7.6 KiB
QMC=Chem : Quantum Monte Carlo for Chemistry
QMC=Chem is the quantum Monte Carlo program of the Toulouse (France) group. It is meant to be used in the post-Full-CI context : a quasi-Full-CI calculation is done with the quantum package, and this wave function is used as a trial wave function for the fixed-node diffusion Monte Carlo algorithm.
- Parallel efficiency of 98.3% on 16_000 cores
- The load balancing is optimal: the workers always work 100% of the time, independently of their respective CPU speeds
- Efficient: 0.96 Pflops/s on 76_800 cores of Curie in 2011
- All network communications are non-blocking, with the ZeroMQ library
- All the implemented algorithms are CPU-bound : the only limit is the available CPU time
- The number of simultaneous worker nodes can be variable during a calculation
- Fully fault-tolerant (crashing nodes don’t stop the running calculation)
- QMC=Chem has been used in grid environments (EGI european grid) and in Cloud environments (France Grilles) coupled to supercomputers
Warnings: * QMC=Chem is under the GPLv2 license. Any modifications to or software including (via compiler) GPL-licensed code must also be made available under the GPL along with build & install instructions. * Pseudopotentials are about to change in the EZFIO database. Current calculations will not be compatible with future versions
Requirements
- Ninja build tool
- OCaml compiler with Opam and Core library
- ZeroMQ high performance communication library
- F77_ZMQ ZeroMQ Fortran interface
- IRPF90 Fortran code generator
- EZFIO Easy Fortran I/O library generator
- GNU C++ Compiler (g++) for ZeroMQ and Ninja
- Python >= 2.6 for install scripts
- Bash for install scripts
- Fortran compiler, Intel Fortran recommended
- Lapack library, Intel MKL recommended
Most of the dependencies are open-source will be downloaded automatically. The Fortran and C++ compilers, Python and Bash interpreters and the Lapack library need to be installed manually by the user.
Installation
The make.config
file contains compiler specific
parameters. You should change them to match your hardware.
The configure.sh
script will first download the Ninja build tool, and will
then run Ninja using the install/build.ninja
file. The
configuration script will work in the install
directory. It
will first download into the install/Downloads
directory
everything that needs to be installed. The building of the dependencies
takes place in the install/_build
directory, and the
packages that are being installed can be followed by looking at the log
files in this directory. When a package was successfully installed, a
*.ok
file is created and the log file is deleted.
If you don’t have an internet connection available, you can execute
the downloading step on another computer and transfer all the downloaded
files into the Downloads
directory.
Before using or compiling QMC=Chem, environment variables need to be
loaded. The environment variables are located in the
qmcchemrc
file:
$ source qmcchemrc
To compile the program, run
$ ninja
Example of a QMC=Chem calculation
Calculation with the quantum package ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
- Create the
xyz
file containing the nuclear coordinates of the system
$ cat > h2o.xyz << EOF
3
Water molecule
O 0. 0. 0.
H 0.9572 0. 0.
H -0.239987 0.926627 0.
EOF
- Choose a suitable basis set and create the EZFIO database
$ qp_create_ezfio_from_xyz -b cc-pvdz h2o.xyz -o h2o
- Run the SCF calculation
$ qp_run SCF h2o
- Run the CIPSI calculation
$ qp_run full_ci h2o
- Transform the input for use in QMC=Chem
$ qp_run save_for_qmcchem h2o
FN-DMC calculation with QMC=Chem ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Before using QMC=Chem, you need to load the environment variables:
$ source qmcchem.rc
In QMC=Chem, everything goes through the use of the
qmcchem
command. When a command is run with no arguments,
it prints a help message. This is mainly the manual of QMC=Chem. For
example:
$ qmcchem
QMC=Chem command
qmcchem SUBCOMMAND
=== subcommands ===
debug Debug ZeroMQ communications
edit Edit input data
md5 Manipulate input MD5 keys
result Displays the results computed in an EZFIO directory.
run Run a calculation
stop Stop a running calculation
version print version information
help explain a given subcommand (perhaps recursively)
missing subcommand for command qmcchem
$ qmcchem edit
Run a calculation
qmcchem run EZFIO_FILE
Run QMC=Chem
=== flags ===
[-a] Add more resources to a running calculation.
[-d] Start a dataserver process on the local host.
[-q <dataserver_addr>] Start a qmc process on the local host.
[-s <host>] Start a qmc process on <host>.
[-help] print this help text and exit
(alias: -?)
missing anonymous argument: EZFIO_FILE
- Set the parameters for a VMC calculation to create initial walker positions
$ qmcchem edit -h
Edit input data
qmcchem edit EZFIO_FILE [INPUT]
Edit input data
=== flags ===
[-c] Clear blocks
[-e energy] Fixed reference energy to normalize DMC weights
[-f 0|1] Correct wave function to verify electron-nucleus cusp
condition
[-j jastrow_type] Type of Jastrow factor [ None | Core | Simple ]
[-l seconds] Length (seconds) of a block
[-m method] QMC Method : [ VMC | DMC ]
[-n norm] Truncation t of the wave function : Remove determinants
with a
contribution to the norm less than t
[-s sampling] Sampling algorithm : [ Langevin | Brownian ]
[-t seconds] Requested simulation time (seconds)
[-ts time_step] Simulation time step
[-w walk_num] Number of walkers per CPU core
[-wt walk_num_tot] Total number of stored walkers for restart
[-help] print this help text and exit
(alias: -?)
$ qmcchem edit h2o -f 1 -m VMC -n 1.e-5 -s Langevin -t 300 -l 10
- Get info on the wave function
$ qmcchem info h2o
- Run the VMC calculation
$ qmcchem run h2o
- Set the correct parameters for FN-DMC
$ qmcchem edit h2o -e -76.438 -m DMC -s Brownian -ts 3.e-4 -t 3600 -l 30
- Run the FN-DMC calculation
$ qmcchem run h2o
- Print the result
$ qmcchem result h2o
References
Quantum Monte Carlo for large chemical systems: Implementing efficient strategies for petascale platforms and beyond > Anthony Scemama , Michel Caffarel , Emmanuel Oseret and William Jalby (2013), in: Journal of Computational Chemistry, 34:11(938–951)
Quantum Monte Carlo with very large multideterminant wavefunctions > Anthony Scemama , Thomas Applencourt , Emmanuel Giner and Michel Caffarel (2015), in: ArXiv ePrints:arXiv:1510.00730v2 [physics.chem-ph]