Tighter multipole-based integral estimates and parallel implementation of linear-scaling AO-MP2 theory

Within an atomic-orbital-based (AO-based) formulation of second-order M?ller-Plesset perturbation theory (MP2), we present a novel screening procedure which allows us to preselect numerically significant two-electron integrals more efficiently, especially for large basis sets. The screening is based on our recently introduced multipole-based integral estimates (MBIE) method [J. Chem. Phys., 2005, 123, 184102], that allows to exploit the 1/R(4) or 1/R(6) coupling between electronic charge distributions in transformed integral products within AO-MP2. In this way, linear scaling is attained with fully-controlled numerical accuracy. Furthermore, a parallel implementation of our linear-scaling AO-MP2 method is described, which also allows us to perform calculations with larger basis sets. First calculations reveal that for e.g. linear alkanes the scaling of the number of required transformed integral products is almost equal for 6-31G* and cc-pVTZ basis sets. Using the improved MBIE screening, the largest parallel calculation was performed for a ribozyme fragment consisting of 497 atoms and 5697 basis functions, while our largest AO-MP2 calculation was performed for a stacked DNA system (16 base pairs) comprising 1052 atoms and 10 674 basis functions on a single processor.     

Distance-dependent Schwarz-based integral estimates for two-electron integrals: reliable tightness vs. rigorous upper bounds

A new integral estimate for four-center two-electron integrals is introduced that accounts for distance information between the bra- and ket-charge distributions describing the two electrons. The screening is denoted as QQR and combines the most important features of the conventional Schwarz screening by Ha?ser and Ahlrichs published in 1989 [J. Comput. Chem. 10, 104 (1989)] and our multipole-based integral estimates (MBIE) introduced in 2005 [D. S. Lambrecht and C. Ochsenfeld, J. Chem. Phys. 123, 184101 (2005)]. At the same time the estimates are not only tighter but also much easier to implement, so that we recommend them instead of our MBIE bounds introduced first for accounting for charge-distance information. The inclusion of distance dependence between charge distributions is not only useful at the SCF level but is particularly important for describing electron-correlation effects, e.g., within AO-MP2 theory, where the decay behavior is at least 1/R(4) or even 1/R(6). In our present work, we focus on studying the efficiency of our QQR estimates within SCF theory and demonstrate the performance for a benchmark set of 44 medium to large molecules, where savings of up to a factor of 2 for exchange integrals are observed for larger systems. Based on the results of the benchmark set we show that reliable tightness of integral estimates is more important for the screening performance than rigorous upper bound properties.     

Multipole-based integral estimates for the rigorous description of distance dependence in two-electron integrals

We derive multipole-based integral estimates (MBIE) as rigorous and tight upper bounds to four-center two-electron integrals in order to account for the 1/R distance decay between the charge distributions, which is missing in the Schwarz screening commonly used in ab initio methods. Our screening criteria are valid for all angular momenta and can be formulated for any order of multipoles. We have found the expansion limited to dipoles to be sufficiently tight for estimating the integrals in Hartree-Fock and density-functional theories, while the screening effort is negligible. For, e.g., a DNA fragment with 1052 atoms and 10,674 basis functions (6-31G*) the exchange part is faster by a factor of 2.1 as compared to the Schwarz screening both within our linear exchange scheme, whereas a smaller factor of 1.3 is gained for the Coulomb part within the continuous fast multipole method. Most importantly, our new MBIE screening is perfectly suited to exploit the strong distance decay of electron-correlation effects of at least 1/R4 in atomic-orbital-based formulations of correlation methods.     

Parallelization of the integral equation formulation of the polarizable continuum model for higher-order response functions

We present a parallel implementation of the integral equation formalism of the polarizable continuum model for Hartree-Fock and density functional theory calculations of energies and linear, quadratic, and cubic response functions. The contributions to the free energy of the solute due to the polarizable continuum have been implemented using a master-slave approach with load balancing to ensure good scalability also on parallel machines with a slow interconnect. We demonstrate the good scaling behavior of the code through calculations of Hartree-Fock energies and linear, quadratic, and cubic response function for a modest-sized sample molecule. We also explore the behavior of the parallelization of the integral equation formulation of the polarizable continuum model code when used in conjunction with a recent scheme for the storage of two-electron integrals in the memory of the different slaves in order to achieve superlinear scaling in the parallel calculations.     

