Application Partnerships
Success Stories

Application advancements through prior work by the FASTMath and QUEST teams enabled next generation codes, faster solutions, and more robust simulations. These advancements include:

Next-generation application codes:

  • We helped develop two new climate application codes based on FASTMath and QUEST libraries to attack scientific questions regarding ice sheet dynamics and their contribution to sea level rise. These codes required developments within many technical areas of FASTMath and QUEST: structured and unstructured mesh, nonlinear solvers, preconditioners, and UQ. This work also drove interoperability among our technologies. With these efforts, DOE now has unique capabilities for ice sheet applications, and this work resulted in the first simulations of Antarctic response to warm-water forcing that is sufficiently resolved to capture dynamics. (Climate Sciences)
  • We combined structured, high-order, mapped-multiblock, finite volume discretizations in Chombo with hypre and PETSc to enable state-of-the-art algorithms in the COGENT edge plasma code. This enabled the first-ever, self-consistent solution of a continuum gyrokinetic system in the edge geometry across the magnetic separatix. This is an important foundational component in the AToM fusion application. (Fusion Energy Sciences)
  • We provided the discretizations, linear and nonlinear solvers, and time integrators for a new high-order adaptive 3V Landau collision integral code through PETSc. This allows scientists to better understand the physics and mitigation strategies of runaway electrons in tokamak plasmas; a key concern for ITER. (Fusion Energy Sciences)
  • We provided the core software infrastructre through BoxLib AMR and its multigrid solver for a Nyx Lyman-alpha forest cosmology simulation at unprecedented resolution. (High Energy Physics)
  • Our POUNDERS optimization algorithm in TAO is used widely by the international nuclear physics community. This code produced energy density functionals and chiral interactions that are the present-day standard for nuclear structure calculations of medium-mass nuclei. (Nuclear Physics)

Faster time to solution:

  • We developed time integrators for nonhydrostatic dynamical core systems and solvers for the implicit part of the integration strategy and developed new partitioning and task placement strategies to increase in-node locality and reduce communication costs by 31% on 16K cores for the CME/HOMME code. This work allows ACME to reach new, critical levels of meso-scale resolution and better utilize LCF systems. (Climate Sciences)
  • We developed an eigensolver for linear response eigenvalue problems in excited state electronic structure calculations. The new solver is at least 2x faster than existing solvers used in quantum chemistry software, e.g., the NWChem suite. We also developed the parallel pole expansion and selected inversion (PEXSI) software to enable fast electronic structure calculation in biological and lithium ion interface dynamics applications. PEXSI enables simulation of ab initio molecular dynamics of low-dimensional systems with more than 10,000 atoms; previous solvers were limited to 100s to 1000s of atoms. (Basic Energy Sciences)
  • We developed a multi-level eigensolver based on the locally optimal, block-preconditioned conjugate algorithm (LOBPCG) to compute eigenpairs of the nuclear configuration interaction Hamiltonian in successive configuration spaces. We used network topology-aware techniques to optimize the performance of the eigensolver on LCFs. Our new solver is 2-3x faster than existing solvers, which enables nuclear physicists to accurately study properties of light nuclei. (Nuclear Physics)
  • We added an accelerated nonlinear solver and higher-order time integration methods to the ParaDiS code resulting in significant speedups using SUNDIALS, which enabled a 55 million dislocation node system on Sequoia, the largest ever run by ParaDiS. (NNSA SciDAC Application)

More robust simulations:

  • We developed the FusionMesh code to generate the field-following unstructured meshes needed by many tokamak modeling codes, which dramatically decreased the time and effort required to generate high-quality meshes for the XCG and M3D-C1 fusion codes. (Fusion Energy Sciences)
  • We provided unstructured mesh infrastructure for the M3D-C1 extended MHD code including linkage to solvers and adaptive meshing. This work combines finite element formulation, error estimation, mesh adaptation, and physics-based preconditioners to substantially improve robustness.  This code is now used to effectively resolve the ELM disruptions in the edge regions of tokamaks. (Fusion Energy Sciences)
  • We worked with material chemists to accelerate the computation of MP2/VMP high-dimensional integrals. We employed low-rank tensor methods for representation of high-dimensional functions coupled with quadrature methods, all deployed in UQTk, and demonstrated computations of relevant integrals with order of magnitude improvements in accuracy and speedup. (Basic Energy Sciences)
  • We developed high-order, curved mesh adaptation procedures for the accelerator simulation code, ACE3P, using the unstructured mesh components PUMI/MeshAdapt. This enabled more optimal accelerator cavity design through increased resolution and fidelity and provided the highly accurate geometrical representations needed for high-order methods. We also developed nonlinear eigensolvers for Omega3P in ACE3P for electromagnetic simulation. This work enabled cavity and accelerator stage design with far less computational expense than was previously possible. (High Energy Physics)