c\right)\right)}

and where

\displaystyle{C=\int\mathcal{D}\varphi\exp\left(i\left(\int d^4x\left(\dfrac{1}{2}(\partial_\mu\varphi)^2-\dfrac{1}{2}m^2\varphi^2\right)\right)\right)}

Note that C is independent of J and will cancel in the generating functional Z(J).

Usint the fact that the classical field configuration satisfies the equations of motion (EOM)


we write

\displaystyle{N_0=C\exp\left(\dfrac{i}{2}\int d^4xJ(x)\phi_c(x)\right)}

Finally, we can write as well


i.e., the propagator of the Klein-Gordon field theory (Green function), since

\phi_c(x)=i\int d^4x\Delta_F(x,x')J(x')

\displaystyle{Z_0=\dfrac{N_0}{C}=\exp\left(-\dfrac{1}{2}\int d^4xd^4x'J(x)\Delta_F(x,x')J(x')\right)}

The Green functions are found by solving its equation in 4-momentum space, and it yields


Let me calculate the 2-point Green function corresponding to this system:

G^{(2)}_0(x_1,x_2)=\langle 0\vert T\phi(x_1)\phi(x_2)\vert 0\rangle=\dfrac{1}{i^2}\left(\dfrac{\delta^2}{\delta J(x_1)\delta J(x_2)}Z_0(J)\right)\vert_{J=0}

If we expand the generating functional in powers of J, we obtain

\displaystyle{Z_0(J)=1-\dfrac{1}{2}\int d^4xd^4x'J(x)\Delta_F(x-x')J(x')+o(J^4)}

The term quadratic in J is the only one that survives in the integral, and thus

\displaystyle{G_0^{(2)}(x_1,x_2)=\dfrac{1}{i^2}\dfrac{\delta^2}{\delta J(x_1)\delta J(x_2)}\left(-\dfrac{1}{2}\int d^4xd^4x'J(x)\Delta_F(x-x')J(x')\right)}

There are two identical terms inside this last expression, depneding on which derivative acts on which J. The final result is simple


Therefore, the Green function (two point correlation function) int the QFT is also the Green function in the usual differential equation sense! It is a cool (consistent!) result! Lut us compute the 4-point correlation function further

\displaystyle{G_0^{(4)}(x_1,x_2,x_3,x_4)=\dfrac{1}{i^4}\left(\dfrac{\delta}{\delta J(x_1)}\cdots\dfrac{\delta}{\delta J(x_4)}\exp\left(-\dfrac{1}{2}\int d^4xd^4x'J(x)\Delta_F(x,x')J(x')\right)\right)\vert_{J=0}}

The only term that survives is the part of the exponential that contributes with four J’s. Thus, the 4-point correlation function for the free particle reads

\displaystyle{\boxed{G_0^{(4)}(x_1,x_2,x_3,x_4)=\dfrac{\delta}{\delta J(x_1)}\cdots\dfrac{\delta}{\delta J(x_4)}\frac{1}{2}\left(-\dfrac{1}{2}\int d^4xd^4x'J(x)\Delta_F(x,x')J(x')\right)^2}}

There, there are 4!=24 terms, corresponding to the number of ways of associating the derivatives with the J’s (sources, quantum operators). In 8 of them, the Green functions which arise are \Delta_F(x_1-x_2)\Delta_F(x_3-x_4), and so on. The final result reads (you can check it with the use of the Wick theorem!):


And the cool thing is that this thing can be represented diagrammatically in a convenient way related to Feynman graphs! That is the hidden power of QFT!

Example 2. Interacting field theory.

