From P.M.C.Wright@ncl.ac.uk Mon Jan 22 07:13:26 1996
Subject: problems with NONMEM and windows 3.1 NONMEM users
I am helping a colleague to implement NONMEM on a PC (486) running windows 3.1. We have followed the memos which accompany the installation materials and had some success. However when we try to run a population analysis and include the lines
IPRED = F
IRES = DV - IPRED
IWRES = IRES/F
then NONEM fails. The search commences but exits almost immediately. When we run identical files in an established UNIX implementation of NONMEM there is no problem. Has anybody had this problem?
Department of Anaesthesia, University of Newcastle upon Tyne, UK
From lewis Mon Jan 22 09:25:21 1996
Subject: Re: problems with NONMEM and windows 3.1 This may not be any help here, but I notice that in the line
IWRES = IRES/F
you are dividing by F - that will generate a zero-divide if F ever takes the value 0 ... perhaps your UNIX-based implementation recovers from a zero-divide gracefully and your PC one does not.
PREDPP knows nothing about IRES and such things, so it cannot protect you from zero-divides (it does, for example, protect you from zero-divides if you mistakenly assign V1 = 0 for example, because it understands volumes and what they mean).
From alison Mon Jan 22 09:31:12 1996
Subject: Problems with NM and Windows 3.1
My guess is that F=0 for some data records. (E.g., if PREDPP is being used, there may be infusion doses rather than bolus doses, so that, for the dose event records, F=0.) A division by zero occurs. It could be that on the Unix system, IWRES is set to Inf and the run continues. This Inf (Infinity) does no harm, because IWRES does not enter into the objective function, and is being calculated only for display purposes. But on the PC, the division by zero causes the run to terminate. (The Fortran 77 standard does not give any guidelines as to how arithmetic exceptions should be handled.) The division by zero can be avoided by code such as the following:
IF (F.EQ.0) DEL=1