Efficient Treatment of Large Active Spaces through Multi-GPU Parallel Implementation of Direct Configuration Interaction

Massively scalable workflows for quantum chemistry: BigChem and ChemCloud

Electronic structure theory, i.e., quantum chemistry, is the fundamental building block for many problems in computational chemistry. We present a new distributed computing framework (BigChem), which allows for an efficient solution of many quantum chemistry problems in parallel. BigChem is designed to be easily composable and leverages industry-standard middleware (e.g., Celery, RabbitMQ, and Redis) for distributed approaches to large scale problems. BigChem can harness any collection of worker nodes, including ones on cloud providers (such as AWS or Azure), local clusters, or supercomputer centers (and any mixture of these). BigChem builds upon MolSSI packages, such as QCEngine to standardize the operation of numerous computational chemistry programs, demonstrated here with Psi4, xtb, geomeTRIC, and TeraChem. BigChem delivers full utilization of compute resources at scale, offers a programable canvas for designing sophisticated quantum chemistry workflows, and is fault tolerant to node failures and network disruptions. We demonstrate linear scalability of BigChem running computational chemistry workloads on up to 125 GPUs. Finally, we present ChemCloud, a web API to BigChem and successor to TeraChem Cloud. ChemCloud delivers scalable and secure access to BigChem over the Internet.

A configuration-based heatbath-CI for spin-adapted multireference electronic structure calculations with large active spaces

This work reports on a spin-pure configuration-based implementation of the heatbath configuration interaction (HCI) algorithm for selective configuration interaction. Besides the obvious advantage of being spin-pure, the presented method combines the compactness of the configurational ansatz with the known efficiency of the HCI algorithm and a variety of algorithmic and conceptual ideas to achieve a high level of performance. In particular, through pruning of the selected configurational space after HCI selection by means of a more strict criterion, a more compact wavefunction representation is obtained. Moreover, the underlying logic of the method allows us to minimize the number of redundant matrix-matrix multiplications while making use of just-in-time compilation to achieve fast diagonalization of the Hamiltonian. The critical search for 2-electron connections within the configurational space is facilitated by a tree-based representation thereof as suggested previously by Gopal et al. Usage of a prefix-based parallelization and batching during the calculation of the PT2-correction leads to a good load balancing and significantly reduced memory requirements for these critical steps of the calculation. In this way, the need for a semistochastic approach to the PT2 correction is avoided even for large configurational spaces. Finally, several test-cases are discussed to demonstrate the strengths and weaknesses of the presented method.

Distributed memory, GPU accelerated Fock construction for hybrid, Gaussian basis density functional theory

With the growing reliance of modern supercomputers on accelerator-based architecture such a graphics processing units (GPUs), the development and optimization of electronic structure methods to exploit these massively parallel resources has become a recent priority. While significant strides have been made in the development GPU accelerated, distributed memory algorithms for many modern electronic structure methods, the primary focus of GPU development for Gaussian basis atomic orbital methods has been for shared memory systems with only a handful of examples pursing massive parallelism. In the present work, we present a set of distributed memory algorithms for the evaluation of the Coulomb and exact exchange matrices for hybrid Kohn-Sham DFT with Gaussian basis sets via direct density-fitted (DF-J-Engine) and seminumerical (sn-K) methods, respectively. The absolute performance and strong scalability of the developed methods are demonstrated on systems ranging from a few hundred to over one thousand atoms using up to 128 NVIDIA A100 GPUs on the Perlmutter supercomputer.

Active Learning Configuration Interaction for Excited-State Calculations of Polycyclic Aromatic Hydrocarbons

We present the active learning configuration interaction (ALCI) method for multiconfigurational calculations based on large active spaces. ALCI leverages the use of an active learning procedure to find important electronic configurations among the full configurational space generated within an active space. We tested it for the calculation of singlet-singlet excited states of acenes and pyrene using different machine learning algorithms. The ALCI method yields excitation energies within 0.2-0.3 eV from those obtained by traditional complete active-space configuration interaction (CASCI) calculations (affordable for active spaces up to 16 electrons in 16 orbitals) by including only a small fraction of the CASCI configuration space in the calculations. For larger active spaces (we tested up to 26 electrons in 26 orbitals), not affordable with traditional CI methods, ALCI captures the trends of experimental excitation energies. Overall, ALCI provides satisfactory approximations to large active-space wave functions with up to 10 orders of magnitude fewer determinants for the systems presented here. These ALCI wave functions are promising and affordable starting points for the subsequent second-order perturbation theory or pair-density functional theory calculations.

Modern multireference methods and their application in transition metal chemistry

Transition metal chemistry is a challenging playground for quantum chemical methods owing to the simultaneous presence of static and dynamic electron correlation effects in many systems. Wavefunction based multireference (MR) methods constitute a physically sound and systematically improvable Ansatz to deal with this complexity but suffer from some conceptual difficulties and high computational costs. The latter problem partially arises from the unfavorable scaling of the Full Configuration Interaction (Full-CI) problem which in the majority of MR methods is solved for a subset of the molecular orbital space, the so-called active space. In the last years multiple methods such as modern variants of selected CI, Full-CI Quantum Monte Carlo (FCIQMC) and the density matrix renormalization group (DMRG) have been developed that solve the Full-CI problem approximately for a fraction of the computational cost required by conventional techniques thereby significantly extending the range of applicability of modern MR methods. This perspective review outlines recent advancements in the field of MR electronic structure methods together with the resulting chances and challenges for theoretical studies in the field of transition metal chemistry. In light of its emerging importance a special focus is put on the selection of adequate active spaces and the concomitant development of numerous selection aides in recent years.

