Allowing Tmin and Tmax to function, automatically selecting the appropriate rate parameters

[alongd]

Consider the following reaction from Lin 2005, for example:

HCO + NO <=> HNO + CO

With the following rates:
Arrhenius(A=(1.85e+12, 'cm^3/(mols)'), n=0.10, Ea=(-481, 'cal/mol'), T0=(1, 'K'), Tmin=(200, 'K'), Tmax=(500, 'K'))
Arrhenius(A=(1.04e+08, 'cm^3/(mols)'), n=1.47, Ea=(-1765, 'cal/mol'), T0=(1, 'K'), Tmin=(500, 'K'), Tmax=(3000, 'K'))

(taken from the SOxNOx library which I haven't merged into master yet, waiting for SAtomTypes to be merged first)

Several of RMG's applications are for low to mid T, and I'd like to allow RMG access to both rates in this example, letting it choose the appropriate one according to the specified T range (i.e., causing the Tmin & Tmax parameters to actually function...)

I have about half a dozen cases like this in this library, I can assume more might pop-up in other libraries (and training reactions!).
My first though was to create two libraries, one of which consists only of the several low T rates, preceding the original library if the user chooses to use it. However, this becomes a bit more complex, since one of the reactions has three(!) T ranges, and in fact all three matter to me... Plus, my "simple" solution won't address training reactions.

I suggest we allow the following syntax:

kinetics = MultiTRange(
        arrhenius = [
            Arrhenius(A=..., n=..., Ea=..., T0=..., Tmin=(300, 'K'), Tmax=(1000, 'K')),
            Arrhenius(A=..., n=..., Ea=..., T0=..., Tmin=(1000, 'K'), Tmax=(3000, 'K')),
        ],
    ),

Unless there's already a similar functionality that I'm not aware of?

[alongd]

But...
what if one has two reactors in the input file, where each should ideally be assigned different kinetic parameters for the same reaction?

Perhaps in a situation like this RMG should get some numerical rate values from both representations in the relevant overall T range, and construct a modified Arrhenius expression using least squares?
Same goes for training reactions, which are added at the beginning of the run.

[mjohnson541]

How do you plan to distribute the T points used for the least squares fit?

Is it possible that in some cases the reason there are two rates is because they doesn't fit well the combined range?

[alongd]

I can't imagine M.C. Lin compromising for an R² value of less than 0.999 in his publications.
But I don't see any other escape for a model that should cover a larger T range. Right, we'll lose some accuracy. If one is very sensitive to such a reaction, one should divide the model according to the T ranges, so each resulting model is more accurate in its range. I hope this will only be required in rare cases.

Regarding T distribution, I guess it should reflect the temperatures in the input file (giving more points than in the input file, but distributing more heavily around the inputted values).

[github-actions[bot]]

This issue is being automatically marked as stale because it has not received any interaction in the last 90 days. Please leave a comment if this is still a relevant issue, otherwise it will automatically be closed in 30 days.