Iterative introduction of integral approximations with error bounds in SCF calculations

Another possible application of a previously reported approximation theory for electron repulsion integrals using rigorous error bounds is considered by incorporating the electron density matrix in the approximation scheme. Error bounds are set on the contribution of a given integral to the total energy for a given molecular wave-function; the wave-function is then refined cyclically and additional integrals are computed exactly if necessary until convergence is achieved.     

Tensor hypercontraction density fitting. I. Quartic scaling second- and third-order Mo̸ller-Plesset perturbation theory

Many approximations have been developed to help deal with the O(N(4)) growth of the electron repulsion integral (ERI) tensor, where N is the number of one-electron basis functions used to represent the electronic wavefunction. Of these, the density fitting (DF) approximation is currently the most widely used despite the fact that it is often incapable of altering the underlying scaling of computational effort with respect to molecular size. We present a method for exploiting sparsity in three-center overlap integrals through tensor decomposition to obtain a low-rank approximation to density fitting (tensor hypercontraction density fitting or THC-DF). This new approximation reduces the 4th-order ERI tensor to a product of five matrices, simultaneously reducing the storage requirement as well as increasing the flexibility to regroup terms and reduce scaling behavior. As an example, we demonstrate such a scaling reduction for second- and third-order perturbation theory (MP2 and MP3), showing that both can be carried out in O(N(4)) operations. This should be compared to the usual scaling behavior of O(N(5)) and O(N(6)) for MP2 and MP3, respectively. The THC-DF technique can also be applied to other methods in electronic structure theory, such as coupled-cluster and configuration interaction, promising significant gains in computational efficiency and storage reduction.     

The limits of local correlation theory: electronic delocalization and chemically smooth potential energy surfaces

Local coupled-cluster theory provides an algorithm for measuring electronic correlation quickly, using only the spatial locality of localized electronic orbitals. Previously, we showed [J. Subotnik et al., J. Chem. Phys. 125, 074116 (2006)] that one may construct a local coupled-cluster singles-doubles theory which (i) yields smooth potential energy surfaces and (ii) achieves near linear scaling. That theory selected which orbitals to correlate based only on the distances between the centers of different, localized orbitals, and the approximate potential energy surfaces were characterized as smooth using only visual identification. This paper now extends our previous algorithm in three important ways. First, locality is now based on both the distances between the centers of orbitals as well as the spatial extent of the orbitals. We find that, by accounting for the spatial extent of a delocalized orbital, one can account for electronic correlation in systems with some electronic delocalization using fast correlation methods designed around orbital locality. Second, we now enforce locality on not just the amplitudes (which measure the exact electron-electron correlation), but also on the two-electron integrals themselves (which measure the bare electron-electron interaction). Our conclusion is that we can bump integrals as well as amplitudes, thereby gaining a tremendous increase in speed and paradoxically increasing the accuracy of our LCCSD approach. Third and finally, we now make a rigorous definition of chemical smoothness as requiring that potential energy surfaces not support artificial maxima, minima, or inflection points. By looking at first and second derivatives from finite difference techniques, we demonstrate complete chemical smoothness of our potential energy surfaces (bumping both amplitudes and integrals). These results are significant both from a theoretical and from a computationally practical point of view.     

Efficient multipole model and linear scaling of NDDO-based methods

Fast growth of computational costs with that of the system's size is a bottleneck for the applications of traditional methods of quantum chemistry to polyatomic molecular systems. This problem is addressed by the development of linear (or almost linear) scaling methods. In the semiempirical domain, it is typically achieved by a series of approximations to the self-consistent field (SCF) solution. By contrast, we propose a route to linear scalability by modifying the trial wave function itself. Our approach is based on variationally determined strictly local one-electron states and a geminal representation of chemical bonds and lone pairs. A serious obstacle previously faced on this route were the numerous transformations of the two-center repulsion integrals characteristic for the neglect of diatomic differential overlap (NDDO) methods. We pass it by replacing the fictitious charge configurations usual for the NDDO scheme by atomic multipoles interacting through semiempirical potentials. It ensures invariance of these integrals and improves the computational efficiency of the whole method. We discuss possible schemes for evaluating the integrals as well as their numerical values. The method proposed is implemented for the most popular modified neglect of diatomic overlap (MNDO), Austin model 1 (AM1), and PM3 parametrization schemes of the NDDO family. Our calculations involving well-justified cutoff procedures for molecular interactions unequivocally show that the proposed scheme provides almost linear scaling of computational costs with the system's size. The numerical results on molecular properties certify that our method is superior with respect to its SCF-based ancestors.     

