Suppose we have the path integral:

\begin{equation} Z=\int \mathcal{D}x\mathcal{D}y\,\exp\left[-\frac{a}{2}\int_0^1 dt\,\left(\,\dot{x}(t)^2-\,\dot{y}(t)^2\right)\right]. \end{equation}

The integration over $x$ is trivial: it is the well known result for a free particle:

\begin{equation} \int \mathcal{D}x\,\exp\left[-\frac{a}{2}\int_0^1 dt\,\dot{x}(t)^2\right]=\sqrt{\frac{a}{2\pi}}\exp\left[-\frac{a}{2}(x_f-x_i)^2\right] \end{equation}

where $x(1)=x_f$, $x(0)=x_i$.

What about the other integration? It has the "wrong" sign. One could think to apply the same formula with $a\to-a$, hence getting an imaginary unit from the negative square root. Would be this correct? How would you compute this path integral?

The integral comes from https://doi.org/10.1103/PhysRevD.43.2572, equation (4.9). One should be able to pick up a convergent contour in the complex plane where the integral should converge. It is however not clear which contour to choose and why.

  • $\begingroup$ Strictly speaking,no it would not be correct. But usually physicists don’t care. The path integral is mathematically ill defined anyway. The physicists approach is - just do it and see what we get. If the final result makes sense, then the manipulation is probably correct. $\endgroup$
    – Prahar
    Mar 28, 2022 at 14:46
  • 1
    $\begingroup$ Assuming $a>0$ the path integral over $y\in\mathbb{R}$ is divergent. Is this from a reference? Which page? $\endgroup$
    – Qmechanic
    Mar 28, 2022 at 15:20
  • $\begingroup$ It is from journals.aps.org/prd/abstract/10.1103/PhysRevD.43.2572, equation (4.9). It is said that the integral can be easily calculated, but they do not show how. They talk about a contour in the complex plane but it is not clear which one to choose and why. $\endgroup$ Mar 28, 2022 at 16:08
  • $\begingroup$ Am I reading something wrong in that paper? In equation (4.9), the action $I_2$ from equation (4.8) has (only) kinetic terms, it comes as $\frac{1}{8N}\int_0^1\mathrm{d}\tau (-\dot{X}^2+\dot{Y}^2)$. $\endgroup$ Apr 1, 2022 at 11:38

1 Answer 1


Hint: For a Gaussian integral $$ \int_{e^{i\phi}\mathbb{R}} \mathrm{d}z ~e^{-\frac{1}{2}re^{i\theta}z^2}~\stackrel{z=e^{i\phi} x}{=}~e^{i\phi} \int_{\mathbb{R}} \mathrm{d}x ~e^{-\frac{1}{2}re^{i(\theta+2\phi)}x^2}~=~\sqrt{\frac{2\pi}{re^{i\theta}}} \tag{1}$$ to be convergent, $$\begin{align}0~\leq&~{\rm Re} (e^{i(\theta+2\phi)})\cr &\Updownarrow\cr \phi~ \in~& \left[-\frac{\pi}{4}- \frac{\theta}{2},\frac{\pi}{4}- \frac{\theta}{2} \right] +\pi\mathbb{Z}. \end{align} \tag{2}$$ Notice that as long as the integral is convergent, it doesn't depend on the contour angle $\phi$ [up to a subtlety with the double-valueness of the square-root in eq. (1)].


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.