A novel machine learning tool can calculate how much energy is needed to make or break a molecule with greater accuracy than conventional means, according to a new study published Tuesday in the journal Nature Communications.
Machine learning tool cracks quantum chemistry quirk
As of yet, the tool can only work with simple molecules, but it carves a path to future advances in quantum chemistry.
"Using machine learning to solve the fundamental equations governing quantum chemistry has been an open problem for several years, and there's a lot of excitement around it right now," said co-creator Giuseppe Carleo, a research scientist at the New-York-City-based Flatiron Institute's Center for Computational Quantum Physics. A greater understanding of the creation and destruction of molecules could be, according to Carleo, a way to unveil the inner workings of the chemical reactions crucial for life.
Carleo — along with collaborators Antonio Mezzacapo of the IBM Thomas J. Watson Research Center and Kenny Choo of the University of Zurich — presented their work on May 12.
How machine learning cracked a quantum conundrum
The team's new machine learning tool estimates the amount of energy required to assemble or rip a molecule like ammonia or water. The calculation requires a determination of the molecule's electronic structure, which is the total behavior of all electrons binding the molecule into one, according to phys.org.
A molecule's electronic structure is not easy to calculate, and forces scientists to determine every potential state the molecule's electron could take — not to mention each state's probability.
Moreover, electrons interact and become quantum-mechanically entangled with one another, which means scientists can't treat them individually. The more electrons in a molecule, the more entanglements happen, and the problem becomes exponentially more complex. This is why exact solutions simply don't exist for molecules with complexity beyond the two electrons found in a pair of simple hydrogen atoms. Even approximations lack accuracy when more than a proverbial handful of electrons are involved. This is why this new discovery — unveiled via machine learning — might one day transform a nigh-impossible challenge into a matter of simply crunching the numbers.