The \(quild\) program has been developed by M. Swart and F. M. Bickelhaupt for enabling calculations through multi-level approaches, in which different computational treatments are used for different regions of the system under study.

Improved optimization routines

The program uses delocalized coordinates, originally developed by Jon Baker et al., and adapted for weak/strong coordinate separation based on ideas by Roland Lindh et al., as described in 10.1002/qua.21049.

Relevant papers

10.1002/jcc.20834
10.1002/qua.21049
10.1002/qua.10463

QUILD uses a model Hessian (see 10.1002/jcc.20834) for TS searches, based on Transition State Reaction Coordinates (TSRC), which the user has to define on input. Typically these involve bonds being broken/formed. The TSRCs are assigned negative force constants, while all other internal coordinates (bonds, angles, dihedrals) are assigned positive ones. Moreover, the TSRCs force constants are coupled to each other within the delocalized coordinates setup, to give one (and only one) negative Hessian eigenvalue at the start.

   

The division of a system under study in a quantum mechanical (QM) and a classical system in QM/MM molecular mechanical calculations is sometimes very natural, but a problem arises in the case of bonds crossing the QM/MM boundary. The AddRemove model uses a capping (link) atom to satisfy the valences of the quantum chemical system is presented, with the position of the capping atom depending on the positions of the real atoms involved in the link bond. Using this method no degrees of freedom for the capping atom are introduced. Moreover, the introduction of this artificial atom is corrected for by subtracting the classical MM interactions with the real QM system it would have if it were a classical atom. That is, the capping atoms are added and removed.

In the case of a DNA dimer, consisting of the base pair and nucleotide, the total energy can be divided into two parts: one for the whole system, and one specifically for π-π stacking energy. For the latter, DFT is known to have problems, although LDA with a large basis set actually gives excellent results. Hence, we can use e.g. BP86 (blue) for the whole system, and then cut out its π-π stacking energy and replace it with LDA (yellow, pink). The QUILD program is set up as flexible as possible and allows such a division of the system.

Also a protein can be simulated with QUILD, in which shells of atoms are treated with different quantum-chemistry or molecular mechanics levels. In this case the whole system could be treated with Amber forcefields (blue), the active site by DFT (yellow) and the active site with some surrounding residues by DFTB or composite-DFT methods (pink).