Modeling of quantum dots using improved AIMD highly parallel methodology and its application to commercial AIMD software
The main challenge that will be addressed with this SOSCIP project is the scalability of the Atomic Works ab initio molecular dynamics (AIMD) code running in parallel on 1000 or more cores. Our aim is to explore the most efficient way to break down large simulation domains and optimize message passing between computational threads to achieve a near linear scale increase in performance. The approach to increasing the size domain of AIMD simulations is to develop a highly efficient parallelized algorithm based on a message passing interface (MPI) framework that can be efficiently scaled. In addition, we will utilize a variable time-step methodology that will allow us to break down the simulation domain into multiple sub-domains, each recalculated at different time-step intervals. This will allow us to concentrate computing resources in the areas of high interest (high kinetic energy, stress, or potential energy) while reducing the focus on less demanding areas of the simulation. The main technical challenge of this SOSCIP project will be the effective combination of space-domain parallelization with the variable time-domain use and optimization of message passing between the domains.
Modeling of quantum dots using improved MD/DFT highly parallel methodology and its application to commercial MD software
Quantum dots are small ligand-protected chunks of semiconductors which have diameters ranging from 2 to 100 nanometers. Because of their small size, their properties are quite different from those of the bulk semiconductor. In particular, due to quantum confinement, their band gap is larger and the band gap increases as the diameter decreases. In solution, the chunks must be protected by legends in order to prevent agglomeration. Modeling the properties of quantum dots is very helpful in understanding and interpreting experimental results. Efficient parallelization of molecular dynamics software forms a cornerstone of simulation modeling of nanoparticles as it allows for simulation of larger space domain over longer periods of time. The software methodology improvements aimed at increasing computational efficiency on large HPC clusters undertakes as part of this research project aim to achieve this goal.
Non-equilibrium green’s function approach to simulations of active photonic nanostructures
The present project is aimed at creating fast and accurate highly-parallel algorithms and the related mathematical models to simulate active photonics devices (e.g., quantum well lasers) using high performance computers. Computational tools for simulation of optoelectronic devices bring to the industry a considerable reduction of the development cost. New models are developed, implemented in software, verified by experiments and tested for new predictions. Such an approach shortens development time of new devices, thus reduces overall price of finished products. The past era of device simulations was based on drift-diffusion approach. However, the nano-era of optoelectronic devices involves new materials and architectures, along with a number of novel phenomena that must be taken into account. Altogether the field is expanding very fast. At the same time modern simulations require extensive computational resources, and there exists significant room for optimization. One of the approaches to optimization is the use of parallel algorithms on the systems like Blue Gene Q. We hope our project will allow Optiwave to increase significantly their global market share in photonic simulations software
Parallel programs for autocorrelation problems (PAAP)
Autocorrelation problems are a rich source of extremely hard computational challenges for which conventional parallel computing has been the only reasonably successful approach. Sequences with constant autocorrelation are called complementary and have a wide range of applications, including: Coding Theory, Telecommunications, Image Compression, and Wireless Communication Protocols.
Over the past decade, the PI has been able to find several new complementary sequences in a series of papers with his collaborators in Canada, the United States, Australia and Europe. It has become apparent, though, that the current algorithms have reached a point of saturation, in that they are unsuitable for producing new results. The advent of hardware accelerator technologies such as the GPU is a promising new direction. In some problems GPU‐enabled algorithms have been reported to exhibit a 2000‐fold speedup, which is quite significant. Therefore, it is clear that hardware accelerator technologies provide vast opportunities for innovation in scientific computing. It is a fortuitous coincidence that our partner, Maplesoft, based in Waterloo Ontario, has recently devoted a large part of their efforts in producing parallel versions of Maple, the flagship Canadian mathematical software product. Our proposal will benefit and complement the efforts of Maplesoft.