[Columbus] Fwd: mcscf.x MO integral read error

[Christopher Malbon]

>From $COLUMBUS/docs/fulldocs/sifs.txt:
info(6) contains information on the format of the packed integral labels
read from the header in the integral file.
0 = 8 bits/label. 1 = 10 bits/label
ifmt is the format of the packed integral labels in the buffer.

Systems with basis functions >256 require >8 bits to encode a basis
function label.
It's essential a consistency check to ensure you are assigning integrals
the appropriate basis function labels. This is at least my understanding.

Best

Christopher L Malbon
Yarkony Group
Department of Chemistry
The Johns Hopkins University
cmalbon1 at jhu.edu

---------- Forwarded message ----------
From: Christopher Malbon <clmalbon at gmail.com>
Date: Wed, Jul 20, 2016 at 3:28 PM
Subject: Re: [Columbus] mcscf.x MO integral read error
To: Scott Brozell <srb at osc.edu>, Christopher Malbon <clmalbon at gmail.com>,
columbus at osc.edu


>From $COLUMBUS/docs/fulldocs/sifs.txt:
info(6) contains information on the format of the packed integral labels
read from the header in the integral file.
0 = 8 bits/label. 1 = 10 bits/label
ifmt is the format of the packed integral labels in the buffer.

Systems with basis functions >256 require >8 bits to encode a basis
function label.
It's essential a consistency check to ensure you are assigning integrals
the appropriate basis function labels. This is at least my understanding.

Best


Christopher L Malbon
Yarkony Group
Department of Chemistry
The Johns Hopkins University
cmalbon1 at jhu.edu

On Wed, Jul 20, 2016 at 2:24 PM, Scott Brozell <srb at osc.edu> wrote:

> Hi,
>
> The error message is new code in file colib8 subroutine sifd1 of
> columbus version 7 that is not in older columbus versions:
>
>       call ulab16( buffer(1), unpack, 4 )
>       num    = unpack(1)
>       lab1   = unpack(2)
>       last   = mod(unpack(4),         4)
>       ifmt   = mod(unpack(4)/2**2,    8)
>       itypeb = mod(unpack(3),      1024)
>       itypea = mod(unpack(3)/2**10,   8)
>       ibvtyp = mod(unpack(3)/2**13,   8)
> c
> c     ifmt is now double checked
> c
>        if (ifmt.ne.info(6)) then
>         call bummer(' sifd1: ifmt.ne.info(6)',0,2)
>        endif
>
> The comment does not indicate why this check is useful.  Perhaps the
> author of the new code can describe its purpose.
>
> scott
>
> On Tue, Jul 19, 2016 at 05:16:47PM -0400, Christopher Malbon wrote:
> > mcscf.x routine fails with the following error message:
> > end of hernew
> >  IGNORING BNDCHK
> > *** End of DALTON calculation ***
> > end of mcdrt
> > end of mcuft
> >  bummer (fatal): sifd1: ifmt.ne.info(6)0
> > bummer
> >
> > This bummer error is called during the read of the integral records after
> > ulab16 unpacks the buffer.
> >
> > I am using the Col7.0_2015_01_15 release, building from source.
> > I am using the cray intel fortran wrappers on the NERSC cori machine.
> > I had to change the cpp flags omitting the -C option. (In gcc 4.8.x, cpp
> -C
> > includes the header file stdc-predef.h which prints the GNU license at
> the
> > top of each preprocessed file commented C-style, which fortran compilers
> do
> > not handle.) Other than this change, no other changes were made - with
> the
> > exception of 'ftn' and 'cc' for 'ifort' and 'icc', respectively.
> >
> > I am installing using machine.cfg/linux64.ifc.byterecl.
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osu.edu/pipermail/columbus/attachments/20160720/6e068a5a/attachment-0001.html>