Systematically improvable three-dimensional tensor-network renormalization group: controlled entanglement filtering of the area-law obstruction and extraction of universal data in 3D lattice models

The entanglement-entropy area law $S=\alpha L-F$ of a block face is the fundamental obstruction that keeps coarse-graining tensor-network renormalization group (TNRG) schemes from acting as faithful Wilsonian renormalization group (RG) maps in three dimensions: the microscopic edge-entanglement (edge-double-line, EDL, structure) multiplies under exact block-tensor coarse-graining, violating the UV-free and RG-flow conditions, and higher-order TRG (HOTRG) truncation errors exceed 30% at the 3D Ising critical point. Until recently no 3D entanglement-filtering scheme was known; the cube-filtering construction of Lyu and Kawashima (Phys. Rev. E 111, 054140 (2025)) tames the area-law term on the cubic-lattice Ising model and yields a 3D critical fixed-point tensor. What remains open is the direction of the source's own concluding guidance: establish a 3D tensor RG that verifiably satisfies the approximate-Z, RG-flow, and UV-free conditions with systematically improvable truncations, and demonstrate it by controlled extraction of universal data (scaling dimensions with their degeneracy structure and critical exponents) in at least one 3D lattice model. Adjacent open strands include lattice-rotation symmetry in 3D TNRG (where the rotation group is non-abelian) and transferring entanglement-filtering ideas to the divergent terms of entropic c-theorems in three-dimensional continuum quantum field theory.

Background

Real-space renormalization group (RG) methods coarse-grain a lattice model — for example, the Ising model on the cubic lattice — by repeatedly replacing blocks of degrees of freedom with effective ones, so that universal data such as critical exponents can be read off from the flow and its fixed points. In the tensor-network formulation, the partition function is written as the contraction of a network of local tensors, and the RG acts directly on the tensors; the prototypical schemes are the tensor renormalization group (TRG) of Levin and Nave in two dimensions and its 3D extension, the higher-order tensor renormalization group (HOTRG), which coarse-grain by singular-value decompositions truncated to a bond dimension $\chi$. A key diagnostic of a coarse-graining step is the entanglement entropy $S$ of a tensor, obtained by treating the tensor as a quantum state and forming the reduced density matrix of the degrees of freedom on one side of a cut. In 2D this entropy saturates to a constant under coarse-graining, but at 2D criticality it grows logarithmically — the reason 2D schemes such as tensor network renormalization (TNR), the tensor-entanglement filtering of Gu and Wen, and graph-independent local truncations (Gilt) introduce explicit disentangling or filtering operations that remove short-range entanglement such as the corner-double-line (CDL) structure.

In 3D the obstruction is qualitatively worse. The entanglement entropy associated with a face of a block obeys the area law

$S(\rho_{s_{x+}}) = \alpha L - F,$

where $L$ is the linear size of the face, $\alpha$ is a non-universal microscopic prefactor, and $F$ is a universal constant (for the 3D Ising universality class, $F \approx 0.0623$). Lyu and Kawashima formalized why this defeats the plain 3D block-tensor map. They define three properties of an apt RG transformation: the [Approximate Z condition] — the partition function is preserved (to controlled accuracy) by the map; the [RG flow condition] — the map exhibits the correct fixed points and flows, for example flowing to the zero-entanglement high-temperature fixed point in the disordered phase and possessing a genuine critical fixed-point tensor; and the [UV-free condition] — short-scale, non-universal physics is filtered out so that finite computational resources are spent on universal data. The area law violates the second and third conditions even for exact maps. A tensor-network toy model with an edge-double-line (EDL) structure — the 3D analog of CDL, built from edge matrices $e$ sitting on the twelve edges of a cube — transforms under the exact block map with rescaling factor $b$ as $A_{\mathrm{EDL}}(e) \to A_{\mathrm{EDL}}(e^{\otimes b})$, so microscopic entanglement multiplies instead of being filtered. In practical HOTRG calculations the truncation error near the 3D Ising critical point grows beyond 30% (about 38% at $\chi=22$), and no critical fixed-point tensor appears. In 2D, entanglement filtering was a luxury for better accuracy; in 3D it is a basic necessity for any apt TNRG scheme.

