Dynamically Optimal Quantum Jump Process

• Dynamically Optimal Quantum Jump Process (DO-QJP) is an adaptive quantum-jump approach that minimizes jump rates in Markovian open quantum systems.
• It employs a resummation of the Lindblad jump series with state-dependent shifts to ensure complete positivity and rapid convergence.
• DO-QJP facilitates efficient numerical simulations and analytic approximations while providing insights into decoherence and the emergence of classicality.

Searching arXiv for papers on Dynamically Optimal Quantum Jump Process and closely related unraveling schemes. The Dynamically Optimal Quantum Jump Process (DO-QJP) is an adaptive quantum-jump construction for Markovian open quantum systems in which the decomposition parameters of a Lindblad generator are chosen at every step so as to minimize instantaneous jump rates. In the formulation of Lucas and Hornberger, the resulting expansion is an adaptive resummation of the Lindblad jump series that remains completely positive order by order and typically converges within the lowest two to five orders, thereby supporting both analytic approximation and efficient numerical simulation (Lucas et al., 2013). Later work uses the same designation for a stochastic unraveling chosen to minimize the short-time growth of the variance of an observable within a parametric family of jump-process schemes (Cao et al., 24 Sep 2025). A plausible implication is that DO-QJP is best understood as a class of state-adaptive optimization principles for quantum-jump representations, rather than as a single fixed algorithm.

1. Markovian setting and jump-series representation

The starting point is a time-local, completely positive and trace-preserving generator $\mathcal{L}(t)$ acting on the density operator $\rho_t$,

$\partial_t \rho_t = \mathcal{L}(t)\rho_t,$

with standard Lindblad form

$$\mathcal{L}(t)\rho = -\frac{i}{\hbar}[H,\rho] + \sum_k L_k \rho L_k^\dagger - \frac{1}{2}\{L_k^\dagger L_k,\rho\}.$$

The exact solution may be written formally as $\rho_t=\mathcal{U}(t,0)\rho_0$, where $$\mathcal{U}(t,0)=\mathcal{T}\exp\int_0^t \mathcal{L}(t')\,dt'$$ (Lucas et al., 2013).

A Dyson-like jump expansion follows from an arbitrary splitting

$\mathcal{L}(t)=\mathcal{L}_0(t)+\mathcal{J}(t),$

which yields the integral equation

$$\rho_t=\mathcal{U}_0(t,0)\rho_0+\int_0^t dt_1\,\mathcal{U}_0(t,t_1)\mathcal{J}(t_1)\rho_{t_1}.$$

Iteration gives

$\rho_t=\sum_{n=0}^{\infty}\rho_t^{(n)},$

with

$$\rho_t^{(0)}=\mathcal{U}_0(t,0)\rho_0,\qquad \rho_t^{(n)}=\int_0^t dt_n\,\mathcal{U}_0(t,t_n)\mathcal{J}(t_n)\rho_{t_n}^{(n-1)}.$$

Each term $\rho_t$0 contains exactly $\rho_t$1 insertions of the jump superoperator $\rho_t$2, while the factors $\rho_t$3 describe continuous evolution between jumps (Lucas et al., 2013).

The nontrivial issue is not the existence of such a series, but the choice of decomposition. Lucas and Hornberger introduce a physically adapted Lindblad decomposition parameterized by complex shifts $\rho_t$4, so that the expansion is manifestly completely positive order by order. This establishes the basic framework in which DO-QJP is defined (Lucas et al., 2013).

2. Adaptive resummation and the dynamically optimal choice

To preserve positivity term by term, the jump operators are shifted as

$\rho_t$5

while the Hamiltonian is compensated by

$\rho_t$6

The Lindblad form is invariant under this transformation. One then defines

$\rho_t$7

and

$\rho_t$8

with

$\rho_t$9

and the anti-commutator-reversed bracket $\partial_t \rho_t = \mathcal{L}(t)\rho_t,$0 (Lucas et al., 2013).

The optimization criterion is phrased in terms of the weights

$\partial_t \rho_t = \mathcal{L}(t)\rho_t,$1

which satisfy $\partial_t \rho_t = \mathcal{L}(t)\rho_t,$2. Their dynamics is governed by the positive jump-rate operator

$\partial_t \rho_t = \mathcal{L}(t)\rho_t,$3

through the cascade equation

$\partial_t \rho_t = \mathcal{L}(t)\rho_t,$4

The expansion is convergent iff the low-$\partial_t \rho_t = \mathcal{L}(t)\rho_t,$5 weights dominate (Lucas et al., 2013).

