Simulating electronic structure with quantum annealing devices and artificial neural networks

Patent metadata

Field	Value
Publication number	US-12033728-B2
Application number	US-202318212646-A
Country	US
Kind code	B2
Filing date	Jun 21, 2023
Priority date	Mar 7, 2019
Publication date	Jul 9, 2024
Grant date	Jul 9, 2024

Abstract

Approaches, techniques, and mechanisms are disclosed for predicting molecular electronic structural information. According to one embodiment, quantum simulation results are generated for a molecule based on a quantum simulation of an electronic structure of the molecule. The quantum simulation of the electronic structure of the molecule is performed with quantum processing units. An input vector comprising data field values derived from the quantum simulation results for the molecule is created. An electronic structural information prediction model is applied to generate, based at least in part on the input vector, predicted electronic structural information for the molecule.

First claim

What is claimed is:

1. A method for predicting electronic structural information of a molecule, the method comprising:
generating quantum simulation results for the molecule based on a quantum simulation of an electronic structure of the molecule, wherein the quantum simulation of the electronic structure of the molecule is performed with one or more quantum processing units (QPUs);
creating an input vector comprising data field values derived from the quantum simulation results for the molecule; and
applying an electronic structural information prediction model to generate, based at least in part on the input vector, predicted electronic structural information for the molecule, wherein the electronic structural information prediction model is trained with training data instances corresponding to other types of molecules than said molecule, wherein a training data instance in the training data instances includes (a) a second quantum simulation result of a second electronic structure of a second molecule generated with a quantum computer and (b) a classical computing result of the second electronic structure of the second molecule generated with a classic computer.

2. The method of claim 1 , wherein the quantum simulation comprises: one or more of: molecular properties of the molecule, atomic properties of atoms present in the molecule, presence indications of atoms in the molecule, numbers of types of atoms present in the molecule, one or more bond lengths, and/or quantum simulation energies.

3. The method of claim 1 , wherein a second electronic structural information prediction model is applied to generate, based at least in part on the quantum simulation results and the predicted electronic structural information generated by the electronic structural information prediction model for the molecule, second electronic structural information for the molecule.

4. The method of claim 1 , wherein the electronic structural information prediction model is implemented with an artificial neural network representing a multi-layer perceptron.

5. The method of claim 1 , wherein the electronic structural information prediction model is trained by training data in a model training phase; wherein the training data is derived at least in part from quantum simulation results for one or more molecules based on quantum simulation of electronic structure of the one or more molecules; wherein the training data is further derived at least in part from classically determined electronic structural information for the one or more molecules generated by one or more classical computers; wherein differences derived from the quantum simulation results for the one or more molecules in the training data and the classically determined electronic structural information are used to optimize the electronic structural information prediction model.

6. The method of claim 1 , wherein the predicted electronic structural information for the molecule at least includes predicted full configuration interaction energies for the molecule.

7. The method of claim 1 , wherein the quantum simulation of electronic structure of the molecule is carried out with a quantum annealing process implemented by the one or more QPUs.

8. One or more non-transitory computer readable media storing a program of instructions that is executable by a device to perform a method for predicting electronic structural information of a molecule, the method comprising:
generating quantum simulation results for the molecule based on a quantum simulation of an electronic structure of the molecule, wherein the quantum simulation of the electronic structure of the molecule is performed with one or more quantum processing units (QPUs);
creating an input vector comprising data field values derived from the quantum simulation results for the molecule; and
applying an electronic structural information prediction model to generate, based at least in part on the input vector, predicted electronic structural information for the molecule, wherein the electronic structural information prediction model is trained with training data instances corresponding to other types of molecules than said molecule, wherein a training data instance in the training data instances includes (a) a second quantum simulation result of a second electronic structure of a second molecule generated with a quantum computer and (b) a classical computing result of the second electronic structure of the second molecule generated with a classic computer.

9. The media of claim 8 , wherein the quantum simulation comprises: one or more of:
molecular properties of the molecule, atomic properties of atoms present in the molecule, presence indications of atoms in the molecule, numbers of types of atoms present in the molecule, one or more bond lengths, and/or quantum simulation energies.

10. The media of claim 8 , wherein a second electronic structural information prediction model is applied to generate, based at least in part on the quantum simulation results and the predicted electronic structural information generated by the electronic structural information prediction model for the molecule, second electronic structural information for the molecule.

11. The media of claim 8 , wherein the electronic structural information prediction model is implemented with an artificial neural network representing a multi-layer perceptron.

12. The media of claim 8 , wherein the electronic structural information prediction model is trained by training data in a model training phase; wherein the training data is derived at least in part from quantum simulation results for one or more molecules based on quantum simulation of electronic structure of the one or more molecules; wherein the training data is further derived at least in part from classically determined electronic structural information for the one or more molecules generated by one or more classical computers; wherein differences derived from the quantum simulation results for the one or more molecules in the training data and the classically determined electronic structural information are used to optimize the electronic structural information prediction model.

13. The media of claim 8 , wherein the predicted electronic structural information for the molecule at least includes predicted full configuration interaction energies for the molecule.

14. The media of claim 8 , wherein the quantum simulation of electronic structure of the molecule is carried out with a quantum annealing process implemented by the one or more QPUs.

15. A system, comprising:
one or more computing processors;
one or more non-transitory computer readable media storing a program of instructions that is executable by the one or more computing processors to perform a method for predicting electronic structural information of a molecule, the method comprising:
generating quantum simulation results for the molecule based on a quantum simulation of an electronic structure of the molecule, wherein the quantum simulation of the electronic structure of the molecule is performed with one or more quantum processing units (QPUs);
creating an input vector comprising data field values derived from the quantum simulation results for the molecule; and
applying an electronic structural information prediction model to generate, based at least in part on the input vector, predicted electronic structural information for the molecule, wherein the electronic structural information prediction model is trained with training data instances corresponding to other types of molecules than said molecule, wherein a training data instance in the training d

Assignees

• Volkswagen Ag

Inventors

• Neukart Florian
• Streif Michael
• Von Dollen David
• Graf Tanja
• Schladt Thomas
• Voigt Arne-Christian
• Mueller Jonathan Edward

Classifications

• G06N10/20

  Models of quantum computing, e.g. quantum circuits or universal quantum computers · CPC title

• G06N3/0499

  Feedforward networks · CPC title

• G06N3/09

  Supervised learning · CPC title

• G06N3/0985

  Hyperparameter optimisation; Meta-learning; Learning-to-learn · CPC title

• G06N3/08

  Learning methods · CPC title