Variational Monte Carlo (VMC) calculations based on tensor networks (TN) have recently achieved competitive accuracy in ground-state calculations of strongly correlated spin and fermionic systems. However, existing tensor network VMC (TN-VMC) algorithms have not been formulated in a manner that can fully utilize GPU acceleration. We tackle the key missing ingredient, namely, the vectorized evaluation of tensor network amplitudes and tensor network operations. This ensures high GPU utilization by batching over computations with identical structure. In particular, we show how to achieve vectorization for the practically relevant case of abelian symmetric tensor networks (which includes fermionic tensor networks) by developing a ``flat'' tensor network formalism for block-sparse tensor representation and contraction. Using this, we construct a GPU-adapted symmetric TN-VMC workflow with batched tensor network computation. In the two-dimensional Fermi--Hubbard model, we demonstrate a GPU speedup of up to $300 \times$ over single core CPU implementations, for fermionic TN variational wavefunctions.Variational Monte Carlo (VMC) calculations based on tensor networks (TN) have recently achieved competitive accuracy in ground-state calculations of strongly correlated spin and fermionic systems. However, existing tensor network VMC (TN-VMC) algorithms have not been formulated in a manner that can fully utilize GPU acceleration. We tackle the key missing ingredient, namely, the vectorized evaluation of tensor network amplitudes and tensor network operations. This ensures high GPU utilization by batching over computations with identical structure. In particular, we show how to achieve vectorization for the practically relevant case of abelian symmetric tensor networks (which includes fermionic tensor networks) by developing a ``flat'' tensor network formalism for block-sparse tensor representation and contraction. Using this, we construct a GPU-adapted symmetric TN-VMC workflow with batched tensor network computation. In the two-dimensional Fermi--Hubbard model, we demonstrate a GPU speedup of up to $300 \times$ over single core CPU implementations, for fermionic TN variational wavefunctions.

physics.comp-phCross-listedarXiv:2608.27861Aug 31, 20260 citations

Tensor Networks

Thibaut Lacroix

Springer thesesJan 1, 20250 citations

Global symmetries in tensor network states: Symmetric tensors versus minimal bond dimension

