Since these are being implemented in ABINIT or CP2K for materials science problems, I would like to know the possibility of using the same with GAMESS.
Knowing almost nothing about GAMESS I can still safely say that it is possible. The question is to what extent your end product will still resemble GAMESS... Here follows a general assessment based on a complete lack of specific knowledge on GAMESS and a substantial (although I say it myself) knowledge of electronic structure codes in general.
An electronic structure program basically does three things
1. Construct a basis and construct the Hamiltonian
2. Diagonalize the Hamiltonian
3. Solve Poisson's equation for the electrostatics
Plane waves excel at 1 and 3 at the expense of being bad for 2, due to the large basis set required.
GAMESS uses (as far as I can understand) standard quantum chemistry basis functions such as gaussians and other atom-centered basis functions, which probably means that you will probably be OK for step 2. But solving Poisson's equation and constructing the Hamiltonian will be completely different. Poisson's equation is not too bad, since that is very simple with plane waves, but the setup of the Hamiltonian is both difficult and very different. In the case of LAPW it involves solving differential equations for the muffin-tin part of the basis functions and matching these to the plane-waves. In the case of "pure" plane waves you have to implement either pseudopotentials (PP) or projector augmented waves (PAW) to select the right energy window for your calculation ("staying orthogonal to the core states" to use a more common phrase). Neither PP's or PAW are trivial, but they allow you to postpone some problems and just go for using some standard PP or PAW library (which amounts to trusting someone else's solutions to the radial differential equations to work for you).
Doing this is a challenging task and if you also want it fast and accurate, it is very, very hard. GAMESS is (I infer from its popularity) a well-tuned piece of software, which means that datastructures and similar are tweaked to be highly efficient for the underlying method. This means that you shouldn't count on being able to re-use a lot of code when you implement your new basis (although here in-depth knowledge of GAMESS is of course required to give a full assessment).
This makes me ask the question: Why do you want to do it? It is a lot of work, you will most likely not benefit much from the code already available and, as you say, other packages are already available.
- Badges
- Science topic
- Similar topics
- Computer Science and Engineering
- Programming Languages