Implementation of the diagonalization‐free algorithm in the self‐consistent field procedure within the four‐component relativistic scheme

A recently developed Thouless‐expansion‐based diagonalization‐free approach for improving the efficiency of self‐consistent field (SCF) methods (Noga and ?imunek, J. Chem. Theory Comput. 2010, 6, 2706) has been adapted to the four‐component relativistic scheme and implemented within the program package ReSpect. In addition to the implementation, the method has been thoroughly analyzed, particularly with respect to cases for which it is difficult or computationally expensive to find a good initial guess. Based on this analysis, several modifications of the original algorithm, refining its stability and efficiency, are proposed. To demonstrate the robustness and efficiency of the improved algorithm, we present the results of four‐component diagonalization‐free SCF calculations on several heavy‐metal complexes, the largest of which contains more than 80 atoms (about 6000 4‐spinor basis functions). The diagonalization‐free procedure is about twice as fast as the corresponding diagonalization. © 2014 Wiley Periodicals, Inc.     

An efficient self-consistent field method for large systems of weakly interacting components

An efficient method for removing the self-consistent field (SCF) diagonalization bottleneck is proposed for systems of weakly interacting components. The method is based on the equations of the locally projected SCF for molecular interactions (SCF MI) which utilize absolutely localized nonorthogonal molecular orbitals expanded in local subsets of the atomic basis set. A generalization of direct inversion in the iterative subspace for nonorthogonal molecular orbitals is formulated to increase the rate of convergence of the SCF MI equations. Single Roothaan step perturbative corrections are developed to improve the accuracy of the SCF MI energies. The resulting energies closely reproduce the conventional SCF energy. Extensive test calculations are performed on water clusters up to several hundred molecules. Compared to conventional SCF, speedups of the order of (N/O)2 have been achieved for the diagonalization step, where N is the size of the atomic orbital basis, and O is the number of occupied molecular orbitals.     

Parallel computation of molecular energy gradients on the loosely coupled array of processors (LCAP)

The implementation of the HONDO program on the Loosely Coupled Array of Processors (LCAP) parallel computer system assembled in our laboratory is presented. We discuss a general strategy used to maintain a high level of compatibility between the serial version and the parallel version of the code. We report the implementation of energy-gradient calculation for SCF wavefunctions. The integral and integral derivative programs display high parallel efficiency, and so does the SCF part in the case of very large basis sets.     

A general ab initio molecular multi-configuration self-consistent field algorithm

The ab initio multiconfiguration self-consistent-field (MC SCF ) techniques and computer programs of Basch [1, 2] and the ab initio configuration interaction (CI ) techniques and programs of Whitten and Hackmeyer [3] have been combined and generalized to form a general technique and program to yield optimized ab initio MC SCF wavefunctions for any set of Slater determinants. The Slater determinants are read in as input data to the program along with the spin parity that is being considered (optional) and the program successively does the CI calculation and one iteration of the SCF calculation, constructing the proper Fock–Hamiltonians by examining the set of Slater determinants and their CI coefficients. The Fock–Hamiltonian matrices are calculated and diagonalized in succession, a single two-dimensional array being used to store these matrices. The basis function integrals are read from a tape only once during each MC SCF iteration (one MC SCF iteration = a CI calculation followed by one iteration of the SCF calculation).     

Convergence optimization of restricted open-shell self-consistent field calculations

Different self-consistent field (SCF) iteration schemes for open-shell systems are discussed. After a brief summary of the well-known level shifting and damping procedure, we describe the quadratically convergent SCF (QCSCF) approach based on the gradient and the Hessian matrix in a space of orbital rotation parameters. An analytical expression for the latter is derived for the general many-shell case. Starting from the expression for the energy change obtained by the QCSCF method, we then present a simplified direct procedure avoiding matrix diagonalization but also the difficulties of the QCSCF method in handling the Hessian matrix. Numerical calculations on some open-shell systems involving transition-metal complexes show that this method leads to rapid and reliable convergence of the iteration process in cases where the usual SCF procedure of iterative diagonalization tends to diverge. © 1997 John Wiley & Sons, Inc. Int J Quant Chem 62: 617–637, 1997     

Development of parallel density functional program using distributed matrix to calculate all-electron canonical wavefunction of large molecules

