Constraint optimization and $\mathcal{SU}(N)$ quantum control landscapes

TL;DR

This paper introduces an embedded gradient method for optimizing quantum control landscapes on the special unitary group, revealing the presence of local extrema that are not global for dimensions N≥5.

Contribution

It develops a new embedded gradient vector field approach for $\\mathcal{SU}(N)$ and fully solves the associated optimization problem, showing the landscape's trap structure.

Findings

The landscape is not trap-free for N≥5.

Existence of kinematic local extrema that are not global.

Complete solution to the optimization problem on $\\mathcal{SU}(N)$.

Abstract

We develop the embedded gradient vector field method, introduced in [8] and [9], for the case of the special unitary group $\mathcal{SU}(N)$ regarded as a constraint submanifold of the unitary group $\mathcal{U}(N)$. The optimization problem associated to the trace fidelity cost function defined on $\mathcal{SU}(N)$ that appears in the context of $\mathcal{SU}(N)$ quantum control landscapes is completely solved using the embedded gradient vector field method. We prove that for $N\geq 5$, the landscape is not $\mathcal{SU}(N)$-trap free, there are always kinematic local extrema that are not global extrema.