Tensor Networks Simplify Physics Simulations

Victor Vanthilt and colleagues have created new software to simplify complex calculations in theoretical physics. TNRKit, a Julia package, is designed for Tensor Network Renormalization (TNR) of both two- and three-dimensional classical statistical models and Euclidean lattice field theories. The package builds upon existing TensorKit tools and offers a framework for creating tensor networks and refining them with algorithms like TRG, HOTRG, and LoopTNR. It allows the extraction of key universal conformal data, such as scaling dimensions and the central charge, directly from fixed-point tensors. This provides a practical and flexible platform for advancing modern tensor renormalization algorithms.

Simplifying complex physical systems via tensor network coarse-graining

Tensor Network Renormalization forms the core of this new software, providing a technique for simplifying complex calculations by representing data as interconnected networks. This approach mirrors the reduction of a complex electrical circuit to a simpler, equivalent form, enabling the tackling of previously intractable problems in theoretical physics. The process begins by constructing a ‘partition function’, a mathematical tool mapping all possible configurations of a system to calculate its probability of being in a particular state.

TNRKit.jl, a new open-source software package built upon TensorKit.jl, was developed to enable tensor network renormalization for two and three-dimensional models and field theories. The software then ‘coarse-grains’ this network, systematically reducing complexity while preserving essential physical information, utilising algorithms like TRG and HOTRG to refine the network until key properties can be readily extracted. This offers a complementary path-integral formalism to the Hamiltonian-based DMRG, particularly suited for lattice gauge theories and quantum field theories.

The package enables the extraction of universal conformal data, such as scaling dimensions and the central charge, directly from fixed-point tensors, allowing detailed analysis of physical systems. A symmetry-aware framework sharply boosts efficiency by using symmetric, block-diagonal tensor networks, providing substantial computational gains. Researchers have expanded TNRKit.jl’s capabilities by implementing models exhibiting discrete symmetries, including the Zq clock model, a generalisation of the Ising model, allowing analysis of systems with more than two spin states.

The set of tools constructs initial tensors via a character expansion, a mathematical technique decomposing the Boltzmann weight, defining the energy of spin configurations, into factors linked by intermediate indices. Not only does this simplify calculations, but it also enforces symmetry directly within the tensor network, preventing errors arising from numerical inaccuracies. Block-diagonal tensor networks manifest this symmetry. Specifically for the Ising model, TNRKit.jl defines a 2x2x2x2 tensor with only eight non-zero entries, reducing computational demands by utilising the Z2 symmetry.

Julia package streamlines tensor network renormalisation for conformal data extraction

The unveiling of TNRKit.jl, a new open-source package for Tensor Network Renormalization (TNR), represents a computational milestone comparable to running demanding calculations on a Windows 95 machine with only 16 megabytes of RAM. This Julia-based set of tools is the first to offer a complete, publicly accessible resource for TNR methods, previously requiring individuals to individually code complex algorithms. While the package currently supports models with binary link variables, extending this framework to accommodate higher-order symmetries and continuous fields remains a key area for future development. Individuals can now investigate how systems behave at different scales, a process known as renormalization, without needing to individually code complex algorithms. By providing a symmetry-aware framework, the package efficiently constructs and refines tensor networks, representations of a system’s many interacting parts. Extracting universal conformal data, such as scaling dimensions and central charge, directly from these networks offers insights into fundamental physical properties.

Extending TNRKit.jl to higher spin states and continuous fields

Tensor network methods are increasingly relied upon by scientists to unravel the behaviour of complex physical systems, spanning materials science to particle physics. Although TNRKit.jl offers a strong step forward by streamlining these calculations, the package presently supports models with binary link variables, restricting its immediate application to systems with more than two spin states. Expanding this framework to encompass higher-order symmetries and continuous fields represents a substantial challenge, demanding new approaches to tensor manipulation and efficient algorithms.

Tensor network renormalisation is a powerful technique for simplifying complex calculations in areas like materials science and high-energy physics, allowing individuals to model many interacting particles more efficiently. The package’s focus on usability and extensibility will accelerate research, enabling scientists to test existing algorithms and develop new ones. This will ultimately allow for more comprehensive modelling of complex physical phenomena.

TNRKit is a new, open-source software package that simplifies tensor network renormalisation calculations for two- and three-dimensional models. This means researchers can now investigate how systems behave at different scales without needing to write complex code themselves. The package efficiently builds and refines tensor networks, and it allows extraction of universal conformal data like scaling dimensions and central charge, providing insights into fundamental physical properties. The authors intend to extend the framework to accommodate higher-order symmetries and continuous fields in future work.