Direct Inversion In The Iterative Subspace Research Articles

Efficient and Robust Ab Initio Self-Consistent Field Acceleration Algorithm Based on a Semiempirical Model Hamiltonian.

A novel doubly iterative self-consistent field (SCF) approach using a semiempirical model Hamiltonian (denoted as the SMH algorithm) is proposed to accelerate the Hartree-Fock (HF) and density functional theory (DFT) calculations. This method first constructs the Fock matrix exactly in each SCF macroiteration, followed by a few SCF microiterations, where the Fock matrix is incrementally updated using an inexpensive semiempirical approximation. This leads to an improved wave function in each SCF macroiteration with minimal additional cost, and therefore a reduced number of exact Fock builds is required for SCF convergence. The SMH algorithm can be combined with conventional SCF convergence techniques such as level shifting, direct inversion in the iterative subspace (DIIS), and energy-DIIS (EDIIS). When integrated with DIIS, SMH enhances the convergence of simple organic molecules by approximately 10% compared to plain DIIS, with speedups of up to 60% for the more challenging transition metal systems compared to EDIIS + DIIS. Our results show that SMH is a reliable SCF accelerator that seldom deteriorates convergence and is highly robust.

An Efficient Algorithm for Constrained CASSCF(1,2) and CASSCF(3,2) Simulations as Relevant to Electron and Hole Transfer Problems.

We propose an efficient algorithm for the recently published electron/hole-transfer Dynamical-weighted State-averaged Constrained CASSCF (eDSC/hDSC) method studying charge transfer states and D1-D0 crossings for systems with odd numbers of electrons. By separating the constrained minimization problem into an unconstrained self-consistent-field (SCF) problem and a constrained nonself-consistent-field (nSCF) problem, as well as accelerating the direct inversion in the iterative subspace (DIIS) technique to solve the SCF problem, the overall computational cost is reduced by a factor of 8-20 compared with directly using sequential quadratic programming (SQP). This approach should be applicable for other constrained minimization problems, and in the immediate future, once gradients are available, the present eDSC/hDSC algorithm should allow for speedy nonadiabatic dynamics simulations.

Exploring alternative approaches to improve the convergence pattern in solving the coupled-cluster equations

The perturbational parameter λ formally appearing in the coupled-cluster (CC) and perturbation theory (PT) frameworks is included explicitly in the Hamiltonian and the CC formalism. Using the Møller–Plesset (MP) partitioning, properties of the resulting λ-dependent amplitude functions are discussed. Truncating the cluster expansion at singles and doubles (CCSD), we use these properties and knowledge of the CCSD amplitudes at values to estimate the CCSD amplitudes without solving the relevant equations at (corresponding to the original, physical problem). This extrapolation scheme is explored, with emphasis on convergence improvement of CC iterations. The approximate amplitudes generated by this procedure are used as the starting point for the CCSD iterations, an alternative to the first order MP (MP1) set. Numerical examples are presented which show that the technique combined with the direct inversion in the iterative subspace (DIIS) method is especially helpful in situations where the standard CCSD iterations converge slowly or not at all. We report cases where starting from MP1, the solution process has an unfavourable convergence pattern even with DIIS applied, while the amplitudes generated by our method provide a much better starting point, overcoming the divergence issues of the original sequence.

Iterative subspace algorithms for finite-temperature solution of Dyson equation.

One-particle Green's functions obtained from the self-consistent solution of the Dyson equation can be employed in the evaluation of spectroscopic and thermodynamic properties for both molecules and solids. However, typical acceleration techniques used in the traditional quantum chemistry self-consistent algorithms cannot be easily deployed for the Green's function methods because of a non-convex grand potential functional and a non-idempotent density matrix. Moreover, the optimization problem can become more challenging due to the inclusion of correlation effects, changing chemical potential, and fluctuations of the number of particles. In this paper, we study acceleration techniques to target the self-consistent solution of the Dyson equation directly. We use the direct inversion in the iterative subspace (DIIS), the least-squared commutator in the iterative subspace (LCIIS), and the Krylov space accelerated inexact Newton method (KAIN). We observe that the definition of the residual has a significant impact on the convergence of the iterative procedure. Based on the Dyson equation, we generalize the concept of the commutator residual used in DIIS and LCIIS and compare it with the difference residual used in DIIS and KAIN. The commutator residuals outperform the difference residuals for all considered molecular and solid systems within both GW and GF2. For a number of bond-breaking problems, we found that an easily obtained high-temperature solution with effectively suppressed correlations is a very effective starting point for reaching convergence of the problematic low-temperature solutions through a sequential reduction of temperature during calculations.

