arXiv subjects

Anthony Scemama

Publications and source records attributed to Anthony Scemama.

At least 73 records · Page 4Linked to original sources

arXiv2018

Excitation energies from diffusion Monte Carlo using selected Configuration Interaction nodes

Quantum Monte Carlo (QMC) is a stochastic method which has been particularly successful for ground-state electronic structure calculations but mostly unexplored for the computation of excited-state energies. Here, we show that, within a Jastrow-free QMC protocol relying on a deterministic and systematic construction of nodal surfaces using selected configuration interaction (sCI) expansions, one is able to obtain accurate excitation energies at the fixed-node diffusion Monte Carlo (FN-DMC) level. This evidences that the fixed-node errors in the ground and excited states obtained with sCI wave functions cancel out to a large extent. Our procedure is tested on two small organic molecules (water and formaldehyde) for which we report all-electron FN-DMC calculations. For both the singlet and triplet manifolds, accurate vertical excitation energies are obtained with relatively compact multideterminant expansions built with small (typically double-$ζ$) basis sets.

physics.chem-ph↗

arXiv2018

Perturbatively selected configuration-interaction wave functions for efficient geometry optimization in quantum Monte Carlo

We investigate the performance of a class of compact and systematically improvable Jastrow-Slater wave functions for the efficient and accurate computation of structural properties, where the determinantal component is expanded with a perturbatively selected configuration interaction scheme (CIPSI). We concurrently optimize the molecular ground-state geometry and full wave function -- Jastrow factor, orbitals, and configuration interaction coefficients-- in variational Monte Carlo (VMC) for the prototypical case of 1,3-trans-butadiene, a small yet theoretically challenging $π$-conjugated system. We find that the CIPSI selection outperforms the conventional scheme of correlating orbitals within active spaces chosen by chemical intuition: it gives significantly better variational and diffusion Monte Carlo energies for all but the smallest expansions, and much smoother convergence of the geometry with the number of determinants. In particular, the optimal bond lengths and bond-length alternation of butadiene are converged to better than one mÅ with just a few thousand determinants, to values very close to the corresponding CCSD(T) results. The combination of CIPSI expansion and VMC optimization represents an affordable tool for the determination of accurate ground-state geometries in quantum Monte Carlo.

physics.chem-ph↗

arXiv2017

Deterministic construction of nodal surfaces within quantum Monte Carlo: the case of FeS

In diffusion Monte Carlo (DMC) methods, the nodes (or zeroes) of the trial wave function dictate the magnitude of the fixed-node (FN) error. Within standard DMC implementations, they emanate from short multideterminant expansions, \textit{stochastically} optimized in the presence of a Jastrow factor. Here, following a recent proposal, we follow an alternative route by considering the nodes of selected configuration interaction (sCI) expansions built with the CIPSI (Configuration Interaction using a Perturbative Selection made Iteratively) algorithm. In contrast to standard implementations, these nodes can be \textit{systematically} and \textit{deterministically} improved by increasing the size of the sCI expansion. The present methodology is used to investigate the properties of the transition metal sulfide molecule FeS. This apparently simple molecule has been shown to be particularly challenging for electronic structure theory methods due to the proximity of two low-energy quintet electronic states of different spatial symmetry. In particular, we show that, at the triple-zeta basis set level, all sCI results --- including those extrapolated at the full CI (FCI) limit --- disagree with experiment, yielding an electronic ground state of $^{5}Σ^+$ symmetry. Performing FN-DMC simulation with sCI nodes, we show that the correct $^{5}Δ$ ground state is obtained if sufficiently large expansions are used. Moreover, we show that one can systematically get accurate potential energy surfaces and reproduce the experimental dissociation energy as well as other spectroscopic constants.

physics.chem-ph↗

arXiv2017

Hybrid stochastic-deterministic calculation of the second-order perturbative contribution of multireference perturbation theory

