MOLECULAR TAILORING APPROACH: TOWARDS PC-BASED AB INITIO TREATMENT OF LARGE MOLECULES

Abstract

The development of a fragmentation-based scheme, viz. molecular tailoring approach (MTA) for ab initio computation of one-electron properties and geometry optimization is described. One-electron properties such as the molecular electrostatic potential (MESP), molecular electron density (MED), and dipole moments are computed by synthesizing the density matrix (DM) of the parent molecule from DMs of its small overlapping fragments. The electron density obtained via MTA was found to be typically within 0.5% of its actual counterpart, while maximum errors of about 2% were noticed in the case of the dipole moment and MESP distribution. An attempt is made to develop MTA-based geometry optimization that involves picking relevant energy gradients from fragment self-consistent field (SCF) calculations, bypassing the CPU and memory extensive SCF step of the complete molecule. This is based on the observation that the MTA gradients mimic the actual ones fairly well. As the calculations on individual fragments are mutually independent, this algorithm is amenable to large-scale parallelization and has been extended to a distributed setup of PCs. The code developed is put to test on γ-cyclodextrin, taxol, and a small albumin-binding protein (1prb) for one-electron properties. Further, molecules such as γ-cyclodextrin, taxol, a silicalite, and 1prb are subjected to MTA-based geometry optimization, on a PC cluster. The results indicate a favorable speedup of two to three times over the actual computations in the initial phase of optimization. Furthermore, it enables computations otherwise not possible on a PC. Preliminary results indicate similar savings with sustained accuracy even for large molecules at the level of Møller–Plesset second order perturbation (MP2) theory.

Molecular tailoring approach in conjunction with MP2 and Ri‐MP2 codes: A comparison with fragment molecular orbital method

“…The MTA code is built inside a locally modified version of GAMESS (hereafter referred to as MTA-enabled GAMESS), which has so far been well-tested for single point energy and gradient calculation, molecular geometry optimization as well as Hessian evaluation. [34][35][36] The fragmentation process is automated and the routines for that purpose are separately written, while the rest of the part is integrated in the standard GAMESS source files in the local version. The stepwise algorithm of MTA is outlined in Figure 1.…”

“…In step 4, actual execution of fragment jobs is performed at the desired level of theory. Further, outputs of fragment jobs are collected and the results of the fragment energies are patched together employing the cardinality-based expression for energy 34,35 …”

“…The process of fragmentation is generally automatic, although it can be supplemented or complemented by employing a general package MeTAStudio 37 developed for this purpose by Ganesh. For a quantitative definition of the quality of fragments, a parameter called R-goodness (R g ) is introduced. 34,35 The R g of atom i in fragment f i is taken as the maximum radius of a sphere centered on i such that all the atoms within this sphere are also included in f i . The goodness of a fragmentation scheme is defined as the minimum of all the atomic R g values.…”

“…[32][33][34][35][36] Details of MTA, determination of R g value for atoms as well as for the scheme, correlation of R g with accuracy of the scheme, cardinality based equations for patching fragment-energies, etc., are discussed in detail in references. [34][35] After fragmentation, cardinality based expressions 34,35 for energy and gradients are set up. These will be briefly discussed in the following Section.…”

WebProp: Web interface forab initiocalculation of molecular one‐electron properties

“…For molecules, which have [50 atoms, WebProp defaults to using a linear-scaling methodology termed Molecular Tailoring Approach (MTA) [11][12][13][14] or Cardinality Guided MTA (CG-MTA) 13 for obtaining the DM, which is then used for evaluating the oneelectron properties. The current implementation of MTA on WebProp restricts the maximum system size to 200 atoms.…”