Limits...
Finding Chemical Reaction Paths with a Multilevel Preconditioning Protocol.

Kale S, Sode O, Weare J, Dinner AR - J Chem Theory Comput (2014)

Bottom Line: Chem.Phys. 2014, 140, 184114) can be used to accelerate quantum-chemical string calculations.The approach also shows promise for free energy calculations when thermal noise can be controlled.

View Article: PubMed Central - PubMed

Affiliation: Department of Chemistry, James Franck Institute, Institute for Biophysical Dynamics, Computation Institute, Department of Statistics, University of Chicago , Chicago, Illinois 60637, United States ; Department of Chemistry, James Franck Institute, Institute for Biophysical Dynamics, Computation Institute, Department of Statistics, University of Chicago , Chicago, Illinois 60637, United States.

ABSTRACT

Finding transition paths for chemical reactions can be computationally costly owing to the level of quantum-chemical theory needed for accuracy. Here, we show that a multilevel preconditioning scheme that was recently introduced (Tempkin et al. J. Chem. Phys. 2014, 140, 184114) can be used to accelerate quantum-chemical string calculations. We demonstrate the method by finding minimum-energy paths for two well-characterized reactions: tautomerization of malonaldehyde and Claissen rearrangement of chorismate to prephanate. For these reactions, we show that preconditioning density functional theory (DFT) with a semiempirical method reduces the computational cost for reaching a converged path that is an optimum under DFT by several fold. The approach also shows promise for free energy calculations when thermal noise can be controlled.

No MeSH data available.


Related in: MedlinePlus

Convergence of ML path refinement of the Claissenreaction. Convergenceis expressed in terms of the norm of the displacement vector as projectedover the CVs and averaged over all images. Abscissa is number of Riterations. The inset shows a version of the BLYP ML curve in whichthe number of iterations is scaled by the number of energy evaluationsper outer loop iteration (6), which provides a comparison in termsof rough computational cost.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4263463&req=5

fig9: Convergence of ML path refinement of the Claissenreaction. Convergenceis expressed in terms of the norm of the displacement vector as projectedover the CVs and averaged over all images. Abscissa is number of Riterations. The inset shows a version of the BLYP ML curve in whichthe number of iterations is scaled by the number of energy evaluationsper outer loop iteration (6), which provides a comparison in termsof rough computational cost.

Mentions: The rate of convergence, as measured by the norm of thedisplacementvector averaged over all images, is shown in Figure 9. In terms of the number of reference model evaluations (outerloop iterations), we see that preconditioning with BLYP leads to agreater acceleration than preconditioning with PM3. For BLYP, an innerloop iteration costs about as much as the reference model’souter loop iteration. Because there are 5 inner loop iterations plusone energy evaluation for the outer loop, we multiply the ML iterationsby 6 in the inset to Figure 9 to obtain a roughestimate of relative computational costs. This shows that the speedupcomes from shifting the work into the inner loop.


Finding Chemical Reaction Paths with a Multilevel Preconditioning Protocol.

Kale S, Sode O, Weare J, Dinner AR - J Chem Theory Comput (2014)

Convergence of ML path refinement of the Claissenreaction. Convergenceis expressed in terms of the norm of the displacement vector as projectedover the CVs and averaged over all images. Abscissa is number of Riterations. The inset shows a version of the BLYP ML curve in whichthe number of iterations is scaled by the number of energy evaluationsper outer loop iteration (6), which provides a comparison in termsof rough computational cost.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4263463&req=5

fig9: Convergence of ML path refinement of the Claissenreaction. Convergenceis expressed in terms of the norm of the displacement vector as projectedover the CVs and averaged over all images. Abscissa is number of Riterations. The inset shows a version of the BLYP ML curve in whichthe number of iterations is scaled by the number of energy evaluationsper outer loop iteration (6), which provides a comparison in termsof rough computational cost.
Mentions: The rate of convergence, as measured by the norm of thedisplacementvector averaged over all images, is shown in Figure 9. In terms of the number of reference model evaluations (outerloop iterations), we see that preconditioning with BLYP leads to agreater acceleration than preconditioning with PM3. For BLYP, an innerloop iteration costs about as much as the reference model’souter loop iteration. Because there are 5 inner loop iterations plusone energy evaluation for the outer loop, we multiply the ML iterationsby 6 in the inset to Figure 9 to obtain a roughestimate of relative computational costs. This shows that the speedupcomes from shifting the work into the inner loop.

Bottom Line: Chem.Phys. 2014, 140, 184114) can be used to accelerate quantum-chemical string calculations.The approach also shows promise for free energy calculations when thermal noise can be controlled.

View Article: PubMed Central - PubMed

Affiliation: Department of Chemistry, James Franck Institute, Institute for Biophysical Dynamics, Computation Institute, Department of Statistics, University of Chicago , Chicago, Illinois 60637, United States ; Department of Chemistry, James Franck Institute, Institute for Biophysical Dynamics, Computation Institute, Department of Statistics, University of Chicago , Chicago, Illinois 60637, United States.

ABSTRACT

Finding transition paths for chemical reactions can be computationally costly owing to the level of quantum-chemical theory needed for accuracy. Here, we show that a multilevel preconditioning scheme that was recently introduced (Tempkin et al. J. Chem. Phys. 2014, 140, 184114) can be used to accelerate quantum-chemical string calculations. We demonstrate the method by finding minimum-energy paths for two well-characterized reactions: tautomerization of malonaldehyde and Claissen rearrangement of chorismate to prephanate. For these reactions, we show that preconditioning density functional theory (DFT) with a semiempirical method reduces the computational cost for reaching a converged path that is an optimum under DFT by several fold. The approach also shows promise for free energy calculations when thermal noise can be controlled.

No MeSH data available.


Related in: MedlinePlus