Support added for Molcas v8.1.15-12-09 (fix to support structure optimizations with v8.1 newer
than about march 2015). Hence, there are two versions of one Columbus module (tran.x, tran.x_81)
which is called as appropriate (i.e. tran.x for v8.0 and old v8.1; tran.x_81 for current v8.1).
bug fix for excited state densities (incorrect normalization check for 2e density)
COLUMBUS/MOLCAS binaries rev. 1.0.4 (feb 9,2016)
readme.pdf (Installation and Trouble Shooting)
Several bugfixes and optimizations in the Columbus part. Switched from mpich1.2.7 to mpich2-1.4.1.
Note, that mpich1.2.7 is script based while mpich2-1.4.1 is partially based on binaries loading
shared libraries. This might produce failures with older operating system distributions.
In case of trouble, drop an email to th.mueller@fz-juelich.de.
The fix from feb 9,2016 prevents a failure due to the incorporated BLAS libraries on a system with
avx2 registers and has missing links added to the mpich2/bin subdirectory.
COLUMBUS/MOLCAS binaries rev. 1.0.5 (april 19,2016)
readme.pdf (Installation and Trouble Shooting)
Several bugfixes and optimizations in the Columbus part,
especially fixing the daio: maxrec exceeded , due to an incorrect file size estimate
(thanks to N. Bogdanov, to point out the problem and provision of a test case).
This version changes for some additional files to compression, thereby reducing the
disk and memory space requirements. Per default the parallel CI code keeps everything
in memory while the seriel CI code uses disk space, instead. Although, it is not yet
moved in to the interface, there is the possibility to drastically reduce the memory
consumption for the subspace vectors forcing them to disk (ciudgin: fileloc=1,0,0 to
be added manually, currently) - for the parallel code, consider using a SSD then.
Vice versa, the seriel code can be forced to run fully in memory (ciudgin: fileloc=1,1,1 to
be added manually).
Early termination of the CI code due to incorrect evaluation of integral distribution size
fixed.
Incorrect printout of the size of the MO basis fixed.
In case of trouble, drop an email to th.mueller@fz-juelich.de.
COLUMBUS/MOLCAS binaries rev. 1.1.0 (august 23,2016)
readme.pdf (Installation and Trouble Shooting)
Major changes in the low-level operation in order to facilitate efficient multi- and many-core operation
Global Array Toolkit is now completely replaced by a small library directly utilizing shared memory functionality
Global shared counter (for dynamic load balancing) replaced by a faster scheme based on semaphores
which does not suffer from saturation effects
New storage format for more efficient storage of integrals, densities and ci vectors reducing disk and memory
usage for extended systems by up to 90%
New parallelization scheme that largely decouples load balancing from the associated communication volume
leading to a reduction of the overall network load by more than one order of magnitude.
COLUMBUS/MOLCAS binaries rev. 1.3.0 (january 3,2017)
readme.pdf (Installation and Trouble Shooting)
Added support for excited state single-point and structure optimizations at MRAQCC, MRACPF and
LRT-MRAQCC level of theory
Test cases 30p through 37p illustrate the usage.
COLUMBUS/OpenMOLCAS binaries rev. 2.1 (sept 10, 2018)
readme.pdf (Installation and Trouble Shooting)
This version constitutes a pre-configured docker image with Columbus 7.1 /OpenMolcas (sept 9,2018 snapshot) which should work fairly independent of the host configuration of the Linux system. Given an installed and running docker and containerd daemon, unzip the tar ball and load it to the docker image repository; now run this image in a docker container interactively and it should work straight away (seriel and parallel on a single host). Please read the readme.pdf file carefully. This docker image is experimental, since documentation is not fully updated, important examples and benchmarks have not yet been included and some scripts may still be missing. The binaries are based on intel 18 compilers, OpenBlas 0.3.0 and mpich2. The OpenMolcas source code used to derive these binaries can be found in /software/OpenMolcas/openmolcas.tgz.
Primarily, rewritten/extended/revised parallel versions of the conventional SCF (RHF,ROHF), MCSCF (now directly supporting frozen core/frozen virtual orbitals, directly constructing density and transition densities, on-disk,in-memory and on-the-fly coupling coefficient storage and evaluation, respectively,), somewhat generalized AO-MO transformation and CI gradient code. Columbus 7.1 is increasingly focussing on speed, parallelism and extensive shared memory usage.