Practical corner-transfer density-matrix renormalization group for three-dimensional classical lattice models

In their 1997 review of the corner transfer matrix renormalization group (CTMRG), Nishino and Okunishi formulated its extension to three-dimensional (3D) classical lattice models: divide the cubic system into eight corner transfer tensors, construct the density submatrix as their tensor product, and renormalize by keeping its dominant eigenvectors. The construction is well defined but was computationally too heavy even for a small number of retained states m, and this has remained the central unresolved issue of the CTMRG program. The open problem is to realize the density-matrix (environment-based) corner-tensor renormalization in three dimensions as a practical, systematically improvable algorithm with controlled convergence on standard benchmarks.

Background

Real-space renormalization-group (RG) methods approximate the partition function $Z$ of a classical lattice model by repeatedly coarse-graining local degrees of freedom while discarding the least important states. White's density matrix renormalization group (DMRG) made this truncation variational for one-dimensional quantum systems: one keeps the eigenvectors of a reduced density matrix (the density submatrix) with the largest eigenvalues. Nishino and Okunishi transplanted this idea to two-dimensional (2D) classical models using Baxter's corner transfer matrix (CTM) — the transfer matrix of one quadrant of the lattice, introduced for the eight-vertex model: the four quadrant CTMs $A$, $B$, $C$, $D$ multiply to give the density submatrix $\rho = ABCD$ of the central region of an infinite system. Keeping the dominant eigenvectors of $\rho$ defines the corner transfer matrix renormalization group (CTMRG), where $m$ denotes the number of retained basis states. Combined with finite-size scaling (FSS — the systematic analysis of how observables of clusters of linear size $N$ approach the thermodynamic limit at criticality), CTMRG delivered high-precision critical exponents for the $q=2$ and $q=3$ Potts models (in a $q$-state Potts model each spin takes $q$ values and interacts with its nearest neighbors) and the latent heat for $q=5$; for $q=4$ a marginally irrelevant operator produces logarithmic corrections to scaling that complicate the analysis. CTMRG also remains the standard environment engine for contracting infinite projected entangled-pair states (iPEPS — variational tensor-network wavefunctions for 2D quantum systems whose contraction requires precisely such corner environments), and the feedback of CTMRG to 1D quantum systems (the product wave function renormalization group, PWFRG) grew into modern infinite-system algorithms.

For three-dimensional (3D) classical models the construction generalizes formally: divide a cubic cluster into eight corner transfer tensors (CTTs — the higher-rank analogues of the CTM), form the density submatrix as their tensor product, and renormalize by keeping its dominant eigenvectors. Unlike the 2D case, the renormalized CTT is not diagonal, and the density submatrix has far higher rank. In their 1997 review Nishino and Okunishi stated that this 3D extension, although well defined, is computationally too heavy even for small $m$, and listed it as the central unresolved issue of the program, alongside secondary implementation issues for anisotropic, antiferromagnetic, and marginally critical ($q=4$) 2D models. Subsequent 3D practice — notably the higher-order tensor renormalization group (HOTRG), which coarse-grains by local higher-order singular value decomposition (HOSVD) truncation rather than by a density matrix — is practical but does not implement the density-matrix (environment) truncation, which in 2D is known to reach a given accuracy at lower cost than local truncation. This mismatch — a well-defined, density-matrix-optimal 3D construction that nobody has been able to run at useful size, versus practical but locally truncated alternatives — is the subject of this problem.

Problem Statement

The problem, in its source formulation (Nishino and Okunishi, 1997), is to make the extension of CTMRG to three-dimensional classical lattice models practical. The extension is defined constructively: represent the partition function of a cubic system through eight corner transfer tensors, build the density submatrix of the cubic cluster as their tensor product, and perform the RG transformation by keeping the dominant eigenvectors of this density submatrix, with $m$ retained states. In 1998 the authors implemented this algorithm but could run it only at $m=2$ for the 3D Ising model, concluding that the 3D calculation is far heavier than 2D CTMRG and must be improved for realistic use; a 2025 study states that the direct 3D CTMRG still suffers from an extremely high order of bond dimension in computational complexity and from the slow decay of the spectrum of the density matrices constructed by corner-transfer tensors.