The same authors have since constructed the first 3D entanglement-filtering scheme — a graph-independent cube filtering applied at the corners of a $2\times2\times2$ block, with lattice-reflection symmetry exploited to reduce the independent filtering matrices from 24 to 3 — and demonstrated it on the cubic-lattice Ising model (Phys. Rev. E 111, 054140 (2025)); a follow-up (Phys. Rev. B 113, 144109 (2026)) supplied a lattice-reflection-symmetry toolkit and a sector-resolved linearization of the RG map in 2D and 3D. These works tamed the RG-error growth (from over 20–30% down to about 2–6%) and produced a 3D critical fixed-point tensor, from which scaling dimensions with a conformal-tower-like organization could be estimated. However, the estimated scaling dimensions carry typical errors of about 5% without clear improvement as the bond dimension increases, descendant-operator estimates remain unreliable, only the nearest-neighbor Ising model has been tested, the universal constant $F$ has not been extracted, no correctness proof or a priori error bound exists, and incorporating lattice-rotation symmetry in 3D (where the rotation group is non-abelian) is open. Rigorous tensor RG in 3D exists so far only away from criticality and without truncation (Ebel's high-temperature analysis of untruncated, infinite-dimensional tensors). Relatedly, entropic c-theorems (the entropic c-theorem of Casini and Huerta, and the F-theorem of Casini, Huerta, Myers and Yale) state that a central-charge-like quantity — the universal part $F$ of the entanglement entropy of a sphere in 3D — decreases along RG trajectories of continuum quantum field theories; the divergent area-law terms complicate those continuum analyses in the same way $\alpha L$ burdens the lattice RG. The open problem is therefore no longer the existence of a 3D entanglement filter, but establishing such schemes as controlled, systematically improvable Wilsonian RG in three dimensions.

Problem Statement

Construct a three-dimensional tensor-network renormalization group map — including an explicit entanglement-filtering step that renormalizes out the area-law prefactor $\alpha$ of the 3D entanglement-entropy area law $S=\alpha L-F$, i.e., that clears the edge-double-line (EDL) type edge-matrix structure carrying short-scale degrees of freedom — that simultaneously satisfies the three apt-RG conditions of the source in a controlled manner: (i) the [Approximate Z condition]: the partition function is preserved to an accuracy that is controlled and does not deteriorate with successive RG steps; (ii) the [RG flow condition]: the map exhibits the correct flow structure, including convergence to the zero-entanglement high-temperature fixed point in the disordered phase and a genuine critical fixed-point tensor at criticality, from which scaling dimensions with their degeneracy structure can be extracted by linearization; and (iii) the [UV-free condition]: short-scale, non-universal physics — the lattice-scale information encoded in $\alpha L$ — is identified and filtered out by the map, so that the finite computational power and memory are saved for the universal properties. 'Controlled manner' means that the truncation parameters of the scheme (bond dimensions and filtering tolerances) can be enlarged so that the approximations systematically improve, and this improvability must be demonstrated rather than asserted: the errors of the extracted universal data decrease as the truncation parameters increase, rather than fluctuating between best-case values. The demonstration must extract universal data — scaling dimensions (with degeneracy structure) and critical exponents — in at least one 3D lattice model with controlled error estimates, following the source's guidance that entanglement filtering in 3D should renormalize out the area-law prefactor $\alpha$ and that the EDL structure provides a good guidance for the design of a better 3D TNRG. Further strands recorded in the source and its follow-ups — importing entanglement-filtering ideas to handle the divergent terms in entropic c-theorem analyses of 3D continuum quantum field theory, and incorporating lattice-rotation symmetry in 3D TNRG (where the rotation group is non-abelian) — are open adjacent directions that are named for fidelity to the source but whose resolution is not part of the completion boundary; a solution that additionally advances either of them would exceed, not merely meet, this problem.

