My question is about the calculation of a functional integral (which looks like a partition function).

If we have the operator $A$ having discrete spectrum, and eigenvectors $\phi_{i}$ and eigenvalues $\lambda_i$, and the eigenvalues have density $\rho(\lambda)=\sum_i \delta(\lambda_i-\lambda)$. Then the functional integral $\int D\phi \exp\{-(\phi,A\phi)\}$ is written as integral over the "fourier" coefficients $a_i = (\phi_i,\phi)$, where we use $\phi = \sum_{i} a_i \phi_i$:

$\int D\phi \exp\{-(\phi,A\phi)\} = \int_{-\infty}^{\infty}\prod_i da_i \exp\{-\lambda a_i^2\}$

This is product of gaussian integrals, so if we put "UV" cutoff on eigenvalue, $\Lambda$, then I think the result of this integral is the product of the inverse square root of eigenvalues:

$ \int^{\Lambda}\prod_i da_i \exp\{-\lambda a_i^2\}= \prod_i^{\Lambda}\lambda_i^{-1/2}$

However, my TA wrote without deriving that the correct result is this:

$N(\Lambda) \ e^{-1/2 \sum_\lambda^{\Lambda}\ln \lambda}$

where $N(\Lambda) = \int \rho(\lambda) \ d\lambda$ is the number of eigenvalues below $\Lambda$.

I don't understand this result, especially this appearance of this counting number $N(\Lambda)$. Why does the functional integral not give the product of eigenvalues, and where is this $N(\Lambda)$ coming from? Is this related to the density of states that is sometimes written in a partition function? I would appreciate any help in understanding this, and any pointer to books that explain as well.


I think you TA is either wrong, or you misunderstood what he wrote. You are correct up the factor of $(\sqrt \pi)^n$ where $n=\int_0^\Lambda \rho(\lambda) d\lambda$ from the $n$ Gaussian integrals. I expect that the TA just meant that there is some normalization factor $N(\Lambda)$ that depends on $\Lambda$, but that factor is the one I gave above, and not $\int_0^\Lambda \rho(\lambda) d\lambda$

| cite | improve this answer | |
  • $\begingroup$ Thank you for your reply. I think it is likely that yes the factor $N(\Lambda)$ is not the integral of the density of eigenvalue. Otherwise it doesn't make sense. In general however, when a sum is transformed into integral over eigenvalues using $\rho$, then the measure will be $\int \rho(\lambda) d\lambda (...)$, right? This means that the integral depends on $N(\Lambda)$ correct? Thank you again. $\endgroup$ – TJNY699 Mar 3 '19 at 18:33
  • $\begingroup$ @TJNY699 The integral depends on the normalization, yes. But the correlators etc. do not, as it cancels out when you divide by the partition function. $\endgroup$ – mike stone Mar 3 '19 at 21:03

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.