Superlinear scaling in master-slave quantum chemical calculations using in-core storage of two-electron integrals

We describe the implementation of a parallel, in-core, integral-direct Hartree-Fock and density functional theory code for the efficient calculation of Hartree-Fock wave functions and density functional theory. The algorithm is based on a parallel master-slave algorithm, and the two-electron integrals calculated by a slave are stored in available local memory. To ensure the greatest computational savings, the master node keeps track of all integral batches stored on the different slaves. The code can reuse undifferentiated two-electron integrals both in the wave function optimization and in the evaluation of second-, third-, and fourth-order molecular properties. Superlinear scaling is achieved in a series of test examples, with speedups of up to 55 achieved for calculations run on medium-sized molecules on 16 processors with respect to the time used on a single processor.     

An efficient implementation of the "cluster-in-molecule" approach for local electron correlation calculations

An efficient implementation of the "cluster-in-molecule" (CIM) approach is presented for performing local electron correlation calculations in a basis of orthogonal occupied and virtual localized molecular orbitals (LMOs). The main idea of this approach is that significant excitation amplitudes can be approximately obtained by solving the coupled cluster (or Moller-Plesset perturbation theory) equations of a series of "clusters," each of which contains a subset of occupied and virtual LMOs. In the present implementation, we have proposed a simple approach for constructing virtual LMOs of clusters, and new ways of constructing clusters and extracting the correlation contributions from calculations on clusters, which are more efficient than those suggested in the original work. More importantly, linear scaling of computational time of the CIM approach is achieved by evaluating the transformed two-electron integrals over LMOs using simple truncation techniques in limited operations (independent of the molecular size). With typical thresholds, for a variety of molecules our test calculations demonstrate that more than 99% of the conventional MP2 or coupled cluster with doubles correlation energies can be recovered in the present CIM approach.     

Big data analysis of ab Initio molecular integrals in the neglect of diatomic differential overlap approximation

Most modern semiempirical quantum-chemical (SQC) methods are based on the neglect of diatomic differential overlap (NDDO) approximation to ab initio molecular integrals. Here, we check the validity of this approximation by computing all relevant integrals for 32 typical organic molecules using Gaussian-type orbitals and various basis sets (from valence-only minimal to all-electron triple-ζ basis sets) covering in total more than 15.6 million one-electron (1-e) and 10.3 billion two-electron (2-e) integrals. The integrals are calculated in the nonorthogonal atomic basis and then transformed by symmetric orthogonalization to the Löwdin basis. In the case of the 1-e integrals, we find strong orthogonalization effects that need to be included in SQC models, for example, by strategies such as those adopted in the available OMx methods. For the valence-only minimal basis, we confirm that the 2-e Coulomb integrals in the Löwdin basis are quantitatively close to their counterparts in the atomic basis and that the 2-e exchange integrals can be safely neglected in line with the NDDO approximation. For larger all-electron basis sets, there are strong multishell orthogonalization effects that lead to more irregular patterns in the transformed 2-e integrals and thus cast doubt on the validity of the NDDO approximation for extended basis sets. Focusing on the valence-only minimal basis, we find that some of the NDDO-neglected integrals are reduced but remain sizable after the transformation to the Löwdin basis; this is true for the two-center 2-e hybrid integrals, the three-center 1-e nuclear attraction integrals, and the corresponding three-center 2-e hybrid integrals. We consider a scheme with a valence-only minimal basis that includes such terms as a possible strategy to go beyond the NDDO integral approximation in attempts to improve SQC methods. © 2018 Wiley Periodicals, Inc.     