Sukhwinder Singh, Guifré Vidal

Tensor networks offer a variational formalism to efficiently represent wave functions of extended quantum many-body systems on a lattice. In a tensor network $\mathcal{N}$ , the dimension $\ensuremath{\chi}$ of the bond indices that connect its tensors controls the number of variational parameters and associated computational costs. In the absence of any symmetry, the minimal bond dimension ${\ensuremath{\chi}}^{\mathrm{min}}$ required to represent a given many-body wave function $|\ensuremath{\Psi}\ensuremath{\rangle}$ leads to the most compact, computationally efficient tensor network description of $|\ensuremath{\Psi}\ensuremath{\rangle}$. In the presence of a global, on-site symmetry, one can use a tensor network ${\mathcal{N}}_{\mathrm{sym}}$ made of symmetric tensors. Symmetric tensors allow one to exactly preserve the symmetry and to target specific quantum numbers, while their sparse structure leads to a compact description and lowers computational costs. In this paper we explore the trade-off between using a tensor network $\mathcal{N}$ with minimal bond dimension ${\ensuremath{\chi}}^{\mathrm{min}}$ and a tensor network ${\mathcal{N}}_{\mathrm{sym}}$ made of symmetric tensors, where the minimal bond dimension ${\ensuremath{\chi}}_{\mathrm{sym}}^{\mathrm{min}}$ might be larger than ${\ensuremath{\chi}}^{\mathrm{min}}$. We present two technical results. First, we show that in a tree tensor network, which is the most general tensor network without loops, the minimal bond dimension can always be achieved with symmetric tensors, so that ${\ensuremath{\chi}}_{\mathrm{sym}}^{\mathrm{min}}={\ensuremath{\chi}}^{\mathrm{min}}$. Second, we provide explicit examples of tensor networks with loops where replacing tensors with symmetric ones necessarily increases the bond dimension, so that ${\ensuremath{\chi}}_{\mathrm{sym}}^{\mathrm{min}}>{\ensuremath{\chi}}^{\mathrm{min}}$. We further argue, however, that in some situations there are important conceptual reasons to prefer a tensor network representation with symmetric tensors (and possibly larger bond dimension) over one with minimal bond dimension.Tensor networks offer a variational formalism to efficiently represent wave functions of extended quantum many-body systems on a lattice. In a tensor network $\mathcal{N}$, the dimension $\ensuremath{\chi}$ of the bond indices that connect its tensors controls the number of variational parameters and associated computational costs. In the absence of any symmetry, the minimal bond dimension ${\ensuremath{\chi}}^{\mathrm{min}}$ required to represent a given many-body wave function $|\ensuremath{\Psi}\ensuremath{\rangle}$ leads to the most compact, computationally efficient tensor network description of $|\ensuremath{\Psi}\ensuremath{\rangle}$. In the presence of a global, on-site symmetry, one can use a tensor network ${\mathcal{N}}_{\mathrm{sym}}$ made of symmetric tensors. Symmetric tensors allow one to exactly preserve the symmetry and to target specific quantum numbers, while their sparse structure leads to a compact description and lowers computational costs. In this paper we explore the trade-off between using a tensor network $\mathcal{N}$ with minimal bond dimension ${\ensuremath{\chi}}^{\mathrm{min}}$ and a tensor network ${\mathcal{N}}_{\mathrm{sym}}$ made of symmetric tensors, where the minimal bond dimension ${\ensuremath{\chi}}_{\mathrm{sym}}^{\mathrm{min}}$ might be larger than ${\ensuremath{\chi}}^{\mathrm{min}}$. We present two technical results. First, we show that in a tree tensor network, which is the most general tensor network without loops, the minimal bond dimension can always be achieved with symmetric tensors, so that ${\ensuremath{\chi}}_{\mathrm{sym}}^{\mathrm{min}}={\ensuremath{\chi}}^{\mathrm{min}}$. Second, we provide explicit examples of tensor networks with loops where replacing tensors with symmetric ones necessarily increases the bond dimension, so that ${\ensuremath{\chi}}_{\mathrm{sym}}^{\mathrm{min}}>{\ensuremath{\chi}}^{\mathrm{min}}$. We further argue, however, that in some situations there are important conceptual reasons to prefer a tensor network representation with symmetric tensors (and possibly larger bond dimension) over one with minimal bond dimension.

arXiv:1307.1522Physical Review BSep 30, 201333 citations

Tensor Network Methods

Timo Felser, Simone Montangero

Graduate texts in physicsJan 1, 20260 citations

Tensor network decompositions in the presence of a global symmetry

Sukhwinder Singh, Robert N. C. Pfeifer, Guifré Vidal

Tensor network decompositions offer an efficient description of certain many-body states of a lattice system and are the basis of a wealth of numerical simulation algorithms. We discuss how to incorporate a global symmetry, given by a compact, completely reducible group $\mathcal{G}$ , in tensor network decompositions and algorithms. This is achieved by considering tensors that are invariant under the action of the group $\mathcal{G}$ . Each symmetric tensor decomposes into two types of tensors: degeneracy tensors, containing all the degrees of freedom, and structural tensors, which only depend on the symmetry group. In numerical calculations, the use of symmetric tensors ensures the preservation of the symmetry, allows selection of a specific symmetry sector, and significantly reduces computational costs. On the other hand, the resulting tensor network can be interpreted as a superposition of exponentially many spin networks. Spin networks are used extensively in loop quantum gravity, where they represent states of quantum geometry. Our work highlights their importance in the context of tensor network algorithms as well, thus setting the stage for cross-fertilization between these two areas of research.Tensor network decompositions offer an efficient description of certain many-body states of a lattice system and are the basis of a wealth of numerical simulation algorithms. We discuss how to incorporate a global symmetry, given by a compact, completely reducible group $\mathcal{G}$, in tensor network decompositions and algorithms. This is achieved by considering tensors that are invariant under the action of the group $\mathcal{G}$. Each symmetric tensor decomposes into two types of tensors: degeneracy tensors, containing all the degrees of freedom, and structural tensors, which only depend on the symmetry group. In numerical calculations, the use of symmetric tensors ensures the preservation of the symmetry, allows selection of a specific symmetry sector, and significantly reduces computational costs. On the other hand, the resulting tensor network can be interpreted as a superposition of exponentially many spin networks. Spin networks are used extensively in loop quantum gravity, where they represent states of quantum geometry. Our work highlights their importance in the context of tensor network algorithms as well, thus setting the stage for cross-fertilization between these two areas of research.

