🔬 condensed matter

Estimating applied potentials in cold atom lattice simulators

This paper proposes a robust experimental protocol that leverages Feshbach resonances to turn off interactions, enabling the high-precision estimation of arbitrary site-dependent potentials in cold atom lattice simulators by analyzing the time evolution of snapshots from a known initial state.

Original authors: Bhavik Kumar, Daniel Malz

Published 2026-10-09

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Bhavik Kumar, Daniel Malz

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

In the quest to understand the complex behavior of matter, physicists often turn to nature's most fundamental building blocks: atoms. By trapping clouds of these ultra-cold atoms in grids of light created by lasers, scientists have built powerful tools known as quantum simulators. These devices act as controllable laboratories where researchers can mimic the behavior of materials that are too difficult to study directly, such as superconductors or exotic magnetic materials. The power of these simulators lies in their precision; scientists can arrange the atoms in specific patterns and adjust the energy levels at each point in the grid to create custom environments. However, a persistent challenge has been that the tools used to shape these light grids are not perfect. Just as a painter cannot place a brushstroke with infinite precision, the lasers used to create these atomic grids have physical limits, meaning the intended energy landscape often differs slightly from the one that actually exists. This small, consistent mismatch creates a bias that accumulates over time, causing the simulation to drift away from the truth and preventing the high-precision measurements needed to solve deep scientific mysteries.

To solve this problem, researchers Bhavik Kumar and Daniel Malz have developed a new method to measure and correct these hidden imperfections. Their work focuses on a specific type of quantum simulator where atoms can move, or "hop," from one spot in the light grid to another. The team realized that even though the exact energy at each spot is unknown, the way the atoms move and spread out over time contains a complete record of that hidden landscape. By turning off the interactions between the atoms—a trick possible with certain types of atoms using magnetic fields—the researchers made the system's behavior mathematically simple enough to predict. They proposed a protocol where scientists prepare a simple, known starting arrangement of atoms, let them evolve for a short period, and then take a series of snapshots of where the atoms are located. By analyzing how the density of atoms changes at each spot over time, the method can work backward to reconstruct the exact potential energy landscape that the atoms actually experienced, effectively revealing the true shape of the light grid.

The researchers tested this idea through detailed computer simulations to see if it could work in a real laboratory setting. They found that the method is remarkably efficient and robust. Instead of needing to measure the system for a long time or in many different ways, the protocol requires only a short evolution period, lasting less than ten times the time it takes for an atom to hop to a neighboring spot. Furthermore, the team discovered that taking measurements at just a handful of different time points is sufficient to recover the entire potential landscape with high accuracy. Even when the initial setup of the atoms is not perfect, or when the rate at which atoms hop is slightly miscalibrated, the method remains effective. In fact, the researchers showed that by adjusting the calculation to account for these uncertainties simultaneously, the system can still pinpoint the correct energy values. This approach is scalable, meaning it works just as well for grids with twenty sites as it does for grids with a hundred, and it does not require any new or exotic equipment to implement.

The implications of this discovery are significant for the future of quantum simulation. Because the method relies on standard imaging techniques already used in many laboratories, it can be adopted immediately by experimentalists to calibrate their devices. The team demonstrated that with a reasonable amount of data—roughly thirty thousand snapshots for a system of one hundred sites—the error in the reconstructed potential can be kept below five percent. This level of precision is enough to enable rigorous testing of complex physical theories, such as those describing how disordered systems behave or how quantum information spreads. By providing a reliable way to verify exactly what potential is being applied, this protocol removes a major barrier to achieving high-fidelity simulations. It transforms the quantum simulator from a device that might be slightly off-target into a precise instrument capable of exploring the most intricate details of quantum physics, paving the way for more accurate models of the natural world.

Technical Summary: Estimating Applied Potentials in Cold Atom Lattice Simulators

Problem Statement
Cold atoms in optical lattices serve as versatile platforms for quantum simulation, capable of realizing Hubbard models with site-resolved readout via quantum gas microscopes (QGM). A critical objective is the implementation of arbitrary, site-dependent optical potentials to enable applications in quantum simulation and information processing. However, due to the diffraction limit of optical lattices (where lattice spacings are typically below the diffraction limit), the precise application and calibration of these potentials remain challenging.

