The differences in GAMESS (US) and PC GAMESS/Firefly behavior are caused by several reasons. First of them, recent GAMESS (US) versions use HONDO integral package for diffuse s, p, and sp shells even if inttyp is set to Pople. This is quite a reasonable approach which helps to improve precision, nevertheless, as Pople package is still used for most sp integrals, the overall accuracy is not very high. This is why we decided not to follow this way, allowing user to explicitly specify integral package to use. Moreover, we plan to completely eliminate both Pople and HONDO integral packages in the future and to replace them by fastints throughout all the code. Meantime, all runs involving old integral packages should explicitly specify inttyp=HONDO in the case of diffuse basis functions and quasi-linear dependences.
Another difference is that for all direct runs, GAMESS (US) internally sets ICUT to MAX(ICUT,10). This allows to use fdiff option even for very large molecules and basis sets, but causes performance degradation for smaller systems, where the default value of ICUT (9) usually works fine. Once again, with PC GAMESS/Firefly it is completely up to user to provide ICUT value. We recommend to use ICUT=10 with fdiff=.t. for large systems and basis sets, especially for those containing relatively large number of d functions and above, and to use default value of ICUT in other cases.
Last updated: March 18, 2009