05.06.2013 Views

PNNL-13501 - Pacific Northwest National Laboratory

PNNL-13501 - Pacific Northwest National Laboratory

PNNL-13501 - Pacific Northwest National Laboratory

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

Study Control Number: PN98013/1259<br />

Combustion Simulation and Modeling<br />

Jeffrey A. Nichols, Robert J. Harrison<br />

High-performance calculations require efficient computer algorithms to achieve efficient modeling of the<br />

thermodynamics and kinetics of gasoline and diesel fuel combustion. Combustion simulation and modeling have<br />

important practical applications for predicting reaction rates and products.<br />

Project Description<br />

The purpose of this project is to develop highperformance<br />

computational chemistry software for<br />

massively parallel processor systems in the area of<br />

coupled cluster (CCSD(T)) methods for treating the<br />

electron correlation problem. Coupled cluster methods<br />

are needed to achieve the accuracies required to replace<br />

experimental measurements for many areas of chemistry,<br />

but especially for the combustion simulation and<br />

modeling community. Unfortunately, the coupled cluster<br />

method scales as N 7 , where N is the number of atoms in<br />

the system. This greatly limits the range of applicability<br />

of this method to combustion and other problems. The<br />

project focuses on developing methodological and<br />

computational approaches to bring about a dramatic<br />

reduction in the computational cost of coupled cluster<br />

calculations. This entails algorithm redesign with the use<br />

of new methods in order to achieve the reduced scaling<br />

and increased massively parallel processor efficiencies<br />

needed to accurately model the thermodynamics and<br />

kinetics of large molecular systems such as those found in<br />

gasoline (C8) and diesel fuel (C16) combustion.<br />

This work is accomplished within the framework of<br />

NWChem (Kendall et al. 2000), a computational<br />

chemistry suite that is part of the Molecular Sciences<br />

Computing Facility in the Environmental Molecular<br />

Sciences <strong>Laboratory</strong>. In addition, POLYRATE (a<br />

chemical dynamics package developed at the University<br />

of Minnesota, in collaboration with <strong>PNNL</strong> staff, and used<br />

to predict rate constants) (Steckler et al. 1995; Corchado<br />

et al. 1998) will be integrated with NWChem in order to<br />

provide improved means for the calculation of chemical<br />

reaction rate constants.<br />

Results and Accomplishments<br />

The integration of POLYRATE with NWChem to<br />

perform direct dynamics has been completed and will be<br />

widely distributed in NWChem version 4.0, which is due<br />

for release in fall 2000. It is already available for<br />

<strong>Laboratory</strong> internal use and is being used by the<br />

atmospheric oxidation study.<br />

Alternative representations of the Møller-Plesset<br />

perturbation theory and coupled cluster wave functions<br />

have been explored with some limited success. Our goals<br />

were to develop a framework to improve the scaling of<br />

high-accuracy calculations in large atomic orbital basis<br />

sets by using a low-rank representation of both the wave<br />

function (coupled-cluster amplitudes) and integrals.<br />

Our investigation into the Laplace-factorized algorithm<br />

for the 4 th -order linear triples contribution is continuing in<br />

collaboration with Professor J.H. van Lenthe, from U.<br />

Utrecht, who visited this summer.<br />

Most of the effort this fiscal year has been devoted to the<br />

new production code for CCSD(T). Several prototypes<br />

were finished earlier this year, with the objectives of<br />

validating the equations and exploring various methods.<br />

Significant highlights of the new CCSD(T) code include<br />

the following:<br />

• Hartree-Focks RHF/ROHF and UHF reference<br />

functions.<br />

• Full use of Abelian point-group symmetry.<br />

• Design goals set by considering calculations on<br />

molecules using cc-pVnZ (n=2-6) basis sets (also<br />

with augmentation) that could be executed in one<br />

week on a 1 TFLOP/s computer with 1 gigabyte of<br />

Computational Science and Engineering 109

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!