A finer description resolves each $\partial_t \rho_t = \mathcal{L}(t)\rho_t,$6-jump contribution into branches labeled by the full jump record $\partial_t \rho_t = \mathcal{L}(t)\rho_t,$7. For each branch, the partial rates are

$\partial_t \rho_t = \mathcal{L}(t)\rho_t,$8

Minimizing these partial rates with respect to $\partial_t \rho_t = \mathcal{L}(t)\rho_t,$9 gives the optimal shift

$$\mathcal{L}(t)\rho = -\frac{i}{\hbar}[H,\rho] + \sum_k L_k \rho L_k^\dagger - \frac{1}{2}\{L_k^\dagger L_k,\rho\}.$$0

and the minimal partial rate becomes

$$\mathcal{L}(t)\rho = -\frac{i}{\hbar}[H,\rho] + \sum_k L_k \rho L_k^\dagger - \frac{1}{2}\{L_k^\dagger L_k,\rho\}.$$1

Accordingly, the optimal jump operators are

$$\mathcal{L}(t)\rho = -\frac{i}{\hbar}[H,\rho] + \sum_k L_k \rho L_k^\dagger - \frac{1}{2}\{L_k^\dagger L_k,\rho\}.$$2

updated adaptively after each jump. Equivalently, one calls the resulting expansion the Dynamically Optimal Quantum Jump Process (Lucas et al., 2013).

3. Convergence mechanism and low-order truncation

The central convergence argument is structural. Because the weights obey the cascade relation above, and because $$\mathcal{L}(t)\rho = -\frac{i}{\hbar}[H,\rho] + \sum_k L_k \rho L_k^\dagger - \frac{1}{2}\{L_k^\dagger L_k,\rho\}.$$3 has been minimized at each step, all higher-order rates $$\mathcal{L}(t)\rho = -\frac{i}{\hbar}[H,\rho] + \sum_k L_k \rho L_k^\dagger - \frac{1}{2}\{L_k^\dagger L_k,\rho\}.$$4 are as small as possible. In many physically relevant problems the system rapidly localizes into a pointer basis of eigenstates of the dominant $$\mathcal{L}(t)\rho = -\frac{i}{\hbar}[H,\rho] + \sum_k L_k \rho L_k^\dagger - \frac{1}{2}\{L_k^\dagger L_k,\rho\}.$$5; for such states, $$\mathcal{L}(t)\rho = -\frac{i}{\hbar}[H,\rho] + \sum_k L_k \rho L_k^\dagger - \frac{1}{2}\{L_k^\dagger L_k,\rho\}.$$6 equals the eigenvalue, and the optimal partial rate vanishes exactly. Even when it does not vanish exactly, it becomes very small after a few jumps. Hence the expansion typically freezes beyond some small $$\mathcal{L}(t)\rho = -\frac{i}{\hbar}[H,\rho] + \sum_k L_k \rho L_k^\dagger - \frac{1}{2}\{L_k^\dagger L_k,\rho\}.$$7 and converges in only two to five orders (Lucas et al., 2013).

Convergence may be quantified by the fidelity

$$\mathcal{L}(t)\rho = -\frac{i}{\hbar}[H,\rho] + \sum_k L_k \rho L_k^\dagger - \frac{1}{2}\{L_k^\dagger L_k,\rho\}.$$8

which in numerics reaches $$\mathcal{L}(t)\rho = -\frac{i}{\hbar}[H,\rho] + \sum_k L_k \rho L_k^\dagger - \frac{1}{2}\{L_k^\dagger L_k,\rho\}.$$9 already for $\rho_t=\mathcal{U}(t,0)\rho_0$0 (Lucas et al., 2013).

Two standard illustrations summarize the reported behavior.

Model	DO-QJP behavior	Un-resummed comparison
Damped harmonic oscillator, $\rho_t=\mathcal{U}(t,0)\rho_0$1, $\rho_t=\mathcal{U}(t,0)\rho_0$2, $\rho_t=\mathcal{U}(t,0)\rho_0$3	$\rho_t=\mathcal{U}(t,0)\rho_0$4 by $\rho_t=\mathcal{U}(t,0)\rho_0$5 jumps	$\rho_t=\mathcal{U}(t,0)\rho_0$6 jumps for the same accuracy
Spatial decoherence	convergence by $\rho_t=\mathcal{U}(t,0)\rho_0$7	$\rho_t=\mathcal{U}(t,0)\rho_0$8 needed without resummation