arXiv:0907.2994Physical Review ANov 8, 2010279 citations

300 \times

over single core CPU implementations, for fermionic TN variational wavefunctions.

Tensor networks offer a variational formalism to efficiently represent wave functions of extended quantum many-body systems on a lattice. In a tensor network

\mathcal{N}

, the dimension $\ensuremath{\chi}$ of the bond indices that connect its tensors controls the number of variational parameters and associated computational costs. In the absence of any symmetry, the minimal bond dimension ${\ensuremath{\chi}}^{\mathrm{min}}$ required to represent a given many-body wave function $|\ensuremath{\Psi}\ensuremath{\rangle}$ leads to the most compact, computationally efficient tensor network description of $|\ensuremath{\Psi}\ensuremath{\rangle}$. In the presence of a global, on-site symmetry, one can use a tensor network

{\mathcal{N}}_{\mathrm{sym}}

made of symmetric tensors. Symmetric tensors allow one to exactly preserve the symmetry and to target specific quantum numbers, while their sparse structure leads to a compact description and lowers computational costs. In this paper we explore the trade-off between using a tensor network

\mathcal{N}

with minimal bond dimension ${\ensuremath{\chi}}^{\mathrm{min}}$ and a tensor network

{\mathcal{N}}_{\mathrm{sym}}

made of symmetric tensors, where the minimal bond dimension ${\ensuremath{\chi}}_{\mathrm{sym}}^{\mathrm{min}}$ might be larger than ${\ensuremath{\chi}}^{\mathrm{min}}$. We present two technical results. First, we show that in a tree tensor network, which is the most general tensor network without loops, the minimal bond dimension can always be achieved with symmetric tensors, so that ${\ensuremath{\chi}}_{\mathrm{sym}}^{\mathrm{min}}={\ensuremath{\chi}}^{\mathrm{min}}$. Second, we provide explicit examples of tensor networks with loops where replacing tensors with symmetric ones necessarily increases the bond dimension, so that ${\ensuremath{\chi}}_{\mathrm{sym}}^{\mathrm{min}}>{\ensuremath{\chi}}^{\mathrm{min}}$. We further argue, however, that in some situations there are important conceptual reasons to prefer a tensor network representation with symmetric tensors (and possibly larger bond dimension) over one with minimal bond dimension.

\mathcal{G}

, in tensor network decompositions and algorithms. This is achieved by considering tensors that are invariant under the action of the group

\mathcal{G}

. Each symmetric tensor decomposes into two types of tensors: degeneracy tensors, containing all the degrees of freedom, and structural tensors, which only depend on the symmetry group. In numerical calculations, the use of symmetric tensors ensures the preservation of the symmetry, allows selection of a specific symmetry sector, and significantly reduces computational costs. On the other hand, the resulting tensor network can be interpreted as a superposition of exponentially many spin networks. Spin networks are used extensively in loop quantum gravity, where they represent states of quantum geometry. Our work highlights their importance in the context of tensor network algorithms as well, thus setting the stage for cross-fertilization between these two areas of research.