A test on analytic continuation of thermal imaginary-time data
Some time ago, Cuniberti et al. have proposed a novel method for analytically continuing thermal imaginary-time correlators
to real time, which requires no model input and should be applicable with finite-precision data as well. Given that these
assertions go against common wisdom, we report on a naive test of the method with an idealized example. We do encounter two
problems, which we spell out in detail; this implies that systematic errors are difficult to quantify. On a more positive
note, the method is simple to implement and allows for an empirical recipe by which a reasonable qualitative estimate for
some transport coefficient may be obtained, if statistical errors of an ultraviolet-subtracted imaginary-time measurement
can be reduced to roughly below the per mille level.