⚛️ quantum physics

Locally optimized variational evolution for quantum many-body systems

This paper introduces a locally optimized variational time-evolution principle for quantum many-body systems that replaces global-state fidelity with a cost function based on local reduced density matrices, enabling efficient simulation of thermalizing dynamics through a hybrid quantum-classical algorithm validated on Quantinuum H2 and IBM Heron processors.

Original authors: Carolin Wille, Max Marvell, Lauren Stewart, Max Murphy, Vinul Wimalaweera, Lesley Gover, Andrew G. Green

Published 2026-09-18

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Carolin Wille, Max Marvell, Lauren Stewart, Max Murphy, Vinul Wimalaweera, Lesley Gover, Andrew G. Green

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 vast landscape of quantum physics, a persistent challenge has long stood in the way of understanding how complex systems change over time. When a collection of particles interacts, they become deeply linked in a way that defies simple description; the state of the whole system grows so intricate that representing it on a classical computer requires resources that explode exponentially as the system gets larger. This complexity suggests that to simulate even a modest number of interacting particles, one would need a computer far more powerful than any currently exists. However, a crucial distinction often gets lost in this mathematical wall: while the global state of the system becomes impossibly complex, the things we can actually measure in the real world are local. We do not observe the entire universe at once; we look at a specific region, a small patch of the system. In many physical situations, particularly those where the system settles into a state of thermal equilibrium, these local regions lose their memory of the microscopic details of the far-away parts of the system. They relax into a predictable state governed by just a few parameters, effectively simplifying the problem of describing them.

This observation forms the bedrock of a new approach developed by researchers at the London Centre for Nanotechnology and University College London. They have devised a method to simulate the evolution of quantum systems that bypasses the need to track the entire, overwhelming global wave function. Instead of trying to keep the whole picture in focus, their technique concentrates exclusively on the local details that matter. By optimizing a mathematical description to match the behavior of small, finite regions of the system, they can accurately predict how local observables change over time. This strategy allows the simulation to remain computationally manageable, even as the system evolves and the global state becomes too complex to handle. The method works by repeatedly projecting the system's state back onto a simplified model that best fits the local data, effectively ignoring the distant, irrelevant complexities that would otherwise bog down the calculation.

The researchers tested this idea, which they call locally optimized matrix-product states, on two different types of quantum systems. First, they looked at a system that is known to be predictable and stable, where the particles interact in a way that preserves a clear, coherent pattern over time. In these early stages of evolution, the new method successfully captured the sharp, rhythmic oscillations of the local particles, matching the results of highly precise theoretical calculations. This demonstrated that the approach does not simply smooth over the interesting details; it retains the coherent dynamics that define the system's behavior in the short term. Next, they turned to a more chaotic system, one where the particles interact in a way that leads to rapid thermalization. In this scenario, the local regions are expected to lose their specific history and settle into a state of maximum disorder, akin to a gas reaching a uniform temperature. Here, the method proved its second strength: as the simulation progressed, the local regions naturally relaxed toward this expected equilibrium. The researchers found that by increasing the size of the local patch they were optimizing, the simulation became even more accurate, suppressing artificial wiggles in the data and converging more closely to the true thermal state.

To prove that this concept could work on actual hardware, not just in theoretical simulations, the team implemented the algorithm on two different quantum processors: the Quantinuum H2 and the IBM Heron. These machines are real, physical devices that are subject to noise and imperfections, unlike the perfect, noiseless computers used in the initial simulations. The researchers encoded their local optimization strategy onto these chips, using a sequence of operations to prepare the quantum state and measure the overlap between the current state and the target local region. Despite the presence of hardware noise and the statistical uncertainty inherent in quantum measurements, the algorithm on both devices successfully recovered the characteristic local dynamics. The results from the noisy quantum hardware aligned with the clean, noiseless simulations, showing that the method is robust enough to handle the realities of current quantum technology.

The significance of this work lies in its shift of perspective regarding what is required to simulate quantum matter. It suggests that the difficulty of predicting local observables is not necessarily tied to the size of the entire system, but rather to the intrinsic complexity of the local evolution itself. By focusing only on the information that remains locally accessible, the researchers have created a tool that scales with this intrinsic complexity rather than the total number of particles. This opens the door to studying larger systems and longer timescales than previously thought possible, particularly in scenarios where thermalization simplifies the local picture. While the method is currently limited by the size of the local patch it can optimize, the results suggest that increasing this patch size systematically improves the accuracy of the prediction. The work provides a concrete path forward for understanding how many-body systems evolve, offering a way to navigate the complexity of the quantum world by looking closely at the small, manageable pieces that define our physical reality.

