Conventional computational costs

The maintenance of a connection to experiment is essential in that reliability is only measurable against experimental results. However, in practice, the computational cost of the most reliable conventional quantum chemical methods has tended to preclude their application to the large, low-symmetry molecules which form liquid crystals. There have however, been several recent steps forward in this area and here we will review some of these newest developments in predictive computer simulation of intramolecular properties of liquid crystals. In the next section we begin with a brief overview of important molecular properties which are the focus of much current computational effort and highlight some specific examples of cases where the molecular electronic origin of macroscopic properties is well established. 

One of the most dependably accurate methods for deriving 95% confidence intervals for cost-effectiveness ratios is the nonparametric bootstrap method. In this method, one resamples from the smdy sample and computes cost-effectiveness ratios in each of the multiple samples. To do so requires one to (1) draw a sample of size n with replacement from the empiric distribution and use it to compute a cost-effectiveness ratio (2) repeat this sampling and calculation of the ratio (by convention, at least 1000 times for confidence intervals) (3) order the repeated estimates of the ratio from lowest (best) to highest (worst) and (4) identify a 95% confidence interval from this rank-ordered distribution. The percentile method is one of the simplest means of identifying a confidence interval, but it may not be as accurate as other methods. When using 1,000... 

Now, in the TF-HK equation, all the potential terms can be set up by conventional plane-wave-basis teehniques with essentially linear scaling. However, for very large systems with more than 5000 nuclei, the computational cost associated with the nuelear-nuelear Coulomb repulsion energy becomes the major bottleneck." In this case, linear-scaling Ewald sirmmation techniques should be utilized. 

Because of the enormous amount of integration required, computation costs on conventional computers would have been prohibitive and thus, all of our computations were performed on the Cray 1 and Cray 2 supercomputers at the University of Minnesota. The results are shown in figure 2 where the forcing frequency w has been scaled by the system s natural frequency co0... 

The purpose of ONIOM is to obtain a better balance between accuracy and computational cost. We illustrate this using the 3Wa cluster. The ONIOM calculation with TD-DFT/medium in the low level takes half the computing time as the target, and has excellent results (error compared to target less than 0.05 eV). ONIOM with TD-HF/medium in the low level takes about a tenth of the CPU time relative to the target. The error with this low level is about 0.2 eV, which is much smaller than the error of 1.2 eV in a conventional TD-UF/medium calculation. 

The use of the periodic boundary conditions in the two directions perpendicular to the interface normal (X and Y) implies that the system has infinite extent in these directions. To make the computational cost reasonable, one must truncate the number of interactions that each molecule experiences. The simplest possible technique is to include, for each molecule i, the interaction with all the other molecules that are within a sphere of radius which is smaller than half the shortest box axis. One selects, from among the infinite possible images of each molecule, the one that is the closest to the molecule i under consideration. This is called the minimum image convention, and more details about its implementation can be found elsewhere [2]. To arrive at the correct bulk properties, any ensemble average calculated by this technique must be corrected for the contribution of the interactions beyond the cutoff distance. The fixed analytical corrections are calculated by assuming some simple form of the statistical mechanics distribution function for distances greater then R. ... 

Density functional theory (DFT) [62] incorporates electron correlation at a very small computational cost, but its suitability for weak interaction still seems a somewhat open question. There have been comparisons of DFT and conventional methods [63 68]. Mostly, an improvement over SCF level treatment seems possible, but there is a clear dependence on the choice of functional and on basis-set size. Also, DFT may be more sensitive to BSSE in smaller basis sets than conventional treatments [64]. A single functional choice for spectroscopic accuracy in treating weakly bound clusters does not yet seem at hand, but that alone does not preclude application of DFT for lower levels of accuracy. With the computational cost advantage of DFT, the capability exists for treating large, extended clusters. 

The solute molecules can, in principle, be treated at any level of QM theory. However, in the majority of QM/MM studies of biologically important systems, C/qM is computed using one of the approximate semiempirical AMI, MNDO, and PM3 methods. The reason for this predominance of semiempirical methods is due solely to the computational cost of conventional ab initio or density functional methods. In fact, semiempirical methods are efficient enough to be used in MD simulations. In the following, we describe the most recent and significant advancements in the development of solvation models based on both semiempirical and ab initio QM/MM methods. 

---