Quantum Stochastic Walks

• Quantum Stochastic Walks (QSWs) are transport models that combine coherent quantum evolution with stochastic, environment-induced dynamics on graph structures.
• They utilize both continuous-time Lindblad master equations and discrete-time step maps to interpolate between pure quantum and classical random walks.
• QSWs have practical applications in modeling energy transfer, refining network rankings, optimizing financial portfolios, and advancing simulation software for complex systems.

Quantum stochastic walk (QSW) is a model of transport on a graph in which a walker evolves under both coherent quantum dynamics and incoherent, environment-induced stochastic dynamics. In the continuous-time setting it is formulated as an open quantum system governed by a Gorini–Kossakowski–Sudarshan–Lindblad master equation; in the discrete-time setting it is formulated as a completely positive trace-preserving step map. QSWs generalize both classical random walks and coherent quantum walks, and they have been used to study transport, ranking, state discrimination, recurrence, and noisy network dynamics on both undirected and directed graphs (Domino et al., 2016, Schuhmacher et al., 2020, Stefanak et al., 15 Jan 2025).

1. Formal definition

In its standard continuous-time form, a QSW evolves a density operator $\rho(t)$ according to

$$\dot{\rho} = -i(1-\omega)[H,\rho] + \omega\sum_k \left( L_k \rho L_k^\dagger - \frac12 \{L_k^\dagger L_k,\rho\} \right),$$

with $\omega\in[0,1]$. Here $H$ generates coherent motion, while the Lindblad operators $L_k$ encode decoherence, dissipation, continuous measurement, or incoherent hopping. The limits are explicit: $\omega=0$ gives a purely coherent continuous-time quantum walk, $\omega=1$ gives purely dissipative dynamics, and $0<\omega<1$ is the genuine QSW regime. Several papers use the equivalent interpolation parameter $p$ instead of $\omega$, with the same role in mixing coherent and incoherent dynamics (Domino et al., 2016, Pozza et al., 2020, Bressanini et al., 2022).

The graph determines the Hilbert-space basis $$\dot{\rho} = -i(1-\omega)[H,\rho] + \omega\sum_k \left( L_k \rho L_k^\dagger - \frac12 \{L_k^\dagger L_k,\rho\} \right),$$0, one basis state per vertex, and it constrains both $$\dot{\rho} = -i(1-\omega)[H,\rho] + \omega\sum_k \left( L_k \rho L_k^\dagger - \frac12 \{L_k^\dagger L_k,\rho\} \right),$$1 and the dissipator. Common choices take $$\dot{\rho} = -i(1-\omega)[H,\rho] + \omega\sum_k \left( L_k \rho L_k^\dagger - \frac12 \{L_k^\dagger L_k,\rho\} \right),$$2 to be the adjacency matrix or the graph Laplacian, depending on whether the aim is tight-binding propagation, continuous-time quantum-walk transport, or explicit comparison with classical continuous-time random walks. The diagonal entries $$\dot{\rho} = -i(1-\omega)[H,\rho] + \omega\sum_k \left( L_k \rho L_k^\dagger - \frac12 \{L_k^\dagger L_k,\rho\} \right),$$3 are node populations, and the off-diagonal entries $$\dot{\rho} = -i(1-\omega)[H,\rho] + \omega\sum_k \left( L_k \rho L_k^\dagger - \frac12 \{L_k^\dagger L_k,\rho\} \right),$$4 encode coherence between vertices (Domino et al., 2016, Falloon et al., 2016).

A discrete-time QSW replaces the Lindblad generator by a step channel. One form used for graph transport is

$$\dot{\rho} = -i(1-\omega)[H,\rho] + \omega\sum_k \left( L_k \rho L_k^\dagger - \frac12 \{L_k^\dagger L_k,\rho\} \right),$$5

where $$\dot{\rho} = -i(1-\omega)[H,\rho] + \omega\sum_k \left( L_k \rho L_k^\dagger - \frac12 \{L_k^\dagger L_k,\rho\} \right),$$6 weights the coherent contribution, $$\dot{\rho} = -i(1-\omega)[H,\rho] + \omega\sum_k \left( L_k \rho L_k^\dagger - \frac12 \{L_k^\dagger L_k,\rho\} \right),$$7 weights incoherent jumps, and $$\dot{\rho} = -i(1-\omega)[H,\rho] + \omega\sum_k \left( L_k \rho L_k^\dagger - \frac12 \{L_k^\dagger L_k,\rho\} \right),$$8 for each origin vertex $$\dot{\rho} = -i(1-\omega)[H,\rho] + \omega\sum_k \left( L_k \rho L_k^\dagger - \frac12 \{L_k^\dagger L_k,\rho\} \right),$$9. Iteration of this map defines a discrete-time quantum stochastic walk, and restricted classes of continuous-time QSWs can be approximated by such discrete-time steps to first order in $\omega\in[0,1]$0 (Schuhmacher et al., 2020).

