It is known that the well-definiteness of the path integral leads to the Feynman's $i\varepsilon$-prescription for the field propagator. I've found many ways of showing this in the literature, but it is precisely the way that I have learned in my QFT course (and which I have not found in literature) that I do not understand.

Context of the problem

Considering the case of a real scalar field for simplicity, one has that the following path integral (evaluated at asymptotic times)

\begin{equation} \lim_{T \rightarrow \infty}\int_{\phi(-T, \vec{x})}^{\phi(T,\vec{x})} \mathcal{D}\phi \ \text{exp} \left( i \int^T_T dt \int d³ x \ ( \mathcal{L}+J \phi) \right)\tag{1} \end{equation}

can be expressed as

\begin{equation} \lim_{T \rightarrow \infty} \sum_{m, n} e^{-i\left(E_n+E_m \right)T} <\phi, T|n, T>_J <n|m>_J <m,-T|\phi, -T>\tag{2} \end{equation}

where $|n>_J$ are eigenstates of the hamiltonian $H$ in the pressence of the source $J$. In order to make this oscillatory exponential converge (and properly define the path integral) one adds to $T$ a small imaginary part $T \rightarrow T(1-i\varepsilon)$. With this, one writes the vacuum persistence amplitude as \begin{equation} <0|0>_J = \frac{1}{N} \lim_{\varepsilon \rightarrow 0} \ \lim_{T \rightarrow \infty(1-i\varepsilon)} \int \mathcal{D} \phi \ \text{exp} \left( i \int_{-T}^T dt \int d³x (\mathcal{L}+J\phi)\ \right) \equiv \frac{1}{N} Z[J]\tag{3} \end{equation} where the constant $N$ is typically taken to be $N=Z[0]$.

My problem

In order to relate $Z[J]$ with the Feynman propagator $D_{F}(x-y)$, one typically writes the argument of the exponential in the Fourier space, then makes the change of variable $$\hat{\phi}(p)'=\hat{\phi}(p)+(p²-m²)^{-1} \hat{J}(p)\tag{4}$$ (which leaves $\mathcal{D}\phi'=\mathcal{D}\phi$) to get

\begin{equation} Z[J]=Z[0] \text{exp} \left(-\frac{i}{2} \int \frac{d⁴ p}{(2\pi)⁴} \hat{J}(p) \frac{1}{p²-m²} \hat{J}(-p)\right).\tag{5} \end{equation} Here I'm using the $(+,-,-,-)$ Minkowski sign convention.

Now here I've been told that the fact of replacing $T\rightarrow (1-i\varepsilon)T$ to define the path integral is equivalent in Fourier space as $p⁰ \rightarrow (1+i\varepsilon)p⁰$. With this, one gets the correct $i \varepsilon$ Feynman prescription for the propagator $D_F(x-y)$

\begin{equation} Z[J]=Z[0] \text{exp} \left(-\frac{i}{2} \int \frac{d⁴ p}{(2\pi)⁴} \hat{J}(p) \frac{1}{p²-m²+i\varepsilon} \hat{J}(-p) \right) = Z[0] \text{exp} \left(-\frac{1}{2} \int d⁴y \int d⁴x \ J(x) D_F(x-y) J(y) \right). \tag{6} \end{equation}

And this is the part that I don't get at all, I've tried but I don't see how the fact that $T\rightarrow (1-i\varepsilon)T$ leads in the previous approach to $p⁰ \rightarrow (1+i\varepsilon)p⁰$ and therefore to the Feynman prescription. I'm having nightmares with this, any help would be really appreciated.

NOTE: I use the following convention for the Fourier transform \begin{equation} \hat{\phi}(p)=\int d⁴ x \ \phi(x) e^{-ip \cdot x}\tag{7} \end{equation}

so that \begin{equation} \phi(x)= \int \frac{d⁴p}{(2 \pi)⁴} \ \hat{\phi}(p) e^{+ip \cdot x}.\tag{8} \end{equation}


TL;DR: No need to lose sleep & having nightmares: The Wick rotation & the Feynman $i\epsilon$-prescription work.

