Selecting parameters for a quantum approximate optimization algorithm (QAOA)
In a general aspect, values of input parameters for a quantum approximate optimization algorithm (QAOA) are selected by a Bayesian optimizer. The QAOA can be configured to solve a combinatorial optimization problem (COP), such as Maximum Cut. A hybrid classical-quantum computing system can be used to execute the QAOA and select the input parameters.
1. A method comprising:
obtaining a first set of output values from a first execution of a quantum approximate optimization algorithm (QAOA), wherein the first set of output values is generated by a quantum processor unit based on a first set of input values for the QAOA;
based on the first set of output values, using a Bayesian optimizer running on one or more classical processor units to select a second set of input values for the QAOA; and
obtaining a second set of output values from a second execution of the QAOA, wherein the second set of output values is generated by the quantum processor unit based on the second set of input values for the QAOA.
2. The method of claim 1 , wherein the Bayesian optimizer comprises a set of distributions for respective input parameters of the QAOA, and the method comprises:
updating the set of distributions based on the first set of input values and the first set of output values, using a Gaussian process model; and
using the updated distributions to select the second set of input values for the QAOA.
3. The method of claim 2 , wherein the distributions for the respective input parameters are Gaussian distributions for respective parameters in the QAOA.
4. The method of claim 2 , further comprising executing an iterative process, where respective iterations of the iterative process include:
obtaining a new set of output values from an execution of the QAOA on the quantum processor unit;
updating the distributions of the Bayesian optimizer based on the new set of output values; and
selecting a new set of input values for the QAOA based on the updated distributions.
5. The method of claim 1 , wherein the QAOA is configured to solve a combinatorial optimization problem (COP), wherein the COP comprises a cost operator, and the COP is encoded in the cost operator.
6. The method of claim 5 , wherein the QAOA is configured to solve a Maximum Cut problem.
7. The method of claim 5 , wherein the cost operator comprises an Ising Hamiltonian, and the method comprises evaluating the Ising Hamiltonian based on the first and second sets of output values.
8. A hybrid computer system comprising a quantum processor unit (QPU) and one or more classical processor units, the hybrid computer system being configured to perform operations comprising:
obtaining a first set of output values from a first execution of a quantum approximate optimization algorithm (QAOA), wherein the first set of output values is generated by the QPU based on a first set of input values for the QAOA;
based on the first set of output values, using a Bayesian optimizer running on the one or more classical processor units to select a second set of input values for the QAOA; and
obtaining a second set of output values from a second execution of the QAOA, wherein the second set of output values is generated by the QPU based on the second set of input values for the QAOA.
9. The system of claim 8 , wherein the Bayesian optimizer comprises a set of distributions for respective input parameters of the QAOA, and the system is further configured to perform operations comprising:
updating the set of distributions based on the first set of input values and the first set of output values, using a Gaussian process model running on the one or more classical processor units; and
using the updated distributions to select the second set of input values for the QAOA.
10. The system of claim 9 , wherein the distributions for the respective input parameters are Gaussian distributions for respective parameters in the QAOA.
11. The system of claim 8 , wherein the hybrid computer system is configured to perform operations further comprising executing an iterative process, where respective iterations of the iterative process include:
obtaining a new set of output values from an execution of the QAOA on the quantum processor unit;
updating, by operation of the one or more classical processor units, the distributions of the Bayesian optimizer based on the new set of output values; and
selecting, by operation of the one or more classical processor units, a new set of input values for the QAOA based on the updated distributions.
12. The system of claim 8 , wherein the QAOA is configured to solve a combinatorial optimization problem (COP), wherein the COP comprises a cost operator, and the COP is encoded in the cost operator.
13. The system of claim 8 , wherein the QAOA is configured to solve a Maximum Cut problem.
14. The system of claim 8 , wherein the cost operator comprises an Ising Hamiltonian, and wherein the hybrid computer system is configured to perform operations further comprising evaluating, by the one or more classical processor unit, the Ising Hamiltonian based on the first and second sets of output values.