CHARMM Development Project Forums CHARMM Community Bug Reports & Fixes PGI problem with CORFUN

I've noticed that there seems to be a problem when computing correlation functions using CHARMM compiled with Linux PGI; large artefacts are observed in the results of CORFUN which are absent when the same calculation is repeated using CHARMM compiled with GNU g77.

Has anyone else noted this problem, and is there a workaround or fix for this?

TIA

Rick,
CORFUN has two methods that should give nearly identical results, DIREct and FFT. The default is FFT, and it's much faster. Could you try both and see if the problem is with the FFT code, or is somewhere else...

Bernie

dear Rick, naive question, does C(tau)=<f(t)f(t-tau)> show the wild artifacts only with certain types of functions f, or at characteristic values of tau, eg, tau ~0. truncation error is possible. how important is this problem...is gnu f77 an obviously better compiler? richard pan, happy to work towards an acceptable resolution...

The problem occurs when the P2 correlation function is computed using FFT. I get the same results with pgf77 (v 4.3 and 5.0) and pgf90 (v5.0) and I believe it to be a compiler bug (which I have not reported to the Portland people). It disappears if you turn off optimization for cordyn.f in correl.mk:
$(LIB)/correl.a(cordyn.o) : $(SRC)/correl/cordyn.src
$(FLX) $(SRC)/correl/cordyn.src
# NB: corfun using fft goes wrong using pgf77 with FC2 LNI
$(FC0) cordyn.f

A workaround without recompiling is to use DIREct instead of FFT for the correlation function calculation.

Lennart