A hybrid stochastic-deterministic approach for computing the second-order perturbative contribution $E^{(2)}$ within multireference perturbation theory (MRPT) is presented. The idea at the heart of our hybrid scheme --- based on a reformulation of $E^{(2)}$ as a sum of elementary contributions associated with each determinant of the MR wave function --- is to split $E^{(2)}$ into a stochastic and a deterministic part. During the simulation, the stochastic part is gradually reduced by dynamically increasing the deterministic part until one reaches the desired accuracy. In sharp contrast with a purely stochastic MC scheme where the error decreases indefinitely as $t^{-1/2}$ (where $t$ is the computational time), the statistical error in our hybrid algorithm displays a polynomial decay $\sim t^{-n}$ with $n=3-4$ in the examples considered here. If desired, the calculation can be carried on until the stochastic part entirely vanishes. In that case, the exact result is obtained with no error bar and no noticeable computational overhead compared to the fully-deterministic calculation. The method is illustrated on the F$_2$ and Cr$_2$ molecules. Even for the largest case corresponding to the Cr$_2$ molecule treated with the cc-pVQZ basis set, very accurate results are obtained for $E^{(2)}$ for an active space of (28e,176o) and a MR wave function including up to $2 \times 10^7$ determinants.

physics.chem-ph↗

arXiv2017

A Jeziorski-Monkhorst fully uncontracted Multi-Reference perturbative treatment I: principles, second-order versions and tests on ground state potential energy curves

The present paper introduces a new multi-reference perturbation approach developed at second order, based on a Jeziorsky-Mokhorst expansion using individual Slater determinants as perturbers. Thanks to this choice of perturbers, an effective Hamiltonian may be built, allowing for the dressing of the Hamiltonian matrix within the reference space, assumed here to be a CAS-CI. Such a formulation accounts then for the coupling between the static and dynamic correlation effects. With our new definition of zeroth-order energies, these two approaches are strictly size-extensive provided that local orbitals are used, as numerically illustrated here and formally demonstrated in the appendix. Also, the present formalism allows for the factorization of all double excitation operators, just as in internally contracted approaches, strongly reducing the computational cost of these two approaches with respect to other determinant-based perturbation theories. The accuracy of these methods has been investigated on ground-state potential curves up to full dissociation limits for a set of six molecules involving single, double and triple bond breaking. The spectroscopic constants obtained with the present methods are found to be in very good agreement with the full configuration interaction (FCI) results. As the present formalism does not use any parameter or numerically unstable operation, the curves obtained with the two methods are smooth all along the dissociation path.

physics.chem-ph↗

arXiv2017

Alternative definition of excitation amplitudes in Multi-Reference state-specific Coupled Cluster

A central difficulty of state-specific Multi-Reference Coupled Cluster (MR-CC) formalisms concerns the definition of the amplitudes of the single and double excitation operators appearing in the exponential wave operator. If the reference space is a complete active space (CAS) the number of these amplitudes is larger than the number of singly and doubly excited determinants on which one may project the eigenequation, and one must impose additional conditions. The present work first defines a state-specific reference-independent operator $\hat{\tilde{T}}^m$ which acting on the CAS component of the wave function $|Ψ_0^m \rangle$ maximizes the overlap between $(1+\hat{\tilde{T}}^m)|Ψ_0^m \rangle$ and the eigenvector of the CAS-SD CI matrix $|Ψ_{\rm CAS-SD}^m \rangle$. This operator may be used to generate approximate coefficients of the Triples and Quadruples, and a dressing of the CAS-SD CI matrix, according to the intermediate Hamiltonian formalism. The process may be iterated to convergence. As a refinement towards a strict Coupled Cluster formalism, one may exploit reference-independent amplitudes provided by $(1+\hat{\tilde{T}}^m)|Ψ_0^m \rangle$ to define a reference-dependent operator $\hat{T}^m$ by fitting the eigenvector of the (dressed) CAS-SD CI matrix. The two variants, which are internally uncontracted, give rather similar results. The new MR-CC version has been tested on the ground state potential energy curves of 6 molecules (up to triple-bond breaking) and a two excited states. The non-parallelism error with respect to the Full-CI curves is of the order of 1 m$E_{\rm h}$.

