When a Driven Quantum Gas Chooses Insulator or Metal

Floquet systems are often described as quantum matter under a rhythmic push. That picture is useful, but it hides the central thermodynamic question: what happens to the injected energy? In many periodically driven systems the answer is simple and disappointing. Keep shaking long enough and the system heats, spreads, and drifts toward classical-looking diffusion. Yet quantum interference can do something stranger. It can stop energy absorption, trapping the motion in momentum space even while the external drive continues to kick.

A new ultracold-atom preprint, “Interaction-enabled metal-insulator phase transition in a driven quantum gas,” by Camilo Cantillano, Karthick Ramanathan, Zekai Chen, Ang Yang, Emilio Aguilera-Valdes, Lei Ying, Manuele Landini, Hanns-Christoph Nägerl, and Yanliang Guo, reports an experiment aimed directly at that boundary. Submitted to arXiv on May 21, 2026, the work studies a periodically kicked three-dimensional cesium Bose gas with tunable interactions. The result is not a heat engine and not an energy generator. It is a controlled laboratory map of when a closed Floquet many-body system refuses to absorb energy, when it absorbs energy diffusively, and how interactions move it between those regimes.

The experiment turns Floquet heating from a nuisance into the observable: the question is not merely whether the gas is driven, but whether many-body coherence can keep the drive’s energy from spreading.

The kicked-rotor problem, upgraded to many bodies

The underlying idea traces back to the quantum kicked rotor, a model in which a particle receives periodic impulses. Classically, repeated kicks can produce chaotic diffusion in momentum. Quantum mechanically, interference can arrest that diffusion. The phenomenon is called dynamical localization, and it is closely related to Anderson localization, where disorder prevents waves from spreading through space.

Atom-optics realizations of the kicked rotor have been a workhorse for studying this physics since the 1990s, including experiments by Moore and colleagues and later observations of dynamical localization and decoherence. The broader localization framework goes back to P. W. Anderson’s 1958 work on the absence of diffusion in random lattices, while Fishman, Grempel, and Prange famously connected the kicked rotor to Anderson localization in 1982.

The new experiment asks a harder question: what if the rotor is not a single particle, but an interacting many-body quantum gas in three dimensions? Interactions usually help systems thermalize. They scramble phases, redistribute energy, and often restore diffusion. But recent work has shown that many-body dynamical localization can survive in special regimes. Cantillano and collaborators push this question into a tunable, experimentally mapped phase diagram.

How the experiment works

The team prepares a nearly pure Bose-Einstein condensate of about 1 × 10^{4 133}Cs atoms in a crossed optical dipole trap, levitated against gravity. A magnetic field tunes the interatomic interaction using a Feshbach resonance, allowing the scattering length to range from essentially zero up to about 1037 Bohr radii. This tunability is the central knob: the same gas can behave almost non-interacting, moderately interacting, or strongly interacting.

The Floquet drive is implemented by repeatedly pulsing a vertical optical lattice. The lattice spacing is reported as 532.2 nanometers; the square pulses last 5.0 microseconds and repeat every 31.0 microseconds. In kicked-rotor language, the pulse area defines a dimensionless kick strength κ. By changing κ and the scattering length, the researchers can explore a two-dimensional landscape: drive strength on one axis, interaction strength on the other.

After a chosen number of kicks, the researchers image the atoms either in situ or after time-of-flight expansion. From the momentum distribution they compute an energy proxy, sensitive to atoms leaving the zero-momentum peak. If the proxy saturates, the gas has stopped absorbing drive energy in the measured momentum channel. If it grows roughly linearly, the gas is diffusing toward higher momenta.

A phase boundary between arrested and diffusive energy flow

The headline result is a sharp localization-delocalization boundary. At moderate kick and interaction strength, for example κ ≈ 0.75 and scattering length around 220 Bohr radii, the momentum distribution reaches a stationary state even after hundreds to one thousand kicks. The gas remains dynamically localized. At stronger driving and stronger interactions, such as κ ≈ 1.5 and scattering length around 775 Bohr radii, atoms spread to higher momentum states and the energy proxy grows diffusively.

