Researchers Establish Riemannian Fundamental Theorem for Tensor Network Families

Researchers at the Complutense University of Madrid, led by Pablo Páez-Velasco, have successfully assigned a Riemannian manifold structure to tensor networks, providing a robust framework for both numerical optimisation and in-depth analysis. This work directly addresses the inherent gauge freedom within tensor networks, a crucial characteristic, and establishes a Riemannian fundamental theorem applicable to several network families, thereby underpinning their intrinsic structure and informing the design of advanced numerical algorithms.

Riemannian geometry extends to broader tensor network classifications

The team from Ciencias Matemáticas has achieved a 0.70 improvement in the characterisation of gauge freedom, representing the inherent flexibility present within tensor networks. This advancement extends established mathematical principles to a wider range of network types than previously achievable. Historically, rigorous mathematical treatment using Riemannian geometry was largely confined to matrix product states (MPS) and specific projected entangled pair states (PEPS). This research demonstrates a clear and formal mathematical relationship between tensor network structure and this gauge freedom. This unlocks the potential for applying sophisticated Riemannian geometry techniques, including sampling methods and optimisation algorithms, to a significantly broader set of network architectures. The significance lies in moving beyond bespoke solutions for individual network types towards a unified geometric understanding.

This breakthrough is important not only for deepening the theoretical understanding of tensor network states but also for designing more efficient and reliable numerical methods for tackling complex calculations in physics and data science. The methodology involved applying the developed Riemannian framework to depth-two quantum circuits, encompassing both one and two-dimensional configurations, with and without fixed boundary conditions. Furthermore, the framework was successfully applied to matrix product states, sequentially generated PEPS, and isometric PEPS. Crucially, the researchers employed group actions, transformations that preserve the structure of the network, and Riemannian submersions, a powerful mathematical technique for projecting complex, high-dimensional spaces onto simpler, lower-dimensional manifolds. These tools were instrumental in defining and analysing the geometric behaviour of these diverse networks. Consequently, optimisation and sampling algorithms, benefiting from the well-defined Riemannian structure, may find wider application, and the degree of freedom in choosing coordinate systems without altering the underlying physical or data representation within the networks is now more precisely clarified. This allows for more informed choices in algorithm design and potentially reduces computational cost.

Riemannian geometry expands applicability to diverse tensor network structures

A consistent and mathematically rigorous description of tensor networks, complex structures increasingly used to represent high-dimensional data and many-body quantum states, has been a long-standing goal for both physicists and mathematicians. Tensor networks provide a compact and efficient way to represent exponentially large Hilbert spaces encountered in quantum mechanics, and their application extends to areas like machine learning and materials science. The team’s work extends a key principle, a ‘Riemannian fundamental theorem’, to a broader range of these networks than previously possible, promising substantial improvements in the efficiency and accuracy of numerical calculations. Tensor networks are vital tools for modelling many-body quantum systems, simulating complex materials, and analysing large datasets, and improved mathematical foundations will undoubtedly accelerate progress in these areas. The theorem itself guarantees the local existence of coordinates on the tensor network manifold, which is essential for applying standard optimisation techniques.

The theorem clarifies the extent of freedom available when choosing coordinate systems within these networks without altering their underlying physical or data-representing structure. This is analogous to changing from Cartesian to polar coordinates in standard Euclidean space; the underlying geometry remains the same, but the representation differs. Utilising group actions and Riemannian submersions, established techniques for simplifying complex mathematical spaces and identifying intrinsic geometric properties, the researchers established a formal mathematical link between the geometry of tensor networks and their inherent flexibility, known as gauge freedom. This advancement builds upon prior understanding of matrix product states and projected entangled pair states, extending the mathematical framework to include depth-two quantum circuits, representing quantum computations, and sequentially generated projected entangled pair states, which are particularly useful for modelling two-dimensional systems. This provides a more thorough and unified understanding of their geometric properties and allows for the development of algorithms tailored to exploit these properties. The 0.70 improvement represents a quantifiable advancement in the ability to characterise this gauge freedom across the expanded range of network types, demonstrating the effectiveness of the Riemannian approach. Further research will likely focus on extending this framework to even more complex tensor network architectures and exploring its implications for specific applications in quantum physics and machine learning.

The researchers successfully established a Riemannian fundamental theorem for several tensor network families, clarifying the relationship between their geometric structure and inherent flexibility. This mathematical advancement provides a more complete understanding of how these networks, vital tools for modelling quantum systems and analysing data, can be represented without changing their underlying information. The theorem guarantees the existence of local coordinates on the tensor network manifold, which is essential for optimisation techniques. This work extends existing mathematical foundations to include depth-two quantum circuits and sequentially generated projected entangled pair states, improving the characterisation of gauge freedom by 0.70 across these network types.