At various points along the path toward productive nanosystems for molecular manufacturing it would be useful to be able to calculate the properties and reactions of assemblies of atoms of various sizes. Within the domain of non-relativistic quantum mechanics, such information is supplied by the Schrödinger equation, but this can only be solved analytically for the hydrogen atom and ions with only one electron. For larger atoms and molecules, numerical solutions require compromises between computational feasibility and accuracy. Recent work from researchers at Argonne National Laboratory suggests that machine learning can be an efficient alternative to numerical computations. A hat tip to KurzweilAI.net for pointing to this New Scientist article by Lisa Grossman “Molecules from scratch without the fiendish physics“:
A SUITE of artificial intelligence algorithms may become the ultimate chemistry set. Software can now quickly predict a property of molecules from their theoretical structure. Similar advances should allow chemists to design new molecules on computers instead of by lengthy trial-and-error.
Our physical understanding of the macroscopic world is so good that everything from bridges to aircraft can be designed and tested on a computer. There’s no need to make every possible design to figure out which ones work. Microscopic molecules are a different story. “Basically, we are still doing chemistry like Thomas Edison,” says Anatole von Lilienfeld of Argonne National Laboratory in Lemont, Illinois.
The chief enemy of computer-aided chemical design is the Schrödinger equation. In theory, this mathematical beast can be solved to give the probability that electrons in an atom or molecule will be in certain positions, giving rise to chemical and physical properties.
But because the equation increases in complexity as more electrons and protons are introduced, exact solutions only exist for the simplest systems: the hydrogen atom, composed of one electron and one proton, and the hydrogen molecule, which has two electrons and two protons. …
The researchers developed a machine learning model to calculate the atomisation energy—the energy of all the bonds holding a molecule together and applied it to a database of 7165 small organic molecules of known structure and atomization energy and containing up to seven atoms of carbon, nitrogen, oxygen, or sulfur, plus the number of hydrogen atoms necessary to saturate the bonds. These molecules had atomization energies ranging from 800 to 2000 kcal/mol. The model was trained on a subset of 1000 compounds and then used to calculate the energies of the remaining molecules in the database. The results showed a mean error of only 9.9 kcal/mol, comparable to the accuracy of methods based upon the Schrödinger equation, but the computations were done in milliseconds rather than hours. The authors suggest that extensions of their approach might permit rational molecule design or molecular dynamics calculations of systems of atoms undergoing chemical reactions.