We developed a new parallel density-functional canonical molecular-orbital program for large molecules based on the resolution of the identity method. In this study, all huge matrices were decomposed and saved to the distributed local memory. The routines of the analytical molecular integrals and numerical integrals of the exchange-correlation terms were parallelized using the single program multiple data method. A conventional linear algebra matrix library, ScaLAPACK, was used for matrix operations, such as diagonalization, multiplication, and inversion. Anderson's mixing method was adopted to accelerate the self-consistent field (SCF) convergence. Using this program, we calculated the canonical wavefunctions of a 306-residue protein, insulin hexamer (26,790 orbitals), and a 133-residue protein, interleukin (11,909 orbitals) by the direct-SCF method. In regard to insulin hexamer, the total parallelization efficiency of the first SCF iteration was estimated to be 82% using 64 Itanium 2 processors connected at 3.2 GB/s (SGI Altix3700), and the calculation successfully converged at the 17-th SCF iteration. By adopting the update method, the computational time of the first and the final SCF loops was 229 min and 156 min, respectively. The whole computational time including the calculation before the SCF loop was 2 days and 17 h. This study put the calculations of the canonical wavefunction of 30,000 orbitals to practical use.     

Direct relativistic MP2: properties of ground state CuF, AgF and AuF

A computer program for the calculation of the MP2 energy correction for a Kramers-restricted Dirac-Hartree-Fock four component wave-function is presented. In the spirit of the integral-driven direct SCF scheme the algorithm has been developed as direct MP2, calculating integrals as they are needed and avoiding the integral storage bottle-neck of conventional MP2. Relativistic MP2 is applied to ground state (¹Σ⁺) CuF, AgF and AuF. Received: 15 December 1996 / Accepted: 2 April 1997     

Parallel pseudospectral electronic structure: I. Hartree–Fock calculations

We present an outline of the parallel implementation of our pseudospectral electronic structure program, Jaguar, including the algorithm and timings for the Hartree–Fock and analytic gradient portions of the program. We also present the parallel algorithm and timings for our Lanczos eigenvector refinement code and demonstrate that its performance is superior to the ScaLAPACK diagonalization routines. The overall efficiency of our code increases as the size of the calculation is increased, demonstrating actual as well as theoretical scalability. For our largest test system, alanine pentapeptide [818 basis functions in the cc-pVTZ(-f) basis set], our Fock matrix assembly procedure has an efficiency of nearly 90% on a 16-processor SP2 partition. The SCF portion for this case (including eigenvector refinement) has an overall efficiency of 87% on a partition of 8 processors and 74% on a partition of 16 processors. Finally, our parallel gradient calculations have a parallel efficiency of 84% on 8 processors for porphine (430 basis functions). © 1998 John Wiley & Sons, Inc. J Comput Chem 19: 1017–1029, 1998     

On the use of symmetry in the ab initio quantum mechanical simulation of nanotubes and related materials

Nanotubes can be characterized by a very high point symmetry, comparable or even larger than the one of the most symmetric crystalline systems (cubic, 48 point symmetry operators). For example, N = 2n rototranslation symmetry operators connect the atoms of the (n,0) nanotubes. This symmetry is fully exploited in the CRYSTAL code. As a result, ab initio quantum mechanical large basis set calculations of carbon nanotubes containing more than 150 atoms in the unit cell become very cheap, because the irreducible part of the unit cell reduces to two atoms only. The nanotube symmetry is exploited at three levels in the present implementation. First, for the automatic generation of the nanotube structure (and then of the input file for the SCF calculation) starting from a two‐dimensional structure (in the specific case, graphene). Second, the nanotube symmetry is used for the calculation of the mono‐ and bi‐electronic integrals that enter into the Fock (Kohn‐Sham) matrix definition. Only the irreducible wedge of the Fock matrix is computed, with a saving factor close to N. Finally, the symmetry is exploited for the diagonalization, where each irreducible representation is separately treated. When M atomic orbitals per carbon atom are used, the diagonalization computing time is close to Nt, where t is the time required for the diagonalization of each 2M × 2M matrix. The efficiency and accuracy of the computational scheme is documented. © 2009 Wiley Periodicals, Inc. J Comput Chem, 2010     

Computational chemistry with transputers: a direct SCF program

By using transputers it is possible to build up networks of parallel processors with varying topology. Due to the architecture of the processors it is appropriate to use the MIMD (multiple instruction multiple data) concept of parallel computing. The most suitable programming language is OCCAM. We investigate the use of transputer networks in computational chemistry, starting with the direct SCF method. The most time consuming step, the calculation of the two electron integrals is executed parallelly. Each node in the network calculates whole batches of integrals. The main program is written in OCCAM. For some large-scale arithmetic processes running on a single node, however, we used FORTRAN subroutines out of standard ab-initio programs to reduce the programming effort. Test calculations show, that the integral calculation step can be parallelized very efficiently. We observe a speed-up of almost 8 using eight network processors. Even in consideration of the scalar part of the SCF iteration, the speed-up is not less than 7.1.     