Two-electron integrations in the quantum theory of atoms in molecules

A method to compute two-electron integrals over arbitrary regions of space is introduced and particularized to the basins appearing in the quantum theory of atoms in molecules. The procedure generalizes the conventional multipolar approach to account for overlapping densities. We show that the approach is always convergent and computationally efficient, scaling as N(4) in the worst, two-center case. Several numerical results supporting our claims are also presented.     

Parallel direct four-index transformations

Summary Two parallel direct integral transformation algorithms are presented. Specific attention is directed to producing transformed integrals containing at least two active orbital indices. The number of active orbitals is typically much less than the total number of molecular orbitals reflecting the requirements of a wide range of correlated electronic structure methods. Sample direct second-order Møller-Plesset theory calculations are reported. For situations where multipassing of the integrals is required, superlinear speedup is obtained by exploiting the increase in global memory. As a consequence, for morphine in a 6-31G basis, a speedup of over 25 is observed in scaling from 32 to 512 processors.Pacific Northwest Laboratory is operated for the U.S. Department of Energy by Battelle Memorial Institute under contract DE-AC06-76RL0 1830     

Evaluation of bielectronic integrals for 1s Slater orbitals by using averages

A new approach for evaluating the four‐center bielectronic integrals (12|34), involving 1s Slater‐type orbitals, is presented. The method uses the multiplication theorem for Bessel functions. The bielectronic integral is expressed in terms of a finite sum of functions, and a scaling parameter is introduced. In the present work, the scaling parameter used is an average. The results show that the first term in the sum is always the principal contribution, and the remainder has a corrective character. The whole scheme and its numerical trend are understood on the basis of a theorem recently proved. © 2004 Wiley Periodicals, Inc. Int J Quantum Chem, 2005     

Tensor decomposition in post-Hartree-Fock methods. I. Two-electron integrals and MP2

A new approximation for post-Hartree-Fock (HF) methods is presented applying tensor decomposition techniques in the canonical product tensor format. In this ansatz, multidimensional tensors like integrals or wavefunction parameters are processed as an expansion in one-dimensional representing vectors. This approach has the potential to decrease the computational effort and the storage requirements of conventional algorithms drastically while allowing for rigorous truncation and error estimation. For post-HF ab initio methods, for example, storage is reduced to O(d·R·n) with d being the number of dimensions of the full tensor, R being the expansion length (rank) of the tensor decomposition, and n being the number of entries in each dimension (i.e., the orbital index). If all tensors are expressed in the canonical format, the computational effort for any subsequent tensor contraction can be reduced to O(R(2)·n). We discuss details of the implementation, especially the decomposition of the two-electron integrals, the AO-MO transformation, the M?ller-Plesset perturbation theory (MP2) energy expression and the perspective for coupled cluster methods. An algorithm for rank reduction is presented that parallelizes trivially. For a set of representative examples, the scaling of the decomposition rank with system and basis set size is found to be O(N(1.8)) for the AO integrals, O(N(1.4)) for the MO integrals, and O(N(1.2)) for the MP2 t(2)-amplitudes (N denotes a measure of system size) if the upper bound of the error in the l(2)-norm is chosen as ε = 10(-2). This leads to an error in the MP2 energy in the order of mHartree.     

Development and parametrization of sindo1 for second-row elements

The semiempirical MO method SINDO 1 is extended to second-row atoms from sodium to chlorine. The basis set has a provision to include d orbitals. To retain rotational invariance in a d orbital set, a number of hybrid integrals has to be included that invalidate the zero differential overlap (ZDO ) assumption even in a symmetrically orthogonalized basis set. The inclusion of d orbitals rendered the set-up of integral calculation of the original INDO method impractical. Instead of one subroutine for each integral, all explicitly calculated integrals (overlap, core, electronic repulsion) are now contained in a single subroutine under unifying aspects. The parametrization scheme includes pseudopotentials and adjusts the total energy under inclusion of zero point energies to experimental heats of formation of ground states. The vibrational frequencies for the calculation of zero point energies are obtained from calculated force constants and G matrix elements by a scaling procedure. The results for geometries, energies, and dipole moments are compared with MNDO data.     