2. Graph encoding and dissipative architectures

The most consequential modeling choice in a QSW is the structure of the dissipator. In local-dissipation constructions, one assigns one Lindblad operator per edge, for example $\omega\in[0,1]$1 or, on a line, operators proportional to nearest-neighbor jumps. In these models the dissipator reproduces a classical continuous-time random walk when the coherent term is removed, and directedness can be encoded directly in the jump operators (Domino et al., 2016, Pozza et al., 2020, Bressanini et al., 2022).

A distinct construction uses a single collective operator. On a line segment, the global operator

$\omega\in[0,1]$2

becomes a symmetric tridiagonal matrix satisfying $\omega\in[0,1]$3, so $\omega\in[0,1]$4 commutes with $\omega\in[0,1]$5. This “global dissipation” corresponds to a global continuous position measurement rather than local damping. In the line model studied analytically, this structural alignment between Hamiltonian and dissipator leads to qualitatively different transport from the local model (Domino et al., 2016).

Directed graphs introduce an additional complication. A naive global Lindblad operator built from the adjacency matrix of a directed graph can generate extra transitions between parents of a common child, a phenomenon called “spontaneous moralization,” because the effective connectivity becomes the moral graph rather than the original directed graph. A corrected construction enlarges the Hilbert space by splitting vertices into orthogonal copies according to indegree, chooses blockwise orthogonal matrices $\omega\in[0,1]$6, and defines a modified global Lindblad operator on the enlarged graph. In that construction, the original directed topology is preserved, premature localization can be corrected with a locally rotating Hamiltonian, and symmetric propagation on line segments can be restored by using two global Lindblad operators (Domino et al., 2017).

Another widely used continuous-time QSW takes the graph Laplacian $\omega\in[0,1]$7 as Hamiltonian and defines jump operators

$\omega\in[0,1]$8

This embeds explicit incoherent hopping along graph edges together with diagonal degree-dependent terms. It is this Laplacian-based construction that is used in several studies of classicalization, where the QSW is treated as a graph-induced interpolation between CTQW and CTRW (Bressanini et al., 2022).

3. Transport, scaling, recurrence, and classicalization

A central diagnostic for continuous-time QSWs is the second central moment

$\omega\in[0,1]$9

together with the asymptotic scaling law $H$0. The exponent $H$1 distinguishes subdiffusive, diffusive, superdiffusive, and ballistic transport. In the global-dissipation line model with single operator $H$2, the exact result

$H$3

implies $H$4 for every $H$5 and $H$6 only at $H$7. In the corresponding local-dissipation model, by contrast, any nonzero $H$8 drives the walk asymptotically to diffusive scaling, $H$9. The same work also proposes $L_k$0 as an operational measure of coherence, while showing that in the single-operator global model purity is a non-increasing function of $L_k$1 even though ballistic transport persists for all $L_k$2 (Domino et al., 2016).

A different viewpoint is provided by the quantum–classical dynamical distance

$L_k$3

with $L_k$4. For Laplacian-based QSWs on complete, cycle, and star graphs, $L_k$5 for every $L_k$6: off-diagonal coherences vanish and the diagonal approaches the classical stationary distribution. In that setting QSW acts as an efficient classicalizer, and increasing the dissipative weight $L_k$7 accelerates convergence without the localization bottleneck seen in the Haken–Strobl model at large dephasing rates (Bressanini et al., 2022).

Discrete-time QSWs exhibit an additional phenomenon: monitored recurrence need not vary monotonically between the unitary and classical limits. For a line walk with Kraus operators

$L_k$8

the recurrence probability can be reduced by introducing classical randomness, despite the fact that the classical random walk returns with certainty. Numerical evaluation of the first-return generating function identifies a threshold $L_k$9: for $\omega=0$0, the recurrence probability initially decreases as $\omega=0$1 increases from zero, and the evidence presented indicates that this suppression persists asymptotically rather than being a transient effect (Stefanak et al., 15 Jan 2025).

Taken together, these results do not support a single universal “quantum-to-classical crossover” for QSWs. Instead, they show that the asymptotic behavior depends sensitively on the graph, the choice of Hamiltonian, the Lindblad architecture, and whether the walk is continuous-time or discrete-time (Domino et al., 2016, Bressanini et al., 2022, Stefanak et al., 15 Jan 2025).

4. Applications

In excitation transport, QSWs have been used to model the Fenna–Matthews–Olson complex as a graph of 7 chromophore sites plus a sink. With the Adolphs–Renger Hamiltonian, initial state $\omega=0$2, and sink operator $\omega=0$3 with $\omega=0$4, three noise models were compared. The reported localization times are $\omega=0$5 fs and $\omega=0$6 fs for pure dephasing at $\omega=0$7 and $\omega=0$8, $\omega=0$9 fs and $\omega=1$0 fs for only incoherence, and $\omega=1$1 fs and $\omega=1$2 fs for dephasing plus incoherence. In that phenomenological setting, pure dephasing yields a longer super-diffusive window at larger $\omega=1$3, whereas models containing incoherent hopping show a marked slowdown (Dudhe et al., 2020).