In practice, an experimenter intending to apply a target potential $V$ often implements a slightly different potential $W$ . The difference $V - W$ constitutes a systematic bias rather than fluctuating noise. This discrepancy leads to a simulation error that accumulates over time, preventing high-precision quantum simulation. While learning the Hamiltonian from dynamics offers a solution, existing proposals often require complex state preparations, measurements in multiple local bases, or classical simulation of many-body dynamics, which are difficult to implement in current ultracold atom setups where measurements are typically restricted to the fixed number basis (particle positions).

Methodology
The authors propose a simple and efficient experimental protocol to estimate the applied potential (and hopping rate) using only particle-number readout. The protocol relies on two key physical and computational ingredients:

Interaction Suppression: The protocol exploits the ability of certain atomic species to turn off interactions via a Feshbach resonance, reducing the system to a non-interacting (free fermion) model. This makes the time evolution of the system classically efficient to compute.
Data Collection: The experiment involves preparing a known initial state (specifically a charge-density wave), allowing the system to evolve under the unknown Hamiltonian for various times $t_j$ , and collecting snapshots of site occupations using a QGM.

The authors present two approaches to reconstruct the potential from the time-resolved local densities $D_i(t) = \langle c^\dagger_i(t)c_i(t) \rangle$ :

Rigorous Analytic Reconstruction: This approach fits a polynomial to the measured data to infer Hamiltonian parameters via the second derivative of the correlation matrix at $t=0$ . While theoretically sound, numerical analysis reveals this method is highly sensitive to statistical noise, requiring an infeasible number of snapshots ( $R \sim 10^7$ ) for accurate reconstruction.
Efficient Heuristic Protocol: To overcome noise sensitivity, the authors propose minimizing a cost function via gradient-based optimization. The cost function $K(v; C(0))$ measures the squared difference between the predicted occupation (based on a trial potential $v$ ) and the measured data. This approach avoids computing higher-order derivatives and utilizes standard gradient descent algorithms (e.g., L-BFGS).

Key Contributions and Results
The paper demonstrates that the heuristic protocol is robust, scalable, and experimentally feasible. Key findings from numerical experiments include:

Optimal Parameters: The reconstruction error saturates for a maximal evolution time of $t_{max} \approx 5$ (in units of hopping time) and a small number of sampled time points ( $S \gtrsim 5$ ). Increasing these parameters further yields negligible improvement.
Scalability: The mean reconstruction error scales as $\epsilon_{MRE} \sim 1/\sqrt{M}$ , where $M$ is the total number of snapshots, and remains largely independent of the system size $N$ . For systems up to 100 sites, a mean reconstruction error of $\le 5\%$ is achievable with approximately $3 \times 10^4$ snapshots.
Robustness to Errors: The protocol remains robust against state-preparation errors and uncertainties in the hopping rate. When the hopping rate is mismatched, performing a joint optimization over both the on-site potentials and the hopping rate effectively mitigates the resulting bias.
Initialization Robustness: The optimization converges to the true potential even when the initial guess differs from the true potential by a significant amount (initialization uncertainty $\lambda \lesssim 1$ ). While large deviations may require multiple initial guesses, the method does not fail.

Significance and Claims
The authors claim that this protocol paves the way toward precision quantum simulation with arbitrary potentials by providing a scalable tool for Hamiltonian calibration. The method is notable for:

Experimental Feasibility: It requires no additional experimental modifications (such as local control) and utilizes standard QGM readout and global quenches.
Classical Efficiency: By leveraging non-interacting dynamics, the post-processing is computationally efficient, avoiding the "curse of dimensionality" associated with many-body simulations.
Low Sample Complexity: The protocol achieves high accuracy with a manageable number of experimental shots, making it suitable for benchmarking many-body localization and randomized protocols.

The paper positions this work as a robust, analog-compatible shadow-tomography scheme that numerically learns the true inversion map for Gaussian free fermions. While the current implementation focuses on non-interacting systems and static potentials, the authors identify extending this approach to interacting systems and time-dependent potentials as a direction for future work. They also suggest the potential utility of "shadow-calibration" methods inspired by randomized measurements, though they do not implement these extensions in the current study.

More like this