Density Operators and Stochastic Evolution

Open-system computation requires a choice of state representation, physical noise model, and numerical solver. This chapter connects those choices without treating them as interchangeable. It teaches finite-dimensional algorithms and their checks; the physical derivation of Markovian evolution belongs to the Lindblad–GKSL treatment.

Enter this chapter

Start with density matrices in code: basis ordering, conjugation, trace, positivity, and partial traces. Then read superoperator vectorization before assembling a Liouvillian. A wrong stacking convention can generate dimensionally correct arrays with incorrect coherent evolution.

For prescribed discrete operations, use quantum channels to validate Kraus and Choi representations, then density-matrix simulation to apply local channels without forming enormous full-register superoperators.

For continuous evolution, Lindblad solvers provides deterministic propagation and steady states. Effective non-Hermitian Hamiltonians supplies conditional survival probabilities; quantum trajectories adds jump records and ensemble uncertainty.

Match the algorithm to the requested output

A density matrix gives ensemble expectations directly, with $d^2$ complex state entries. A pure-state trajectory needs $d$ entries, but an ensemble of $R$ trajectories introduces typical observable uncertainty proportional to $R^{-1/2}$. Which method is cheaper depends on the required accuracy and the number of observables, not dimension alone.

If the question concerns an individual monitored record, the unraveling is part of the model. Diffusive stochastic equations describes Wiener-driven pure states; continuous-measurement feedback handles conditional filters and causal control. An unconditioned master equation by itself does not specify a unique measurement record.

Quantum noise simulation treats rates and temporal correlations. Circuit noise models treats error placement, coherent overrotations, readout confusion, and leakage assumptions. Measurement sampling converts exact probabilities into finite-shot statistics.

A shared benchmark that separates errors

Use an undriven qubit with $L=\sqrt\gamma|0\rangle\langle1|$. Starting from the excited state, every approach should reproduce

$$P_e(t)=e^{-\gamma t}.$$

The exact finite-time channel has damping probability $p=1-e^{-\gamma t}$. A master-equation integrator approximates this curve with a tolerance-dependent bias. A jump ensemble estimates it with variance $P_e(1-P_e)/R$. Terminal measurement shots add a distinct sampling layer if requested.

Next use a complex superposition and nonzero level splitting. That extension tests off-diagonal phases and conjugation errors invisible to a population-only benchmark. For diffusive monitoring, use a dephasing qubit and compare the averaged conditional coherence with deterministic dephasing.

Count independent records before counting shots

Suppose trajectory $r$ produces a conditional event probability $q_r$, and one samples $S$ independent terminal shots from that state. Different trajectories are independent, but shots that reuse one trajectory share the random $q_r$. For the average event frequency over $R$ trajectories, the law of total variance gives

$$\operatorname{Var}(\hat p)= \frac{\operatorname{Var}(q_r)}{R} +\frac{\mathbb E[q_r(1-q_r)]}{RS}.$$

The first term is trajectory uncertainty and survives as $S\to\infty$. Using the exact probability $q_r$ instead of terminal shots removes the second term without changing the ensemble mean. The measurement-sampling page treats the categorical and conditional-state algorithms.

For the excited-state amplitude-damping jump benchmark, a trajectory ends in either $|1\rangle$ or $|0\rangle$, so $q_r$ is exactly one or zero. Repeated energy measurements on copies of that same conditional state add no new information about the trajectory ensemble. At $P_e=1/2$, $R=10^4$ independent trajectories give standard error $0.005$, whether one draws one or one hundred terminal shots per trajectory. Treating the million resulting shots as independent would understate that error by a factor of ten. A deterministic density-matrix calculation followed by independently prepared terminal shots has a different sampling structure and the ordinary binomial variance.

What counts as a controlled result

Keep four questions separate: Is the model physically appropriate? Is its finite-dimensional representation converged? Is the numerical propagation accurate? Is the stochastic estimate precise enough? Trace and positivity address part of the third question; they do not establish the first two.

A useful report states the collapse operators with their rates, basis cutoff, integration settings, independent trajectory count or shot count, and uncertainty in the requested observable. Show refinement in each relevant parameter. Preserve raw diagnostic failures before any renormalization or positivity repair.

References

• Dalibard, J., Castin, Y., and Mølmer, K., “Wave-function approach to dissipative processes in quantum optics,” Physical Review Letters 68, 580–583 (1992), doi:10.1103/PhysRevLett.68.580.
• Jacobs, K., and Steck, D. A., “A straightforward introduction to continuous quantum measurement,” Contemporary Physics 47, 279–303 (2006), arXiv:quant-ph/0611067.
• QuTiP developers, “Time Evolution and Quantum System Dynamics”, QuTiP documentation — available solver families.