New approach improves accuracy of quantum chemistry simulations using machine learning

A new tip for modeling molecules with quantum precision takes a step towards the revelation of the equation at the center of a popular simulation approach, which is used in fundamental studies of chemistry and materials.

The effort to understand the chemical materials and reactions eat almost a third of the time of supermarkets of national laboratory in the United States, the ordeal for precision is the quantum problem with several bodies, which can tell you what is happening at the level of individual electrons. It is the key to chemical and material behavior because the electrons are responsible for chemical reactivity and links, electrical properties and more. However, quantum calculations with several bodies are so difficult that scientists can only use them to calculate atoms and molecules with a handful of electrons at the same time.

The functional theory of density, or DFT, is easier – the computer resources necessary for its calculation scale with the number of electrons in cubes, rather than rising exponentially with each new electron. Instead of following each individual electron, this theory calculates the densities of electrons – where the electrons are more likely to be located in space. In this way, it can be used to simulate the behavior of several hundred atoms.

A key problem for DFT users is the exchange function, which describes how electrons interact with each other, following quantum mechanical rules. Until now, researchers have had to be satisfied with approximation of the XC function for their particular application.

“We know that there is a universal functional – it does not matter that the electrons are in a molecular system, a piece of metal or a semiconductor. But we do not know what Scientific advances.

Due to the importance of the DFT for future materials as well as for the fundamental sciences, the Ministry of Energy provided funding and supercomputer for the quest for the UM team to approach this Universal XC functional.

The researchers began by studying individual atoms and small molecules with a quantum theory with several bodies so that they can return the DFT problem. Instead of adding the approximate XC function to give the behavior of the electrons in atoms and molecules, they determine it – using automatic learning – which the XC function will give the behavior of electrons as calculated by quantum theory with several bodies.

“Many body theories give us the right answer for the right reason, but at an unreasonable cost of calculation. Our team has translated results with several bodies into a simpler and faster form which keeps most of its precision,” said Paul Zimmerman, professor of UM chemistry, who led the quantum calculations of many bodies with the chemistry PH.D. Student Jeffrey Hatch.

The Zimmerman group has created a set of training data of five atoms and two molecules, in particular lithium, carbon, nitrogen, oxygen, neon, dihydrogen and lithium hydride. They tried to add fluorine and water, but these additions did not improve the XC function – the team believes that it was already as good as it was going to get out of data from light atoms and molecules.

However, DFT calculations using this XC function were already much better than expected for its level of complexity. DFT precision is described as a set of chess in a scale. In the most elementary leading form, electrons are considered existing in a uniform cloud. In the second row version, the Gavini team used, the electron cloud changes density, considered a gradient.

For the third level, researchers add more information on electrons, such as their kinetic energies. This generally means bringing simplified versions of the difficult wave difficulty to several electrons, which can better describe what is happening with the electrons. However, by calculating a better XC functional, Gavini’s team obtained third -plan details.

“The use of a precise XC function is as diverse as chemistry itself, precisely because it is agnostic materials. It is also relevant for researchers who try to find better battery materials to those who discover new drugs for those who build quantum computers,” said Bikash Kanungo, assistant researcher UM in mechanical engineering and first author of the study.

Researchers can directly use the XC function discovered by the group or experiment with the team’s approach. For example, Gavini says they started with light atoms and molecules, and then he would like to explore solid materials.

Again, the XC function should have a universal form, but the delicate part is to determine what it is. Does the XC work functional that its team will work well for solids? Would a new functional calculated for solids be more effective? And could they build a combined function that worked well for the two sets of materials?

The other improvement that the team would like to continue is higher precision. This would mean that instead of collectively looking at the electrons, as electron densities, they should include the individual orbitals in which the electrons move. In this case, their tip in reversing the problem to obtain the XC functional becomes a much more difficult calculation. Even with the density gradients, they had to do the calculations on one of the largest superordinators in the United States, so this avenue would require more computer time.