⚛️ quantum physics

Optimizing quantum encodings for analog simulation through dynamical algebra reachability

This paper introduces a geometry-independent framework that optimizes analog quantum encodings by aligning the target Hamiltonian's spectral algebra with the device's native control capabilities through Riemannian gradient descent, thereby significantly improving simulation fidelity and reducing sensitivity to physical geometry, as demonstrated on a Rydberg-atom processor simulating the deuteron.

Original authors: Mariane Mangin-Brinet

Published 2026-09-11

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Mariane Mangin-Brinet

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 universe, scientists often turn to quantum computers, machines that harness the strange rules of the subatomic world to solve problems too complex for any classical supercomputer. While digital quantum computers break tasks down into a sequence of tiny, discrete steps, analog quantum computers take a different path. Instead of chopping a problem into pieces, they build a physical model that mimics the target system directly, letting the machine evolve naturally over time. This approach is particularly powerful for studying how groups of particles interact, such as in the heart of an atom or in new materials. However, there is a catch. The physical parts of an analog computer are not infinitely flexible; they can only push and pull in specific ways. If the way a scientist maps a problem onto the machine does not align with these natural movements, the computer simply cannot reach the answer, no matter how long it runs or how carefully the controls are tuned.

The core challenge lies in the translation process. To simulate a physical system, researchers must first encode it into the language of the computer's qubits, the basic units of information. There are many ways to do this translation, and while they all describe the same underlying physics, they differ drastically in how well they fit the machine's native controls. A poor choice of encoding can place the necessary dynamics in a direction the hardware cannot reach, rendering the simulation impossible. A good choice, however, can make the impossible possible. The question is how to find the right translation before the experiment begins, without having to guess and check every physical configuration of the machine.

A researcher has developed a new method to solve this puzzle by looking at the mathematical "shape" of the problem and the machine's capabilities. They treat the target problem and the computer's controls as two different collections of directions in a high-dimensional space. The goal is to rotate the representation of the problem until its directions line up as closely as possible with the directions the machine can actually move. This is not about changing the physics of the problem, which remains exactly the same, but about changing the perspective from which the problem is viewed. By finding the orientation that best matches the machine's natural abilities, the researcher can ensure that the simulation is not fighting against the hardware.

To test this idea, the researcher applied their method to a specific problem from nuclear physics: simulating a deuteron, which is a simple atomic nucleus made of one proton and one neutron. They used a model of a Rydberg-atom computer, a type of analog machine where atoms are excited to high energy states and interact with each other. These machines have a specific set of controls: they can drive all atoms at once with a global signal, adjust their energy levels, and change the phase of the signal. The researcher first calculated the "reachable space" of this machine—the set of all possible states it can generate using only these global controls, ignoring for a moment the specific arrangement of the atoms. They then compared this reachable space to the mathematical structure of the deuteron problem.

The results revealed a sharp divide between small and larger systems. For the smallest versions of the problem, involving three or four qubits, the standard way of encoding the deuteron was completely misaligned with the machine's capabilities. The mathematical directions required to describe the nucleus simply did not exist within the set of moves the machine could make. No amount of tuning the pulses or adjusting the timing could fix this; the simulation was fundamentally blocked. However, when the researcher applied their new optimization method, they found a different way to encode the same problem. This new encoding rotated the mathematical structure so that it fit perfectly within the machine's reachable space. In simulations, this change allowed the machine to reach the exact ground state of the deuteron with perfect accuracy, a feat that was impossible with the standard encoding.

For larger systems, involving up to eight qubits, the story became more nuanced. Even with the optimized encoding, the machine could not mathematically reach every single direction required to describe the full complexity of the nucleus. The alignment was partial, not perfect. Yet, when the researcher allowed the physical arrangement of the atoms to vary as well, the machine could still reach the specific state they wanted with extremely high precision. This finding clarified a crucial distinction: while it is ideal for the entire mathematical structure of a problem to fit the machine's controls, it is not strictly necessary to prepare a single specific state, like the lowest energy level. The physical interactions between the atoms, which depend on their geometry, provide extra resources that can compensate for the mathematical mismatch.

The most significant practical benefit of this work emerged when the researcher tested the optimized encoding on different physical layouts of the atoms. When using the standard, unoptimized encoding, the success of the simulation depended heavily on the specific arrangement of the atoms; some layouts worked well, while others failed completely. In contrast, the single optimized encoding, which was determined without knowing the specific layout, worked robustly across all the different arrangements tested. It consistently produced high-fidelity results, regardless of how the atoms were spaced. This suggests that the new method acts as a powerful preprocessing step. It allows scientists to choose a representation of their problem that is structurally compatible with the machine's basic controls before they even decide on the physical details of the experiment.

This approach offers a clear path forward for analog quantum simulation. It separates the problem of "how to write the code" from the problem of "how to build the machine." By optimizing the encoding first, researchers can eliminate the most fundamental barriers to success, ensuring that the problem they are trying to solve is actually reachable by the hardware they have. While the method does not replace the need to fine-tune the physical controls or the geometry of the atoms, it removes the guesswork from the initial design. It ensures that the simulation starts on solid ground, maximizing the chances that the complex dance of quantum particles can be successfully emulated by the analog computer.