physics.chem-ph↗

arXiv2016

Using CIPSI nodes in diffusion Monte Carlo

Several aspects of the recently proposed DMC-CIPSI approach consisting in using selected Configuration Interaction (SCI) approaches such as CIPSI (Configuration Interaction using a Perturbative Selection done Iteratively) to build accurate nodes for diffusion Monte Carlo (DMC) calculations are presented and discussed. The main ideas are illustrated with a number of calculations for diatomics molecules and for the benchmark G1 set.

physics.chem-ph↗

arXiv2016

Toward an improved control of the fixed-node error in quantum Monte Carlo: The case of the water molecule

All-electron Fixed-node Diffusion Monte Carlo (FN-DMC) calculations for the nonrelativistic ground-state energy of the water molecule at equilibrium geometry are presented. The determinantal part of the trial wavefunction is obtained from a perturbatively selected Configuration Interaction calculation (CIPSI method) including up to about 1.4 million of determinants. Calculations are made using the cc-pCV$n$Z family of basis sets, with $n$=2 to 5. In contrast with most QMC works no re-optimization of the determinantal part in presence of a Jastrow is performed. For the largest cc-pCV5Z basis set the lowest upper bound for the ground-state energy reported so far of -76.43744(18) is obtained. The fixed-node energy is found to decrease regularly as a function of the cardinal number $n$ and the Complete Basis Set limit (CBS) associated with {\it exact nodes} is easily extracted. The resulting energy of -76.43894(12) -in perfect agreement with the best experimentally derived value- is the most accurate theoretical estimate reported so far. We emphasize that employing selected CI nodes of increasing quality in a given family of basis sets may represent a simple, deterministic, reproducible, and systematic way of controlling the fixed-node error in DMC.

physics.chem-ph↗

arXiv2015

Quantum Monte Carlo with very large multideterminant wavefunctions

An algorithm to compute efficiently the first two derivatives of (very) large multideterminant wavefunctions for quantum Monte Carlo calculations is presented. The calculation of determinants and their derivatives is performed using the Sherman-Morrison formula for updating the inverse Slater matrix. An improved implementation based on the reduction of the number of column substitutions and on a very efficient implementation of the calculation of the scalar products involved is presented. It is emphasized that multideterminant expansions contain in general a large number of identical spin-specific determinants: for typical configuration interaction-type wavefunctions the number of unique spin-specific determinants $N_{\rm det}^σ$ ($σ=\uparrow,\downarrow$) with a non-negligible weight in the expansion is of order ${\cal O}(\sqrt{N_{\rm det}})$. We show that a careful implementation of the calculation of the $N_{\rm det}$-dependent contributions can make this step negligible enough so that in practice the algorithm scales as the total number of unique spin-specific determinants, $\; N_{\rm det}^\uparrow + N_{\rm det}^\downarrow$, over a wide range of total number of determinants (here, $N_{\rm det}$ up to about one million), thus greatly reducing the total computational cost. Finally, a new truncation scheme for the multideterminant expansion is proposed so that larger expansions can be considered without increasing the computational time. The algorithm is illustrated with all-electron Fixed-Node Diffusion Monte Carlo calculations of the total energy of the chlorine atom. Calculations using a trial wavefunction including about 750 000 determinants with a computational increase of $\sim$ 400 compared to a single-determinant calculation are shown to be feasible.

physics.chem-ph↗

arXiv2015

Efficiency of a Multi-Reference Coupled Cluster method

The multi-reference Coupled Cluster method first proposed by Meller et al (J. Chem. Phys. 1996) has been implemented and tested. Guess values of the amplitudes of the single and double excitations (the ${\hat T}$ operator) on the top of the references are extracted from the knowledge of the coefficients of the Multi Reference Singles and Doubles Configuration Interaction (MRSDCI) matrix. The multiple parentage problem is solved by scaling these amplitudes on the interaction between the references and the Singles and Doubles. Then one proceeds to a dressing of the MRSDCI matrix under the effect of the Triples and Quadruples, the coefficients of which are estimated from the action of ${\hat T}^2$. This dressing follows the logics of the intermediate effective Hamiltonian formalism. The dressed MRSDCI matrix is diagonalized and the process is iterated to convergence. The method is tested on a series of benchmark systems from Complete Active Spaces (CAS) involving 2 or 4 active electrons up to bond breakings. The comparison with Full Configuration Interaction (FCI) results shows that the errors are of the order of a few milli-hartree, five times smaller than those of the CASSDCI. The method is totally uncontracted, parallelizable, and extremely flexible since it may be applied to selected MR and/or selected SDCI. Some potential generalizations are briefly discussed.

