Yu Zhuang

Direct chemical dynamics simulations: coupling of classical and quasiclassical trajectories with electronic structure theory

In classical and quasiclassical trajectory chemical dynamics simulations, the atomistic dynamics of collisions, chemical reactions, and energy transfer are studied by solving the classical equations of motion. These equations require the potential energy and its gradient for the chemical system under study, and they may be obtained directly from an electronic structure theory. This article reviews such direct dynamics simulations. The accuracy of classical chemical dynamics is considered, with simulations highlighted for the F− + CH3OOH reaction and of energy transfer in collisions of CO2 with a perfluorinated self‐assembled monolayer (F‐SAM) surface. Procedures for interfacing chemical dynamics and electronic structure theory computer codes are discussed. A Hessian‐based predictor–corrector algorithm and high‐accuracy Hessian updating algorithm, for enhancing the efficiency of direct dynamics simulations, are described. In these simulations, an ensemble of trajectories is calculated which represents the experimental and chemical system under study. Algorithms are described for selecting the appropriate initial conditions for bimolecular and unimolecular reactions, gas‐surface collisions, and initializing trajectories at transition states and conical intersections. Illustrative direct dynamics simulations are presented for the Cl− + CH3I SN2 reaction, unimolecular decomposition of the epoxy resin constituent CH3NHCHCHCH3 versus temperature, collisions and reactions of N‐protonated diglycine with a F‐SAM surface that has a reactive head group, and the product energy partitioning for the post‐transition state dynamics of C2H5F → HF + C2H4 dissociation.

Direct dynamics simulations using Hessian-based predictor-corrector integration algorithms

In previous research [J. Chem. Phys. 111, 3800 (1999)] a Hessian-based integration algorithm was derived for performing direct dynamics simulations. In the work presented here, improvements to this algorithm are described. The algorithm has a predictor step based on a local second-order Taylor expansion of the potential in Cartesian coordinates, within a trust radius, and a fifth-order correction to this predicted trajectory. The current algorithm determines the predicted trajectory in Cartesian coordinates, instead of the instantaneous normal mode coordinates used previously, to ensure angular momentum conservation. For the previous algorithm the corrected step was evaluated in rotated Cartesian coordinates. Since the local potential expanded in Cartesian coordinates is not invariant to rotation, the constants of motion are not necessarily conserved during the corrector step. An approximate correction to this shortcoming was made by projecting translation and rotation out of the rotated coordinates. For the current algorithm unrotated Cartesian coordinates are used for the corrected step to assure the constants of motion are conserved. An algorithm is proposed for updating the trust radius to enhance the accuracy and efficiency of the numerical integration. This modified Hessian-based integration algorithm, with its new components, has been implemented into the VENUS/NWChem software package and compared with the velocity-Verlet algorithm for the H(2)CO-->H(2)+CO, O(3)+C(3)H(6), and F(-)+CH(3)OOH chemical reactions.

Higher-accuracy schemes for approximating the Hessian from electronic structure calculations in chemical dynamics simulations

In this paper, we present a family of generally applicable schemes for updating the Hessian from electronic structure calculations based on an equation derived with compact finite difference (CFD). The CFD-based equation is of higher accuracy than the quasi-Newton equation on which existing generally applicable Hessian update schemes are based. Direct tests of Hessian update schemes, as well as dynamics simulations using an integrator incorporating Hessian update schemes, have shown four of the new schemes produce reliably higher accuracy than existing Hessian update schemes.

Evaluating the Accuracy of Hessian Approximations for Direct Dynamics Simulations.

Direct dynamics simulations are a very useful and general approach for studying the atomistic properties of complex chemical systems, since an electronic structure theory representation of a systems potential energy surface is possible without the need for fitting an analytic potential energy function. In this paper, recently introduced compact finite difference (CFD) schemes for approximating the Hessian [J. Chem. Phys.2010, 133, 074101] are tested by employing the monodromy matrix equations of motion. Several systems, including carbon dioxide and benzene, are simulated, using both analytic potential energy surfaces and on-the-fly direct dynamics. The results show, depending on the molecular system, that electronic structure theory Hessian direct dynamics can be accelerated up to 2 orders of magnitude. The CFD approximation is found to be robust enough to deal with chaotic motion, concomitant with floppy and stiff mode dynamics, Fermi resonances, and other kinds of molecular couplings. Finally, the CFD approximations allow parametrical tuning of different CFD parameters to attain the best possible accuracy for different molecular systems. Thus, a direct dynamics simulation requiring the Hessian at every integration step may be replaced with an approximate Hessian updating by tuning the appropriate accuracy.

Memory-conscious collective I/O for extreme scale HPC systems

The continuing decrease in memory capacity per core and the increasing disparity between core count and off-chip memory bandwidth create significant challenges for I/O operations in exascale systems. The exascale challenges require rethinking collective I/O for the effective exploitation of the correlation among I/O accesses in the exascale system. In this study we introduce a Memory-Conscious Collective I/O considering the constraint of the memory space. 1)Restricts aggregation data traffic within disjointed subgroups 2)Coordinates I/O accesses in intra-node and inter-node layer 3)Determines I/O aggregators at run time considering data distribution and memory consumption among processes.