Strategies to vectorize conventional SCF-Cl algorithms

Recent high-performance computers, especially supercomputers, achieve very high-speed operations but bring about serious I/O problems in quantum chemical computations. Strategies to vectorize conventional SCF-CI algorithms are discussed relating to the I/O problems. The conventional SCF-CI algorithm which is proposed here reduces I/O processing by eliminating all sorting routines and redundant integral files and generates directly nonzero and nonredundant PK integrals with a vectorizable canonically-ordered list. The new implementation has been undertaken and successfully realized as a program system named GSCF3. The vector to scalar acceleration rate of GSCF3 on the HITAC S-810 are as follows: 2.5 5 in the AO integral evaluation, 5 12 in the SCF calculation, 15 30 in the four-index integral transformation, 10 20 in the CI matrix diagonalization, and overall 510 through SCF-CI.     

A new direct CI method for large CI expansions in a small orbital space

A new direct CI method is presented, which is particularly suited for large CI expansions in a small orbital space. These are the type of expansions which are common in the CAS SCF method. Only one-electron coupling coefficients are stored, which leads to reduced elapsed times and storage requirements compared to earlier approaches. The two-electron coupling coefficients are implicitly created in the diagonalization step. The algorithm for updating the CI vector is formulated as the trace of a product of three matrices, Tr(A · D · I). By ordering the one-electron coupling coefficients (A) in a certain way the matrix D is easilly created as a sparse scalar product between these coefficients and the trial CI vector. The main computational step is then a simple matrix multiplication between the matrix D and the symmetry blocked integral matrix (1). This operation vectorizes very well on most vector processors. Another sparse scalar product between the resultant matrix and the coupling coefficients leads to the update of the CI coefficients. In a calculation on CRAY-1 with 30700 configurations, the two-electron part in a CI iteration required 10 s of which half went into the handling of the one-electron formula tape.     

An exact reformulation of the diagonalization step in electronic structure calculations as a set of second order nonlinear equations

A new formulation of the diagonalization step in self-consistent-field (SCF) electronic structure calculations is presented. It exactly replaces the diagonalization of the effective Hamiltonian with the solution of a set of second order nonlinear equations. The density matrix and/or the new set of occupied orbitals can be directly obtained from the resulting solution. This formulation may offer interesting possibilities for new approaches to efficient SCF calculations. The working equations can be derived either from energy minimization with respect to a Cayley-type parametrization of a unitary matrix, or from a similarity transformation approach.     

The trust-region self-consistent field method: towards a black-box optimization in Hartree-Fock and Kohn-Sham theories

The trust-region self-consistent field (TRSCF) method is presented for optimizing the total energy E(SCF) of Hartree-Fock theory and Kohn-Sham density-functional theory. In the TRSCF method, both the Fock/Kohn-Sham matrix diagonalization step to obtain a new density matrix and the step to determine the optimal density matrix in the subspace of the density matrices of the preceding diagonalization steps have been improved. The improvements follow from the recognition that local models to E(SCF) may be introduced by carrying out a Taylor expansion of the energy about the current density matrix. At the point of expansion, the local models have the same gradient as E(SCF) but only an approximate Hessian. The local models are therefore valid only in a restricted region-the trust region-and steps can only be taken with confidence within this region. By restricting the steps of the TRSCF model to be inside the trust region, a monotonic and significant reduction of the total energy is ensured in each iteration of the TRSCF method. Examples are given where the TRSCF method converges monotonically and smoothly, but where the standard DIIS method diverges.     

MC SCF optimization using the direct,restricted step,second-order norm-extended optimization method

The potential of a direct, restricted step, second order MC SCF computer program is illustrated with a large configuration space calculation containing 24156 configuration state functions on the ³B₁ state of methylene.     