The researchers then map the phase diagram using measurements after 500 kicks across κ values from about 0.62 to 1.75 and interaction strengths from 0 to 1037 Bohr radii. The insulating region appears at weak and intermediate interactions with moderate kick strengths. Stronger interactions lower the critical drive strength needed to delocalize the gas. In plain language: interactions do not simply destroy localization everywhere, but they can carve out a boundary where the same Floquet platform switches from quantum-arrested motion to classical-like absorption.

To characterize the transition, the team applies finite-time scaling, the toolset normally used when experiments cannot run to infinite time but still want to infer critical behavior. At a fixed scattering length of 220 Bohr radii, their scaling collapse identifies a critical kick strength of κ_c = 1.162(13). They interpret the data as consistent with a second-order quantum phase transition, with a diverging localization length as the boundary is approached.

This is why the result matters for quantum energy science: it gives an experimental handle on the point where periodic driving stops being a controllable Floquet resource and becomes a route to irreversible-looking energy diffusion.

Why this matters for Floquet thermodynamics

For practical Floquet engineering, heating is the recurring tax. Periodic driving can create synthetic magnetic fields, topological bands, new quasiparticle spectra, and exotic nonequilibrium phases. But if the drive simply dumps energy into all available degrees of freedom, the engineered phase may be short-lived or thermodynamically expensive.

This experiment does not remove that tax; instead, it measures when the tax is suppressed. Many-body dynamical localization is a way for coherent interference to limit energy absorption in a closed driven system. The delocalized side shows the opposite: interactions and driving combine to produce diffusion. The transition between the two is a potential design principle for future Floquet devices, especially those that must balance useful drive-induced structure against unwanted heating.

Three lessons for energy-focused readers

• Energy absorption can be phase-like. The gas is not merely “heating faster” or “heating slower”; the data support distinct localized and diffusive regimes separated by a boundary.
• Interactions are not only a source of loss. They can enable the transition itself, and at moderate values localization can still persist.
• Floquet control needs thermodynamic maps. Knowing the drive frequency and amplitude is not enough. The many-body interaction landscape determines whether injected energy remains bounded or spreads.

There is also an important caveat. Dynamical localization is not free work extraction, and it does not imply a violation of the second law or of Carnot-style bounds. The optical lattice drive, magnetic-field control, trapping beams, and measurement protocol all sit outside the idealized gas dynamics. For energy applications, the full cost of creating and stabilizing the Floquet drive must still be counted. The value of the experiment is more fundamental: it shows how a quantum many-body system can either absorb or refuse energy from a periodic drive depending on controllable parameters.

Connections to the broader literature

The paper sits at the intersection of several established threads. Anderson’s localization theory and the scaling theory of localization by Abrahams, Anderson, Licciardello, and Ramakrishnan provide the language of metal-insulator transitions. The quantum kicked rotor connects that language to periodically driven dynamics, with classic theoretical work by Casati and collaborators and the Fishman-Grempel-Prange mapping. Atom-optics experiments then made those ideas physical, including the kicked-rotor demonstrations by Moore et al. and later work on decoherence and amplitude noise.

More recent many-body work includes interaction-driven breakdown of dynamical localization in a kicked quantum gas reported by Cao and collaborators in Nature Physics in 2022, the 2025 observation of many-body dynamical localization by Guo and colleagues in Science, and theoretical studies such as Rylands, Rozenbaum, Galitski, and Konik on many-body dynamical localization in a kicked Lieb-Liniger gas. The 2026 experiment is notable because it maps a three-dimensional, tunably interacting phase diagram and tests reversibility by ramping the interactions during the kicking sequence.

What to watch next

The authors point to several open directions: how the transition changes with dimensionality, whether tighter transverse confinement would move the system toward a strictly one-dimensional regime, and how temperature or initial energy might reshape the boundary. Those questions matter beyond cold atoms. Any future Floquet material, quantum battery, or driven quantum processor will face an analogous design problem: identify the operating window where periodic driving creates useful structure without uncontrolled energy spreading.

For now, the experiment is best read as a precise quantum simulator of Floquet energy flow. It does not promise a new power source. It does something arguably more useful for a young field: it provides a measurable map of when a driven quantum many-body system behaves like an energy insulator, when it behaves like an energy metal, and how interactions decide the outcome.