For the damped harmonic oscillator at finite temperature,

$\rho_t=\mathcal{U}(t,0)\rho_0$9

with jump operators $$\mathcal{U}(t,0)=\mathcal{T}\exp\int_0^t \mathcal{L}(t')\,dt'$$0 and $$\mathcal{U}(t,0)=\mathcal{T}\exp\int_0^t \mathcal{L}(t')\,dt'$$1, the adaptive resummation gives the reported fidelity gain at low order (Lucas et al., 2013). For spatial decoherence,

$$\mathcal{U}(t,0)=\mathcal{T}\exp\int_0^t \mathcal{L}(t')\,dt'$$2

one again sees low-order convergence under DO-QJP (Lucas et al., 2013).

The significance of these examples is not merely numerical acceleration. The rapid suppression of higher-order terms reflects the underlying localization mechanism into pointer-like sectors, which is also why the same formalism is informative about decoherence structure and the emergence of classicality (Lucas et al., 2013).

4. Numerical realization and relation to trajectory methods

For the damped harmonic oscillator example, the implementation is described explicitly: one draws jump times $$\mathcal{U}(t,0)=\mathcal{T}\exp\int_0^t \mathcal{L}(t')\,dt'$$3, for example by Monte-Carlo sampling, propagates $$\mathcal{U}(t,0)=\mathcal{T}\exp\int_0^t \mathcal{L}(t')\,dt'$$4 under $$\mathcal{U}(t,0)=\mathcal{T}\exp\int_0^t \mathcal{L}(t')\,dt'$$5 from $$\mathcal{U}(t,0)=\mathcal{T}\exp\int_0^t \mathcal{L}(t')\,dt'$$6 to $$\mathcal{U}(t,0)=\mathcal{T}\exp\int_0^t \mathcal{L}(t')\,dt'$$7, then applies $$\mathcal{U}(t,0)=\mathcal{T}\exp\int_0^t \mathcal{L}(t')\,dt'$$8, and continues analogously; $$\mathcal{U}(t,0)=\mathcal{T}\exp\int_0^t \mathcal{L}(t')\,dt'$$9 is estimated as the average over many such realizations (Lucas et al., 2013). This produces a numerical scheme for efficient simulation, while retaining a direct connection to the analytic jump expansion.

The 2013 construction differs in emphasis from conventional pure-state trajectory approaches. It complements quantum-trajectory methods by lumping branches and optimizing mixed-state contributions rather than single pure-state trajectories (Lucas et al., 2013). This distinction matters because the optimization target is the convergence of the expansion weights $\mathcal{L}(t)=\mathcal{L}_0(t)+\mathcal{J}(t),$0, not the variance of a Monte Carlo estimator.

A later formulation places DO-QJP directly inside the theory of stochastic unravelings of Lindblad equations. In that setting one considers piecewise-deterministic Markov processes on the unit sphere,

$\mathcal{L}(t)=\mathcal{L}_0(t)+\mathcal{J}(t),$1

where $\mathcal{L}(t)=\mathcal{L}_0(t)+\mathcal{J}(t),$2 is a scalar Poisson process with state-dependent rate $\mathcal{L}(t)=\mathcal{L}_0(t)+\mathcal{J}(t),$3, and the average over pure-state trajectories reconstructs the density matrix,

$\mathcal{L}(t)=\mathcal{L}_0(t)+\mathcal{J}(t),$4

For one Lindblad operator and one noise term, a parametric family of norm-preserving jump-process unravelings is characterized in terms of functions $\mathcal{L}(t)=\mathcal{L}_0(t)+\mathcal{J}(t),$5, $\mathcal{L}(t)=\mathcal{L}_0(t)+\mathcal{J}(t),$6, and $\mathcal{L}(t)=\mathcal{L}_0(t)+\mathcal{J}(t),$7; the conventional QJP of Dalibard–Castin–Mølmer corresponds to $\mathcal{L}(t)=\mathcal{L}_0(t)+\mathcal{J}(t),$8, $\mathcal{L}(t)=\mathcal{L}_0(t)+\mathcal{J}(t),$9, $$\rho_t=\mathcal{U}_0(t,0)\rho_0+\int_0^t dt_1\,\mathcal{U}_0(t,t_1)\mathcal{J}(t_1)\rho_{t_1}.$$0 (Cao et al., 24 Sep 2025).