  1. Advice: Use the $(-,+,+,+)$ Minkowski signature convention$^1$, cf. my Phys.SE answer here. [The opposite sign convention is also possible but there are more surprises in store along the way.]

  2. The Wick rotation treats all contravariant 4-vectors the same way. In particular, time $x^0$ and energy $p^0$ rotate in the same direction, cf. my Phys.SE answer here.

  3. The Feynman $i\epsilon$-prescription can be viewed as stopping the Wick rotation just before a full $\frac{\pi}{2}$-rotation in the complex plane, cf. my Phys.SE answer here. E.g., the energy transforms as follows $$ p^0_E~=~ie^{-i\epsilon}p^0_M \qquad \Rightarrow \qquad -(p^0_M)^2~=~ e^{2i\epsilon}\underbrace{(p^0_E)^2}_{\geq 0} ~=~(p^0_E)^2+i\epsilon $$ $$\qquad \Leftrightarrow \qquad (p^0_E)^2~=~-(p^0_M)^2-i\epsilon \qquad \Leftrightarrow \qquad p^2_E~=~p^2_M-i\epsilon.\tag{A} $$

  4. Perhaps it is most convincing to perform the Gaussian integration in the Euclidean formulation: $$\begin{align}Z_0[J]&~~~~~~=~\int\! {\cal D}\phi \exp\left\{ \frac{i}{\hbar}\int \! d^4x_M \left( J(x_M)\phi(x_M)+\frac{1}{2}\phi(x_M)(\Box_M-m^2/\hbar^2 +i\epsilon)\phi(x_M) \right)\right\} \cr ~&\stackrel{x^0_E=ie^{-i\epsilon}x^0_M}{=}~\int\! {\cal D}\phi \exp\left\{ \frac{1}{\hbar}\int_{\mathbb{R}^4} \! d^4x_E \left( J(x_E)\phi(x_E)- \frac{1}{2}\phi(x_E)(-\Box_E+m^2/\hbar^2)\phi(x_E) \right)\right\}\cr &~\stackrel{\text{Gauss. int.}}{\sim} ~\exp\left\{ \frac{1}{2\hbar}\int_{\mathbb{R}^4} \! d^4x_E ~ J(x_E)\frac{1}{-\Box_E+m^2/\hbar^2}J(x_E) \right\}\cr &~~~\stackrel{\text{Fourier}}{=}~\exp\left\{ \frac{1}{2\hbar}\int_{\mathbb{R}^4} \! \frac{d^4k_E}{(2\pi)^4} ~ \frac{\widetilde{J}(k_E)\widetilde{J}(-k_E)}{k^2_E+m^2/\hbar^2}\right\} \cr &\stackrel{k^0_E=ie^{-i\epsilon}k^0_M}{=}~\exp\left\{ \frac{i}{2\hbar}\int \! \frac{d^4k_M}{(2\pi)^4} ~ \frac{\widetilde{J}(k_M)\widetilde{J}(-k_M)}{k^2_M+m^2/\hbar^2 -i\epsilon}\right\}.\end{align}\tag{B}$$


  1. M. Srednicki, QFT, 2007; Chapter 8, p. 55. A prepublication draft PDF file is available here.


$^1$ Conventions & notations: In this answer we use the $(-,+,+,+)$ Minkowski signature convention and the speed of light is $c=1$. The subscripts $E$ and $M$ means Euclidean and Minkowskian, respectively.

  • $\begingroup$ Wow, thank you very much for writting such an elaborated answer, I had lost the chance to find the solution, now I will sleep well :) So if I am right, taking the integral limit $T \rightarrow \infty (1-i\varepsilon) $ is equivalent to considering a term like $+i\varepsilon \phi (x_M)²$ inside the lagrangian? In any case, I will have a look at Srednicki and your other answers ^^ $\endgroup$ – Guillermo Franco Abellán Feb 11 at 23:02
  • $\begingroup$ @Qmechanic, can you please shed some light on the 'surprises', if $(+,-,-,-)$ sign convention is adopted. $\endgroup$ – MadMax Apr 18 at 14:11
  • $\begingroup$ See e.g. my Phys.SE answer here. $\endgroup$ – Qmechanic Apr 18 at 20:14

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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