Dear all, I am trying to use the maxent code from ALPS. In the document hybdoc.pdf, it is being suggested that "it helps to continue the self-energies instead of the Green's functions". But the CT-QMC hybridization code only returns the imaginary frequency self-energy, and it does not mention error associated with it. While it is suggested that one should use data with error bars for the analytic continuation.
In the sample parameter file in the hybdoc.pdf, even though DATASPACE=frequency, the data points(X_i) are taken as real numbers. In general it should be a complex numbers.
Can somebody clarify this as to how to input the self-energy (complex number) and what should one do about the errors associated with the self-energy? I tried to give input as X_i=(0.5,0.5), but parameter2xml gives error when it tries to convert the input file to xml format. It says - Caught exception: parameter parse error at "{ PARTICLE_HOLE_SYMMETRY = 1 X_0".
So, can someone send some sample input file for the analytic continuation of imaginary frequency data (complex numbers) to the real frequency.
With best regards, Viveka Nand Singh
The second question is easy to answer: For a particle hole-symmetric system (i.e. PARTICLE_HOLE_SYMMETRY=1), the self-energy has an imaginary part only, the real part only consists of the constant Hartree term, which must be subtracted. In this case, X_i corresponds to Im(S(i*wn)), und SIGMA_i to the corresponding errorbar (to be discussed later). If you have a general situation (PARTICLE_HOLE_SYMMETRY=0) with a complex self-energy, the structure is as follows: X_0 = Re(S(i*w0))-Hartree, X_1 = Im(S(i*w0)), X_2= Re(S(i*w1))-Hatree, X_3 = Im(...), ... and correspondingly the SIGMA_i. Now to the error bars. The code unfortunately does not provide them directly, so you need to analyze for example the time series. To obtain statistically independent measurements, it is however better to perform a series of independent simulations with different seeds and then perform a standard statistical analysis on them, providing a mean and error. Hope that helps. I apologize for the bad shape the docu is in, but I did not have the time yet to write a better one and/or provide tutorials for the use of maxent. Best TP
comp-phys-alps-users@lists.phys.ethz.ch