The ability to accurately compute long-range dependencies in quantum systems has been advanced by research into how information flows through complex states despite limited internal memory. A large amount of conditional mutual information, describing the relationship between measurement outcomes, can persist even when the underlying quantum state is relatively simple, represented using an exact matrix product state with a bond dimension of two.

The research reveals that quantum systems can maintain complex relationships between measurements even when they possess limited internal “memory”. Conditional mutual information, a measure of how much useful information persists after multiple observations, can remain substantial despite a simple underlying quantum structure.

Keyly, this long-range connection doesn’t necessitate extensive computational resources; incorporating minimal memory into machine learning models improved their ability to learn these intricate patterns within quantum states. Researchers demonstrated this by examining one-dimensional rotated cluster states, a simplified way to represent complicated quantum behaviour, much like zipping a large computer file to reduce its size.

They measured *conditional mutual information* (CMI), which assesses how much knowing one measurement outcome tells you about another, factoring in other influences, think of deducing if your friend enjoys coffee based on whether they also drink tea. Keyly, the team found substantial CMI could exist with minimal computational resources and incorporated that principle into an artificial intelligence model designed to mimic the structure of quantum memory.

Conditional Mutual Information from Low Bond Dimension Cluster States

Exact matrix product state (MPS) contraction compresses information about complicated quantum systems, reducing size for computer processing and enabling precise calculations of how much useful information persists between measurements within rotated one-dimensional cluster states. The technique enabled work with systems possessing limited internal “memory”, represented by an MPS bond dimension of two. This approach was key because it allowed the team to calculate *conditional mutual information* (CMI), assessing relationships between measurement outcomes while accounting for other influencing factors, all without computationally expensive simulations of larger or more complex structures.

Conditional mutual information (CMI), a method to assess these relationships, was calculated using one-dimensional rotated cluster states at a bond dimension of two; this simplified calculations due to its limited internal “memory”. The team compared performance on sixty-four qubits at a rotation angle of 0.35 radians, utilising an AKLT chain as a control.

Low Bond Dimension Matrix Product States Capture High Fidelity Quantum Information

This represents a leap beyond previous methods unable to accurately represent quantum states at such low computational cost. Even systems represented using exact matrix product state with minimal internal memory can maintain substantial conditional mutual information (CMI), challenging assumptions about system complexity and informational dependencies.

Specifically, the AKLT chain control exhibited predictable CMI length equal to its two-point correlation length, confirming methodological accuracy and providing a baseline for comparison.

Limited State Memory Enables Complex Measurement Correlations

Researchers at the Institute and University of Waterloo have demonstrated that complex relationships between measurements can persist in simplified quantum systems possessing limited internal “memory”. Their work currently focuses on rotated one-dimensional cluster states, prompting questions about whether these findings broadly apply to other, more disordered or higher-dimensional, quantum scenarios; this is particularly relevant given competing approaches like those using generative models for broader reconstruction.

Despite focusing on this specific type of system, their results are important because they demonstrate that vast internal “memory” isn’t necessarily required for complex measurement correlations. The team showed a small virtual memory built into machine learning models eases optimisation for these systems.

The Institute and University of Waterloo’s findings suggest complex connections between measurements don’t demand extensive internal “memory” within a quantum state, potentially simplifying future designs in quantum information processing. Their work shows substantial long-range connections, measured as conditional mutual information, can exist even with limited internal “memory”. This challenges the expectation that intricate correlations require significant resources to maintain; they achieved this using rotated one-dimensional cluster states represented with an exceptionally low bond dimension of two in matrix product state form. Incorporating this principle into machine learning models improved their ability to learn patterns and eased optimisation during training, suggesting new avenues for efficient algorithms handling quantum data.

Researchers demonstrated that complex relationships between measurements can be maintained in simplified quantum systems without requiring extensive internal memory. Using rotated one-dimensional cluster states with a minimal bond dimension of two, they showed long-range connections, measured as conditional mutual information, can persist despite limited “memory” within the system. This challenges expectations about resource requirements for maintaining intricate correlations; incorporating these principles into machine learning models aided optimization and pattern recognition. The team plans further investigation into whether these findings extend to more disordered or higher-dimensional quantum scenarios.

👉 More information
🗞 Conditional Information Flow and Neural Representability of One-Dimensional Rotated Cluster States
✍️ Ziqi Xu
🧠 ArXiv: https://arxiv.org/abs/2610.39392

More like this