All experiments

Quantum many-body physics

Caesium quantum gas microscope

We image single caesium atoms in an optical lattice with 383.5 nm spacing and measure particle currents on individual bonds between lattice sites.

Two fluorescence images of single caesium atoms in an optical lattice. Left: a densely and regularly filled region of the lattice. Right: the same region after a quench, with atoms spread irregularly over the sites.
Single caesium atoms in the optical lattice before (left) and after a quench (right), from the experiment reported in Karch et al., arXiv:2604.24744 (2026, preprint). Image: Quantum Munich

What can site-resolved images of currents and fluctuations tell us about strongly interacting bosons in optical lattices?

The platform

We operate a quantum gas microscope for bosonic ¹³³Cs atoms at LMU, together with Monika Aidelsburger’s group. Two-dimensional optical superlattices with short and long spacings of 383.5 nm and 767 nm hold the atoms, and a 0.8 NA objective images them by fluorescence. The short spacing is less than half the Rayleigh resolution of about 850 nm, so a deep neural network reconstructs the occupation of each site. A digital micromirror device projects programmable potentials, a Feshbach resonance tunes the contact interactions, and laser-assisted tunnelling imprints a synthetic magnetic flux.

Single atoms closer than the imaging resolution

Our lattice spacing is less than half the imaging resolution, so the light from neighbouring atoms overlaps on the camera. To read out the occupation of each site, we developed a deep convolutional neural network that learns directly from experimental fluorescence images, without labelled training data. It reconstructs images with several thousand lattice sites and reaches fidelities of about 96 % or better at all fillings.

Four panels: a raw fluorescence image of a square Mott insulator, a sketch of the encoder network, the deconvolved site counts with a two-peaked histogram, and the reconstructed occupation of every lattice site.Enlarge (opens the full-size figure)
From the raw fluorescence image of a Mott insulator with about 2,500 atoms (a), the network computes deconvolved site counts (b, c) and the occupation of every lattice site (d). Impertro et al., Commun. Phys. 6, 166 (2023), Fig. 2, cropped · CC BY 4.0

Fluctuations and return probabilities in quantum dynamics

We asked whether large-scale fluctuations in an isolated quantum system are described by fluctuating hydrodynamics, a framework developed for classical stochastic systems. Starting from a charge-density wave of hard-core bosons in 20 independent ladders of up to 2 × 50 sites, we recorded the full counting statistics of the atom number in subsystems as the system evolved. By tuning the coupling between the two legs, we moved from integrable chains, where fluctuations grow ballistically, to chaotic ladders, where they grow diffusively. In the chaotic case the growth of fluctuations followed the predictions of fluctuating hydrodynamics, and the diffusion constant we extracted from it agreed with equilibrium linear-response estimates.

In a 2025 preprint we measure a subsystem Loschmidt echo, the probability that a subsystem returns to its initial configuration. In a one-dimensional Bose-Hubbard chain it reveals a dynamical quantum phase transition at short times. At long times it measures the effective dimension of the Hilbert space the system explores and shows ergodicity breaking caused by Hilbert-space fragmentation.

Two fluorescence images of 20 ladder systems in an optical lattice. Left: the initial state with a regular stripe pattern in the averaged inset. Right: the evolved state with irregular bright and dark patches.Enlarge (opens the full-size figure)
Fluorescence images of 20 ladder systems in the initial charge-density wave (a) and after time evolution (b). The insets show the averaged site occupation. Wienand et al., Nat. Phys. 20, 1732 (2024), Fig. 2, cropped · CC BY 4.0

Currents in synthetic magnetic fields

Quantum gas microscopes usually record only where atoms sit. We isolate pairs of sites in double wells and rotate the measurement basis, which gives us particle currents and kinetic energy on individual bonds. Laser-assisted tunnelling adds a synthetic magnetic flux of π/2 per plaquette. With both tools we realised the strongly interacting Mott-Meissner phase in bosonic flux ladders of 48 sites at half filling, imaged its chiral currents with local resolution and found the interaction-induced suppression of the chiral current that is characteristic of this phase.

Grid of square plaquettes with an arrow on every bond: blue arrows along the legs and red arrows on the rungs, circulating around each plaquette.Enlarge (opens the full-size figure)
Ground-state currents measured bond by bond in an array of 140 isolated plaquettes with synthetic magnetic flux. Leg currents are blue, rung currents red. Impertro et al., Nat. Phys. 21, 895 (2025), Fig. 2a, cropped · CC BY 4.0

A U(1) quantum spin liquid from non-equilibrium dynamics

In a 2026 preprint we realise a two-dimensional U(1) lattice gauge theory on more than 3,000 lattice sites. In a tilted square superlattice with a large staggered potential, atoms can only tunnel in pairs: two atoms on neighbouring high-energy sites (monomers) hop together and form a doubly occupied site (a dimer) on the link between them. This kinetically constrained dynamics maps onto a monomer-dimer model, equivalent to a quantum link model of quantum electrodynamics in two dimensions with dynamical matter, which obeys a Gauss’s law at every vertex. A new imaging step converts doubly occupied sites into single atoms with 98(1) % fidelity, so monomers and dimers appear in the same snapshot and we can check Gauss’s law directly after a quench.