Technical Summary: Locally Optimized Variational Evolution for Quantum Many-Body Systems

Problem Statement
Conventional quantum simulation of many-body dynamics faces a fundamental barrier: the global wavefunction's entanglement grows rapidly, leading to an extensive exponential complexity that makes classical representation intractable. While quantum computers are motivated by this difficulty, the paper argues that physically accessible observables are local in nature. Their complexity does not necessarily inherit the extensive complexity of the global state. Instead, local observables are governed by an "intrinsic local complexity" that is independent of the total system size, particularly in thermalizing systems where local subsystems lose memory of microscopic details and relax toward equilibrium. The challenge is to simulate these local dynamics without paying the cost of tracking the full global wavefunction.

Methodology
The authors propose a variational time-evolution principle called Locally Optimized Matrix-Product States (LOMPS). Unlike established methods such as Time-Evolving Block Decimation (TEBD) or the Time-Dependent Variational Principle (TDVP), which optimize for global state fidelity, LOMPS optimizes a cost function defined on local reduced density matrices (RDMs).

The algorithm operates on a uniform Matrix Product State (MPS) manifold and proceeds via a two-step protocol for each time step $\Delta t$ :

Time Evolution: The state is evolved by a short Trotter step unitary $U(\Delta t)$ .
Local Projection: The evolved state is projected back onto the variational manifold by minimizing the Hilbert-Schmidt distance between the $L$ -site reduced density matrix of the evolved state and the $L$ -site RDM of the ansatz state.

The cost function $C_L(A')$ is defined as:
$C_L(A') = \frac{1}{2} \|\rho_L[A'] - \rho_L^{\text{evol}}(t + \Delta t)\|_{\text{HS}}^2$
where $\rho_L^{\text{evol}}$ is the target RDM and $\rho_L[A']$ is the ansatz RDM. This minimization leads to closed-form equations of motion for the variational parameters, analogous to TDVP but induced by the metric of the local RDMs rather than the global wavefunction.

The framework is extended to a quantum-classical hybrid algorithm. In this implementation:

The cost function (specifically local fidelity and ansatz purity) is evaluated on a quantum processor using sequential circuit preparation of MPS and destructive SWAP tests to measure RDM overlaps.
The optimization of parameters is performed classically.
The quantum circuit uses a resource-efficient parametrization (e.g., bond dimension $D=2$ ) and open boundary conditions sufficient to construct the local RDMs while tracing out the environment.

Key Contributions

Variational Principle: Introduction of a variational principle that replaces global fidelity with a local cost function, explicitly trading global wavefunction accuracy for an accurate description of local observables.
Algorithmic Implementation: Development of the LOMPS algorithm, which admits closed-form equations of motion and allows for systematic improvement by increasing the patch size $L$ .
Hybrid Quantum-Classical Scheme: Demonstration of a noise-robust quantum-classical counterpart where the local cost is evaluated on hardware, allowing for the exploration of potential intensive quantum advantages.
Theoretical Framework: Provision of a geometric perspective on local thermalization and intrinsic complexity, supported by continuous-time equations of motion.

Results
The paper presents proof-of-principle results on both classical simulations and quantum hardware (Quantinuum H2 and IBM Heron):

Early-Time Coherence: In an integrable transverse-field Ising model quench, LOMPS accurately reproduces the coherent local dynamics, specifically the local Loschmidt echo, matching quasi-exact free-fermion calculations.
Late-Time Thermalization: In a non-integrable tilted-field Ising quench, the algorithm captures the relaxation of local observables toward the infinite-temperature thermal state. Increasing the patch size $L$ systematically suppresses artificial late-time oscillations and improves convergence to the expected thermal steady state.
Hardware Performance: Experiments on IBM and Quantinuum processors successfully recovered qualitative features of local dynamics (local Loschmidt rates) despite shot noise and hardware imperfections. The results showed that the quantum-classical algorithm could track the trajectory defined by the local cost function.
Systematic Improvement: Analysis of the non-integrable Ising quench showed that increasing the patch size $L$ extends the validity time of the simulation before deviation from reference (TEBD) results and exponentially reduces the late-time error relative to the thermal state.

Significance and Claims
The authors claim that this work supports a shift in perspective regarding the simulation of many-body dynamics: the exponential complexity of a global wavefunction need not dictate the difficulty of predicting local observables. Instead, the difficulty is determined by the "intrinsic local complexity"—the spatial range and variational resources required to retain dynamically relevant information.

The paper posits that LOMPS provides a computational strategy to exploit the separation between global complexity and local simplicity, particularly in thermalizing systems. The quantum-classical implementation offers a route to simulate regimes where representing local RDMs becomes classically costly (e.g., in two dimensions). However, the authors remain modest regarding practical quantum advantage, noting that establishing scaling laws with patch size and bond dimension beyond the current proof-of-principle is necessary before such an advantage can be assessed. The work also identifies the principal limitation of finite-window evolution: matching a local RDM does not uniquely determine future evolution, making the identification of the regime of validity (where unresolved information remains irrelevant) a central challenge.

Technical Summary: Locally Optimized Variational Evolution for Quantum Many-Body Systems

More like this