Summation over Matsubara axis#
In many cases, we want to perform the summation of a Greens-function-like object \(f(\mathrm{i}\nu)\) over the Matsubara axis. Here, \(\nu = n\pi/\beta\) runs over the fermionic (odd \(n\)) or bosonic (even \(n\)) Matsubara frequencies; the letter \(\omega\) is reserved for real frequencies and \(A\) for the spectral function (see Notation and conventions).
The Fourier transform of \(f\) reads
This leads to the following the two formula:
We now expand \(f(\mathrm{i}\nu)\) at high frequencies as
As discussed in Sec. B3 of E. Gull’s Ph. D thesis, \(f(\tau=0^+) = f(\tau=0^-)\) if and only if \(c_1 = 0\). This condition is equivalent that \(f(\mathrm{i}\nu)\) vanishes at high frequencies faster than \(O(1/{\mathrm{i}\nu})\). More precisely, \(f(\tau=0^+) - f(\tau=0^-) = -c_1\): the Green’s function of an elementary operator has \(c_1 = 1\) and jumps by \(-1\) at \(\tau = 0\) for both statistics. If \(c_1 \neq 0\), the summation does NOT converge without a convergence factor and thus \( \sum_{\nu} f(\mathrm{i}\nu) e^{\mathrm{i}\nu 0^+} \neq \sum_{\nu} f(\mathrm{i}\nu) e^{\mathrm{i}\nu 0^-}\).