Pseudospectral time-dependent density functional theory

Time-dependent density functional theory (TDDFT) is implemented within the Tamm-Dancoff approximation (TDA) using a pseudospectral approach to evaluate two-electron repulsion integrals. The pseudospectral approximation uses a split representation with both spectral basis functions and a physical space grid to achieve a reduction in the scaling behavior of electronic structure methods. We demonstrate here that exceptionally sparse grids may be used in the excitation energy calculation, following earlier work employing the pseudospectral approximation for determining correlation energies in wavefunction-based methods with similar conclusions. The pseudospectral TDA-TDDFT method is shown to be up to ten times faster than a conventional algorithm for hybrid functionals without sacrificing chemical accuracy.     

Overlap integrals of B functions

Two different methods for the evaluation of overlap integrals of B functions with different scaling parameters are analyzed critically. The first method consists of an infinite series expansion in terms of overlap integrals with equal scaling parameters [14]. The second method consists of an integral representation for the overlap integral which has to be evaluated numerically. Bhattacharya and Dhabal [13] recommend the use of Gauss-Legendre quadrature for this purpose. However, we show that Gauss-Jacobi quadrature gives better results, in particular for larger quantum number. We also show that the convergence of the infinite series can be improved if suitable convergence accelerators are applied. Since an internal error analysis can be done quite easily in the case of an infinite series even if it is accelerated, whereas it is very costly in the case of Gauss quadratures, the infinite series is probably more efficient than the integral representation. Overlap integrals of all commonly occurring exponentially declining basis functions such as Slater-type functions, can be expressed by finite sums of overlap integrals of B functions, because these basis functions can be represented by linear combinations of B functions.Dedicated to Professor J. Koutecký on the occasion of his 65th birthday     

Fast evaluation of scaled opposite spin second-order Møller-Plesset correlation energies using auxiliary basis expansions and exploiting sparsity

The scaled opposite spin Møller–Plesset method (SOS‐MP2) is an economical way of obtaining correlation energies that are computationally cheaper, and yet, in a statistical sense, of higher quality than standard MP2 theory, by introducing one empirical parameter. But SOS‐MP2 still has a fourth‐order scaling step that makes the method inapplicable to very large molecular systems. We reduce the scaling of SOS‐MP2 by exploiting the sparsity of expansion coefficients and local integral matrices, by performing local auxiliary basis expansions for the occupied‐virtual product distributions. To exploit sparsity of 3‐index local quantities, we use a blocking scheme in which entire zero‐rows and columns, for a given third global index, are deleted by comparison against a numerical threshold. This approach minimizes sparse matrix book‐keeping overhead, and also provides sufficiently large submatrices after blocking, to allow efficient matrix–matrix multiplies. The resulting algorithm is formally cubic scaling, and requires only moderate computational resources (quadratic memory and disk space) and, in favorable cases, is shown to yield effective quadratic scaling behavior in the size regime we can apply it to. Errors associated with local fitting using the attenuated Coulomb metric and numerical thresholds in the blocking procedure are found to be insignificant in terms of the predicted relative energies. A diverse set of test calculations shows that the size of system where significant computational savings can be achieved depends strongly on the dimensionality of the system, and the extent of localizability of the molecular orbitals. © 2007 Wiley Periodicals, Inc. J Comput Chem 2007     

Linear-scaling method for calculating nuclear magnetic resonance chemical shifts using gauge-including atomic orbitals within Hartree-Fock and density-functional theory

Details of a new density matrix-based formulation for calculating nuclear magnetic resonance chemical shifts at both Hartree-Fock and density functional theory levels are presented. For systems with a nonvanishing highest occupied molecular orbital-lowest unoccupied molecular orbital gap, the method allows us to reduce the asymptotic scaling order of the computational effort from cubic to linear, so that molecular systems with 1000 and more atoms can be tackled with today's computers. The key feature is a reformulation of the coupled-perturbed self-consistent field (CPSCF) theory in terms of the one-particle density matrix (D-CPSCF), which avoids entirely the use of canonical MOs. By means of a direct solution for the required perturbed density matrices and the adaptation of linear-scaling integral contraction schemes, the overall scaling of the computational effort is reduced to linear. A particular focus of our formulation is to ensure numerical stability when sparse-algebra routines are used to obtain an overall linear-scaling behavior.