4 ms·
Disclaimer: didn’t read the paper and only have a very basic understanding of chemistry (and sorry for simplifying a complex topic), but started wondering wheth
by krmblg 3y ago
Disclaimer: didn’t read the paper and only have a very basic understanding of chemistry (and sorry for simplifying a complex topic), but started wondering whether we should be able to discover useful reactions via pure calculation by now (ignoring quantum effects)?
Despite being a specific domain, wouldn’t reasoning about reactions (and their efficiency in an industrial, large-scale setting) be something that "models" should be able to do quite "easily" (given the fact that bonding forces, energy requirements and catalytic effects are sort of well-known, i.e. just some more dimensions to deal with)?
- spenczar5 3y agoYes, but its not easy. The devil is in the details. https://en.m.wikipedia.org/wiki/Computational_chemistry https://en.m.wikipedia.org/wiki/Computational_chemistry
- philipkglass 3y agoThe only kind of bonds we can model reasonably well while ignoring quantum effects are hydrogen bonds. Hydrogen bonds can be modeled as parameterized effects in Newtonian mechanics, and molecular dynamics simulations using Newtonian mechanics can reproduce some interesting properties of molecules in solution. The vast majority of chemical reactions involve changes in electronic configuration that are only described by quantum mechanics. The most accurate algorithmic approximations of electron quantum mechanics have terrible scaling properties - O(N^7) or worse. Due to that terrible scaling, the largest high-accuracy calculations that are tractable now are not that much larger than those that could be completed in the 1990s, despite much greater processing power. There are other ways of approximating quantum effects in chemistry that scale better, but they all have tradeoffs and weaknesses of their own. They can used in limited domains or used to guide experimental design, but they're not accurate enough to discover useful reactions via pure calculation. They need to work in tandem with experimental validation and it requires domain experts in both the experimental and theoretical work. The methods are not simple enough for a bench chemist to use them as a black box for reaction discovery.
- krmblg 3y agoThanks for the reply, really didn't know such a vast amount of chemical reactions and outcomes was grounded in QM.
- mkesper 3y agoProbably because our school atom model is too simple. https://www.quantamagazine.org/inside-the-proton-the-most-complicated-thing-imaginable-20221019/ https://www.quantamagazine.org/inside-the-proton-the-most-co...
- jabl 3y agoBroadly speaking, "computational materials science" has been a thing for quite a while, and has been responsible for the majority of academic supercomputer time allocations for decades.
- barbarr 3y agoYes, there are many research groups working on this at the moment. We can (roughly) screen through chemical reactions performed in vaccuum. Even this is difficult since you either need to simulate atoms bouncing around until you observe a reaction (extremely slow to perform) [0], or you need to numerically search for a viable reaction pathway (still quite slow to perform) [1]. The main problem is that the best methods scale badly with the number of atoms you're simulating, so you need to trade off accuracy for speed by using less-accurate methods. Screening through reactions in the real world is particularly hard, since you not only need to worry about the inaccuracy of your simulation method, but you also need to take solvent/environment effects into account. You need to trade off even more accuracy for speed if you want to do so. As computing power advances, there will be less pressure to make these tradeoffs, but a lot of work in comp chem at the moment is focused on either exploring or expanding the speed-accuracy frontier. [0] https://en.wikipedia.org/wiki/Transition_path_sampling https://en.wikipedia.org/wiki/Transition_path_sampling [1] https://people.chem.ucsb.edu/kahn/kalju/chem126/public/qm_ts_optim.html https://people.chem.ucsb.edu/kahn/kalju/chem126/public/qm_ts... [2] https://en.wikipedia.org/wiki/Ab_initio_quantum_chemistry_methods#Accuracy_and_scaling https://en.wikipedia.org/wiki/Ab_initio_quantum_chemistry_me...
- krmblg 3y agoThanks for the reply and the links. Really have to read up a little.
- lumb63 3y agoNot a chemist, but I think trying to do chemistry without modeling quantum effects is like trying to do physics without math. The latter is necessary to express the former.
- krmblg 3y agoYeah, reading the sibling comments made me aware that my mental model when it comes to chemistry (and the role QM plays even in "simple" scenarios) needs an update. Thanks for the angle!