This later perspective recasts DO-QJP as an optimization problem over stochastic trajectories themselves. The conceptual continuity with the adaptive resummation approach is the state-dependent modification of jump structure, but the objective function is different.

5. Variance-optimal unravelings and generalized rate-operator schemes

In the variance-based formulation, the aim is to minimize the instantaneous growth of the classical variance

$$\rho_t=\mathcal{U}_0(t,0)\rho_0+\int_0^t dt_1\,\mathcal{U}_0(t,t_1)\mathcal{J}(t_1)\rho_{t_1}.$$1

for a fixed Hermitian observable $$\rho_t=\mathcal{U}_0(t,0)\rho_0+\int_0^t dt_1\,\mathcal{U}_0(t,t_1)\mathcal{J}(t_1)\rho_{t_1}.$$2. The analysis separates the observable-dependent part fixed by the Lindblad generator from a remainder $$\rho_t=\mathcal{U}_0(t,0)\rho_0+\int_0^t dt_1\,\mathcal{U}_0(t,t_1)\mathcal{J}(t_1)\rho_{t_1}.$$3 that depends on the unraveling. Dynamical optimality means minimizing $$\rho_t=\mathcal{U}_0(t,0)\rho_0+\int_0^t dt_1\,\mathcal{U}_0(t,t_1)\mathcal{J}(t_1)\rho_{t_1}.$$4 pointwise in $$\rho_t=\mathcal{U}_0(t,0)\rho_0+\int_0^t dt_1\,\mathcal{U}_0(t,t_1)\mathcal{J}(t_1)\rho_{t_1}.$$5 (Cao et al., 24 Sep 2025).

For the jump-process ansatz, the resulting DO-QJP is obtained by choosing $$\rho_t=\mathcal{U}_0(t,0)\rho_0+\int_0^t dt_1\,\mathcal{U}_0(t,t_1)\mathcal{J}(t_1)\rho_{t_1}.$$6 through a scalar minimization problem, saturating a user-prescribed maximum jump rate $$\rho_t=\mathcal{U}_0(t,0)\rho_0+\int_0^t dt_1\,\mathcal{U}_0(t,t_1)\mathcal{J}(t_1)\rho_{t_1}.$$7 via

$$\rho_t=\mathcal{U}_0(t,0)\rho_0+\int_0^t dt_1\,\mathcal{U}_0(t,t_1)\mathcal{J}(t_1)\rho_{t_1}.$$8

and then constructing $$\rho_t=\mathcal{U}_0(t,0)\rho_0+\int_0^t dt_1\,\mathcal{U}_0(t,t_1)\mathcal{J}(t_1)\rho_{t_1}.$$9, $\rho_t=\sum_{n=0}^{\infty}\rho_t^{(n)},$0, and $\rho_t=\sum_{n=0}^{\infty}\rho_t^{(n)},$1 accordingly. This choice makes the first quadratic term in the variance-growth expression vanish exactly and drives the residual jump term as small as allowed by the bound $\rho_t=\sum_{n=0}^{\infty}\rho_t^{(n)},$2 (Cao et al., 24 Sep 2025).

The same work also derives dynamically optimal quantum state diffusion (DO-QSD) and proves the local bound

$\rho_t=\sum_{n=0}^{\infty}\rho_t^{(n)},$3

so that, locally in time, even the best-tuned jump process cannot beat the variance growth of the optimal diffusion unraveling. Numerical experiments in that paper focus on DO-QSD rather than DO-QJP; DO-QJP was not tested numerically there because it requires extra tuning of $\rho_t=\sum_{n=0}^{\infty}\rho_t^{(n)},$4 and more complex formulas for $\rho_t=\sum_{n=0}^{\infty}\rho_t^{(n)},$5 (Cao et al., 24 Sep 2025).

A different extension is provided by generalized Rate-Operator quantum jumps via realization-dependent transformations. There the master equation is split into a jump part $\rho_t=\sum_{n=0}^{\infty}\rho_t^{(n)},$6 and a no-jump driving part $\rho_t=\sum_{n=0}^{\infty}\rho_t^{(n)},$7, and one exploits the freedom to add a realization-dependent counter-term $\rho_t=\sum_{n=0}^{\infty}\rho_t^{(n)},$8. The generalized rate operator is

$\rho_t=\sum_{n=0}^{\infty}\rho_t^{(n)},$9

with effective Hamiltonian

$$\rho_t^{(0)}=\mathcal{U}_0(t,0)\rho_0,\qquad \rho_t^{(n)}=\int_0^t dt_n\,\mathcal{U}_0(t,t_n)\mathcal{J}(t_n)\rho_{t_n}^{(n-1)}.$$0

