Gate generation for open quantum systems via a monotonic algorithm with time optimization

TL;DR

This monotonic algorithm, based on Lyapunov control techniques, is shown to have a straightforward interpretation in terms of optimal control: its stationary conditions coincides with the first-order optimality conditions for a cost depending linearly on the final values of the quantum states.

Abstract

We present a monotonic numerical algorithm including time optimization for generating quantum gates for open systems. Such systems are assumed to be governed by Lindblad master equations for the density operators on a large Hilbert-space whereas the quantum gates are relative to a sub-space of small dimension. Starting from an initial seed of the control input, this algorithm consists in the repetition of the following two steps producing a new control input: (A) backwards integration of adjoint Lindblad-Master equations (in the Heisenberg-picture) from a set of final conditions encoding the quantum gate to generate; (B) forward integration of Lindblad-Master equations in closed-loop where a Lyapunov based control produced the new control input. The numerical stability is ensured by the stability of both the open-loop adjoint backward system and the forward closed-loop system. A clock-control input can be added to the usual control input. The obtained monotonic algorithm allows then to optimise not only the shape of the control imput, but also the gate time. Preliminary numerical implementations indicate that this algorithm is well suited for cat-qubit gates, where Hilbert-space dimensions (2 for the Z-gate and 4 for the CNOT-gate) are much smaller than the dimension of the physical Hilbert-space involving mainly Fock-states (typically 20 or larger for a single cat-qubit). This monotonic algorithm, based on Lyapunov control techniques, is shown to have a straightforward interpretation in terms of optimal control: its stationary conditions coincides with the first-order optimality conditions for a cost depending linearly on the final values of the quantum states.

Figures (4)

• Figure 1: Z-gate generation. Left side: evolution of $T_f$ and of infidelity $\mathcal{I}$ (see \ref{['eq:infidelity']}). In these simulations $T_{f}^{(0)} = 5$, $T_{f}^{(0)} = 0.85$, and $T_{f}^{(0)}=0.5$ ($\kappa_1=1/100$). Right side: infidelity that is obtained for each iteration of the algorithm for $T_{f}^{(0)} = 0.85$, and $T_{f}^{(0)}=0.5$.
• Figure 2: Z-gate generation. Top: Final control pulses for $T_{f}^{(0)} = 5$, for $T_{f}^{(0)} = 0.85$, and for $T_{f}^{(0)} =0.5$ ($\kappa_1=1/100$). Botton: The seed inputs for each case, $T_{f}^{(0)} = 5$, for $T_{f}^{(0)} = 0.85$, and for $T_{f}^{(0)} =0.5$ (seed inputs are slightly perturbed constant adiabatic control \ref{['eq:adiabcontrol']}).
• Figure 3: CNOT-gate. Top: Evolution of gate-infidelity along the steps of the algorithm. The (uncorrected) infidelity (see \ref{['eq:infidelity']}) of the last step is given by $0.9784\times 10^{-3}$. The corrected infidelity of the last step is $0.98655\times 10^{-3}$ (the variation between the corrected and the uncorrected value is $\approx 0.84\%$). Botton: The evolution of the gate-time $T_f$ along the steps of the algorithm. The gate-time for the last step is $T_f = 1.259$.
• Figure 4: CNOT-gate. Top: Final control pulses generated by the algorithm of section refsClock. The constant adiabatic control for the same $T_f$ of the final control pulses is also presented along with the seed of the algorithm, which is a slightly perturbed adiabatic control defined for a initial gate-time $T_{f}^{(0)} = 1.5 s$. Botton: Infidelity that was obtained after simulating a constant adiabatic control presented as a function of gate-time $T_f$. The minimum infidelity is close to $0.00128$ for $T_f =1.8$, whereas the infidelity obtained with algorithm is $0.00098655$ with a shorteroptimized time $T_f = 1.259$ .