Careers | Phone Book | A - Z Index

symPACK Sparse Symmetric Direct Solver

sympack logo

We developed and a new direct linear solver for sparse symmetric positive definite matrices, symPACK. By exploiting the symmetry explicitly, symPACK achieves low storage costs. » Read More

SuperLU Sparse Unsymmetric Direct Solver

nlpK2D merged 3

The recently developed communication-avoiding 3D sparse LU factorization in SuperLU_DIST reduces latency, achieving speedups up to 27x for planar graphs and up to 2.5x for non-planar graphs over the baseline 2D SuperLU_DIST when run on 24,000 cores of a Cray XC30, Edison at NERSC. » Read More

Sparse Low-rank Hierarchical Matrix Preconditioner

strumpack rotator

STRUMPACK (STRUctured Matrices PACKage) software uses faster algorithms developed in SSG using hierarchical low-rank matrix algebra, which generalizes fast multipole methods and leads to factorization with quasi-linear arithmetic and memory complexity. » Read More

Solvers for Edge Plasma Simulation

3dtokamaksimulation

Researchers in the Scalable Solvers Group at LBNL help the SciDAC application partnership Center for Edge Plasma Simulation (EPSI) by developing solvers for advanced fusion simulation codes based upon first-principles physical modes in realistic magnetic separatrix geometry, to provide insight into multi-physics in the edge region of tokamak reactors and their nonlocal interaction with core plasma dynamics. Contact: Mark Adams, mfadams@lbl.gov

High Performance Geometric Multigrid

Screen Shot 2015 03 04 at 3.30.54 PM

Research scientists at LBNL and Argonne National Laboratory have developed a new high performance computing metric intended to provide the community with both a ranking metric for the world's largest general purpose computers and a simple, compact, standalone benchmark that pressures machines in a balanced way to proxy modern application and their scaling demands. Contact: Mark Adams, mfadams@lbl.gov.

Accelerating Eigenpair Computations

Li318nbnd2000highlight

Novel approach developed by SciDAC FASTMath Institute accelerates computations of large numbers of extreme eigenpairs of Hermitian operators. The new eigenvalue solver has been tested in the framework of Quantum Espresso codes for electronic structure calculations. Current experimental parallel implementations show a 2x speedup over existing state-of-the-art methods. Contact: Chao Yang, cyang@lbl.gov

HipGISAXS

scaling

The recently developed HipGISAXS software is an unique massively parallel code to simulate and analyze custom and complex morphologies from Grazing Incidence Small Angle X-ray Scattering (GISAXS) experiments characterizing the nanostructural features of materials, particularly at surfaces and interfaces. Using C++ augmented with MPI, Nvidia CUDA, OpenMP, and parallel-HDF5 libraries on large-scale clusters of multicores and GPUs. Contact: Sherry Li, xsli@lbl.gov


Our Mission

Develop efficient linear and eigensolver algorithms and fast, scalable, library implementations. Integrate the new algorithms and software into DOE applications.

Current expertise in SSG includes:

  • Dense linear algebra
  • Factorization-based sparse linear solvers
  • Multigrid methods
  • Sparse eigensolvers
  • Optimization
  • Communication-avoiding algorithms (see, e.g., http://bebop.cs.berkeley.edu/)
  • Mathematical software and libraries

Group Leader:  X. Sherry Li


 NEWS

June 2017

Roel Van Beeumen won one of two best poster blitz prizes at the Householder Symposium XX on Numerical Linear Algebra at Virginia Tech, June 18-23, 2017.

May 2017

Congratulations to Mathias Jacquelin whose joint paper with Bert De Jong (CCMC), Eric Bylaska (PNNL), Jeff Hammond, and Michael Klemm (Intel),“Performance Evaluation of NWChem Ab-Initio Molecular Dynamics (AIMD) Simulations on the Intel Xeon Phi Processor” has been given the “Best Paper Award” at the IXPUG workshop held in conjunction with the ISC 2017 conference.

March 2017

Lin Lin has been honored with a 2017 SIAM Activity Group on Computational Science and Engineering (SIAG/CSE) Early Career Prize.

»More News