Photonic Quantum Walks Reveal How Driven Open Systems Choose Their Relaxation Path

A new photonic experiment turns a deceptively small three-node quantum walk into a laboratory for Floquet thermodynamics. By tuning synthetic magnetic flux and dephasing, the researchers directly observe how an open driven system switches between distinct routes to equilibrium.

The paper, “Experimental Observation of Dynamical Phase Transitions in a Dephased Photonic Quantum Walk”, was submitted to arXiv on June 14, 2026 by Xiaojian Huang, Lei Xiao, Bingzi Huo, Xiaowei Wang, Stefano Longhi and Peng Xue. It reports a controlled single-photon experiment in which a discrete-time quantum walk is repeatedly driven, partially dephased and tomographically reconstructed. The result is not a heat engine or a battery by itself. It is something that quantum-energy engineers need just as much: a clean experimental map of how relaxation is selected in a periodically driven open quantum system.

That distinction matters. Most useful quantum-energy devices are neither perfectly isolated nor simply thermal. A Floquet heat engine, a driven quantum battery, a time-modulated thermal router or a dissipative sensor must run through repeated cycles while leaking information and energy into its environment. The central question becomes less romantic than “can we make a new phase of matter?” and more practical: can we decide how fast, by which mode and toward which stationary behavior the device relaxes?

The experiment shows that relaxation is not just a nuisance to be minimized. In a Floquet device, the spectrum of the open-system evolution can be engineered, measured and switched between qualitatively different dynamical regimes.

Why a three-node walk is enough to teach a big lesson

The platform is a discrete-time open quantum walk on a triangular graph with three nodes. In each step, a single photon undergoes a coherent unitary evolution and then experiences controllable dephasing in the node basis. The nodes are encoded in a photon’s polarization and spatial modes. Beam displacers and wave plates synthesize the desired unitary; calibrated dephasing acts like an unread which-node measurement; avalanche photodiodes measure the output populations after each step.

The coherent part of the step is governed by hopping amplitudes around the triangle. Two hoppings are set equal, while the third carries a tunable phase, denoted by the paper as a synthetic gauge flux ϕ. When ϕ = 0, the effective dynamics preserves time-reversal symmetry and detailed balance. When ϕ is nonzero, detailed balance is broken. This lets the team ask a sharp experimental question: what happens to relaxation when the same driven open system is moved from a time-reversal-symmetric regime to a flux-biased, non-Hermitian regime?

The control knob for coherent evolution is a dimensionless step strength β. The dephasing strength is q, ranging from q = 0 for fully coherent unitary evolution to q = 1 for the fully dephased Markovian limit. Importantly, the paper emphasizes that this is a genuine finite-step Floquet channel, not merely a Trotterized version of a continuous-time Lindblad equation. The stroboscopic nature of the walk is part of the physics.

The spectrum behind relaxation

To understand the result, replace the usual picture of “population simply decays” with a spectral one. A closed quantum system has a Hamiltonian and energy eigenstates. An open driven system has a one-step quantum map. Taking the logarithm of that map defines an effective Floquet–Liouvillian superoperator, whose eigenvalues tell us which relaxation modes decay slowly, which decay quickly and which acquire oscillatory phases.

In the fully dephased limit, the one-step map reduces to a classical population transfer matrix. Because the coherent step came from a unitary, the stationary distribution is uniform: each of the three nodes tends to probability 1/3. But the route toward that uniform state is controlled by the non-stationary eigenmodes. As β is tuned, the identity of the slowest relaxation mode can change. That change is the relaxation analogue of a phase transition.

This language builds on recent theory connecting relaxation criticality to eigenvalue crossings in Markov generators. A 2023 Physical Review Letters paper by Teza, Yaacoby and Raz argued that relaxation dynamics can show phase-transition-like behavior without requiring the conventional closing of a Liouvillian spectral gap. In the photonic experiment, that idea becomes directly measurable: the researchers reconstruct the relevant transition matrix or full superoperator and track its eigenvalues and eigenvectors.

First-order: a sudden switch under detailed balance

With detailed balance preserved, at ϕ = 0 and q = 1, the transition matrix is real symmetric. The non-stationary eigenvalues are real, and the allowed transition is a first-order dynamical phase transition produced by a diabolic crossing. In plain English, the slow relaxation branch switches abruptly.

The experiment observes this by comparing population relaxation below and above the critical β. At β = 0.70, the population on one node reaches its stationary value after only a few steps while the other two relax more slowly. At β = 0.83, all three populations relax on more comparable timescales. The difference is not that the nodes themselves have different physical decay constants. It is that the initial state projects differently onto the slow and fast collective relaxation modes.

The reconstructed spectrum puts a number on the switch: the real parts of two non-stationary Floquet exponents cross at β_c = 0.82 ± 0.01. The authors define an order parameter from the node-1 weight of the slow relaxation mode and report a jump size of Δm = 0.662 ± 0.008. For a three-node device, that is an unusually crisp demonstration of a phase-transition concept.

