Quantum Control and Feedback

Quantum control uses drives, pulses, measurements, and engineered couplings to produce a desired state or operation. In an open system, the target must be reached despite noise and with finite amplitude, bandwidth, time, and detector efficiency. Sometimes dissipation limits the control; sometimes a designed dissipative process is the tool that prepares the state.

Required background. Time-Dependent Hamiltonians supplies controlled coherent motion; Lindblad–GKSL Equation supplies a common reduced model. Noise Spectra and Measurement Records describe environmental fluctuations and observed signals.

Specify the target and the allowed controls

A common starting model is

$$\dot\rho={}-\frac{i}{\hbar} \left[H_0+\sum_j u_j(t)H_j,\rho\right] +\sum_k\mathcal D[L_k]\rho.$$

with rates absorbed into the jump operators. The functions $u_j(t)$ must obey the actual control constraints. Strong driving can also change the correct dissipative model, so retaining fixed jump operators under arbitrary pulses is an assumption, not a universal rule.

A state-preparation target concerns specified input states and a desired output. A gate target concerns the operation on a sufficiently rich set of inputs, including its action on coherence. A pulse that transfers one state perfectly has not thereby implemented a high-fidelity gate.

Start with Driven Open Systems and Rabi and Ramsey Control. Then use Optimal Control to formulate the objective, parameter uncertainty, leakage, and waveform constraints before optimizing.

Pulses suppress some noise, not every noise process

For longitudinal classical dephasing in an ideal toggling frame,

$$\begin{gathered} H_{\rm noise}(t)=\frac{\hbar}{2}y(t)\xi(t)Z,\\ \phi(T)=\int_0^T y(t)\xi(t)\,dt. \end{gathered}$$

A single ideal echo pulse switches $y$ from $+1$ to $-1$ halfway through the interval. A constant unknown detuning then contributes equal opposite phases and cancels. Rapid independent fluctuations need not cancel, and finite pulses bring their own imperfections.

Dynamical Decoupling explains the frequency filter; Pulse Sequences compares pulse timing, phases, and error sensitivity. Ramsey, echo, and more elaborate sequences sample different parts of the spectrum, so their coherence times are protocol dependent. Ordinary phase-refocusing pulses do not generally undo irreversible Markovian energy relaxation.

Choose whether the controller uses a record

Strategy	Information and physical action
Open-loop driving	A waveform is chosen before the run, with no contemporaneous detector record.
Measurement-Based Feedback	A classical record determines later controls, with backaction, estimation error, efficiency, and delay included.
Coherent Feedback	A field or auxiliary quantum system is routed back without first converting that signal into a classical record.

For feedback, use only data available when a control is applied. A smoothed estimate that uses future data cannot serve as a real-time control signal. Feedback from Measurement Records connects filtering to causal control.

Coherent feedback is also constrained by propagation delay, port normalization, loss, and stability. It does not remove all noise merely because it avoids an intermediate measurement. The appropriate model may require retaining auxiliary modes or delay instead of eliminating them into instantaneous coefficients.

Use dissipation to prepare or constrain motion

Reservoir Engineering designs environmental couplings, often through lossy auxiliaries. Dissipative State Preparation asks whether the desired state is stationary, reachable from the intended initial set, and attractive on a useful timescale.

These are distinct conditions. A dark state may coexist with other dark states, and a Hamiltonian can move it out of the dark sector. Even a unique attractive state can be reached too slowly to overcome unwanted noise. Compare the preparation rate, competing errors, and relevant initial states.

Quantum Zeno Dynamics uses frequent measurement or strong dissipation to suppress transitions out of a subspace. The constrained system can still move within that subspace; “Zeno” need not mean that all dynamics freezes. Control Limits from Noise collects the limitations shared by these strategies.

For a numerical experiment, Optimal Control Toy Problems compares candidate controls against explicit targets and constraints. Test the optimized waveform on a refined solver and perturbed parameters, rather than only on the model and grid used by the optimizer.

Exercises

Echo and white noise. Let $\langle\xi(t)\xi(s)\rangle=2D\delta(t-s)$ and $|y(t)|=1$. Find $\operatorname{Var}\phi(T)$ for ideal sign-flip pulses and compare with no pulse.

Solution

The double integral over the correlation gives $\operatorname{Var}\phi=2D\int_0^T y(t)^2dt=2DT$. Ideal sign flips do not change this variance, unlike their cancellation of a quasi-static detuning. For Gaussian noise, the coherence envelope is $e^{-\operatorname{Var}\phi/2}=e^{-DT}$ in this convention.

A target that is not uniquely prepared. With $H=0$ and collective dephasing jump $L=Z_1+Z_2$, can dissipation alone prepare $(|01\rangle+|10\rangle)/\sqrt2$ from every state?

Solution

No. Both $|01\rangle$ and $|10\rangle$ are annihilated by $L$, so every state supported on their span is unchanged. In particular, an initial $|01\rangle$ stays there rather than becoming the target superposition. Protection of a subspace does not choose a unique state within it.

References

• D’Alessandro, D. Introduction to Quantum Control and Dynamics. Chapman & Hall/CRC, 2007. Control objectives and coherent dynamics.
• Jacobs, K. Quantum Measurement Theory and its Applications. Cambridge University Press, 2014. State estimation and measurement-based control.
• Wiseman, H. M., and G. J. Milburn. Quantum Measurement and Control. Cambridge University Press, 2010. Feedback, noise, and engineered open dynamics.