Search NASA⌕ Search

SEARCH · Search NASA

Results for “initial guess”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 109 records · Page 6
All sourcesNASA NTRSDOE OSTI
DOE OSTI2022

A deterministic gradient-based approach to avoid saddle points

Abstract Loss functions with a large number of saddle points are one of the major obstacles for training modern machine learning (ML) models efficiently. First-order methods such as gradient descent (GD) are usually the methods of choice for training ML models. However, these methods converge to saddle points for certain choices of initial guesses. In this paper, we propose a modification of the recently proposed Laplacian smoothing gradient descent (LSGD) [Osher et al., arXiv:1806.06317 ], called modified LSGD (mLSGD), and demonstrate its potential to avoid saddle points without sacrificing the convergence rate. Our analysis is based on the attraction region, formed by all starting points for which the considered numerical scheme converges to a saddle point. We investigate the attraction region’s dimension both analytically and numerically. For a canonical class of quadratic functions, we show that the dimension of the attraction region for mLSGD is $\lfloor (n-1)/2\rfloor$ , and hence it is significantly smaller than that of GD whose dimension is $n-1$ .

Mathematics↗
DOE OSTI2022

Pseudodiagonalization Method for Accelerating Nonlinear Subspace Diagonalization in Density Functional Theory

In density functional theory, each self-consistent field (SCF) nonlinear step updates the discretized Kohn-Sham orbitals by solving a linear eigenvalue problem. The concept of pseudodiagonalization is to solve this linear eigenvalue problem approximately, and specifically utilizing a method involving a small number of Jacobi rotations that takes advantage of the good initial guess to the solution given by the approximation to the orbitals from the previous SCF iteration. The approximate solution to the linear eigenvalue problem can be very rapid, particularly for those steps near SCF convergence. Here, we adapt pseudodiagonalization to finite-temperature and metallic systems, where partially-occupied orbitals must be individually resolved with some accuracy. We apply pseudodiagonalization to the subspace eigenvalue problem that arises in Chebyshev-filtered subspace iteration. In tests on metallic and other systems for a range of temperatures, we show that pseudodiagonalization achieves similar rates of SCF convergence to exact diagonalization.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2024

Strategies to Obtain Reliable Energy Landscapes from Embedded Multireference Correlated Wavefunction Methods for Surface Reactions

Embedded correlated wavefunction (ECW) theory is a powerful tool for studying ground- and excited-state reaction mechanisms and associated energetics in heterogeneous catalysis. Several factors are important to obtaining reliable ECW energies, critically the construction of consistent active spaces (ASs) along reaction pathways when using a multireference correlated wavefunction (CW) method that relies on a subset of orbital spaces in the configuration interaction expansion to account for static electron correlation, e.g., complete AS self-consistent field theory, in addition to the adequate partitioning of the system into a cluster and environment, as well as the choice of a suitable basis set and number of states included in excited-state simulations. Here, in this work, we conducted a series of systematic studies to develop best-practice guidelines for ground- and excited-state ECW theory simulations, utilizing the decomposition of NH 3 on Pd(111) as an example. We determine that ECW theory results are relatively insensitive to cluster size, the aug-cc-pVDZ basis set provides an adequate compromise between computational complexity and accuracy, and that a fixed-clean-surface approximation holds well for the derivation of the embedding potential. Additionally, we demonstrate that a merging approach, which involves generating ASs from the molecular fragments at each configuration, is preferable to a creeping approach, which utilizes ASs from adjacent structures as an initial guess, for the generation of consistent potential energy curves involving open-d-shell metal surfaces, and, finally, we show that it is essential to include bands of excited states in their entirety when simulating excited-state reaction pathways.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2024

Efficient Mixed-Precision Matrix Factorization of the Inverse Overlap Matrix in Electronic Structure Calculations with AI-Hardware and GPUs

In recent years, a new kind of accelerated hardware has gained popularity in the artificial intelligence (AI) community which enables extremely high-performance tensor contractions in reduced precision for deep neural network calculations. In this article, we exploit Nvidia Tensor cores, a prototypical example of such AI-hardware, to develop a mixed precision approach for computing a dense matrix factorization of the inverse overlap matrix in electronic structure theory, S –1 . This factorization of S –1 , written as ZZT = S –1 , is used to transform the general matrix eigenvalue problem into a standard matrix eigenvalue problem. Here we present a mixed precision iterative refinement algorithm where Z is given recursively using matrix–matrix multiplications and can be computed with high performance on Tensor cores. To understand the performance and accuracy of Tensor cores, comparisons are made to GPU-only implementations in single and double precision. Additionally, we propose a nonparametric stopping criteria which is robust in the face of lower precision floating point operations. The algorithm is particularly useful when we have a good initial guess to Z, for example, from previous time steps in quantum-mechanical molecular dynamics simulations or from a previous iteration in a geometry optimization.