Unfortunately, it is not possible to compute exactly the generating functional Z(J) in general. It is necessary to use approximate methods or sometimes, systems with a big symmetry to compute any exact Z(J). Let us use perturbation theory, and we consider a scalar field theory with \lambda\phi^4 interactions. This is called the \phi^4 theory as well. It is defined by a lagrangian density


Then, the generating functional reads

\displaystyle{Z(J)=C\int \mathcal{D}\phi\exp\left(i\left(\int d^4x\left(\dfrac{1}{2}(\partial_\mu\phi)^2-\dfrac{1}{2}m^2\phi^2-\dfrac{\lambda}{4!}\phi^4+J\phi\right)\right)\right)}

In the next step, we change the higher-order term by a functional derivative with respect to J:

\displaystyle{C\int\mathcal{D}\phi\exp\left(-\dfrac{i\lambda}{4!}\int d^4x\phi^4\right)\exp\left(i\left(\int d^4x\left(\dfrac{1}{2}(\partial_\mu\phi)^2-\dfrac{1}{2}m^2\phi^2+J\phi\right)\right)\right)}

or equivalently

\displaystyle{C\int\mathcal{D}\phi\exp\left(-\dfrac{i\lambda}{4!}\int d^4x\left(\dfrac{\delta}{i\delta J(x)}\right)^4\right)\exp\left(i\left(\int d^4x\left(\dfrac{1}{2}(\partial_\mu\phi)^2-\dfrac{1}{2}m^2\phi^2+J\phi\right)\right)\right)}

We can push out the first exponential from the integral and the remaining functional derivative is that of Z_0(J). We adjust the constant C so that Z_0(J=0)=0 and we get

\displaystyle{Z(J)=\dfrac{\exp\left(-\dfrac{i\lambda}{4!}\int d^4x\left(\dfrac{\delta}{i\delta J(x)}\right)^4\right)\exp\left(-\frac{1}{2}\int d^4xd^4x'J(x)\Delta_F(x,x')J(x') \right)}{\left(\exp\left(-\dfrac{i\lambda}{4!}\int d^4x\left(\dfrac{\delta}{i\delta J(x)}\right)^4\right)\exp\left(-\frac{1}{2}\int d^4xd^4x'J(x)\Delta_F(x,x')J(x')\right)\right)\Bigg| _{J=0}}}

We can write a perturbative expansion for any Green function/propagator/correlation function from this last expression!

The perturbative expansion for the two point Green function is “simple”. Let me look at G^{(2)}(x_1,x_2) to the first non-zero order in \lambda. We find that

\displaystyle{G^{(2)}(x_1,x_2)=\dfrac{\left(\dfrac{1}{i^2}\dfrac{\delta^2}{\delta J(x_1)\delta J(x_2)}\exp\left(-\dfrac{i\lambda}{4!}\int d^4x\left(\dfrac{1}{i}\dfrac{\delta}{\delta J(x)}\right)^4\right)\exp\left(-\dfrac{1}{2}\langle J_a\Delta_{Fab}J_b\rangle\right)\right)\Bigg| _{J=0}}{\left(\exp\left(-\dfrac{i\lambda}{4!}\int d^4x\left(\dfrac{1}{i}\dfrac{\delta}{\delta J(x)}\right)^4\right)\exp\left(-\dfrac{1}{2}\langle J_a\Delta_{Fab}J_b\rangle\right)\right)\Bigg| _{J=0}}}

where the brackets imply integration over the positions of the J’s. In both numerator and denominator, we can expand both exponentials and the only remaining terms are those which the same total number of derivatives and J’s. Let me look at the linear term in \lambda. There are 6 functional derivatives, so we need the term from the expansion of the second exponential with 6 J’s. For this term, we get the following expression in the numerator

\displaystyle{-\dfrac{\delta}{\delta J(x_1)\delta J(x_2)}\left(-\dfrac{i\lambda}{4!}\right)\int d^4x\left(\dfrac{\delta}{\delta J(x)}\right)^4\dfrac{1}{3!}\left(-\dfrac{1}{2}\right)^3\langle J_a\Delta_{Fab}J_b\rangle\langle J_c\Delta_{Fcd}J_d\rangle\langle J_e\Delta_{Fef}J_f\rangle}

There are now 720=6! terms. However, there are only two distinct analytical expressions in the result. The first of these arises if the functional derivatives at x_1 and x_2 act on different \langle\ldots\rangle brackets. Combinatorics tells us that there are 576 of such terms, yielding the expression