The open problem is to construct a renormalization-group algorithm for 3D classical lattice models whose state truncation is the density-matrix (environment) truncation of the eight-corner-transfer-tensor construction: the RG step keeps the dominant eigenvectors of a density submatrix built as a tensor product of corner transfer tensors of the cubic cluster. Algorithmically improved realizations are admissible — the authors themselves invited an improved numerical algorithm for realistic use in 3D — provided the improved algorithm still performs this environment-based (density-matrix) truncation of the corner-transfer-tensor construction rather than replacing it by a purely local truncation. The method must be practical: the computational cost is manageable as a function of the number of retained states $m$, and controlled convergence in $m$ is demonstrated on standard 3D benchmarks (for example, the cubic-lattice Ising and $q\geq3$ Potts models, comparing free energies, critical couplings, critical exponents, or latent heats against established reference values).

A method that coarse-grains 3D tensor networks by purely local truncation (such as HOTRG) does not by itself answer this question, since the problem is precisely whether the density-matrix (environment) character of CTMRG can be retained in three dimensions.

References

1. T. Nishino and K. Okunishi, "Density matrix and renormalization for classical lattice models", in Strongly Correlated Magnetic and Superconducting Systems, Lecture Notes in Physics Vol. 478 (Springer, Berlin, 1997), pp. 167-183. DOI: 10.1007/bfb0104638, https://doi.org/10.1007/bfb0104638; preprint: arXiv:cond-mat/9610107, https://arxiv.org/abs/cond-mat/9610107
2. S. R. White, "Density matrix formulation for quantum renormalization groups", Physical Review Letters 69, 2863-2866 (1992). DOI: 10.1103/PhysRevLett.69.2863, https://doi.org/10.1103/PhysRevLett.69.2863
3. R. J. Baxter, "Corner transfer matrices of the eight-vertex model. I. Low-temperature expansions and conjectured properties", Journal of Statistical Physics 15, 485-503 (1976). DOI: 10.1007/BF01020802, https://doi.org/10.1007/BF01020802
4. T. Nishino and K. Okunishi, "Product Wave Function Renormalization Group", Journal of the Physical Society of Japan 64, 4084-4087 (1995). DOI: 10.1143/JPSJ.64.4084, https://arxiv.org/abs/cond-mat/9510004
5. T. Nishino and K. Okunishi, "Corner Transfer Matrix Renormalization Group Method", Journal of the Physical Society of Japan 65, 891-894 (1996). DOI: 10.1143/JPSJ.65.891, https://arxiv.org/abs/cond-mat/9507087
6. T. Nishino and K. Okunishi, "A Density Matrix Algorithm for 3D Classical Models", Journal of the Physical Society of Japan 67, 3066-3072 (1998). DOI: 10.1143/jpsj.67.3066, https://arxiv.org/abs/cond-mat/9804134
7. K. Okunishi and T. Nishino, "Kramers-Wannier Approximation for the 3D Ising Model", Progress of Theoretical Physics 103, 541-548 (2000). DOI: 10.1143/PTP.103.541, https://arxiv.org/abs/cond-mat/9909097
8. T. Nishino, K. Okunishi, Y. Hieida, N. Maeshima, and Y. Akutsu, "Self-consistent tensor product variational approximation for 3D classical models", Nuclear Physics B 575, 504-512 (2000). DOI: 10.1016/S0550-3213(00)00133-4, https://arxiv.org/abs/cond-mat/0001083
9. N. Maeshima, Y. Hieida, Y. Akutsu, T. Nishino, and K. Okunishi, "Vertical density matrix algorithm: A higher-dimensional numerical renormalization scheme based on the tensor product state ansatz", Physical Review E 64, 016705 (2001). DOI: 10.1103/PhysRevE.64.016705, https://doi.org/10.1103/PhysRevE.64.016705
