Chemistry with quantum Monte Carlo – QMC=Chem

Various strategies for removing the theoretical and practical bottlenecks of QMC approaches in computational chemistry are explored. Regarding theoretical aspects, we propose to implement a new type of trial wavefunctions and to investigate a robust and accurate framework for computing forces. Regarding technical aspects, the project includes the implementation of a massively parallel version of the code and the development of efficient techniques for dealing with large systems.

A first massively parallel version of the QMC=Chem code including the efficient treatment of large systems and various optimization techniques has been realized. Large-scale simulations with unprecedented accuracy has been carried out for a series of molecular systems including the ß-amyloïd peptide involved in a number of degenerative diseases. Simulations have been performed thanks to a PRACE allocation of resources on the petaflopic supercomputer CURIE. Another part of the project aiming at constructing a new type of trial wavefunctions allowing to make automatized simulations has also been completed. A number of applications illustrating the advantage of using such trial wavefunctions have been performed.

The greatest part of the project has been realized. We are now in the last stage of this project. We are now working hard on the development of an efficient and accurate algorithm for computing forces and also on the design of a fully-automated version of our code.

A publicationr presenting the first implementation of our code on the supercomputer CURIE has been published in « Lecture Notes in Computer Science » (2013). In this paper the scalability of the code up to 80 000 compute nodes is discussed. The various scientific and algorithmic aspects are presented in a work published in J. Comp. Chem. (2013). A publication on the new type of trial wavefunctions proposed here appeared in Can. J. Chem (2013). Three other publications presenting the performance of the method on selected applications have been submitted for publication. Finally, a general public presentation of our scientific project can be found in « La Recherche (Nov. 2012).

The objective of this project is to make important advances in quantum Monte Carlo (QMC) for computational chemistry.
Our long-term motivation is to explore whether or not QMC could become a reference method of the field in the years to come.
As known, the present approaches of computational chemistry are the Density Functional Theory approaches and various types of
post-Hartree-Fock methods (Coupled Cluster, Configuration Interaction, perturbative approaches, etc.).
However, despite remarkable results a number of limitations still restrict their use either in terms of predictability
(e.g, choice of the exchange-correlation potential) and/or computational cost as a function of the system size.
Quantum Monte Carlo methods have been developed as a promising alternative to these standard approaches.
In computational physics they are extensively employed and considered as mature methods. However, it is not the case in
computational chemistry, although a number of interesting results on real systems have been obtained. In this project
we propose to make important progress on what we consider to be the most important practical and theoretical
bottlenecks responsible for the present limited use of QMC in chemistry. From a theoretical point of view, we thus
propose to work on:
(i) The absence of a simple, general, and systematic strategy for constructing accurate enough trial wave functions
for general molecular systems. QMC requiring such an input, present algorithms rest too much on expertise and lengthy
optimizations, thus avoiding the practical use of QMC by the «ordinary» (computational) chemist not expert in QMC.
(ii) The absence of a stable and accurate algorithm for computing forces and allowing large-scale geometry optimizations.
This work will be the continuation of a thesis work presently underway. In particular, linear-scaling techniques (O(N) approach)
to treat large systems will be developed and implemented.
Besides these methodological aspects, the following important practical aspects will also be pursued:
(iii.) To provide a general-public, general-purpose, and easy-to-use version of our code QMC=Chem including the two previous aspects.
Such a version will be delivered at the project end. Interfaced with the standard codes of quantum chemistry (GAUSSIAN, MOLPRO, etc.)
it will allow to treat easily and accurately large electronic systems using our multi-scale strategy with minimal and automated trial
wave function re-optimization, linear-scaling O(N), and efficient force computations.
(iv.) To deliver a parallelized version of the code with maximal efficiency. Monte Carlo methods are known to be intrinsically suited
to parallelism and, more generally, to High Performance Computing. Common believe is that this practical but fundamental feature
will certainly be a key feature for the success of such methods.
Finally, to promote QMC within the theoretical chemistry community we propose to demonstrate its high potential via selected QMC «critical»
applications for problems of high scientific interest and particularly difficult to study with standard methods. We also propose to organize an international workshop at the end of the project to foster the diffusion of our results.
The core of the project will be realized at LCPQ (Toulouse). The financing of a PhD thesis is asked for the implementation of our general multi-scale strategy. Critical QMC applications will be done in collaboration with the Professor A. Ramírez-Solís from Mexico. Regarding the development of large-scale QMC geometry optimizations using our newly developed forces and multi-scale wave functions, and the implementation of O(N) linear-scaling techniques a 36-month post doctoral financing is demanded. This post-doctoral work will be done on a shared basis: Half-time at LCPQ (QMC, localized molecular orbitals) and half-time at the SRMC laboratory at Nancy under the supervision of Dr. A. Monari.