Second-order: an exceptional point when flux breaks symmetry

When the synthetic flux is turned on, the story changes. At ϕ = π/3, time-reversal symmetry and detailed balance are broken. The relevant relaxation spectrum can become complex. Instead of a simple crossing of two real modes, the experiment observes a second-order dynamical phase transition at an exceptional point, where eigenvalues and eigenvectors coalesce.

The dynamical signature is intuitive: relaxation changes from monotonic decay to damped oscillation. Below the critical point, the populations slide toward the stationary distribution without ringing. Above it, the reconstructed spectrum contains a complex-conjugate pair, so the long-time deviations behave like an exponentially damped cosine and sine. The team checks this against a control case at ϕ = 0 and finds no comparable oscillatory component, supporting the interpretation that the oscillations come from the flux-induced spectral structure rather than a technical artifact.

The second-order transition occurs near β_c ≃ 0.74 in the fully dephased broken-symmetry benchmark. The researchers also define an eigenmode overlap g: two distinct eigenvectors give g below 1, while coalescence at an exceptional point drives g toward 1. Near the exceptional point, the splitting follows the characteristic square-root behavior associated with non-Hermitian degeneracies.

The quantum-regime test

A skeptical reader might ask whether all of this is merely classical Markov-chain physics dressed up in photonic language. The paper addresses that directly by moving away from full dephasing. At q = 0.8 and q = 0.5, coherences are only partially damped, so the full one-step superoperator must be reconstructed rather than just the population matrix.

The transition signatures persist in this partially coherent regime. In the time-reversal-symmetric case, the system still shows real-eigenvalue crossing behavior. In the flux-biased case, eigenvalue and eigenvector coalescence still signal an exceptional point. As dephasing is reduced, the critical point shifts toward smaller β, and for weak dephasing around q = 0.1 the signatures become strongly smeared. That is a useful engineering lesson: the effect is neither purely classical nor infinitely robust. It lives in a hybrid regime where coherent Floquet stepping and dissipation are both important.

For quantum technologies, the sweet spot is often not “as isolated as possible.” It is the regime where coherent control and dissipation are balanced well enough to shape the desired spectrum.

Why this belongs in quantum energy

Quantum thermodynamics is often discussed in terms of efficiency, work extraction and entropy production. But before a driven device can be optimized, its relaxation architecture must be understood. Which mode carries memory of the initial state? Which perturbation changes the approach to the steady state from monotonic to oscillatory? Can a phase knob reroute relaxation without changing the whole hardware platform?

The photonic-walk experiment gives concrete answers in a minimal system. A gauge phase switches the universality class of relaxation. Dephasing controls the crossover from coherent quantum walk to classicalized Markov dynamics. Tomography extracts the spectrum that a thermodynamic model would need in order to predict cycle stability, relaxation time and sensitivity.

This matters for Floquet heat engines, because engine cycles depend on returning to a useful state after each stroke. It matters for quantum batteries, because stored energy is valuable only if it can remain in structured, extractable form rather than relaxing immediately into passive heat. It matters for thermal routing and quantum sensing, because exceptional points and non-Hermitian spectra can amplify parameter sensitivity while also introducing fragility. The paper does not claim a new beyond-Carnot machine. Instead, it supplies a measured control principle for the driven-dissipative layer beneath such machines.

What to watch next

The natural next step is scale. A three-node photonic walk is ideal for spectral reconstruction, but quantum-energy applications will need larger graphs, many-body Hilbert spaces and real reservoirs with energy-selective couplings. The authors point toward non-Hermitian spectral topology, bulk-boundary correspondence in open systems, Liouvillian skin effects, non-Markovian exceptional points, Mpemba-like speedups and fast state preparation in scalable photonic simulators.

For floquet.ca, the key takeaway is that Floquet engineering is maturing beyond the construction of exotic isolated Hamiltonians. The field is learning to engineer open-system maps: not only what a system does during a drive period, but how it forgets, how it relaxes and how that relaxation can be reprogrammed. That is exactly the kind of control needed if periodically driven quantum matter is to become a practical energy platform.

Sources and further reading

• Xiaojian Huang, Lei Xiao, Bingzi Huo, Xiaowei Wang, Stefano Longhi and Peng Xue, “Experimental Observation of Dynamical Phase Transitions in a Dephased Photonic Quantum Walk,” arXiv:2606.15935, submitted June 14, 2026. Read the arXiv abstract.
• G. Teza, R. Yaacoby and O. Raz, “Eigenvalue crossing as a phase transition in relaxation dynamics,” Physical Review Letters 130, 207103 (2023).
• F. Minganti, A. Biella, N. Bartolo and C. Ciuti, “Spectral theory of Liouvillians for dissipative phase transitions,” Physical Review A 98, 042118 (2018).
• K. Wang et al., “Simulating dynamic quantum phase transitions in photonic quantum walks,” Physical Review Letters 122, 020501 (2019).
• H. Gao et al., “Experimental observation of the Yang-Lee quantum criticality in open quantum systems,” Physical Review Letters 132, 176601 (2024).