36 MATERIALS SCIENCE↗
DOE OSTI2017

Numerical Optimization of Density Functional Tight Binding Models: Application to Molecules Containing Carbon, Hydrogen, Nitrogen, and Oxygen

New parametrizations for semiempirical density functional tight binding (DFTB) theory have been developed by the numerical optimization of adjustable parameters to minimize errors in the atomization energy and interatomic forces with respect to ab initio calculated data. Initial guesses for the radial dependences of the Slater- Koster bond integrals and overlap integrals were obtained from minimum basis density functional theory calculations. The radial dependences of the pair potentials and the bond and overlap integrals were represented by simple analytic functions. The adjustable parameters in these functions were optimized by simulated annealing and steepest descent algorithms to minimize the value of an objective function that quantifies the error between the DFTB model and ab initio calculated data. The accuracy and transferability of the resulting DFTB models for the C, H, N, and O system were assessed by comparing the predicted atomization energies and equilibrium molecular geometries of small molecules that were not included in the training data from DFTB to ab initio data. The DFTB models provide accurate predictions of the properties of hydrocarbons and more complex molecules containing C, H, N, and O.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2019

Graph Theory Approach to High-Throughput Surface Adsorption Structure Generation

We present a methodology for graph based enumeration of surfaces and unique chemical adsorption structures bonded to those surfaces. Utilizing the graph produced from a bulk structure, we create a unique graph representation for any general slab cleave and further extend that representation to include a large variety of catalytically relevant adsorbed molecules. Here, we also demonstrate simple geometric procedures to generate 3D initial guesses of these enumerated structures. While generally useful for generating a wide variety of structures used in computational surface science and heterogeneous catalysis, these techniques are also key to facilitating an informatics approach to the high-throughput search for more effective catalysts.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2021

Computational Framework for Machine-Learning-Enabled 13 C Fluxomics

13 C metabolic flux analysis (MFA) has emerged as a powerful tool for synthetic biology. This optimization-based approach suffers long computation time and unstable solutions depending on the initial guess. Here, we develop a machine-learning-based framework for 13 C fluxomics. Specifically, training and test data sets are generated by metabolic network decomposition and flux sampling, in which flux ratios at metabolic nodes and simulated labeling patterns of metabolites are used as training targets and features, respectively. To improve prediction accuracy and simplify the model, automated processes are developed for flux ratio selection based on solvability and feature screening based on importance. We found that predictive performance can be significantly improved using both amino acids and central carbon metabolites in comparison with amino acids alone. Together with measured external fluxes, the predicted flux ratios determine the mass balance system, yielding global flux distributions. This approach is validated by flux estimation using both simulated and experimental data in comparison with canonical 13 C MFA. The approach represents a reliable fluxomics method readily applicable to high-throughput metabolic phenotyping, which highlights the advances of intelligent learning algorithms in synthetic biology, specifically in the Test and Learn stage of the Design-Build-Test-Learn cycle.

13C metabolic flux analysis↗
DOE OSTI2013

A Case for Soft Error Detection and Correction in Computational Chemistry

High performance computing platforms are expected to deliver 10(18) floating operations per second by the year 2022 through the deployment of millions of cores. Even if every core is highly reliable the sheer number of the them will mean that the mean time between failures will become so short that most applications runs will suffer at least one fault. In particular soft errors caused by intermittent incorrect behavior of the hardware are a concern as they lead to silent data corruption. In this paper we investigate the impact of soft errors on optimization algorithms using Hartree-Fock as a particular example. Optimization algorithms iteratively reduce the error in the initial guess to reach the intended solution. Therefore they may intuitively appear to be resilient to soft errors. Our results show that this is true for soft errors of small magnitudes but not for large errors. We suggest error detection and correction mechanisms for different classes of data structures. The results obtained with these mechanisms indicate that we can correct more than 95% of the soft errors at moderate increases in the computational cost.

