Fig. 3: Breakdown of AB caging by interactions and the crossover to bound-pair dynamics.

a, Atom pair A and B prepared in tweezers with a spatial separation of R_AB. The dipolar exchange interaction for states \(\left\vert i\right\rangle\) and \(\left\vert\; j\right\rangle\) is V_ij. We scaled all interactions to V = V₆₇ with calculated C₃ coefficients (Supplementary Information). b, Eigenenergy distribution for atom pairs in the 12-state rhombic lattice (with ϕ = π and θ = 0) under different interaction-to-coupling ratios V/Ω. Here n is the eigenstate index. c, Time evolution of the population in \(\left\vert 6\right\rangle\), P₆, for different V/Ω ratios from ideal numerical simulations. The dashed line indicates the single-atom revival time. d, Measured crossovers from caging to delocalization to pairs with reduced mobility. e, Experimentally measured time evolution of the SPAM-corrected population in \(\left\vert 6\right\rangle\) for different V/Ω. The vertical grey dashed line again corresponds to the single-atom revival time. f, Population distributions for the central five sites at the first revival time t = h/2Ω for different V/Ω. The red and white bars correspond to experimental measurements and ideal numerical simulations, respectively. The blue bars indicate the full numerical simulations that include all interactions (including state-changing terms), SPAM errors and finite-temperature effects. Solid lines in d and e are simulations based on equation (1) with resonant flip-flop interactions, whereas the dashed lines and their shaded confidence intervals are Monte Carlo simulations that also include state-changing interaction terms and thermal variations of R_AB, along with contributions from single-atom dynamics due to state preparation infidelity. All error bars are the standard error of several independent datasets.