A New Data Sieving Approach for High Performance I/O

Many scientific computing applications and engineering simulations exhibit noncontiguous I/O access patterns. Data sieving is an important technique to improve the performance of noncontiguous I/O accesses by combining small and noncontiguous requests into a large and contiguous request. It has been proven effective even though more data is potentially accessed than demanded. In this study, we propose a new data sieving approach namely Performance Model Directed Data Sieving, or PMD data sieving in short. It improves the existing data sieving approach from two aspects: (1) dynamically determines when it is beneficial to perform data sieving; and (2) dynamically determines how to perform data sieving if beneficial. It improves the performance of the existing data sieving approach and reduces the memory consumption as verified by experimental results. Given the importance of supporting noncontiguous accesses effectively and reducing the memory pressure in a large-scale system, the proposed PMD data sieving approach in this research holds a promise and will have an impact on high performance I/O systems.

A Limited-Iteration Bisecting K-Means for Fast Clustering Large Datasets

Bisecting K-means (BKM) clustering, with or without refinement, has been shown to exhibit higher computing efficiency, better clustering quality, and low susceptibility to initial cluster centers, when compared with the basic K-means clustering algorithm. For bisecting K-means with refinement, in this paper, we investigate a variant that increases the efficiency while trying to maintain clustering quality. Our approach is to limit the number of iterations of the two-means (the K-means with K=2) in bisecting a data subset. We experimented with one, two, and three iterations for the two-means, and compared them with the original BKMs unlimited iterations which end when two clusters no longer change in the two-means. We carried out experimental studies on three datasets and found that three and unlimited iterations for the two-means produced almost the same clustering qualities on all test cases, leading us to think that three iterations might be adequate. The experimental data also show that the limited-iteration BKM with three iterations led to higher computing efficiency when compared with the BKM, suggesting that limiting the iterations in bisecting K-means has the potential of achieving higher efficiency while maintaining clustering quality.

Performance model-directed data sieving for high-performance I/O

Many scientific computing applications and engineering simulations exhibit noncontiguous I/O access patterns. Data sieving is an important technique to improve the performance of noncontiguous I/O accesses by combining small and noncontiguous requests into a large and contiguous request. It has been proven effective even though more data are potentially accessed than demanded. In this study, we propose a new data sieving approach namely performance model-directed data sieving, or PMD data sieving in short. It improves the existing data sieving approach from two aspects: (1) dynamically determines when it is beneficial to perform data sieving; and (2) dynamically determines how to perform data sieving if beneficial. It improves the performance of the existing data sieving approach considerably and reduces the memory consumption as verified by both theoretical analysis and experimental results. Given the importance of supporting noncontiguous accesses effectively and reducing the memory pressure in a large-scale system, the proposed PMD data sieving approach in this research holds a great promise and will have an impact on high-performance I/O systems.

A Load-Balancing Force Decomposition Scheme for Parallel Simulation of Chemical Dynamics with Multiple Inter-atomic Force Models

Force evaluation is the most computationally intensive part in a chemical dynamics simulation, and hence most parallel simulation algorithms choose the force calculation as the main target for parallelization. The majority of existing parallel algorithms assume a uniform force-evaluation cost for all atom pairs. For dynamics with considerable bonded interactions, different evaluation formulas are usually used for forces between different atom pairs, and this complicates the load balancing for the simulation of chemical dynamics on parallel computers. In this paper, we present a load-balancing scheme that takes into account differences of inter-atomic force models for different atom pairs. By considering different force models, the load partitioning of our algorithm can effectively handle the differences in computation costs for calculating different inter-atomic interactions when atom-tailored force models are used for different atom pairs, which is usually the case for bonded interactions. A parallel simulation algorithm for bonded-interaction-dominated dynamics was developed that employs the load partitioning scheme, and the algorithm was implemented and tested on different ensembles of atoms, and produced good performances for the testing problems.

Abstract: Memory-Conscious Collective I/O for Extreme-Scale HPC Systems

The continuing decrease in memory capacity per core and the increasing disparity between core count and off-chip memory bandwidth create significant challenges for I/O operations in exascale systems. The exascale challenges require rethinking collective I/O for the effective exploitation of the correlation among I/O accesses in the exascale system. In this study, considering the major constraint of the memory space, we introduce a MemoryConscious collective I/O. Given the importance of I/O aggregator in improving the performance of collective I/O, the new collective I/O strategy restricts aggregation data traffic within disjointed subgroups, coordinates I/O accesses in intra-node and inter-node layer and determines I/O aggregators at run time considering data distribution and memory consumption among processes. The preliminary results have demonstrated that the new collective I/O strategy holds promise in substantially reducing the amount of memory pressure, alleviating contention for memory bandwidth and improving the I/O performance for extreme-scale systems.