Computational predictions from chemical systems computations are difficult to analyze, e.g. to assign cause-and-effect relationships and advance understanding, given the nonlinearity, stiffness, and intricate coupling among reaction processes. This is particularly so in results from large-scale computations with complex chemical systems. Accordingly, there is a strong need for analysis software tools that offer fast and effective capabilities tailored to the analysis of chemical systems.

Developments under ECC

In this part of the project we are building a Kokkos-enabled C++ chemical analysis software toolkit tailored for exascale architectures and aim to demonstrate effective performance on heterogeneous architectures and large scale chemical data. Results of CSP analysis are generally useful for understanding cause-and-effect relationships in chemically reacting flows, for describing the general dynamical structure of the chemical system, and for providing means of estimation of important reactions/species, thereby providing key tools necessary for chemical model simplification. The method has been applied for analysis of large-scale computational chemistry databases, including simulations of homogeneous ignition as well as multidimensional chemically reacting flow computations. CSP analysis relies on eigenanalysis of the chemical system Jacobian matrix, and on the projection of the source term onto the resulting eigenvectors. The Jacobian eigenanalysis is an expensive operation, and is the key bottleneck in the computational performance of CSP analysis computations. 


  • Lead
    • Habib N. Najm (SNL)
  • Senior researchers
    • Kyungjoo Kim (SNL)
  • Postdoc
    • Oscar H. Diaz-Ibarra (SNL)
  • Advisors 
    • Eric T. Phipps (SNL)
    • Christian R. Trott (SNL)