In engineering, the ability to optimize materials properties often requires a deep understanding of the relationship between chemical-atomistic structures and the physical properties of the material itself, even both in its solid and fluid phases. Monte Carlo methods allow for a reliable calculation of thermodynamic properties to better predict chemical and physical properties of materials, which is useful not only for engineering but also for science.
One of the main scopes of TREX is to meet the increasing demand for highly reliable and robust calculations in regards with materials design meeting today’s and tomorrows’ complex scientific and industrial applications needs. The TREX project aims therefore to increase the reach of simulations with both scientific and industrial applications, speeding up upcoming innovation projects while positioning the European Union as leaders in computational science.
|TREX, bringing together computational scientists with high-performance computing (HPC) experts, is developing a platform that combines the upcoming exascale HPC and HTC architectures for stochastic quantum chemical simulations of unprecedented accuracy. The software and services will be designed for ease of use to ensure widespread utilisation, spurring a new age of discovery in molecular simulations.|
Achieving Transformation in the Coding for New Materials Design
In words of Mr Pleiter “there are at least two important things which I think this project will achieve:
- The first one is that we will have a larger number of codes, which we can expect to scale on these very big upcoming exascale systems.
- The second one is to achieve a transformation in the coding community, thanks to the different TREX applications. This will help computational scientists focus on their algorithms development, as well as to improve the performance of critical parts of the code, which need to be specifically optimized for different kinds of supercomputers.”
The TREX project, gathering HPC stakeholders, European scientists, and SMEs working on quantum chemistry simulations using QMC methods, aims to develop software solutions that will accelerate the diffusion in both industry and academia of unique computational instruments.
Therefore, TREX’s contribution to the future is that it will permit to enhance the scientific scopes’ tools for designing new materials and understanding the fundamental properties of the matter.
TREX developments target six different codes belonging to the scientific domain of quantum chemistry, not all of them being exclusively QMC codes: TurboRVB, CHAMP, QMC=Chem, NECI, Quantum Package, and GammCor.
We can categorize them as:
Variational and diffusion QMC (VMC and DMC) for molecular and extended systems:
TurboRVB https://people.sissa.it/~sorella/TurboRVB_Manual/build/html/index.html: Developed by the group of Sorella (SISSA), it features sophisticated wave functions to describe strong correlations, efficient algorithms to compute ionic forces, pressure, and stress tensor via algorithmic differentiation, the possibility to variationally optimized the wave function in a fully ab initio fashion, and efficient treatment of bulk materials. It has been deployed on several HPC systems and runs on up to 262,144 cores, reaching 90% of the peak performance on the K-computer in Japan in 2015. It has been used in several PRACE projects with allocations of more than 50M each.
CHAMP https://www.utwente.nl/en/tnw/ccp/research/CHAMP.html: Developed by Filippi (UT) and Moroni (SISSA), it enables the very efficient wave function optimization for ground and excited states, a compact formulation for a fast evaluation of multi determinant expansions and their derivatives2, and multiscale schemes to perform QMC calculations in classical point charges, polarizable continuum model, and polarizable force fields. The code is routinely run on the SURFsara Tier1 supercomputer with yearly allocations between 5M and 10M. It is currently been developed with computer scientists of the Netherlands eScience Center for its complete transition to Fortran2008 by mid-2020.
QMC=Chem https://anr.fr/Project-ANR-11-BS08-0004: It is the main deliverable of an ANR project of the CNRS/Toulouse group, developed by Scemama and optimized for the x86 architecture in close collaboration with the HPC experts of the UVSQ partner laboratory.4 It is designed to be used on HPC resources as well as grid infrastructures.5 In 2011, QMC=Chem was selected for a grand challenge on the Curie Tier0 supercomputer, where a sustained performance of 0.96 petaflops/s was measured with 76800 cores. It was involved in a PRACE project, is part of the benchmark set of the CALMIP (Tier2) regional center, and was also part of benchmarks for French Tier1 machines via GENCI. QMC=Chem is distributed under the GPLv2 license.
Full configuration quantum Monte Carlo (FCIQMC) for molecular and extended systems:
NECI https://github.com/ghb24/NECI_STABLE: is developed in the Alavi group (MPG) under a GNU GPLv3. It performs stochastic FCIQMC simulations of quantum chemical Hamiltonians in 2nd quantised representations, aiming at FCI accuracy for the energy and wave function. The algorithm can be used to calculate ground and excited states of a given symmetry, spin-adaptation, one and two-body reduced density matrices, relativistic Hamiltonians, real-time propagation with spectroscopic applications, and non-Hermitian Hamiltonians. At present NECI uses MPI and is parallelizable with near-ideal scalability up to 512 nodes (~10000 Xeon cores) on the HPC machines of the Max Planck Society in Garching.
Deterministic quantum chemical molecular codes:
Quantum Package https://quantum-package.readthedocs.io/en/master/: is an open-source program to obtain near-FCI-quality wave functions, energies, and one- and two-electron reduced density matrices for ground and excited states with a stochastic variant of a selected-CI (CIPSI) algorithm.8 It does not need any input from another program and can provide one- and two-electron integrals to external codes. It is developed under the GNU AGPLv3 in Toulouse (Scemama/CNRS), Paris (France), and Argonne National Laboratory (USA). It is involved in an INCITE project and was part of the pool of benchmark programs of the CALMIP (Tier2) regional center. It implements an asynchronous task-based multi-level parallelism with OpenMP/MPI/ZeroMQ and has proven to be scalable up to 12000 Xeon cores. This code proposes a dynamic resource management and can be run on grid infrastructures9.
GammCor https://github.com/pernalk/GAMMCOR: is developed in the group of Pernal (TUL) and performs calculations of interaction energy using multireference symmetry adapted perturbation theory method (for ground and excited states), the adiabatic connection (AC and AC0) correlation energy for ground- and excited-state multireference wave functions, range-separated multiconfiguration DFT energy with the long-range AC/AC0 correlation correction. All the implemented multireference methods rely only on one- and two-electron reduced density matrices, which are computed externally.