The Institute for Security Education and Research (INSER) is a joint Texas A&M University, Texas A&M Engineering Experiment Station (TEES), and Texas A&M Agrilife Extension Service center. INSER is funded by Lawrence Livermore National Laboratory as part of Texas A&M’s role as a affiliate of Livermore National Security LLC, which manages the laboratory. The center has one educational component and three research components, one of which is computational methods development in support of nuclear weapons stockpile stewardship. This component is executed through CLASS as a partner with INSER. There are currently three major research projects with tasks as follows:
- Strength of Materials:
Using first-principles-based interaction potentials within large-scale atomistic simulations Texas A&M will investigate structure, energetics, and mobility of dislocation cores, with input from particle based simulations. Texas A&M will also perform microscopic simulations of dislocation-dislocation interactions using tessellation-based dislocation dynamics. As a first step towards modeling of assemblies of grains/polycrystalline plasticity, TEES will use both atomistic simulations of grain boundaries and develop a 2.5-dimensional dislocation-dynamics program.
- Phase Transformations:
In this area, Texas A&M will use an integrated effort combining detailed first-principles atomistic simulations with phase-field and thermodynamics modeling. This task proposes a novel integrated computational study by employing state-of-the-art multiscale modeling and simulations to address the solid-solid phase transformations in plutonium and its alloys. The four subtasks associated with this task are
- Ab-initio calculations on phases of Plutonium and its alloys
- MD simulations on phases of Plutonium and its alloys
- Thermodynamic and kinetic modeling of Plutonium and alloys
- Phase field modeling of phase transformations and microstructure evolution in Plutonium and its alloys
Massively Parallel Radiation-Hydrodynamics including Uncertainty Quantification:
This project contains tasks in discretization methods and algorithms for coupled physics, computer science support of massively parallel calculations, and assessment of the accuracy of uncertainty-quantification methodologies. These tasks are integrated into a coordinated project that will couple radiation transport and hydrodynamics in massively parallel software and will assess uncertainty-quantification methodologies applied to such radiation-hydrodynamics simulations. The four tasks associated with this project are
- Multiphysics Coupling and Operator Splitting
Radiation-hydrodynamics codes have traditionally used simple operator splitting as a time-integration technique. The advantage of simple operator splitting is that it can be easily implemented. The disadvantage is that operator splitting is only first-order accurate, and it can be unreliable in that false convergence can occur as the time step is decreased. A first step in developing a successful second-order operator splitting technique is to develop a time discretization for the radiation transport equation with properties lacking in current second-order discretizations. Texas A&M will develop a second-order accurate non-oscillatory time discretization for the multifrequency discrete ordinates equations that can be efficiently solved using Krylov methods preconditioned with the linear multifrequency-grey acceleration technique.
- Massively Parallel Computing
- Development of the ARMI communication libraryAdaptive Remote Method Invocation (ARMI) is a communication library that uses the remote method invocation (RMI) communication abstraction to hide the lower level implementations (e.g., MPI, OpenMP, PThreads, the machine’s native communication primitives, etc.). TEES will design and develop improved methods for implementing global distributed locks, including support for recursive locks. TEES will provide support for additional communication patterns and group-based versions of the operations that select subgroups of threads for group-based communication.
- Development of a distributed graph and partitioning toolsThe pGraph is a parallel and distributed graph data structure that is being developed as one of the core pContainers of the STAPL project. Texas A&M will further develop the pGraph and related scheduling and partitioning tools.
- Implementation of radiation-hydrodynamics softwareOther tasks associated with massively parallel computing involve discretization and coupling techniques for radiation transport and for hydrodynamics. These techniques will be implemented and tested in a massively parallel radiation-hydrodynamics code that TEES will assemble.
- Uncertainty QuantificationWe are entering an era in which methodologies for uncertainty quantification (UQ) are becoming more complicated, and are being used to support high-consequence decisions. It is therefore important to assess how accurately these methods quantify uncertainty or predictive capability. Texas A&M will use a “method of manufactured universes” (MMU) to quantitatively assess methodologies that are designed to quantitatively assess predictive capability. The MMU concept is newly developed at Texas A&M and is based upon the idea of defining “laws of physics” for the “universe” that can be exactly modeled and using “approximate models” to predict the outcome of “experiments” generated according to these laws.
Hydrodynamics and Transport Methods:
This project contains tasks in hydrodynamics methods and radiation transport methods. Each task is described below.
- High Order Hydrodynamics Methods
Texas A&M will develop a discontinuous Galerkin method in r-z geometry based upon a quadratic trial space for an unstructured triangular mesh. Rather than upwind in the usual manner, Texas A&M will use central averaging in conjunction with a new type of entropy-based artificial viscosity that has recently been developed by Texas A&M researchers. In generalizing this artificial viscosity to r-z geometry, TEES will focus upon preserving spherical symmetry.
- Parallel Radiation Transport
The fundamental problem with transport scaling is that sweeps are serial in one dimension (the direction of radiation propagation). Texas A&M will investigate long-characteristic methods that yield a true 1-D equation along each characteristic. This is in contrast to standard Sn spatial discretizations that couple cells laterally to the direction of flow. Reduced lateral coupling eliminates a substantial portion of cell-to-cell dependencies and thus simplifies parallelization. In addition to parallelism over the characteristic rays, substantial parallelism is also possible for the solution process along each characteristic, via “cyclic reduction” or “parallel prefix” techniques. The basic idea of cyclic reduction is to locally eliminate half of the unknowns in each of a series of reduction steps. Eventually, the system becomes small enough to directly solve, after which the entire solution is obtained via a succession of back-substitution steps that mirror the original reduction steps. The application of cyclic reduction can produce logarithmic or multigrid-type scaling along the direction of flow, thereby substantially improving upon the sequential scaling of current algorithms in this direction.