\displaystyle{-\dfrac{i\lambda}{2}\int d^4x\Delta_F(x_1-x)\Delta_F(x-x)\Delta_F(x-x_2)}

The only different term to this one arises when the derivatives at x_1 and x_2 act on the same bracket. This accounts for 144=12^2 terms (note that 720=576+144). The analytic form of this term reads

\displaystyle{-\dfrac{i\lambda}{8}\Delta_F(x_1-x_2)\int d^4x\Delta_F(x-x)^2}

The denominator can be evaluated in a similar way. The Green function to order \lambda is finally recasted into

\displaystyle{G^{(2)}(x_1,x_2)=\dfrac{\Delta_F(x_1-x_2)-\frac{i\lambda}{2}\int d^4x\Delta_F(x_1-x)\Delta_F(x-x)\Delta_F(x-x_2)-\frac{i\lambda}{8}\Delta_F(x_1-x)\int d^4x\Delta_F(x-x)^2+o(\lambda^2)}{1-\frac{i\lambda}{8}\int d^4x\Delta_F(x-x)^2+o(\lambda^2)}}

Since we have only computed the numerator and denominator up to order \lambda, we can rewrite this last expression in the following manner:

\displaystyle{G^{(2)}(x_1,x_2)=\dfrac{\left(\Delta_F(x_1-x_2)-\frac{i\lambda}{2}\int d^4x\Delta_F(x_1-x)\Delta_F(x-x)\Delta_F(x-x_2)+o(\lambda^2)\right)\cdot \left(1-\frac{i\lambda}{8}\int d^4x\Delta_F(x-x)^2+o(\lambda^2)\right)}{1-\frac{i\lambda}{8}\int d^4x\Delta_F(x-x)^2+o(\lambda^2)}}

We can now cancel the second factor in the numerator against the denominator, as you can easily see, up to order \lambda, and we get as final ultimate result that

\displaystyle{\boxed{G^{(2)}(x_1,x_2)=\Delta_F(x_1-x_2)-\dfrac{i\lambda}{2}\int d^4x\Delta_F(x_1-x)\Delta_F(x-x)\Delta_F(x-x_2)+o(\lambda^2)}}

This factorization of the numerator into a part containing no factors independent of the external position times the denominator occurs to all orders, as you can be prove yourself easily if you have understood these notes. Thus, we can define the “disconnected” parts of the Green function as the parts of the diagrams not connected to any external line. The conclusion is that this disconnected parts cancel from the Green functions!

Additional exercises for you

Exercise 1. Show that if a functional is given by any expansion of the type

\displaystyle{I(f)=\sum_n\dfrac{1}{n!}\int dx_1dx_2\cdots dx_n F_n(x_1,\ldots,x_n)f(x_1)\cdots f(x_n)}

then you can get the functional derivative

\displaystyle{\dfrac{\delta I(f)}{\delta f(x)}=\sum_{n=0}^\infty\dfrac{1}{n!}\int dx_1\cdots dx_n F_{n+1}(x,x_1,\ldots,x_n)f(x_1)\cdots f(x_n)}

Exercise 2. Compute, using the definition of functional derivative you have studied in this thread, the functional derivatives of

a) \displaystyle{I(f)=\int_{-1}^{1}dxf(x)}


\dfrac{\delta I(f)}{\delta f(y)}


\dfrac{\delta I(f^2)}{\delta f(y)}

b) I_z(f)=f(z),  I(f)=\int dx f(x)g(x) with g(x) a fixed function, and I(f)=\exp\left(\int dxf(x)g(x)\right) IF

\dfrac{\delta I(f)}{\delta f(y)}

is the functional derivative.

Exercise 3. Compute explicitly the integral

\displaystyle{I(A,a)=\int dx_1\cdots dx_n\exp\left(-\dfrac{1}{2}\sum_{i,j}x_iA_{ij}x_j+\sum_ia_ix_i\right)=\int dx\exp\left(-\dfrac{1}{2}X\cdot A\cdot X+a\cdot X\right)}

where A is any complex, symmetric matrix with eigenvalues \lambda_i such as Re(\lambda_i)>0,\lambda_i\neq 0.

