4
$\begingroup$

Given that it seems every answer to this question is that it is "impossible to predict the outcome of a chemical reaction." Is chemistry just trial and error? Given how fundamental and revolutionary a simulator capable of predicting possible outcomes of a reaction would be, why has it not been done yet? Do you think it will ever happen?

$\endgroup$
  • 1
    $\begingroup$ A useful computational method like CCSD(T) scales with $N^7$. Even if the Computers get more and more power, those methods will aways be extremely expensive $\endgroup$ – Fl.pf. Oct 28 '17 at 10:18
  • 2
    $\begingroup$ Possible duplicate of Software for predicting chemical reactions $\endgroup$ – paracetamol Oct 28 '17 at 11:41
  • $\begingroup$ This isn't necessarily a duplicate if one focuses on the machine learning aspect. $\endgroup$ – pentavalentcarbon Oct 28 '17 at 15:42
  • 1
    $\begingroup$ @paracetamol I disagree. Machine learning now plays a large role in chemistry, and can predict reaction outcomes. See dx.doi.org/10.1021/acscentsci.7b00064 for just one of many examples. $\endgroup$ – pentavalentcarbon Oct 28 '17 at 17:04
  • 2
    $\begingroup$ Machine Learning does not circumvent the computational complexity of the problem. In other words, ML doesn't transform the traveling salesman problem from NP-complete to P, unless P=NP. A product prediction routine has to explore all possible reaction pathways between all possible stable compounds, which grows exponentially with the number of atoms. Or at least find all convex stability regions and evaluate the thermodynamics correctly, which still explodes combinatorially. $\endgroup$ – Deathbreath Nov 1 '17 at 15:16
1
$\begingroup$

It's not an all-or-nothing situation, where either we can reliably prediction reaction products, or we have not tried to.

For instance, drug companies, for many years, have been making extensive use of increasingly powerful, sophisticated, and refined in silico predictive models to choose likely candidates for in vitro testing. And I would be surprised if they've not also begun to develop, evaluate, and incorporate ML.

At the same time, these models are not perfect. So the answer is that it is being done, it's been done for a while, we're getting better and better at it, but we still have a long way to go.

It will be interesting to see what happens with quantum computing, which is currently in its infancy.

$\endgroup$

Not the answer you're looking for? Browse other questions tagged or ask your own question.