Computational chemistry↗
DOE OSTI2024

Analytical ab initio hessian from a deep learning potential for transition state optimization

Identifying transition states—saddle points on the potential energy surface connecting reactant and product minima—is central to predicting kinetic barriers and understanding chemical reaction mechanisms. In this work, we train a fully differentiable equivariant neural network potential, NewtonNet, on thousands of organic reactions and derive the analytical Hessians. By reducing the computational cost by several orders of magnitude relative to the density functional theory (DFT) ab initio source, we can afford to use the learned Hessians at every step for the saddle point optimizations. We show that the full machine learned (ML) Hessian robustly finds the transition states of 240 unseen organic reactions, even when the quality of the initial guess structures are degraded, while reducing the number of optimization steps to convergence by 2–3× compared to the quasi-Newton DFT and ML methods. All data generation, NewtonNet model, and ML transition state finding methods are available in an automated workflow.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2022

Optical emissivity dataset of multi-material heterogeneous designs generated with automated figure extraction

Optical device design is typically an iterative optimization process based on a good initial guess from prior reports. Optical properties databases are useful in this process but difficult to compile because their parsing requires finding relevant papers and manually converting graphical emissivity curves to data tables. Here, we present two contributions: one is a dataset of thermal emissivity records with design-related parameters, and the other is a software tool for automated colored curve data extraction from scientific plots. We manually collected 64 papers with 176 figures reporting thermal emissivity and automatically retrieved 153 colored curve data records. The automated figure analysis software pipeline uses Faster R-CNN for axes and legend object detection, EasyOCR for axes numbering recognition, and k-means clustering for colored curve retrieval. Additionally, we manually extracted geometry, materials, and method information from the text to add necessary metadata to each emissivity curve. Finally, we analyzed the dataset to determine the dominant classes of emissivity curves and determine the underlying design parameters leading to a type of emissivity profile.

47 OTHER INSTRUMENTATION↗
DOE OSTI2019

Inertial extended-Lagrangian scheme for solving charge equilibration models

The inertial extended Lagrangian/self-consistent field scheme (iEL–SCF) has been adopted for solving charge equilibration in LAMMPS as part of the reactive force field ReaxFF, which due to the charge conservation constraint requires solving two sets of linear system of equations for the new charges at each molecular dynamics time-step. Therefore, the extended Lagrangian for charge equilibration is comprised of two auxiliary variables for the intermediate charges which serve as an initial guess for the real charges. We demonstrate that the iEL–SCF is able to reduce the number of SCF cycles by 50–80% of the original conjugate gradient self-consistent field solver as tested across diverse systems including water, ferric hydroxide, nitramine RDX, and hexanitrostilbene.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2011

A course-grained model for polyethylene glycol polymer

A coarse-grained (CG) model of polyethylene glycol (PEG) was developed and implemented in CG molecular dynamics (MD) simulations of PEG chains with degree of polymerization (DP) 20 and 40. In the model, two repeat units of PEG are grouped as one CG bead. Atomistic MD simulation of PEG chains with DP = 20 was first conducted to obtain the bonded structural probability distribution functions (PDFs) and nonbonded pair correlation function (PCF) of the CG beads. The bonded CG potentials are obtained by simple inversion of the corresponding PDFs. The CG nonbonded potential is parameterized to the PCF using both an inversion procedure based on the Ornstein-Zernike equation with the Percus-Yevick approximation (OZPY{sup -1}) and a combination of OZPY{sup -1} with the iterative Boltzmann inversion (IBI) method (OZPY{sup -1}+IBI). As a simple one step method, the OZPY{sup -1} method possesses an advantage in computational efficiency. Using the potential from OZPY{sup -1} as an initial guess, the IBI method shows fast convergence. The coarse-grained molecular dynamics (CGMD) simulations of PEG chains with DP = 20 using potentials from both methods satisfactorily reproduce the structural properties from atomistic MD simulation of the same systems. The OZPY{sup -1}+IBI method yields better agreement than the OZPY{sup -1} method alone. The new CG model and CG potentials from OZPY{sup -1}+IBI method was further tested through CGMD simulation of PEG with DP = 40 system. No significant changes are observed in the comparison of PCFs from CGMD simulations of PEG with DP = 20 and 40 systems indicating that the potential is independent of chain length.

36 MATERIALS SCIENCE↗
DOE OSTI2014

Adiabatic state preparation study of methylene

Quantum computers attract much attention as they promise to outperform their classical counterparts in solving certain type of problems. One of them with practical applications in quantum chemistry is simulation of complex quantum systems. An essential ingredient of efficient quantum simulation algorithms are initial guesses of the exact wave functions with high enough fidelity. As was proposed in Aspuru-Guzik et al. [Science 309, 1704 (2005)], the exact ground states can in principle be prepared by the adiabatic state preparation method. Here, we apply this approach to preparation of the lowest lying multireference singlet electronic state of methylene and numerically investigate preparation of this state at different molecular geometries. We then propose modifications that lead to speeding up the preparation process. Finally, we decompose the minimal adiabatic state preparation employing the direct mapping in terms of two-qubit interactions.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2015

Efficient preconditioning of the electronic structure problem in large scale ab initio molecular dynamics simulations