physics.chem-ph↗

arXiv2014

Fixed-Node Diffusion Monte Carlo potential energy curve of the fluorine molecule F2 using selected configuration interaction trial wavefunctions

The potential energy curve of the F$_2$ molecule is calculated with Fixed-Node Diffusion Monte Carlo (FN-DMC) using Configuration Interaction (CI)-type trial wavefunctions. To keep the number of determinants reasonable (the first and second derivatives of the trial wavefunction need to be calculated at each step of FN-DMC), the CI expansion is restricted to those determinants that contribute the most to the total energy. The selection of the determinants is made using the so-called CIPSI approach (Configuration Interaction using a Perturbative Selection made Iteratively). Quite remarkably, the nodes of CIPSI wavefunctions are found to be systematically improved when increasing the number of selected determinants. To reduce the non-parallelism error of the potential energy curve a scheme based on the use of a $R$-dependent number of determinants is introduced. Numerical results show that improved FN-DMC energy curves for the F$_2$ molecule are obtained when employing CIPSI trial wavefunctions. Using the Dunning's cc-pVDZ basis set the FN-DMC energy curve is of a quality similar to that obtained with FCI/cc-pVQZ. A key advantage of using selected CI in FN-DMC is the possibility of improving nodes in a systematic and automatic way without resorting to a preliminary multi-parameter stochastic optimization of the trial wavefunction performed at the Variational Monte Carlo level as usually done in FN-DMC.

physics.chem-ph↗

arXiv2014

Spin density distribution in open-shell transition metal systems: A comparative post-Hartree-Fock, Density Functional Theory and quantum Monte Carlo study of the CuCl2 molecule

We present a comparative study of the spatial distribution of the spin density (SD) of the ground state of CuCl2 using Density Functional Theory (DFT), quantum Monte Carlo (QMC), and post-Hartree-Fock wavefunction theory (WFT). A number of studies have shown that an accurate description of the electronic structure of the lowest-lying states of this molecule is particularly challenging due to the interplay between the strong dynamical correlation effects in the 3d shell of the copper atom and the delocalization of the 3d hole over the chlorine atoms. It is shown here that qualitatively different results for SD are obtained from these various quantum-chemical approaches. At the DFT level, the spin density distribution is directly related to the amount of Hartree-Fock exchange introduced in hybrid functionals. At the QMC level, Fixed-node Diffusion Monte Carlo (FN-DMC) results for SD are strongly dependent on the nodal structure of the trial wavefunction employed (here, Hartree-Fock or Kohn-Sham with a particular amount of HF exchange) : in the case of this open-shell system, the 3N -dimensional nodes are mainly determined by the 3-dimensional nodes of the singly occupied molecular orbital. Regarding wavefunction approaches, HF and CASSCF lead to strongly localized spin density on the copper atom, in sharp contrast with DFT. To get a more reliable description and shed some light on the connections between the various theoretical descriptions, Full CI-type (FCI) calculations are performed. To make them feasible for this case a perturbatively selected CI approach generating multi-determinantal expansions of reasonable size and a small tractable basis set are employed. Although semi-quantitative, these near-FCI calculations allow to clarify how the spin density distribution evolves upon inclusion of dynamic correlation effects. A plausible scenario about the nature of the SD is proposed.

physics.chem-ph↗

arXiv2014

A Sparse SCF algorithm and its parallel implementation: Application to DFTB

We present an algorithm and its parallel implementation for solving a self consistent problem as encountered in Hartree Fock or Density Functional Theory. The algorithm takes advantage of the sparsity of matrices through the use of local molecular orbitals. The implementation allows to exploit efficiently modern symmetric multiprocessing (SMP) computer architectures. As a first application, the algorithm is used within the density functional based tight binding method, for which most of the computational time is spent in the linear algebra routines (diagonalization of the Fock/Kohn-Sham matrix). We show that with this algorithm (i) single point calculations on very large systems (millions of atoms) can be performed on large SMP machines (ii) calculations involving intermediate size systems (1~000--100~000 atoms) are also strongly accelerated and can run efficiently on standard servers (iii) the error on the total energy due to the use of a cut-off in the molecular orbital coefficients can be controlled such that it remains smaller than the SCF convergence criterion.

