WIP: Implemented sin w->t grids

[marm314]

Dear all,

I have implemented the construction of the sine w->t grids that were missing. As suggested in Issue 17 (#161), the sin_w_to_t grids have been implemented as optional in minimax_grids.F90 to avoid problems with other codes already using greenX. I have also updated the test_gx_minimax_grid.f90, which now shows how to request these grids. Please, let me know what/how test(s) should be incorporated (or, feel free to directly add them).

Looking forward to hearing from you,

Mauricio

[marm314]

@JWilhelm @fdelesma

[fdelesma]

Left a few comments to address. I’ll be on a business trip for a bit, but I’ll check back once I return.

[marm314]

Hello,

Thank @fdelesma for the revision and your comments. I have updated the files as suggested.
Looking forward to hearing from you if there are more comments or suggestions to address.

Bests,

[fdelesma]

@marm314 Thanks for your changes. I worked on the regression test for the sine transformation duality error, both with and without regularization. It should now meet the requirements for this MR. @moritzleucke, could review it.

[fdelesma]

Looks good to me - let’s try to merge!

[fdelesma]

Looks like the Python version in the Actions might be causing the problem.
@moritzleucke @marm314, it works fine locally for me with Python 3.10.12.
Could you run the ctest to check if you see the same issue?

[marm314]

@fdelesma I have an externally managed python evironment and cannot install pygreenx. I am/was just using the Fortran parts.

[fdelesma]

I compile greenX in a difference machine. Here are my findings
The reference sine duality error without regularization is:

    ref_errors = np.array([[6, 2.50000000E+02, 5.56242714E-02, 2.36021543E-01, 8.74792753E-01],
                           [8, 2.50000000E+02, 1.49383520E-02, 1.11607214E-01, 8.80901314E-01],
                           [10, 2.50000000E+02, 3.87442807E-03, 2.61545954E-02, 9.62458551E-01],
                           [12, 2.50000000E+02, 9.86318209E-04, 1.36267244E-02, 9.39611514E-01],
                           [14, 2.50000000E+02, 2.46521564E-04, 4.19391179E-03, 9.55740373E-01],
                           [16, 2.50000000E+02, 7.70304736E-05, 5.97815737E-04, 9.42329775E-01],
                           [18, 2.50000000E+02, 1.94942501E-05, 1.39775660E-04, 9.45743358E-01],
                           [20, 2.50000000E+02, 4.88850816E-06, 3.73426583E-05, 9.40568100E-01],
                           [22, 2.50000000E+02, 2.68402244E-07, 2.38208860E-05, 1.01253368E+00],
                           [24, 2.50000000E+02, 1.31534818E-07, 2.01595323E-06, 2.04590429E+00],
                           [26, 2.50000000E+02, 3.78405247E-08, 6.37343151E-07, 8.82902975E+00],
                           [28, 2.50000000E+02, 4.30436009E-08, 8.53519608E-07, 1.59367298E+05],
                           [30, 2.50000000E+02, 5.14232643E-08, 1.07148550E-06, 1.05986563E+06],
                           [32, 2.50000000E+02, 4.94132300E-08, 1.02545263E-06, 1.12950579E+06],
                           [34, 2.50000000E+02, 6.66088497E-08, 1.45835462E-06, 5.71640192E+06]])

I got

#n   e_range           freq_to_tau_error tau_to_freq_error duality_error
6    2.50000000E+02    5.56242714E-02    2.36021543E-01    8.74792753E-01
8    2.50000000E+02    1.49383520E-02    1.11607214E-01    8.80901314E-01
10    2.50000000E+02    3.87442807E-03    2.61545954E-02    9.62458551E-01
12    2.50000000E+02    9.86318209E-04    1.36267244E-02    9.39611514E-01
14    2.50000000E+02    2.46521564E-04    4.19391179E-03    9.55740373E-01
16    2.50000000E+02    7.70304736E-05    5.97815737E-04    9.42329775E-01
18    2.50000000E+02    1.94942501E-05    1.39775660E-04    9.45743358E-01
20    2.50000000E+02    4.88850816E-06    3.73426583E-05    9.40568100E-01
22    2.50000000E+02    2.68402246E-07    2.38208860E-05    1.01253368E+00
24    2.50000000E+02    1.31534823E-07    2.01595323E-06    2.04590429E+00
26    2.50000000E+02    3.78405286E-08    6.37343525E-07    8.82903928E+00
28    2.50000000E+02    4.30436019E-08    8.54592843E-07    1.58948056E+05
30    2.50000000E+02    5.14232635E-08    1.07518952E-06    5.40678062E+05
32    2.50000000E+02    4.94132291E-08    1.02033059E-06    7.64394471E+05
34    2.50000000E+02    6.66088490E-08    1.43209368E-06    5.15145142E+06

It seems that the duality errors for the larger grids (28–34) are too large and unstable. When regularization is applied, everything looks fine. I’ll work on a fix for that.

[marm314]

I also noticed the same for the duality errors; in practice, I am using at least 1e-5 for the regularization in my calculations using these grids.

[moritzleucke]

looks good to me 👍