The Effectiveness of Newton's Method for ImprovingAb InitioForce Fields with Applications to CO2and H2CO
β Scribed by Kristy L. Mardis; Edwin L. Sibert III
- Publisher
- Elsevier Science
- Year
- 1998
- Tongue
- English
- Weight
- 350 KB
- Volume
- 187
- Category
- Article
- ISSN
- 0022-2852
No coin nor oath required. For personal study only.
β¦ Synopsis
Ab initio force field parameters are refined using modifications to Newton's method in which a figure of merit function is approximated as a Taylor Series truncated at second order. We investigate two versions of Newton's method: the Discrete Newton method where the Hessian is approximated by finite differences and the Gauss-Newton method where the Hessian is approximated as the product of first derivatives. The applicability of both rests on the capability of current ab initio methods to calculate quartic force fields that accurately reproduce experimental observables such as band centers and inertial constants. As examples, we calculate refined potential energy surfaces for CO 2 and H 2 CO. We show that an algorithm depending solely on Newton's methods requires only a small number of iterations. Additionally, Newton-based methods provide a great deal of information about the sensitivity of the force field parameters to the observables being fit. We also demonstrate that the less computationally demanding Gauss-Newton method gives results similar to the less approximate Discrete Newton method.
π SIMILAR VOLUMES
The rdt, cocii~c~cnts and product eon dlsrrlbutlons for rhc reactions ot OH\* and tlzO+ u ith N2\_ 02 h'0 X20 Xc, CO, CO? H2S Jnd 112 hd\c been dctermlned it 194 + 2 K usmg a selcctcd ton ilo\. rubc (SIFT) dppclrnrus These reactIons urre generally found to proLccd b\~ athcr proton or charge trsnsfcr
The implementation of the reduced gradient following (RGF) method into the COLUMBUS quantum-chemical program system is reported using the newly developed analytic MR-CISD/AQCC gradient feature. By this combination a very useful tool has been developed for general searches of stationary points on gro