Technical Summary: Optimizing Quantum Encodings for Analog Simulation through Dynamical Algebra Reachability

Problem Statement
Analog quantum computers simulate many-body dynamics by directly exploiting the continuous evolution of a physical system. However, a fundamental limitation arises because the native control Hamiltonians of an analog processor generate only a restricted operator space. Consequently, the fidelity with which a target Hamiltonian's dynamics can be reproduced depends not merely on spectral agreement (energy levels) but on whether the physical controls can actually generate the required evolution. A given physical problem can be mapped onto a quantum register in many unitarily equivalent ways (encodings) that share the same spectrum but differ drastically in their compatibility with the device's native controls. A poorly chosen encoding may place relevant dynamics along directions in Hilbert space that the analog controls cannot reach, regardless of pulse sequence or geometry optimization. This work addresses the need to diagnose and optimize this compatibility before device-specific geometry and pulse optimization are attempted.

Methodology
The author introduces a geometry-independent framework based on dynamical algebra reachability. The core methodology involves:

Spectral Algebra Definition: The target model is characterized not just by its Hamiltonian $H$ , but by the spectral algebra $\mathcal{A}(H) = \text{span}\{I, H, H^2, \dots, H^{r-1}\}$ , where $r$ is the number of distinct eigenvalues.
Krylov Space Construction: The analog device is characterized by a Krylov-type operator space $\mathcal{K}$ , generated by the repeated commutators of the device's independently tunable control Hamiltonians (e.g., global Rabi drive, detuning, phase) acting on a fixed initial state. Crucially, this construction excludes geometry-dependent interaction terms (like Ising interactions dependent on atomic positions) to create a geometry-independent diagnostic.
Principal Angle Optimization: The compatibility between an encoding and the hardware is quantified by the principal angles between $\mathcal{A}(H)$ and $\mathcal{K}$ . Since unitary conjugation preserves the spectrum, the encoding problem is formulated as an optimization over the unitary orbit of $H$ ( $H(U) = UHU^\dagger$ ).
Riemannian Gradient Descent: The author maximizes a smooth subspace-overlap functional $F(U) = -\sum \sigma_i^2$ , where $\sigma_i$ are the singular values (cosines of principal angles) of the overlap matrix between the transported spectral algebra and the Krylov space. This is solved using Riemannian gradient descent on the unitary manifold $U(d)$ .

Key Contributions

Geometry-Independent Preprocessing: The paper establishes a criterion to identify representation-level incompatibilities before determining the specific atomic geometry or pulse sequence. This separates the representation problem from the hardware/control optimization.
Distinction Between Reachability Levels: The work clarifies the difference between algebra-level reachability (access to the full spectral algebra generated by $H$ ) and state-level reachability (access to a specific eigenstate, such as the ground state). It demonstrates that full algebraic containment is sufficient but not necessary for high-fidelity state preparation when geometry-dependent resources are available.
Optimization Framework: It provides a constructive method to find a spectrally equivalent representation whose target algebra is better aligned with native controls, using continuous optimization rather than discrete encoding selection (e.g., standard binary vs. Gray code).

Results
The framework is applied to the deuteron Hamiltonian (restricted to the Hamming-weight-one sector) encoded on a Rydberg-atom analog processor with global controls.

Small Systems ( $N=3, 4$ ): The standard binary (SB) encoding is strongly misaligned with the geometry-independent Krylov space (only 1 out of 3 or 4 principal angles has a cosine of 1). Direct dynamical optimization of controls fails to achieve high fidelity. However, principal-angle optimization drives all principal angles to unity, placing the full spectral algebra within the accessible space. This allows for exact ground-state preparation (fidelity $\approx 100\%$ ) even with fixed geometry.
Larger Systems ( $N=5, \dots, 8$ ): For these sizes, the SB encoding does not achieve full algebraic containment (intersection dimension remains 1), and optimization cannot fully align the entire algebra. Nevertheless, joint optimization of geometry and pulses still achieves fidelities $>0.999$ . This confirms that while algebraic containment is a strong indicator, it is not strictly necessary for preparing a single state if geometry-dependent interactions are optimized.
Robustness to Geometry: A single geometry-independent optimized encoding was tested across five different fixed atomic geometries. For the unoptimized SB encoding, fidelity varied wildly (from $<10^{-9}$ to $>0.99$ ) depending on geometry. The optimized encoding maintained high fidelity ( $>0.94$ to $>0.9999$ ) across all geometries, significantly reducing sensitivity to the physical layout.

Significance
The paper argues that spectral equivalence alone is insufficient for analog quantum simulation; the encoding must be compatible with the hardware's native operator space. The proposed principal-angle optimization serves as a systematic preprocessing step that:

Identifies and removes representation-level incompatibilities before expensive device-specific optimizations.
Clarifies when full encoding optimization is necessary (small systems with algebraic obstructions) versus when geometry-dependent control resources can compensate (larger systems).
Offers a low-cost, offline design principle for current analog platforms (like Rydberg atoms and trapped ions) that natively implement global controls and Ising-type interactions, converting models with mismatched native interactions (e.g., XY hopping) into forms reachable by the available hardware.

The author concludes that this approach provides a robust route for mapping target Hamiltonians onto constrained analog quantum hardware, distinguishing between algebraic reachability (essential for full spectral control) and state-level reachability (sufficient for specific eigenstate preparation).

Technical Summary: Optimizing Quantum Encodings for Analog Simulation through Dynamical Algebra Reachability

More like this