From: "Nandy, Partha" <Partha.Nandy@pharma.com>
Subject: MSDOS versus VisualNM
Date: Thu, 16 Aug 2001 10:07:12 -0400
We have recently installed VISUAL NM and I am running into some problems. I
was running a ADVAN2 TRAN 2 on a Dell Pentium 3 machine and got Min OBJ
Function of 1694 with the following estimates for Cl, V and Ka (Cl = 78.7
V=705 and Ka=0.296). Recently I tried running the same program and data
file using Visual NM and got very very different results. The Min OBJ
Function was 1982 and Cl, V and Ka were 25000, 672, and 0.277, respectively.
Needless to say that the parameter estimates were not only off, the
predictions are way off as well.
Can someone suggest how to rectify this problem or what might be the source
of this problem?
Thanks in advance for all your help.
From: "Bachman, William" <email@example.com>
Subject: RE: MSDOS versus VisualNM & NONMEM installation
Date: Thu, 16 Aug 2001 11:29:10 -0400
The critical factors determining the output of a NONMEM run on Wintel*
1. NONMEM version and level
2. FORTRAN compiler brand, its version and update level, and compiler
Factors NOT influencing the output are:
1. PC manufacturer (Dell, Gateway, etc.)
2. PC processor (pentium 1/2/3/etc.)
3. Use of an interface (Visual NM, etc.)
4. Windows operating system (Win95, 98, Me, NT, 2000)
Therefore, if the compiler and its options and NONMEM version and level are
the identical on two different machines, use of Visual-NM will not influence
your output using identical control streams and data files. Visual-NM is
merely a shell that operates on top of the core NONMEM program (installed
using a specified compiler and options).
Also, testing the installation using CONTROL3 and DATA3 is NOT SUFFICIENT to
prove correct installation. The most common installation error made is
using the default arguments for setup.bat (usually by not specifying any
arguments on the command line). The default setup specifies the
now-obsolete MS Powerstation compiler (fl32). The problem lies in that
DIGITAL/Compaq compilers have a Powerstation emulation mode that allows
installation using this default and that the emulation mode is defective.
Runs with CONTROL3 and DATA3 will actually run correctly, BUT, other runs
will fail. Testing with CONTROL5 and THEOPP will FAIL under this defective
install. (This is why Atul suggested "real world" testing in an earlier
1. If installing from the command line, specify all arguments to setup.bat:
setup a c nmv df y link (for DIGITAL/Compaq installations)
If installing through Visual-NM make sure that the option for DIGITAL/Compaq
compiler (df) is used and the compiler options are the same as used for your
command line installation.
2. Test installation using CONTROL5 and THEOPP.
*Windows/Intel is specified here so as to keep apples/oranges comparisons to
a minimum. This statement was made as specific as possible to properly
address the issues at hand and does NOT imply that output from different
compilers or different platforms will necessarily be different.
7250 Parkway Drive, Suite 430
Hanover, MD 21076
Voice: (410) 782-2205
FAX: (410) 712-0737
From: Nick Holford <firstname.lastname@example.org>
Subject: Re: MSDOS versus VisualNM
Date: Fri, 17 Aug 2001 06:48:19 +1200
You do not say if you are using the same compiler with the same compiler options for both cases. The compiler and its options can influence how NONMEM searches for a minimum. Visual NM cannot directly change the way that NONMEM performs but if you have compiled NONMEM for Visual NM with a different compiler or different options compared with the when you used NONMEM without Visual NM then this may be the cause of the differences.
Nick Holford, Divn Pharmacology & Clinical Pharmacology
University of Auckland, 85 Park Rd, Private Bag 92019, Auckland, New Zealand
email:email@example.com tel:+64(9)373-7599x6730 fax:373-7556