10. A. Gendiar and T. Nishino, "Latent heat calculation of the three-dimensional $q=3$, 4, and 5 Potts models by the tensor product variational approach", Physical Review E 65, 046702 (2002). DOI: 10.1103/PhysRevE.65.046702, https://arxiv.org/abs/cond-mat/0102425
11. I. P. McCulloch, "Infinite size density matrix renormalization group, revisited", arXiv:0804.2509 (2008). https://arxiv.org/abs/0804.2509
12. R. Orús and G. Vidal, "Infinite time-evolving block decimation algorithm beyond unitary evolution", Physical Review B 78, 155117 (2008). DOI: 10.1103/PhysRevB.78.155117, https://arxiv.org/abs/0711.3960
13. Z. Y. Xie, H. C. Jiang, Q. N. Chen, Z. Y. Weng, and T. Xiang, "Second Renormalization of Tensor-Network States", Physical Review Letters 103, 160601 (2009). DOI: 10.1103/PhysRevLett.103.160601, https://arxiv.org/abs/0809.0182
14. Z. Y. Xie, J. Chen, M. P. Qin, J. W. Zhu, L. P. Yang, and T. Xiang, "Coarse-graining renormalization by higher-order singular value decomposition", Physical Review B 86, 045139 (2012). DOI: 10.1103/PhysRevB.86.045139, https://arxiv.org/abs/1201.1144
15. R. Orús, "Exploring corner transfer matrices and corner tensors for the classical simulation of quantum lattice systems", Physical Review B 85, 205117 (2012). DOI: 10.1103/PhysRevB.85.205117, https://arxiv.org/abs/1112.4101
16. H. W. J. Blöte, W. Guo, and M. P. Nightingale, "Scaling in the vicinity of the four-state Potts fixed point", Journal of Physics A: Mathematical and Theoretical 50, 324001 (2017). DOI: 10.1088/1751-8121/aa7b53, https://arxiv.org/abs/1710.06088
17. S. Morita and N. Kawashima, "Global optimization of tensor renormalization group using the corner transfer matrix", Physical Review B 103, 045131 (2021). DOI: 10.1103/PhysRevB.103.045131, https://arxiv.org/abs/2009.01997
18. P. C. G. Vlaar and P. Corboz, "Simulation of three-dimensional quantum systems with projected entangled-pair states", Physical Review B 103, 205137 (2021). DOI: 10.1103/PhysRevB.103.205137, https://arxiv.org/abs/2102.06715
19. X. F. Liu, Y. F. Fu, W. Q. Yu, J. F. Yu, and Z. Y. Xie, "Variational corner transfer matrix renormalization group method for classical statistical models", Chinese Physics Letters 39, 067502 (2022). DOI: 10.1088/0256-307X/39/6/067502, https://arxiv.org/abs/2203.17098
20. M. Polackova and A. Gendiar, "Anisotropic deformation of the 6-state clock model: Tricritical-point classification", Physica A: Statistical Mechanics and its Applications 624, 128907 (2023). DOI: 10.1016/j.physa.2023.128907, https://arxiv.org/abs/2207.09132
21. S. Nyckees, A. Rufino, F. Mila, and J. Colbois, "Critical line of the triangular Ising antiferromagnet in a field from a $C_3$-symmetric corner transfer matrix algorithm", Physical Review E 108, 064132 (2023). DOI: 10.1103/PhysRevE.108.064132, https://arxiv.org/abs/2306.09046
22. L.-P. Yang, Y. F. Fu, Z. Y. Xie, and T. Xiang, "Efficient calculation of three-dimensional tensor networks", Physical Review B 107, 165127 (2023). DOI: 10.1103/PhysRevB.107.165127, https://arxiv.org/abs/2210.09896
23. X.-Z. Xu, T.-Y. Lin, and G.-M. Zhang, "Efficient optimization of variational tensor-network approach to three-dimensional statistical systems", Physical Review B 112, 134403 (2025). DOI: 10.1103/7dbj-w7cn, https://doi.org/10.1103/7dbj-w7cn; preprint: arXiv:2506.19339, https://arxiv.org/abs/2506.19339