I have been trying to calculate the massless scalar field propagator in position space by directly Fourier transforming the momentum space propagator. $$\int{d^2p\frac{1}{(p^0)^2-(p^1)^2}e^{-i(p^0t-p^1x)}}$$

Upon referring to multiple sources (linked below), I realize that the answer is actually proportional to $ln|x|$ but I don't see how this integral will result in that answer. All of these sources obtain that answer by finding the massive propagator and then taking the $m\rightarrow 0$ limit. I don't see what I am missing by directly doing doing the above integral.

To see how the above integral does not give $ln|x|$:

Evaluate the $dp^0$ integral using the Feynman prescription for avoiding the poles and this will give: $$\int\frac{i}{2\pi p^1}e^{-ip^1 (t-x)}dp^1$$This integral is actually a constant multipled by a step function.

I also head into a similar problem in the (1+3)-D case where a direct Fourier transform gives a different answer from the known propagator and from the answer got by taking the limit on the massive case. So, what am I missing by directly Fourier transforming the propagator from momentum space?


  • $\begingroup$ Just Wick-rotate the integral and then observe that $d^2p=pdpd\theta$. $\endgroup$
    – Jon
    Commented Jul 8, 2019 at 15:13

2 Answers 2


The idea for this kind of computation is the following. Firstly, add a mass term to the propagator. This will yield $$ \int \frac{d^2p}{(2\pi)^2}\frac{1}{p^2-m^2}e^{ip\cdot x}. $$ This integral can be evaluated provided we make the rotation $p_0\rightarrow ip_0$ that yields $$ i\int \frac{d^2p}{(2\pi)^2}\frac{1}{p^2+m^2}e^{ip\cdot x}. $$ Now, one us $d^2p=pdpd\theta$ and $p\cdot x = pr\cos\theta$ and one has to evaluate the integral $$ \frac{1}{4\pi^2}\int_0^\infty dp\int_0^{2\pi}d\theta\frac{p}{p^2+m^2}e^{ipr\cos\theta}. $$ Firtstly, we integrate on $\theta$. This can be done remembering that $$ e^{ia\cos\theta}=\sum_{n=0}^\infty i^nJ_n(a)e^{in\theta} $$ being $J_n$ the Bessel functions of the first kind of integer order. Integration in $\theta$ leaves just $J_0$ and so, our integral becomes $$ -\frac{1}{4\pi^2i}\int_0^\infty dp\frac{p}{p^2+m^2}J_0(pr). $$ This integral can be evaluated with techniques in complex integration, with a proper choice of the integration path, yielding $$ G(r)=-\frac{1}{2\pi}K_0(mr) $$ being $K_0$ the modified Bessel function of 0 order. This is the point where the references you cite bring you. The next step is to note that, for $m\rightarrow 0$, the massless limit, $$ K_0(mr)\sim -\ln r $$ and you are done. Note the presence of an infinite constant, $\ln m$, that is generally omitted taking the massless limit. The reason is that, in the massless limit, one can always add an arbitrary constant to the propagator.

  • $\begingroup$ Is there a way to get the same answer without adding the mass term? $\endgroup$
    – adithya
    Commented Jul 18, 2019 at 8:00
  • 2
    $\begingroup$ That is an interesting question. I guess so but note that, without the mass term, after integrating on the angle you are left with a divergent integral. This is reflected by the infinite constant $\ln m$ that one gets with the mass term. $\endgroup$
    – Jon
    Commented Jul 18, 2019 at 11:25

There is already an accepted answer, but the OP is asking whether one can get the same answer without adding a mass term, so let me show how this is done.

An alternative to using mass regularization is to use regularization in the scaling dimension. Let us assume for a minute that we are not dealing with a free field, but with an interacting field in a conformal field theory. Then the 2-point function would be $$\int d^2p \left[ (p^0)^2 - (p^1)^2 + i \epsilon \right]^{\Delta - 1} e^{-i (p^0 t - p^1 x)}$$ where $\Delta$ is the scaling dimension of the interacting field. The free field theory limit is $\Delta \to 0$. The only subtlety here is that we need to define what it means to take a non-integer power of a quantity that can be negative. This is why I have added an $i \epsilon$ prescription in the propagator. This was implicit before, but somehow it was already needed to specify how to deal with the singular points where $(p^0)^2 = (p^1)^2$.

Now we can apply the same the steps as in the other answer by Jon:

  1. Wick rotation $p^0 \to i p^0$ (with $t \to i t$ as well): $$i \int d^2p \left[ -(p^0)^2 - (p^1)^2 + i \epsilon \right]^{\Delta - 1} e^{i (p^0 t + p^1 x)}$$ The $i \epsilon$ prescription tells us that this is equal to $$i e^{i \pi (\Delta - 1)} \int d^2p \left[ (p^0)^2 + (p^1)^2 \right]^{\Delta - 1} e^{i (p^0 t + p^1 x)}$$
  2. Spherical coordinates: $$ i e^{i \pi (\Delta - 1)} \int_0^\infty dp \int_0^{2\pi} d\theta p^{2\Delta - 1} e^{i p r \cos\theta}$$
  3. Integral over $\theta$: $$ i e^{i \pi (\Delta - 1)} \int_0^\infty dp ~ p^{2\Delta - 1} J^0(p r)$$
  4. Integral over $p$: first rescale $p \to p/r$ to get $$ i e^{i \pi (\Delta - 1)} r^{-2 \Delta} \int_0^\infty dp ~ p^{2\Delta - 1} J^0(p)$$ When $\Delta$ is positive but small enough this is a convergent integral that gives $$ i e^{i \pi (\Delta - 1)} r^{-2 \Delta} \frac{2^{2\Delta - 1} \Gamma(\Delta)}{\Gamma(\Delta - 1)}$$

The result is divergent in the limit $\Delta \to 0$: the $\Gamma$ function has a pole there. But if we expand in powers of $\Delta$, we get $$i \left[ -\frac{1}{2\Delta} + \log(r) + \ldots \right]$$ This is exactly the same result as with a massive field: an infinite constant, plus the log term (the constant can be absorbed in the definition of what is the "physical" scale).


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.