A self-consistent field formulation of excited state mean field theory.

We show that, as in Hartree-Fock theory, the orbitals for excited state mean field theory can be optimized via a self-consistent one-electron equation in which electron-electron repulsion is accounted for through mean field operators. In addition to showing that this excited state ansatz is sufficiently close to a mean field product state to admit a one-electron formulation, this approach brings the orbital optimization speed to within roughly a factor of two of ground state mean field theory. The approach parallels Hartree Fock theory in multiple ways, including the presence of a commutator condition, a one-electron mean-field working equation, and acceleration via direct inversion in the iterative subspace. When combined with a configuration interaction singles Davidson solver for the excitation coefficients, the self-consistent field formulation dramatically reduces the cost of the theory compared to previous approaches based on quasi-Newton descent.

Robust mixing in self-consistent linearized augmented planewave calculations

We devise a mixing algorithm for full-potential (FP) all-electron calculations in the linearized augmented planewave (LAPW) method. Pulay’s direct inversion in the iterative subspace is complemented with the Kerker preconditioner and further improvements to achieve smooth convergence, avoiding charge sloshing and noise in the exchange–correlation potential. As the Kerker preconditioner was originally designed for the planewave basis, we have adapted it to the FP-LAPW method and implemented in the exciting code. Applications to the 2 × 2 Au(111) surface with a vacancy and to the Pd(111) surface demonstrate that this approach and our implementation work reliably with both density and potential mixing.

Gaussian Basis Sets for Crystalline Solids: All-Purpose Basis Set Libraries vs System-Specific Optimizations.

It is customary in molecular quantum chemistry to adopt basis set libraries in which the basis sets are classified according to either their size (triple-ζ, quadruple-ζ, ...) and the method/property they are optimal for (correlation-consistent, linear-response, ...) but not according to the chemistry of the system to be studied. In fact the vast majority of molecules is quite homogeneous in terms of density (i.e., atomic distances) and types of bond involved (covalent or dispersive). The situation is not the same for solids, in which the same chemical element can be found having metallic, ionic, covalent, or dispersively bound character in different crystalline forms or compounds, with different packings. This situation calls for a different approach to the choice of basis sets, namely a system-specific optimization of the basis set that requires a practical algorithm that could be used on a routine basis. In this work we develop a basis set optimization method based on an algorithm–similar to the direct inversion in the iterative subspace–that we name BDIIS. The total energy of the system is minimized together with the condition number of the overlap matrix as proposed by VandeVondele et al. [VandeVondele et al. J. Chem. Phys.2007, 227, 114105]. The details of the method are here presented, and its performance in optimizing valence orbitals is shown. As demonstrative systems we consider simple prototypical solids such as diamond, graphene sodium chloride, and LiH, and we show how basis set optimizations have certain advantages also toward the use of large (quadruple-ζ) basis sets in solids, both at the DFT and Hartree–Fock level.

Efficient evaluation of three-centre two-electron integrals over London orbitals

The nonperturbative calculation of molecular properties in magnetic fields requires the evaluation of integrals over complex-valued Gaussian-type London atomic orbitals (LAOs). With these orbitals, the calculation of four-centre electron-repulsion integrals (ERIs) is particularly demanding, because their permutational symmetry is lowered, and because complex algebra is required. We have implemented the resolution-of-the-identity (RI) approximation for LAOs in the TURBOMOLE program package. With respect to LAOs, employing the RI approximation is particularly beneficial, because the auxiliary basis set may always be chosen to be real-valued. As a consequence, the two-centre integrals in the RI approximation remain real-valued, and the three-centre integrals possess the same permutational symmetry as their real-valued counterparts. Compared to a direct calculation of four-centre ERIs over LAOs, using the RI approximation thus not only reduces the scaling of the integral evaluation, but also increases the efficiency by an additional factor of at least two. By using other well-established methods such as Cauchy–Schwarz screening, the difference-density approach, and Pulay's direct inversion in the iterative subspace (DIIS), the efficiency of nonperturbative calculations in magnetic fields can be increased even further.

