My computational physics research agenda

As a showcase, I have already established my own cluster using 3 PC linked with a switch (this cluster is called 'comsics',you can access it at http://comsics.usm.my/), which currently has a total of only 6 CPU. With the experience of successfully setting up comsics, I am now quite confident that setting up a larger cluster is not technically difficult at all. It is in principle quite easy to scale up the cluster to have more CPUs. Ａcluster is different from a workstation or a server which are generally faster but more expansive. A cluster is comprised of many units of commercial PCs linked up by network switch. The major advantage of a cluster (as compared to a single PC or a workstation) is that we can run parallel job for large computational calculation, which may takes weeks on a single PC or a single workstation. The other advantage is that clusters are much cheaper in terms of performance/cost, and is a general trend in computational physics community (people don't buy more expansive high-end workstation, instead they buy more ordinary PCs to build larger cluster). This is the most ideal option for us in USM School of Physics - good computational power at affordable cost. In addition, the cluster is flexible: the component desktop PC can be used by an individual user when the cluster is not running parallel job. So, in this sense, we have many new PCs for our users, and in time of need, the PCs can be turned into a parallel machines that run large computational jobs. We can also run the calculation via remote log-in despite the fact that the cluster is located physically in the theory lab. For example, I submit my diffusion monte carlo code to run on comsics (physically located in theory lab, USM) via internet (while I am in the Netherlands).

My consultants in the grid lab in computer science constantly advise me on the technicalities related to the setting up of the cluster computational facility as proposed above. According to the quotation obtained by my consultants in grid lab, their supplier earlier quoted them RM 20k for a 32 core cluster consisting of 8 units clone desktop PC. To set up a large cluster, we also need space to accommodate the desktop PCs, some extra power points and good airconditioning. For budgeting purpose, I propose RM 80k for setting up a cluster computing facilities for intensive computational simulation in theoretical physics projects.

Any proposal to buy such computing facilities must be supplemented with justification. The main justification comes from the our need to run calculation for our projects. For my own case, I do have many research projects that require intensive use of heavy computational resource (discuss below).

I take this opportunity to propose an agenda I have in mind: a proposal to set up a Computational Physics Expertise within the Theory Group. This is a plan to develop professional expertise that focuses on computational physics simulation projects. Currently I have many projects proposal (on-going ones as well as under plan), e.g. first principle calculation of ZnO native defect formation, direct and monte carlo molecular dynamics of structural phase transition of phenol-amines adducts, diffusion quantum monte carlo calculation of Josephson Junction Arrays as qubit, first principle calculation of interaction between electromagnetic waves with crystalline system, and others (you can check my project proposals uploaded in my webpage, http://www2.fizik.usm.my/tlyoon/research.html, but there are still some more inspired research proposals in my mind that have yet to be uploaded). In the long run, I wish to train up experts (including anyone from within the theory group) who wish to master hardcore techniques in computational method, e.g. parallel programming and other linux-based programming skills. The proposed expertise unit will function as an expert unit capable of performing numerical calculation for real physics problems, such as molecular dynamics simulation, first principle calculation for quantum chemistry and crystalline systems, monte carlo simulation, quantum monte carlo calculation, computational material science, Finite Difference Time-Domain (FDTD) calculation for propagation of electromagnetic waves in diffractive medium by solving Maxwell equations from first principle, etc. These mentioned calculation methods are projects I am currently undertaking.

The other major intention for the proposed computational expertise unit is to create strategic collaboration with other experimental groups in USM (X-rays cyrstallography group, NOR lab, quantum chemistry group in PPJJ and PPSK). We will help them to model the experimental results. The collaboration between experimentalists and theorists are not a norm in USM. The proposed effort of setting up a computational physics expertise unit within the theory group can help bridging this gap. I am confident that there are many experimental data from the NOR lab and X-ray lab can be modeled with first principle calculations or other computational techniques. For example, the structural phase transition observed in phenol-amines adducts is in principle model-able using either monte carlo or direct molecular dynamics. The XRD spectrum or Raman spectrum on GaN or ZnO samples can be quite routinely simulated by constructing supercell models with DFT code. So are those novel X-ray cyrstallographic structures of organic crystal solved routinely in the X-ray lab can be calculated using DFT codes. The proposed expertise unit will develop all the necessary expertise, in particular first principle calculation methods, to compliment the research investigations carried out in our experimentalist colleagues' lab.