Stochastically accelerated perturbative triples correction in coupled cluster calculations

We introduce a novel algorithm that leverages stochastic sampling techniques to compute the perturbative triples correction in the coupled-cluster framework. By combining elements of randomness and determinism, our algorithm achieves a favorable balance between accuracy and computational cost. The main advantage of this algorithm is that it allows for the calculation to be stopped at any time, providing an unbiased estimate, with a statistical error that goes to zero as the exact calculation is approached. We provide evidence that our semi-stochastic algorithm achieves substantial computational savings compared to traditional deterministic methods. Specifically, we demonstrate that a precision of 0.5 millihartree can be attained with only 10% of the computational effort required by the full calculation. This work opens up new avenues for efficient and accurate computations, enabling investigations of complex molecular systems that were previously computationally prohibitive.

Tensor Hypercontraction Form of the Perturbative Triples Energy in Coupled-Cluster Theory

We present the working equations for a reduced-scaling method of evaluating the perturbative triples (T) energy in coupled-cluster theory, through the tensor hypercontraction (THC) of the triples amplitudes (t_ijk^abc). Through our method, we can reduce the scaling of the (T) energy from the traditional O(N7) to a more modest O(N5). We also discuss implementation details to aid future research, development, and software realization of this method. Additionally, we show that this method yields submillihartree (mEh) differences from CCSD(T) when evaluating absolute energies and sub-0.1 kcal/mol energy differences when evaluating relative energies. Finally, we demonstrate that this method converges to the true CCSD(T) energy through the systematic increasing of the rank or eigenvalue tolerance of the orthogonal projector, as well as exhibiting sublinear to linear error growth with respect to system size.

Mixed-Precision Implementation of the Density Matrix Renormalization Group

The mixed-precision optimization is an effective emerging technique for quantum chemistry methods to obtain better computational performance and maintain the chemical accuracy. Here, we developed a two-level mixed-precision implementation for the density matrix renormalization group (DMRG) method. This implementation is based on the idea that the DMRG process is an iterative process. Therefore, the first several iteration steps can be executed in single precision. A feasible single-precision DMRG may generate moderate accuracy, and when a few double-precision cleanup sweeps are added, the double-precision accuracy will be recovered. In the double-precision sweeps, we developed a mixed-precision diagonalization method that can run the most time-consuming step in single precision and maintain the double-precision accuracy. By combining these two mixed-precision schemes, we implemented our mixed-precision DMRG method. The benchmark result shows that our mixed-precision implementation achieved a good performance. A speed-up of up to 2.31 is achieved, and the accuracy is preserved within 0.01 kcal/mol.

GPU acceleration of rank-reduced coupled-cluster singles and doubles

We have developed a graphical processing unit (GPU) accelerated implementation of our recently introduced rank-reduced coupled-cluster singles and doubles (RR-CCSD) method. RR-CCSD introduces a low-rank approximation of the doubles amplitudes. This is combined with a low-rank approximation of the electron repulsion integrals via Cholesky decomposition. The result of these two low-rank approximations is the replacement of the usual fourth-order CCSD tensors with products of second- and third-order tensors. In our implementation, only a single fourth-order tensor must be constructed as an intermediate during the solution of the amplitude equations. Owing in large part to the compression of the doubles amplitudes, the GPU-accelerated implementation shows excellent parallel efficiency (95% on eight GPUs). Our implementation can solve the RR-CCSD equations for up to 400 electrons and 1550 basis functions-roughly 50% larger than the largest canonical CCSD computations that have been performed on any hardware. In addition to increased scalability, the RR-CCSD computations are faster than the corresponding CCSD computations for all but the smallest molecules. We test the accuracy of RR-CCSD for a variety of chemical systems including up to 1000 basis functions and determine that accuracy to better than 0.1% error in the correlation energy can be achieved with roughly 95% compression of the ov space for the largest systems considered. We also demonstrate that conformational energies can be predicted to be within 0.1 kcal mol^-1 with efficient compression applied to the wavefunction. Finally, we find that low-rank approximations of the CCSD doubles amplitudes used in the similarity transformation of the Hamiltonian prior to a conventional equation-of-motion CCSD computation will not introduce significant errors (on the order of a few hundredths of an electronvolt) into the resulting excitation energies.