Exercise 4. Show that, for a gaussian distribution function, the generating function IS

\langle \exp (a\cdot X)\rangle=\left(2\pi\right)^{n/2}\left(\det A\right)^{-1/2}\exp\left(\dfrac{1}{2}a\cdot A^{-1}\cdot a\right)

Exercise 5. Compute the n-point functions for the harmonic oscillator (i.e., the 1-point, 2-point,…correlation functions).

Exercise 6. Compute with the saddle point approximation the integral corresponding to


in the case that a) \lambda=i\lambda', i.e., imaginary coupling constant, and b)\lambda=\lambda', where the coupling constant is real or complex in general. Compare the results with the results of a perturbation expansion assuming that b\rightarrow 0, i.e., with the result of a tiny anharmonic term.

Exercise 7. In quantum field theory with interacting \lambda\phi^4 coupling, compute the Green function G^{(2)} (2-point correlation function) up to order \lambda^2. Compute the 4-point correlation function (Green function) G^{(4)} to order \lambda^2. Compare the expressions you obtain with the results given in this thread.

Exercise 8. In QFT with a interaction term of type \dfrac{g}{3!}\phi^3, compute the 2-point Green function to order \lambda^2 and the 4-point correlation function to the same order. Compare the results with those of the \lambda\phi^4 theory.

Some references for further study

Let me finish this interesting thread with some basic and well known references:

1) The Dirac original idea for the role of the action in QM is inside the article

Physikalishche Zeitschrift der Sowjetunion, Band 3, Heft 1 (1933). P.A.M.Dirac.

2) Feynman pioneer works about the path integral are mainly here

i) Reviews of Modern Physics 20, 367.(1948). R.P.Feynman.

ii) Quantum Mechanics and Path Integrals, McGraw Hill, 1965. R.P.Feynman and A.R.Gibbs.

iii) Statistical Mechanics: A set of lectures, 1972. R.P.Feynman. Benjamin eds.

3) Some classical works about the path integral:

i) Lectures in Quantum Mechanics, Benjaming-Cummings, 1973. G.Baynn.

ii) L.S.Schulman, Techniques and Applications of Path Integration, John Wiley and Sons, 1981.

4) Field theory and path integrals are treated in some basic texts about this hard subject. For instance:

i) Field theory, renormalization group and critical phenomena, 2nd edition, World Scientific eds. 1984. D.J.Amit.

ii) L.H.Ryder. Quantum Field Theory, Cambridge University Press, 1985.

iii) P.Ramond, Field Theory: a modern Primer, 2nd edition. Addison-Wesley, 1990.

iv) E.S.Abers and B.W.Lee, Physics Reports 9C, 1,1973.

v) B.Sakita, Quantum Theory of Many Variable Systems and Fields, World Scientific, 1985.

5) More advanced topics in QFT can be found here:

i) S.Coleman, Aspects of Symmetry, C.U.P. 1985. I own and admire this book, written by one of the fathers of the wormhole concept and its application into physics!

ii) Quantum Field Theory in Condensed Matter Physics, Cambridge University Press (C.U.P.), 1995 by A.M.Tsvelik (purely soviet and russian style!).

iii) V.N.Popov, Functional integrals and Collective Excitations, C.U.P. 1987. Other great book by one of the best russian QFT theorists.

iv) Field Theories of Condensed Matter Systems, Addison-Wesley, 1991. E.Fradkin. A great book describing the links between Statistical Mechanics, Condensed Matter and the methods of field theory, by one of the best masters in the subject!

Today, you can find even more modern and better lectures (in some cases, not always) in the internet. But the above references are classical and deserve at least a minimal fast reading.

See you in my next SPECIAL AND SUPER(HYPER)GEEKNERD TSOR post, the 150th!!!!!

View ratings
Rate this article

Leave a Reply

Your e-mail address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.