physics.chem-ph↗

arXiv2013

An efficient implementation of Slater-Condon rules

Slater-Condon rules are at the heart of any quantum chemistry method as they allow to simplify $3N$-dimensional integrals as sums of 3- or 6-dimensional integrals. In this paper, we propose an efficient implementation of those rules in order to identify very rapidly which integrals are involved in a matrix element expressed in the determinant basis set. This implementation takes advantage of the bit manipulation instructions on x86 architectures that were introduced in 2008 with the SSE4.2 instruction set. Finding which spin-orbitals are involved in the calculation of a matrix element doesn't depend on the number of electrons of the system.

physics.comp-ph↗

arXiv2012

Quantum Monte Carlo for large chemical systems: Implementing efficient strategies for petascale platforms and beyond

Various strategies to implement efficiently QMC simulations for large chemical systems are presented. These include: i.) the introduction of an efficient algorithm to calculate the computationally expensive Slater matrices. This novel scheme is based on the use of the highly localized character of atomic Gaussian basis functions (not the molecular orbitals as usually done), ii.) the possibility of keeping the memory footprint minimal, iii.) the important enhancement of single-core performance when efficient optimization tools are employed, and iv.) the definition of a universal, dynamic, fault-tolerant, and load-balanced computational framework adapted to all kinds of computational platforms (massively parallel machines, clusters, or distributed grids). These strategies have been implemented in the QMC=Chem code developed at Toulouse and illustrated with numerical applications on small peptides of increasing sizes (158, 434, 1056 and 1731 electrons). Using 10k-80k computing cores of the Curie machine (GENCI-TGCC-CEA, France) QMC=Chem has been shown to be capable of running at the petascale level, thus demonstrating that for this machine a large part of the peak performance can be achieved. Implementation of large-scale QMC simulations for future exascale platforms with a comparable level of efficiency is expected to be feasible.

cs.PF↗

arXiv2009

IRPF90: a programming environment for high performance computing

IRPF90 is a Fortran programming environment which helps the development of large Fortran codes. In Fortran programs, the programmer has to focus on the order of the instructions: before using a variable, the programmer has to be sure that it has already been computed in all possible situations. For large codes, it is common source of error. In IRPF90 most of the order of instructions is handled by the pre-processor, and an automatic mechanism guarantees that every entity is built before being used. This mechanism relies on the {needs/needed by} relations between the entities, which are built automatically. Codes written with IRPF90 execute often faster than Fortran programs, are faster to write and easier to maintain.

cs.SE↗

arXiv2006

Efficient Monte Carlo Calculations of the One-Body Density

An alternative Monte Carlo estimator for the one-body density rho(r) is presented. This estimator has a simple form and can be readily used in any type of Monte Carlo simulation. Comparisons with the usual regularization of the delta-function on a grid show that the statistical errors are greatly reduced. Furthermore, our expression allows accurate calculations of the density at any point in space, even in the regions never visited during the Monte Carlo simulation. The method is illustrated with the computation of accurate Variational Monte Carlo electronic densities for the Helium atom (1D curve) and for the water dimer (3D grid containing up to 51x51x51=132651 points).

physics.comp-ph↗

arXiv2006

An efficient sampling algorithm for Variational Monte Carlo

We propose a new algorithm for sampling the $N$-body density $|Ψ({\bf R})|^2/\int_{\mathbb{R}^{3N}} |Ψ|^2$ in the Variational Monte Carlo (VMC) framework. This algorithm is based upon a modified Ricci-Ciccotti discretization of the Langevin dynamics in the phase space $({\bf R},{\bf P})$ improved by a Metropolis acceptation/rejection step. We show through some representative numerical examples (Lithium, Fluorine and Copper atoms, and phenol molecule), that this algorithm is superior to the standard sampling algorithm based on the biased random walk (importance sampling).

cond-mat.other↗