PFGW 3.3.5 and PFGW 3.3.6 release
- A bug was discovered in gwnum that triggers a carry propagation issue. In turn gwnum returns what appears to be a valid result, but one that is actually invalid. This predominantly occurs with special modular reduction for bases > 2. I have patched gwnum (based upon George's recommendations) to choose a larger FFT size so that the carry propagation issue can be avoided. George is working on a permanent fix.
I have modified PFGW 3.3.5 to support a new switch -F. This switch will output the FFT size of the test using the buggy version of gwnum. PFGW 3.3.6 has the patch. By comparing the output from the -F switch you can determine which candidate numbers need to be retested. Based upon some preliminary estimates, the number of retests for b=58 is about .3%. The number of retest for b=1029 is 12%. Bases that are multiples of 2 are not affected.
You can d/l PFGW 3.3.5 and 3.3.6 from here, http://sourceforge.net/projects/openpfgw/
I'm sorry for those of you inconvenienced by this. It is most painful on the CRUS and GW/GC projects out there, of which I have spent more time than most.