Its eigenvalues $$\rho_t^{(0)}=\mathcal{U}_0(t,0)\rho_0,\qquad \rho_t^{(n)}=\int_0^t dt_n\,\mathcal{U}_0(t,t_n)\mathcal{J}(t_n)\rho_{t_n}^{(n-1)}.$$1 give instantaneous jump rates and its eigenvectors the post-jump states. The hard requirement for a positive unraveling is that all $$\rho_t^{(0)}=\mathcal{U}_0(t,0)\rho_0,\qquad \rho_t^{(n)}=\int_0^t dt_n\,\mathcal{U}_0(t,t_n)\mathcal{J}(t_n)\rho_{t_n}^{(n-1)}.$$2 for the relevant realizations (Settimo et al., 2024).

That framework does not introduce a single compact cost functional $$\rho_t^{(0)}=\mathcal{U}_0(t,0)\rho_0,\qquad \rho_t^{(n)}=\int_0^t dt_n\,\mathcal{U}_0(t,t_n)\mathcal{J}(t_n)\rho_{t_n}^{(n-1)}.$$3 or derive Euler–Lagrange equations. Instead, the freedom in $$\rho_t^{(0)}=\mathcal{U}_0(t,0)\rho_0,\qquad \rho_t^{(n)}=\int_0^t dt_n\,\mathcal{U}_0(t,t_n)\mathcal{J}(t_n)\rho_{t_n}^{(n-1)}.$$4 is used to optimize physically motivated performance measures such as Shannon entropy of occupation probabilities, total number of jumps per trajectory, classical memory required to track the ensemble, and maximal smoothness so that jumps are rare and far apart. In qubit examples, one chooses a one-parameter family $$\rho_t^{(0)}=\mathcal{U}_0(t,0)\rho_0,\qquad \rho_t^{(n)}=\int_0^t dt_n\,\mathcal{U}_0(t,t_n)\mathcal{J}(t_n)\rho_{t_n}^{(n-1)}.$$5, solves analytically for positivity conditions, and then selects $$\rho_t^{(0)}=\mathcal{U}_0(t,0)\rho_0,\qquad \rho_t^{(n)}=\int_0^t dt_n\,\mathcal{U}_0(t,t_n)\mathcal{J}(t_n)\rho_{t_n}^{(n-1)}.$$6 in the feasible interval to optimize the chosen metric (Settimo et al., 2024).

6. Interpretation, applications, and scope of the term

Several implications recur across these formulations. In the adaptive resummation picture, because the expansion adapts itself to the actual state at each jump, it automatically zeroes out further jumps once a pointer basis is reached. This renders otherwise large Hilbert-space problems rapidly tractable to analytic approximation in the lowest orders or to efficient numerical implementation (Lucas et al., 2013). The same machinery yields insight into pointer-state structure, minimal entropy production, and the emergence of classicality, and potential applications include steady-state engineering by incoherent control, large-scale open-system simulation, quantum metrology under continuous monitoring, and analytic modeling of decoherence in mesoscopic systems (Lucas et al., 2013).

In generalized rate-operator approaches, the state-dependent transformation can also be chosen so that all jump rates remain positive even in example cases where the corresponding dynamical map breaks the property of P-divisibility, thus allowing positive unravelings without reverse quantum jumps and without auxiliary degrees of freedom in several strongly non-Markovian examples (Settimo et al., 2024). This broadens the domain in which dynamically optimized jump descriptions may be constructed.

A common source of confusion is the meaning of “optimal.” In the 2013 jump expansion, optimality refers to minimizing instantaneous partial jump rates and thereby accelerating convergence of the resummed series (Lucas et al., 2013). In the 2025 unraveling theory, optimality refers to minimizing the short-time growth of the variance of an observable over a parametric family of Poisson-driven pure-state processes (Cao et al., 24 Sep 2025). In the generalized rate-operator formalism, no single variational principle is specified; instead, one optimizes a selected figure of merit under positivity constraints (Settimo et al., 2024). This suggests that DO-QJP is not a uniquely defined object across the literature, but a family resemblance term for state-adaptive quantum-jump constructions optimized for a stated criterion.

Under that reading, the unifying content of DO-QJP is the use of realization-dependent or record-dependent freedom in the jump description to improve either convergence, variance, positivity, or computational efficiency while preserving the target open-system dynamics on average.