Is large-scale ab initio Hartree-Fock calculation chemically accurate? Toward improved calculation of biological molecule properties

Numerical errors in total energy values in large-scale Hartree–Fock calculations are discussed. To obtain total energy values within chemical accuracy, 0.01 kcal/mol, stricter numerical accuracy is required as basis size increases. In molecules with 10,000 basis sizes, such as proteins, numerical accuracy for total energy values must be retained to at least 11 digits (i.e., to the order of 1.0D-10) to keep accumulation of numerical errors less than the chemical accuracy (0.01 kcal/mol). With this criterion, we examined the sensitivity analysis of numerical accuracy in Hartree–Fock calculation by uniformly replacing the last bit of the mantissa part of a double-precision real number by zero in the Fock matrix construction step, the total energy calculation step, and the Fock matrix diagonalization step. Using a partial summation technique in the Fock matrix generation step, the numerical error for total energy value of molecules with basis size greater than 10,000 was within chemical accuracy (0.01 kcal/mol), whereas with the conventional method the numerical error with several thousand basis sets was larger than chemical accuracy. Computation of one Fock matrix element with parallel machines can include the partial summation technique automatically, so that parallel calculation yields not only high-performance computing but also more precise numerical solutions than the conventional sequential algorithm. We also found that the numerical error of the Householder-QR diagonalization routine is equal to or less than chemical accuracy, even with a matrix size of 10,000. ©1999 John Wiley & Sons, Inc. J Comput Chem 20: 443–454, 1999     

Ab initio SCF calculations on [V10O28]6−: A benchmark for the classical calculation and processing of molecular integrals on large Gaussian basis sets

A vector efficient implementation of the McMurchie and Davidson algorithm for the calculation of one- and two-electron molecular integrals is presented, as available in the Cray version of the ASTERIX program system. The implementation and performance of a vector-oriented strategy for the generation and processing of the P supermatrix is also discussed. This program system has been applied to the ab initio SCF computation of the ground-state wave function for the [V₁₀O₂₈]^6? ion, with a basis set of triple-zeta quality for the valence shell of oxygen generating 1404 GTOS and 574 CGTOS for the complete system. The performance and the bottlenecks of the integral calculation are discussed as a function of the integral classes. Two-dimensional maps of the electrostatic potential are presented for this molecule and compared to experimental information about proton fixation.     

An efficient ab initio gradient program

An ab initio Hartree-Fock gradient program is described. It is characterized by (1) efficiency of the gradient evaluation, and (2) capability of handling higher angular momentum (d andf) basis functions. The latter are constructed from shifted Cartesian Gaussian p-type primitives. A satisfactory solution is presented for the problems connected with the neglect of small integrals in a gradient program. Methods for increasing the efficiency of the SCF procedure are discussed.     

Semidirect parallel self-consistent field: the load balancing problem in the input/output intensive self-consistent field iterations

The full capacity of contemporary parallel computers can, in the context of iterative ab initio procedures like, for example, self-consistent field (SCF) and multiconfigurational SCF, only be utilized if the disk and input/output (I/O) capacity are fully exploited before the implementation turns to an integral direct strategy. In a recent report on parallel semidirect SCF http://www.tc.cornell.edu/er/media/1996/collabrate.html, http://www.fp.mcs.anl.gd/grand-challenges/chem/nondirect/index.html it was demonstrated that super-linear speedups are achievable for algorithms that exploit scalable parallel I/O. In the I/O-intensive SCF iterations of this implementation a static load balancing, however, was employed, dictated by the initial iteration in which integral evaluation dominates the central processing unit activity and thus determines the load balancing. In the present paper we present the first implementation in which load balancing is achieved throughout the whole SCF procedure, i.e. also in subsequent iterations. The improved scalability of our new algorithm is demonstrated in some test calculations, for example, for 63-node calculation a speedup of 104 was observed in the computation of the two-electron integral contribution to the Fock matrix.Contribution to the Björn Roos Honorary Issue Acknowledgement.We thank J. Nieplocha for valuable help and making the toolkit (including ChemIO) available to us. R.L. acknowledges the Intelligent Modeling Laboratory and the University of Tokyo for financial support during his stay in Japan.     

Parallel algorithm for integral transformations and GUGA MCSCF

Summary An algorithm for the parallelization of the atomic to molecular integral transformation and the subsequent steps in a GUGA based MCSCF calculation is presented. Timing data shows that the transformation and diagonalization steps are well parallelized and that several of the other portions of the MCSCF code are moderately parallel. Remaining sequential bottlenecks are identified.