Fast divide-and-conquer algorithm for evaluating polarization in classical force fields.

Evaluation of the self-consistent polarization energy forms a major computational bottleneck in polarizable force fields. In large systems, the linear polarization equations are typically solved iteratively with techniques based on Jacobi iterations (JI) or preconditioned conjugate gradients (PCG). Two new variants of JI are proposed here that exploit domain decomposition to accelerate the convergence of the induced dipoles. The first, divide-and-conquer JI (DC-JI), is a block Jacobi algorithm which solves the polarization equations within non-overlapping sub-clusters of atoms directly via Cholesky decomposition, and iterates to capture interactions between sub-clusters. The second, fuzzy DC-JI, achieves further acceleration by employing overlapping blocks. Fuzzy DC-JI is analogous to an additive Schwarz method, but with distance-based weighting when averaging the fuzzy dipoles from different blocks. Key to the success of these algorithms is the use of K-means clustering to identify natural atomic sub-clusters automatically for both algorithms and to determine the appropriate weights in fuzzy DC-JI. The algorithm employs knowledge of the 3-D spatial interactions to group important elements in the 2-D polarization matrix. When coupled with direct inversion in the iterative subspace (DIIS) extrapolation, fuzzy DC-JI/DIIS in particular converges in a comparable number of iterations as PCG, but with lower computational cost per iteration. In the end, the new algorithms demonstrated here accelerate the evaluation of the polarization energy by 2-3 fold compared to existing implementations of PCG or JI/DIIS.

An efficient optimization method for geminal-based wave functions

Antisymmetric product of strongly orthogonal geminal (APSG) and its simple variant, generalized valence bond (GVB) perfect pairing are capable of producing the static (intrageminal) electron correlation energy in a chemical entity, and recently it has been proposed how to recover the dynamical (intergeminal) electron correlation energy in terms of the transition density matrix elements by solving the extended random phase approximation equation. The problem of these wave functions is, however, the slow convergence in the optimization. In order to optimize efficiently both orbitals and expansion coefficients of the APSG (or GVB), two different kinds of algorithms are introduced and tested. They employ direct inversion in the iterative subspace along with Broyden–Fletcher–Goldfarb–Shanno methods.

Accelerating the weighted histogram analysis method by direct inversion in the iterative subspace

The weighted histogram analysis method (WHAM) for free energy calculations is a valuable tool to produce free energy differences with the minimal errors. Given multiple simulations, WHAM obtains from the distribution overlaps the optimal statistical estimator of the density of states, from which the free energy differences can be computed. The WHAM equations are often solved by an iterative procedure. In this work, we use a well-known linear algebra algorithm which allows for more rapid convergence to the solution. We find that the computational complexity of the iterative solution to WHAM, and the closely-related multiple Bennett acceptance ratio method can be improved using the method of direct inversion in the iterative subspace. We give examples from a lattice model, a simple liquid and an aqueous protein solution.

Assessment of self-consistent field convergence in spin-dependent relativistic calculations

This Letter assesses the self-consistent field (SCF) convergence behavior in the generalized Hartree–Fock (GHF) method. Four acceleration algorithms were implemented for efficient SCF convergence in the GHF method: the damping algorithm, the conventional direct inversion in the iterative subspace (DIIS), the energy-DIIS (EDIIS), and a combination of DIIS and EDIIS. Four different systems with varying complexity were used to investigate the SCF convergence using these algorithms, ranging from atomic systems to metal complexes. The numerical assessments demonstrated the effectiveness of a combination of DIIS and EDIIS for GHF calculations in comparison with the other discussed algorithms.