In quantum information, QSWs have been used as trainable dynamical networks for quantum state discrimination. One implementation uses layered $\omega=1$4 graphs, sink nodes as output ports, and optimization over coherent couplings $\omega=1$5, incoherent transition probabilities $\omega=1$6, and the interpolation parameter $\omega=1$7. For the $\omega=1$8 example the optimal theoretical bound is reported as $\omega=1$9; for a $0<\omega<1$0 example it is $0<\omega<1$1. The independently optimized scheme performs best among the four QSW schemes considered, but none reaches the optimal POVM bound (Pozza et al., 2020).

In ranking theory, QSWs have been used to refine Google search. With Lindblad operators built from the Google matrix, both “only incoherence” and “dephasing with incoherence” schemes are reported to resolve degeneracies that are unresolvable via classical PageRank, with convergence time comparable to CPR and, for some networks, lower than CPR together with an almost degeneracy-free ranking (Benjamin et al., 2022).

Recent work has extended QSWs to finance by embedding assets in weighted graphs and deriving portfolio weights from the stationary distribution. In that construction the GKLS generator mixes a Hamiltonian built from normalized covariances with Lindblad rates given by a Google matrix on the financial network. Across the reported back-tests on top 100 S&P 500 constituents over 2016–2024 and random 100-stock subsets, the QSW optimizer raises the annualized Sharpe ratio by $0<\omega<1$2 and cuts turnover by $0<\omega<1$3 relative to classical optimisation, while respecting the UCITS 5/10/40 rule (Chang et al., 5 Jul 2025).

5. Simulation, software, and physical realizability

QSW research has been accompanied by dedicated software. The Mathematica package QSWalk simulates time evolution on arbitrary directed and weighted graphs by constructing the sparse $0<\omega<1$4 Liouvillian superoperator and evaluating $0<\omega<1$5 with Mathematica’s MatrixExp; the package illustrates line-graph transport, dephasing, photosynthetic energy transfer, and quantum PageRank (Falloon et al., 2016). The Julia package QSWalk.jl provides analogous functionality together with explicit support for local, global, and nonmoralizing walks on arbitrary directed graphs, sparse/dense dispatch, vectorization utilities, and expmv-based evolution on large systems (Glos et al., 2018).

Discrete-time QSWs can also be simulated by trajectory-based protocols. One proposal uses one ancilla per vertex, a system–ancilla initialization Hamiltonian

$0<\omega<1$6

coherent graph evolution under $0<\omega<1$7, ancilla measurements, and feed-forward conditioned on the measurement outcomes. Averaging the resulting trajectories reproduces the target discrete-time QSW step map, and restricted classes of continuous-time QSWs can be approximated by this scheme to first order in $0<\omega<1$8 (Schuhmacher et al., 2020).

A recurring caution in the literature is that mathematical validity in Lindblad form does not automatically imply straightforward laboratory realizability. Under standard weak-coupling, Markovian, secular assumptions, general implementations of graph QSWs can require complete solution of the underlying unitary dynamics together with sophisticated reservoir engineering. In particular, realizing arbitrary directed incoherent exchange while preserving excitation number is strongly constrained by the microscopic structure of the system–bath coupling. This has made physical realizability a standing conceptual issue rather than a settled engineering detail (Taketani et al., 2016).

6. Conceptual extensions and open problems

QSWs have also become a testing ground for broader dynamical notions. One line of work studies the temporal growth of the von Neumann entropy $0<\omega<1$9 and proposes the logarithmic growth rate

$p$0

as a dissipative generalization of the information dimension. In the classical limit, $p$1, where $p$2 is the spectral dimension of the network; for line and Sierpinski-gasket examples, the near-coherent regime yields values close to twice the classical one (Schijven et al., 2014).

Another extension places the QSW in a stochastic environment. On a ring geometry with bias and random local transition rates, the Lindbladian spectrum exhibits a delocalization transition as the bias is increased beyond a critical value. A distinctive result is that effective disorder is enhanced due to coherent hopping, and the dependence of the relaxation spectrum on the coherent hopping rate is non-monotonic (Avnit et al., 2023).

A further development combines QSW decoherence with postselection-induced nonlinear Lindblad dynamics. On heterogeneous networks, postselection can break dynamical balance under QSW decoherence and induce robust localization preferentially at low-degree nodes, while the localized steady state maintains finite quantum coherence. In the corresponding many-body spin extension, degree heterogeneity similarly stabilizes localization of spin-up excitations and enhances entanglement preservation (J et al., 18 Mar 2026).

Natural extensions identified in the literature include higher-dimensional lattices and general graphs, more general forms of global Lindblad operators, relations between asymptotic scaling and other quantumness measures, and experimental realizations in which structured global dissipation can be engineered. The subject therefore remains divided between exact solvable models, graph-theoretic constructions, and application-driven open-system design, with the central theme unchanged: the behavior of a quantum stochastic walk is determined not only by the amount of dissipation, but by its graph-theoretic and operator-theoretic structure (Domino et al., 2016, Domino et al., 2017).