We present an improved preconditioning scheme for electronic structure calculations based on the orbital transformation method. First, a preconditioner is developed which includes information from the full Kohn-Sham matrix but avoids computationally demanding diagonalisation steps in its construction. This reduces the computational cost of its construction, eliminating a bottleneck in large scale simulations, while maintaining rapid convergence. In addition, a modified form of Hotelling’s iterative inversion is introduced to replace the exact inversion of the preconditioner matrix. This method is highly effective during molecular dynamics (MD), as the solution obtained in earlier MD steps is a suitable initial guess. Filtering small elements during sparse matrix multiplication leads to linear scaling inversion, while retaining robustness, already for relatively small systems. For system sizes ranging from a few hundred to a few thousand atoms, which are typical for many practical applications, the improvements to the algorithm lead to a 2-5 fold speedup per MD step.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2015

An efficient and stable hybrid extended Lagrangian/self-consistent field scheme for solving classical mutual induction

We have adapted a hybrid extended Lagrangian self-consistent field (EL/SCF) approach, developed for time reversible Born Oppenheimer molecular dynamics for quantum electronic degrees of freedom, to the problem of classical polarization. In this context, the initial guess for the mutual induction calculation is treated by auxiliary induced dipole variables evolved via a time-reversible velocity Verlet scheme. However, we find numerical instability, which is manifested as an accumulation in the auxiliary velocity variables, that in turn results in an unacceptable increase in the number of SCF cycles to meet even loose convergence tolerances for the real induced dipoles over the course of a 1 ns trajectory of the AMOEBA14 water model. By diagnosing the numerical instability as a problem of resonances that corrupt the dynamics, we introduce a simple thermostating scheme, illustrated using Berendsen weak coupling and Nose-Hoover chain thermostats, applied to the auxiliary dipole velocities. We find that the inertial EL/SCF (iEL/SCF) method provides superior energy conservation with less stringent convergence thresholds and a correspondingly small number of SCF cycles, to reproduce all properties of the polarization model in the NVT and NVE ensembles accurately. Our iEL/SCF approach is a clear improvement over standard SCF approaches to classical mutual induction calculations and would be worth investigating for application to ab initio molecular dynamics as well.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2015

Self-consistent one dimension in space and three dimension in velocity kinetic trajectory simulation model of magnetized plasma-wall transition

We have developed a self-consistent 1d3v (one dimension in space and three dimension in velocity) Kinetic Trajectory Simulation (KTS) model, which can be used for modeling various situations of interest and yields results of high accuracy. Exact ion trajectories are followed, to calculate along them the ion distribution function, assuming an arbitrary injection ion distribution. The electrons, on the other hand, are assumed to have a cut-off Maxwellian velocity distribution at injection and their density distribution is obtained analytically. Starting from an initial guess, the potential profile is iterated towards the final time-independent self-consistent state. We have used it to study plasma sheath region formed in presence of an oblique magnetic field. Our results agree well with previous works from other models, and hence, we expect our 1d3v KTS model to provide a basis for the studying of all types of magnetized plasmas, yielding more accurate results.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗
DOE OSTI2016

Sparse maps—A systematic infrastructure for reduced-scaling electronic structure methods. II. Linear scaling domain based pair natural orbital coupled cluster theory

Domain based local pair natural orbital coupled cluster theory with single-, double-, and perturbative triple excitations (DLPNO-CCSD(T)) is a highly efficient local correlation method. It is known to be accurate and robust and can be used in a black box fashion in order to obtain coupled cluster quality total energies for large molecules with several hundred atoms. While previous implementations showed near linear scaling up to a few hundred atoms, several nonlinear scaling steps limited the applicability of the method for very large systems. In this work, these limitations are overcome and a linear scaling DLPNO-CCSD(T) method for closed shell systems is reported. The new implementation is based on the concept of sparse maps that was introduced in Part I of this series [P. Pinski, C. Riplinger, E. F. Valeev, and F. Neese, J. Chem. Phys. 143, 034108 (2015)]. Using the sparse map infrastructure, all essential computational steps (integral transformation and storage, initial guess, pair natural orbital construction, amplitude iterations, triples correction) are achieved in a linear scaling fashion. In addition, a number of additional algorithmic improvements are reported that lead to significant speedups of the method. The new, linear-scaling DLPNO-CCSD(T) implementation typically is 7 times faster than the previous implementation and consumes 4 times less disk space for large three-dimensional systems. For linear systems, the performance gains and memory savings are substantially larger. Calculations with more than 20 000 basis functions and 1000 atoms are reported in this work. In all cases, the time required for the coupled cluster step is comparable to or lower than for the preceding Hartree-Fock calculation, even if this is carried out with the efficient resolution-of-the-identity and chain-of-spheres approximations. The new implementation even reduces the error in absolute correlation energies by about a factor of two, compared to the already accurate previous implementation.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