Periodic Pulay method for robust and efficient convergence acceleration of self-consistent field iterations

Pulay's Direct Inversion in the Iterative Subspace (DIIS) method is one of the most widely used mixing schemes for accelerating the self-consistent solution of electronic structure problems. In this work, we propose a simple generalization of DIIS in which Pulay extrapolation is performed at periodic intervals rather than on every self-consistent field iteration, and linear mixing is performed on all other iterations. We demonstrate through numerical tests on a wide variety of materials systems in the framework of density functional theory that the proposed generalization of Pulay's method significantly improves its robustness and efficiency.

Restarted Pulay mixing for efficient and robust acceleration of fixed-point iterations

We present a variant of the restarted Pulay's Direct Inversion in the Iterative Subspace (DIIS) method for efficiently and robustly accelerating the convergence of fixed-point iterations. Specifically, we propose a simple modification of DIIS without any additional parameters, which we refer to as the r-Pulay method. We demonstrate the efficacy of r-Pulay in the context of the Jacobi iteration for solving large linear systems of equations, as well as in the Self Consistent Field (SCF) approach for Density Functional Theory (DFT) calculations. Overall, we find r-Pulay to be an attractive version of the restarted DIIS method.

Scalable evaluation of polarization energy and associated forces in polarizable molecular dynamics: II. Toward massively parallel computations using smooth particle mesh Ewald.

In this article, we present a parallel implementation of point dipole-based polarizable force fields for molecular dynamics (MD) simulations with periodic boundary conditions (PBC). The smooth particle mesh Ewald technique is combined with two optimal iterative strategies, namely, a preconditioned conjugate gradient solver and a Jacobi solver in conjunction with the direct inversion in the iterative subspace for convergence acceleration, to solve the polarization equations. We show that both solvers exhibit very good parallel performances and overall very competitive timings in an energy and force computation needed to perform a MD step. Various tests on large systems are provided in the context of the polarizable AMOEBA force field as implemented in the newly developed Tinker-HP package, which is the first implementation of a polarizable model that makes large-scale experiments for massively parallel PBC point dipole models possible. We show that using a large number of cores offers a significant acceleration of the overall process involving the iterative methods within the context of SPME and a noticeable improvement of the memory management, giving access to very large systems (hundreds of thousands of atoms) as the algorithm naturally distributes the data on different cores. Coupled with advanced MD techniques, gains ranging from 2 to 3 orders of magnitude in time are now possible compared to nonoptimized, sequential implementations, giving new directions for polarizable molecular dynamics with periodic boundary conditions using massively parallel implementations.

Scalable Evaluation of Polarization Energy and Associated Forces in Polarizable Molecular Dynamics: II.Towards Massively Parallel Computations using Smooth Particle Mesh Ewald.

In this paper, we present a scalable and efficient implementation of point dipole-based polarizable force fields for molecular dynamics (MD) simulations with periodic boundary conditions (PBC). The Smooth Particle-Mesh Ewald technique is combined with two optimal iterative strategies, namely, a preconditioned conjugate gradient solver and a Jacobi solver in conjunction with the Direct Inversion in the Iterative Subspace for convergence acceleration, to solve the polarization equations. We show that both solvers exhibit very good parallel performances and overall very competitive timings in an energy-force computation needed to perform a MD step. Various tests on large systems are provided in the context of the polarizable AMOEBA force field as implemented in the newly developed Tinker-HP package which is the first implementation for a polarizable model making large scale experiments for massively parallel PBC point dipole models possible. We show that using a large number of cores offers a significant acceleration of the overall process involving the iterative methods within the context of spme and a noticeable improvement of the memory management giving access to very large systems (hundreds of thousands of atoms) as the algorithm naturally distributes the data on different cores. Coupled with advanced MD techniques, gains ranging from 2 to 3 orders of magnitude in time are now possible compared to non-optimized, sequential implementations giving new directions for polarizable molecular dynamics in periodic boundary conditions using massively parallel implementations.