Open-Source Multi-GPU-Accelerated QM/MM Simulations with AMBER and QUICK

The quantum mechanics/molecular mechanics (QM/MM) approach is an essential and well-established tool in computational chemistry that has been widely applied in a myriad of biomolecular problems in the literature. In this publication, we report the integration of the QUantum Interaction Computational Kernel (QUICK) program as an engine to perform electronic structure calculations in QM/MM simulations with AMBER. This integration is available through either a file-based interface (FBI) or an application programming interface (API). Since QUICK is an open-source GPU-accelerated code with multi-GPU parallelization, users can take advantage of "free of charge" GPU-acceleration in their QM/MM simulations. In this work, we discuss implementation details and give usage examples. We also investigate energy conservation in typical QM/MM simulations performed at the microcanonical ensemble. Finally, benchmark results for two representative systems in bulk water, the N-methylacetamide (NMA) molecule and the photoactive yellow protein (PYP), show the performance of QM/MM simulations with QUICK and AMBER using a varying number of CPU cores and GPUs. Our results highlight the acceleration obtained from a single or multiple GPUs; we observed speedups of up to 53× between a single GPU vs a single CPU core and of up to 2.6× when comparing four GPUs to a single GPU. Results also reveal speedups of up to 3.5× when the API is used instead of FBI.

CAS without SCF-Why to use CASCI and where to get the orbitals

The complete active space self-consistent field (CASSCF) method has seen broad adoption due to its ability to describe the electronic structure of both the ground and excited states of molecules over a broader swath of the potential energy surface than is possible with the simpler Hartree-Fock approximation. However, it also has a reputation for being unwieldy, computationally costly, and un-black-box. Here, we discuss a class of alternatives, complete active space configuration interaction (CASCI) methods, paying particular attention to their application to electronic excited states. The goal of this Perspective is fourfold. First, we argue that CASCI is not merely an approximation to CASSCF, in that it can be designed to have important qualitative advantages over CASSCF. Second, we present several insights drawn from our experience experimenting with different schemes for computing orbitals to be employed in CASCI. Third, we argue that CASCI is well suited for application to nanomaterials. Finally, we reason that, with the rise in new low-scaling approaches for describing multireference systems, there is a greater need than ever to develop new methods for defining orbitals that provide an efficient and accurate description of both static correlation and electronic excitations in a limited active space.

Estimates of electron correlation based on density expansions

Two methods for estimating the correlation energy of molecules and other electronic systems are discussed based on the assumption that the correlation energy can be partitioned between atomic regions. In the first method, the electron density is expanded in terms of atomic contributions using rigorous electron repulsion bounds, and in the second method, correlation contributions are associated with basis function pairs. These methods do not consider the detailed nature of localized excitations but instead define a correlation energy per electron factor that is unique to a specific atom. The correlation factors are basis function dependent and are determined by configuration interaction (CI) calculations on diatomic and hydride molecules. The correlation energy estimates are compared with the results of high-level CI calculations for a test set of 27 molecules representing a wide range of bonding environments (average error of 2.6%). An extension based on truncated CI calculations in which d-type and hydrogen p-type functions are eliminated from the virtual space combined with estimates of dynamical correlation contributions using atomic correlation factors is discussed and applied to the dissociation of several molecules.

TeraChem: Accelerating electronic structure and ab initio molecular dynamics with graphical processing units

Developed over the past decade, TeraChem is an electronic structure and ab initio molecular dynamics software package designed from the ground up to leverage graphics processing units (GPUs) to perform large-scale ground and excited state quantum chemistry calculations in the gas and the condensed phase. TeraChem's speed stems from the reformulation of conventional electronic structure theories in terms of a set of individually optimized high-performance electronic structure operations (e.g., Coulomb and exchange matrix builds, one- and two-particle density matrix builds) and rank-reduction techniques (e.g., tensor hypercontraction). Recent efforts have encapsulated these core operations and provided language-agnostic interfaces. This greatly increases the accessibility and flexibility of TeraChem as a platform to develop new electronic structure methods on GPUs and provides clear optimization targets for emerging parallel computing architectures.

Fast transformations between configuration state function and Slater determinant bases for direct configuration interaction

A hybrid configuration state function (CSF) and Slater determinant (SD) basis full configuration interaction (CI) program was developed to simultaneously take advantage of fast SD basis algorithms for σ = Hc formation and the smaller CI vector length and more robust convergence offered by a CSF basis. Graphical processing unit acceleration of the direct CSF-SD and SD-CSF basis transformation algorithms ensures that the combined transformation time per iteration relative to σ formation is small (∼15%). In addition to the obvious benefits of reducing the memory footprint of the CI vector, additional computational savings are demonstrated that rely directly on the size of the CI basis, in one particular case reducing the CI time-to-solution of a HF-CAS-(16,16)-CI/6-31G calculation of ethylene from 1954.79 s to 956 s by using a CSF basis, a 2.0× speedup.