[molpro-user] disk space in CCSD(T) and integral thresholds

Fabio Mariotti Fabio.Mariotti at ir.phys.chem.ethz.ch
Fri Aug 12 10:56:31 BST 2005

	Dear Molpro users,

	I need to know if there is any little trick
	I can use to reduce just a bit the disk space.

	I can use a nice cluster with queues of max 8h
	and local scratch disk of about 80Gb (on the same
	Linux OS physical disk)
	and I have a CCSD(T) calculation which takes
	8h30min (quite long time spent in I/O)
	and ~85Gb of scratch. (and cannot use symmetry:( )

	So the questions:
	1) Default compression algorithm is aoint,compress=1.
	is there any different option to improve compression?

	2) the code writes first files with names of the type:
	which if I understood correctly are the main
	integral files, then files of the type
	which should be the sort files.
	after the sort, it starts to create files with names of the type:
	which seem to be a copy of the integral files.
	I removed the files eaf_T2100004985.TMPx001 by hand
	while the program was generating the eaf_T3100004985.TMPx001 files.
	and the calculation ended without any problem and using about 20Gb
	less of disk space.
	Can I avoid this extra copy of integrals?

	3) Which one is the correct option to set the integral threshold?
	I have an input with both:
	but on output I get:
 Contracted 2-electron integrals neglected if value below      1.0D-12
 AO integral compression algorithm  1   Integral accuracy      1.0D-12

	Am I misunderstanding or I have a problem with my compiled version?

	thanks in advance

Dr. F.Mariotti - mariotti at ir.phys.chem.ethz.ch
ETH Zurich - Honggerberg, Zurich, Switzerland

More information about the Molpro-user mailing list