Computational methods intermediate approaches

When investigating mechanistic proposals, one has to perform a large number of calculations to find different transition states and intermediates to test the various scenarios. Consequently, the computational scheme used has to be fast and robust enough to allow this. At the same time, the accuracy of the approximations made in the models has to be higher than or comparable to the accuracy of the underlying computational method. One very fruitful approach has been to cut out a relatively small model of enzyme around the active site and treat it at a quite high level of theory. The effects of the parts of the enzyme that are not included in the quantum model are modeled using different approximations. 

Hamiltonian is not known and, as for the nonrelativistic case, further approximations have to be introduced in the wavefunction, it is tempting to derive approximate computational schemes which are still sufficiently accurate but more efficient. Here we will only summarize those approximate methods that have been used frequently to obtain information about the electronic structure of molecules with lanthanide atoms, i.e. relativistically corrected density-functional approach, pseudopotential method, intermediate neglect of differential overlap method, extended Huckel theory, and ligand field theory. 

This problem formulation takes the model and the side conditions, represented by equalities and inequalities as explicit—typically vector valued— constraints, and is called all-at-once approach. Notice again that such an approach is typically not possible in the context of direct methods and meta-heuristics, as the side conditions are hard to implement in definition of populations. Moreover, this format is rather general also from the point of view of numerical simulations. Indeed, only once we approach the optimum in an iterative method— yet to be described—we need to satisfy the model equation and the constraints at the optimum only, whereas if we use a code to compute the state y t) for any feasible t, we need to implement the constraint into the code, leading to the black-box optimization problem discussed above, or proceed with an intermediate approach, where we provide a solution operator S such that S(u) solves the model problem for a given control u. Then we obtain what has come to be known as reduced optimization problem ... 

There are two possible methods of approach. Either the space integral rates Siu and the associated derivative terms a are evaluated at the beginning of the time interval, and the dependent variables are updated together at the end of the interval or the appropriate 5, and Sjy a are evaluated immediately prior to the solution of each equation and each dependent variable is updated immediately after the solution of its equation. When using the latter method, the order of solution of the equations becomes important. The most satisfactory order treats the most reactive free radicals first, then less reactive intermediates, initial reactants, reaction products and inert species in that order, and finally the enthalpy. The latter method is potentially the more economical computationally, but it forfeits precise chemical conservation during single time steps on the approach to the steady state. This has been found to cause instability in certain diffusion flame problems to be outlined in Section 7.2(b). The former method does not suffer from this disadvantage. 

Since 5 is a function of all the intermediate coordinates, a large scale optimization problem is to be expected. For illustration purposes consider a molecular system of 100 degrees of freedom. To account for 1000 time points we need to optimize 5 as a function of 100,000 independent variables ( ). As a result, the use of a large time step is not only a computational benefit but is also a necessity for the proposed approach. The use of a small time step to obtain a trajectory with accuracy comparable to that of Molecular Dynamics is not practical for systems with more than a few degrees of freedom. Fbr small time steps, ordinary solution of classical trajectories is the method of choice. 

---