# Harmonic oscillator path integral: regularizing the functional determinant

From Polchinski's Vol. 1 Appendix A, we can reduce the Euclidean path integral for the 1D harmonic oscillator to computing $$(\det\frac{\Delta}{2\pi})^{-1/2}$$ where $$\Delta = -\partial_u^2 + \omega^2.$$

Since $$f_j(u) = \sqrt{2/U} \sin(j\pi u/U)$$ is an eigenbasis, we compute

$$\det \frac{\Delta}{2\pi} = \prod_{j=1}^{\infty} \frac{j^2\pi^2 + \omega^2 U^2}{2\pi U^2}.$$

Here Polchinski regularizes using Pauli-Villars; he divides by the amplitude for an oscillator with very high frequency $$\Omega$$ and gets

$$\det \frac{\Delta}{2\pi} \to \prod_{j=1} \frac{j^2 \pi^2 + \omega^2 U^2}{j^2 \pi^2 + \Omega^2 U^2} = \frac{\Omega \sinh \omega U}{\omega \sinh \Omega U}.$$

Now he takes the $$\Omega \to \infty$$ limit. Why is that the appropriate limit to take at the end?

He obtains $$(\det \frac{\Delta}{2\pi})^{-1/2}\approx \left(\frac{\omega}{\sinh\omega U}\right)^{1/2} e^{\frac12(\Omega U - \ln \Omega)}$$ and subtracts divergences and gets the right answer. But why is $$\Omega \to \infty$$ the right limit to take? When we regularize, we typically take a limit at the end to "undo" it. For instance, in QFT we take $$d= 4-\epsilon$$ and then take $$\epsilon \to 0$$ at the end.

You’re only interested in the dependence of the determinant on $$U$$ here (the numeric value for a given $$U$$ depends on the normalization constant in the functional measure and is therefore unphysical).
The expression for the determinant that you’ve given exhibits $$U$$-independence for $$\omega \rightarrow \infty$$ (the $$j$$-dependent term becomes small compared to the other summand in the numerator, and $$U^2$$ cancel the $$U^2$$ in the denominator). Hence, in the limit of infinite frequency you’re dividing by a $$U$$-independent constant — an operation which is equivalent to redefining the normalization constant. That is no longer true for finite frequencies. Intuitively, setting the frequency to infinity undoes the regularization like you expected.
• I think I got it. We don't care about the overall normalization, so we feel free to divide by some (infinite) constant. The regularization divides by a $U$-dependent function, and taking $\Omega \to \infty$ makes this function a harmless (infinite) constant which won't change the physics. Then, ultimately, Polchinski subtracts the counterterm and matches onto the $U\to \infty$ limit of the problem, which is easily solved, to determine the correct overall normalization. Jeeze, isn't there an easier way to do this?? Sep 6, 2019 at 23:05