(616e) Transitory Sensitivities and Automated Analysis of Large Chemical Mechanisms

Large chemical kinetic mechanisms are vital to understanding many complex chemical processes. While recent advances have made it feasible to generate and, in many cases, simulate these large mechanisms, technology for analyzing them efficiently has lagged significantly. At these sizes, manual analysis is far too time consuming to be practical. While automated flux analysis tools are available, they miss vital low flux pathways. Sensitivity analysis, a common go-to method for smaller systems, rapidly becomes very expensive in these systems. Even adjoint sensitivity methods can take weeks to finish. Forward sensitivity methods can be accelerated very significantly through parallelization, but with a significantly larger associated computational cost that makes it impractical to deploy regularly and very impractical to scale up. We present a fast, novel, time-local alternative sensitivity approach we call transitory sensitivity analysis. Applying this technique to a medium sized, ~450 species, biofuel ignition simulation, we are able to demonstrate a speed up in identifying sensitive parameters from roughly one day using adjoint sensitivity methods to only seconds using transitory sensitivities. Applying this technique to a truly large mechanism, ~1500 species, we are able to demonstrate that this method takes only minutes to run. We further demonstrate that this method can be fused with traditional flux analyses to go beyond sensitivity analysis. This allows automatic tracing of sensitive pathways, enabling identification of associated sensitivity-adjacent parameters that may be important due to their uncertainties even if they are not sensitive. This is incredibly important given the large uncertainties in kinetic parameters. We present this algorithm as solid groundwork for efficient automated kinetic mechanism analysis and refinement.