Starting from a crystal of monomers, we ramp the monomer mass from negative to positive values. A semi-adiabatic ramp, slow compared with monomer excitations but fast compared with the tiny energy splittings between dimer coverings, turns most monomers into dimers. It prepares regions in an equal-amplitude superposition of all dimer coverings, the Rokhsar-Kivelson state of a U(1) quantum spin liquid, although our Hamiltonian does not host this state as its ground state.

Left: lattice diagram with green monomers on vertices and blue dimers on links, with an inset showing two atoms tunnelling together into one site. Right: three fluorescence images of about 60 by 60 lattice sites, a dense monomer crystal, a sparse parity-projected image after the sweep, and a dimer-resolved image with red outlines around spin-liquid regions, above a sketch of the sweep and of the superposition of dimer coverings.Enlarge (opens the full-size figure)
(a) Monomer-dimer model: monomers (green) on vertices, dimers (blue) on links, formed when two atoms tunnel together. (b) From an ordered monomer state, a non-equilibrium sweep prepares regions of U(1) quantum spin liquid. Standard imaging shows doubly occupied sites as empty; dimer-resolved imaging reveals patches of nearly perfect dimer coverings (red outlines) (preprint). Karch et al., arXiv:2604.24744 (2026), Fig. 1 (preprint) · arXiv.org perpetual, non-exclusive licence

Pinch points and many-body coherence

After the ramp, 83.2(2) % of the vertices fulfil Gauss’s law. Within regions of near-perfect dimer covering, all distinct local dimer patterns occur with about equal probability, as expected for the Rokhsar-Kivelson state. Dimer-dimer correlations are consistent with the 1/r² decay expected for this critical state, and the electric-field structure factor shows a pinch point at zero momentum, the momentum-space signature of Gauss’s law known from neutron scattering on spin ice.

An incoherent mixture of dimer coverings would show the same signatures. To test coherence, we reverse the ramp: only a coherent superposition returns efficiently to the monomer crystal, and large monomer patches reappear in the images. A short dephasing pulse before the reverse ramp suppresses these patches, and imprinting a phase between horizontal and vertical dimers makes the return probability oscillate. From the round-trip snapshots we find coherent spin-liquid regions extending over about 100 lattice sites.

Four panels: a 35 by 35-site snapshot of dimers and monomers with blue dimer domains and red defect regions; bar charts of nearly equal probabilities for two-, three- and four-dimer coverings; a map and radial plot of dimer-dimer correlations falling as one over r squared; and a red-and-blue structure factor map with a bow-tie shaped pinch point at the centre.Enlarge (opens the full-size figure)
(a) Snapshot after the ramp (35 × 35 sites): domains of near-perfect dimer covering (blue) and regions with monomers and Gauss’s-law defects (red). (b) Distinct local dimer coverings occur with nearly equal probability. (c) Dimer-dimer correlations decay approximately as 1/r². (d) Electric-field structure factor with a pinch point at zero momentum (preprint). Karch et al., arXiv:2604.24744 (2026), Fig. 4 (preprint) · arXiv.org perpetual, non-exclusive licence

Research directions

Next we want to study how excitations move in two-dimensional lattice gauge theories and add plaquette interactions through a hybrid digital-analogue scheme. Extending flux ladders to many legs is a route towards fractional Chern insulators in two dimensions, and we plan to use subsystem Loschmidt echoes in two-dimensional systems as well.

Selected results & publications

Dynamical preparation of U(1) quantum spin liquids in an analogue quantum simulatorarXiv:2604.24744 (2026) · PreprintStrongly interacting Meissner phases in large bosonic flux laddersNature Physics 21, 895–901 (2025) · Journal ArticleEmergence of fluctuating hydrodynamics in chaotic quantum systemsNature Physics 20, 1732–1737 (2024) · Journal ArticleLocal readout and control of current and kinetic energy operators in optical latticesPhysical Review Letters 133, 063401 (2024) · Journal ArticleAn unsupervised deep learning algorithm for single-site reconstruction in quantum gas microscopesCommunications Physics 6, 166 (2023) · Journal Article

Research with this team

Discuss student projects, doctoral research or postdoctoral opportunities with the scientific contact; project scope and availability are agreed individually.

Methods and topics: Quantum gas microscopy, Optical lattices, Synthetic magnetic flux, Single-site reconstruction.

Explore research opportunities

The team

Who works on this experiment

Keep exploring

Strontium Rydberg lab

Next experiment