Implement multi-step generic reactions

[shoubhikraj]

Right now autodE only models single step elementary reactions automatically (for multi-step the intermediates must be known). If there are intermediates, they are ignored. Which is a problem for modelling mult-step reactions where intermediate geometries are not available, and cases where it's unknown if there are intermediates (e.g. metal complexes)

Chem. Sci., 2018, 9, 825 proposes a way of going around that by constructing reaction networks by making and breaking all possible bonds involving some selected active atoms. This allows forming a network of possible paths, which can be filtered by checking the number of bonds being broken/made which gives a rough likelihood of reaction, and then by using low-level methods. This seems to be very close to the autodE model and much of the required functionality is already present (molecular graphs etc.)

The unknown intermediate can be solved by using IRC runs. Gradient-only IRC algorithms are cheap and inaccurate in terms of reaction path integral but will end in the correct minima. This would allow checking whether there are hidden intermediates in a single autodE Reaction and make it robust against choosing wrong TS from the adaptive algorithm.

[eneas77]

Hi!

Those are great news!

Let's not let ourselves get seducted a lot about it. Although it is a normal evolution in creating more complex "reaction microscopes", going step-by-step in basic things is wiser.

My suggestion is, keeping in mind that horizon, sort out the current issues and move forward v.2.0.

Progressión is becoming stiff and is time to wrap up most of open bugs/issues.

Just my POV

Hipe it helps...

E77