Plausible Chemical Reactions/Processes via Similarity

[josh]

There will be a lot of tweaks needed to refine theory, parameters of “similarity”, longer searchs where intermediate steps are not cost beneficial but they add up to a big win, special conditions, etc. Impossible to cover all reactions, so expert users need to be able to knowledge engineer additions that are close to the areas of special interest.

Plan on a Debian like stable, testing, devel kind of usage stagger.

[josh]

Based on my preliminary searchs, I believe that the full value, full automaticy realization of the computationally modeled process will involve breaking new ground in description & statical prediction of what matters in the laboratory/PChem setup of each experimental stage. What is the spatial, temporal, temperature, pressure, concentration, etc. distribution of materials & how is it being hoisted into place???

Some thoughts on model form:

At the biz end of reactions, we can use FE analysis of the right granularity to describe all of the chem relevant variables that should affect yields,etc. N-d FE analysis of whatever matters, matched to the experimental data & theory/predictive analytics that are available, with error bars too.

We can think of each reaction state as being constructed by a sequence of build-up, run, harvest, tear-down operators that have the correct syntactic form & which combine to provide all the info we need for the FE analysis, plus cost analyses of various sorts.

[josh]

We believe in a future of efficient robotics process pipelines which will enable a lot of cost, speed, & precision refinements in chemical process pipelines. As one small aspect of that, consider heat, stirring, delivered by energy & hypersonic beams aimed with sensor feedback.

[josh]

Focusing on good choices of practical examples improves work prospects. So put some thought into example areas:

a) Optimization of process has a steep gradients in economic value

b) Lots of alternatives of interest can be easily tested and filling in the model space will add value

c) In statistics, it is common to validate models from finite data using “leave-one-out” or cross-validation analysis – build models of known data like that were the predictive approaches are validated against testing performance of what they would say if known results were not available to influence the predictions as they are being made.

d) Reactions where many by products may be harvested for economic value after tweaking – how it is controlled/optimized in practice

[josh]

One type of model which strikes me as interesting is something like this:

At each FE grid cell, there is a Markov/graph of chemical types with possible transitions between them – the chance of a transition at the current time point is a function of concentrations, temperature, etc. The overall rate of change in the solution is an integration over all the grid points.

[Author]

Posts