Efficient computation of ignition delay time

[jiweiqi]

• Ji, W., Ren, Z. and Law, C.K., 2019. Evolution of sensitivity directions during autoignition. Proceedings of the Combustion Institute, 37(1), pp.807-815.
• Gururajan, V. and Egolfopoulos, F.N., 2019. Direct sensitivity analysis for ignition delay times. Combustion and Flame, 209, pp.478-480.
• Almohammadia, Saja, Mireille Hantoucheb, Olivier P. Le Maıtrec, and Omar M. Knioa. "A Tangent Linear Approximation of the Ignition Delay Time. I: Sensitivity to Rate Parameters."

[jiweiqi]

Gururajan, V. and Egolfopoulos, F.N., 2019. Direct sensitivity analysis for ignition delay times. Combustion and Flame, 209, pp.478-480.

1. In sensBVP, L is first minimized to zero using a BVP solver.
   This step is actually the real bottleneck. But it turns out this
   step can be skipped completely and the result shouldn't change.
   (It was included in the original code as a sanity check for
   reviewers.)
2. Apart from the above, \partial L to \partial A is computed
   via numerical finite differences. I don't believe that this step
   is a major bottleneck, but I'm sure automatic differentiation
   can accomplish the same faster and importantly: more accurately!
   I'm curious to see how autodiff scales with mechanism size.

[jiweiqi]

Core codes
https://github.com/jiweiqi/sensBVP/blob/112b2b823b5322c22e866e70f454281bc6492f4f/src/sensBVP.cpp#L550-L666