arXiv:2610.26725
Quantum Optimal Control Primer
Ignacio R. Sola
Abstract
This is a set of detailed notes that are part of a Ph.D. Course on Quantum Control delivered at the Department of Chemistry of Seoul National University in 2014. Part of this course was also taught in the Department of Chemistry of the Friedrich Schiller University Jena. It focuses on quantum optimal control and quantum optimal control algorithms based on the gradient and extended gradient methods.
AI Summary
These lecture notes, based on a 2014 Ph.D. course, explain how to formulate quantum optimal control and construct iterative pulse-optimization algorithms. The focus is coherent control of isolated molecular systems by classical, linearly polarized laser fields, using the dipole Hamiltonian $H(t)=H_0-\mu\epsilon(t)$. The notes derive the field gradient for a final-time objective $\langle\psi(T)|P|\psi(T)\rangle$ with a quadratic fluence penalty, where $P$ is positive semidefinite. Rather than introducing Lagrange multipliers initially, the derivation explicitly relates state variations to field variations through the inhomogeneous Schrödinger equation. A backward-propagated adjoint state gives the implicit stationarity condition $\epsilon_{\mathrm{oc}}(t)=-Tf^2(t)\operatorname{Im}\langle\chi(t)|\mu|\psi(t)\rangle$. The algorithm discussion covers gradient ascent with line search, Krotov and nested-Krotov updates, Zhu–Rabitz schemes, and the two-point boundary-value quantum control paradigm. Finite differences of the objective and dynamical invariants are used to explain monotonic-improvement conditions, including a two-parameter Krotov family with parameters in $[0,2]$. Local control and tracking are also derived and contrasted with final-time optimization. The treatment excludes quantum fields, decoherence, and spatial laser structure, and emphasizes state-transfer or observable objectives rather than full gate synthesis. It notes that naive fixed-point iteration generally fails, finite-time-step implementations require care, local control is generally not optimal, and tracking can become singular when the field-dependent response vanishes. No numerical benchmarks or experimental demonstrations are reported.
Error found: Lecture 2 Eq. (grad2l2) omits a factor of two in the derivative of the quadratic field penalty, so the update defined using that equation is not the stated functional's gradient-ascent update. The Krotov monotonicity calculation omits the cross term −g_tilde epsilon^(k); the correct integrand is (Delta epsilon)²/(T f²), which supports nondecrease but not the claimed strict increase at fixed points. The TBQCP fast-kick-off prescription prints the impossible parameter interval 0 ≥ gamma ≥ 1.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.