Electrostatics of proteins in dielectric solvent continua. II. Hamiltonian reaction field dynamics

In Paper I of this work [S. Bauer, G. Mathias, and P. Tavan, J. Chem. Phys. 140, 104102 (2014)] we have presented a reaction field (RF) method, which accurately solves the Poisson equation for proteins embedded in dielectric solvent continua at a computational effort comparable to that of polarizable molecular mechanics (MM) force fields. Building upon these results, here we suggest a method for linearly scaling Hamiltonian RF/MM molecular dynamics (MD) simulations, which we call "Hamiltonian dielectric solvent" (HADES). First, we derive analytical expressions for the RF forces acting on the solute atoms. These forces properly account for all those conditions, which have to be self-consistently fulfilled by RF quantities introduced in Paper I. Next we provide details on the implementation, i.e., we show how our RF approach is combined with a fast multipole method and how the self-consistency iterations are accelerated by the use of the so-called direct inversion in the iterative subspace. Finally we demonstrate that the method and its implementation enable Hamiltonian, i.e., energy and momentum conserving HADES-MD, and compare in a sample application on Ac-Ala-NHMe the HADES-MD free energy landscape at 300K with that obtained in Paper I by scanning of configurations and with one obtained from an explicit solvent simulation.

Transition State Search Using a Guided Direct Inversion in the Iterative Subspace Method.

A transition state optimization method using a guided energy-represented direct inversion in the iterative subspace (gEn-DIIS) algorithm is introduced and compared with the quasi-Newton rational function optimization (RFO) method. A hybrid technique that employs a combination of RFO and guided DIIS methods at various stages of convergence is presented. A set of test molecules is optimized for comparison using the hybrid method with gEn-DIIS and the traditional RFO methods. The gEn-DIIS method presented here exhibits fast optimization and is shown to be advantageous for difficult optimizations where the reaction path is flat.

Efficient treatment of solvation shells in 3D molecular theory of solvation

We developed a technique to decrease memory requirements when solving the integral equations of three-dimensional (3D) molecular theory of solvation, a.k.a. 3D reference interaction site model (3D-RISM), using the modified direct inversion in the iterative subspace (MDIIS) numerical method of generalized minimal residual type. The latter provides robust convergence, in particular, for charged systems and electrolyte solutions with strong associative effects for which damped iterations do not converge. The MDIIS solver (typically, with 2 × 10 iterative vectors of argument and residual for fast convergence) treats the solute excluded volume (core), while handling the solvation shells in the 3D box with two vectors coupled with MDIIS iteratively and incorporating the electrostatic asymptotics outside the box analytically. For solvated systems from small to large macromolecules and solid-liquid interfaces, this results in 6- to 16-fold memory reduction and corresponding CPU load decrease in MDIIS. We illustrated the new technique on solvated systems of chemical and biomolecular relevance with different dimensionality, both in ambient water and aqueous electrolyte solution, by solving the 3D-RISM equations with the Kovalenko-Hirata (KH) closure, and the hypernetted chain (HNC) closure where convergent. This core-shell-asymptotics technique coupling MDIIS for the excluded volume core with iteration of the solvation shells converges as efficiently as MDIIS for the whole 3D box and yields the solvation structure and thermodynamics without loss of accuracy. Although being of benefit for solutes of any size, this memory reduction becomes critical in 3D-RISM calculations for large solvated systems, such as macromolecules in solution with ions, ligands, and other cofactors.

Accelerating convergence in the antisymmetric product of strongly orthogonal geminals method

AbstractAntisymmetric product of strongly orthogonal geminals (APSG) method is a wavefunction theory that can effectively treat the static electron correlation using two‐electron wavefunctions, called geminals. However, the APSG method has the problem of slow convergence in the optimization of the geminal function. In this study, we introduced the direct inversion in the iterative subspace (DIIS) method, for both closed‐ and open‐shell systems, for accelerating its convergence. Two types of error vectors, that is, unitary transformation and orbital gradient, were examined for the DIIS procedure. Numerical assessments revealed that the orbital‐gradient error vector shows better performance than the unitary‐transformation one. © 2012 Wiley Periodicals, Inc.