Sandia National Labs FY20 LDRD Annual Report

FY20 ANNUAL REPORT

Efficient tomography of many qubit quantum processors is a state-of-the-art advance. The development and evaluation of efficient characterization of many-qubit quantum processors are key to enabling quantum computing hardware. This project focused on the development of multi-qubit tomography protocols that allow for increasing the size (number of qubits) and quality (and hence utility) of quantum processors. The ability to perform useful tomography on 10+ qubits represents a significant advance in the state-of-the-art, which was only two qubits prior to this LDRD. The work will

benefit national security missions through its potential to achieve and accelerate applications of quantum computing including cryptography, simulation, and computing beyond Moore’s Law. (PI: Erik Nielsen)

Pictorial view of approach: PyGSTi provides efficient forward simulations of quantum circuits, including classes of cloud noise, that enables comparison to experimental data from emerging quantum processors.

Learning hidden structure in multifidelity information sources for efficient uncertainty quantification. An automated learning framework was developed for efficient analysis and prediction of complex systems that uses multiple sources of data of varying trustworthiness and cost. The Sandia team’s approach maximized confidence in predictions by employing approximation

strategies that fuse heterogeneous experimental and numerical simulation data in a way that maximizes information gain subject to resource limitations. Predictions are endowed with probabilistic estimates of error, which can be used to create trusted analytics and inform the future design of experiments. (PI: John Davis Jakeman) Multifidelity (MF) approximation built using both 1D and 2D low-fidelity and lower-fidelity information sources. The green dashed line is the MF approximation of the high-fidelity function (black line) using two high-fidelity evaluations (black dots) and 10 evaluations of each low-fidelity information source (red and blue dots). The gray region is the original uncertainty, and the green region is the new uncertainty.

Solving specific mixed-integer partial differential equation constrained optimization problems. Sandia and collaborator Georgia Tech created the first highly parallel code for solving mixed-integer PDE-constrained optimization (MIPDECO). Partial differential equations, or PDEs, describe the behavior of objects under constraints such as physics and chemistry. PDE-constrained optimization involves making choices of parameters to solve an inverse, control or design problem involving a physical system. When some of the variables in PDE-constrained optimization must take discrete values, the problem becomes MIPDECO. For example, discrete choices might involve a fixed set of controls on an additive manufacturing machine, selecting materials or making yes/ no decisions. The code, called ROL-PEBBL, is C++ and MPI-based. It combines a code to efficiently search over integer choices (PEBBL = Parallel Enumeration Branch-and-Bound Library) and a code for efficient nonlinear optimization, including PDE-constrained optimization (ROL = Rapid Optimization

48

LABORATORY DIRECTED RESEARCH & DEVELOPMENT