References

1. Xinliang Lyu, Naoki Kawashima, "Essential difference between 2D and 3D from the perspective of real-space renormalization group", Journal of Statistical Physics 193, 32 (2026). DOI 10.1007/s10955-026-03587-1; arXiv:2311.05891, https://arxiv.org/abs/2311.05891
2. Xinliang Lyu, Naoki Kawashima, "Three-dimensional real-space renormalization group with well-controlled approximations", Physical Review E 111, 054140 (2025). DOI 10.1103/PhysRevE.111.054140; arXiv:2412.13758, https://arxiv.org/abs/2412.13758
3. Xinliang Lyu, Naoki Kawashima, "Lattice-reflection symmetry in tensor-network renormalization group with entanglement filtering in two and three dimensions", Physical Review B 113, 144109 (2026). DOI 10.1103/z7sk-7sq8; arXiv:2510.19428, https://arxiv.org/abs/2510.19428
4. Xinliang Lyu, "Lattice and PT symmetries in tensor-network renormalization group: Case study of a hard-square lattice gas model", Physical Review B 114, 034111 (2026). DOI 10.1103/vl23-xsyr; arXiv:2603.25492, https://arxiv.org/abs/2603.25492
5. Nikolay Ebel, "3D Tensor Renormalisation Group at High Temperatures", Annales Henri Poincaré 26, 1291 (2025). DOI 10.1007/s00023-024-01464-9; arXiv:2401.04229, https://arxiv.org/abs/2401.04229
6. Z. Y. Xie, J. Chen, M. P. Qin, J. W. Zhu, L. P. Yang, T. Xiang, "Coarse-graining renormalization by higher-order singular value decomposition", Physical Review B 86, 045139 (2012). DOI 10.1103/PhysRevB.86.045139; arXiv:1201.1144, https://arxiv.org/abs/1201.1144
7. Michael Levin, Cody P. Nave, "Tensor Renormalization Group Approach to Two-Dimensional Classical Lattice Models", Physical Review Letters 99, 120601 (2007). DOI 10.1103/PhysRevLett.99.120601; arXiv:cond-mat/0611687, https://arxiv.org/abs/cond-mat/0611687
8. Glen Evenbly, Guifre Vidal, "Tensor Network Renormalization", Physical Review Letters 115, 180405 (2015). DOI 10.1103/PhysRevLett.115.180405; arXiv:1412.0732, https://arxiv.org/abs/1412.0732
9. Markus Hauru, Clement Delcamp, Sebastian Mizera, "Renormalization of tensor networks using graph independent local truncations", Physical Review B 97, 045111 (2018). DOI 10.1103/PhysRevB.97.045111; arXiv:1709.07460, https://arxiv.org/abs/1709.07460
10. Zheng-Cheng Gu, Xiao-Gang Wen, "Tensor-Entanglement-Filtering Renormalization Approach and Symmetry Protected Topological Order", Physical Review B 80, 155131 (2009). DOI 10.1103/PhysRevB.80.155131; arXiv:0903.1069, https://arxiv.org/abs/0903.1069
11. Tom Kennedy, Slava Rychkov, "Tensor RG Approach to High-Temperature Fixed Point", Journal of Statistical Physics 187, 33 (2022). DOI 10.1007/s10955-022-02924-4; arXiv:2107.11464, https://arxiv.org/abs/2107.11464
12. Nikolay Ebel, Tom Kennedy, Slava Rychkov, "Rotations, Negative Eigenvalues, and Newton Method in Tensor Network Renormalization Group", Physical Review X 15, 031047 (2025). DOI 10.1103/y3xz-t2w8; arXiv:2408.10312, https://arxiv.org/abs/2408.10312
13. H. Casini, M. Huerta, "A c-theorem for the entanglement entropy", Journal of Physics A: Mathematical and Theoretical 40, 7031 (2007). DOI 10.1088/1751-8113-40-25-S57; arXiv:cond-mat/0610375, https://arxiv.org/abs/cond-mat/0610375
14. Horacio Casini, Marina Huerta, Robert C. Myers, Alexandre Yale, "Mutual information and the F-theorem", Journal of High Energy Physics 2015, 003 (2015). DOI 10.1007/JHEP10(2015)003; arXiv:1506.06195, https://arxiv.org/abs/1506.06195
15. Tatsuma Nishioka, "Entanglement entropy: holography and renormalization group", Reviews of Modern Physics 90, 035007 (2018). DOI 10.1103/RevModPhys.90.035007; arXiv:1801.10352, https://arxiv.org/abs/1801.10352
16. W. Zhu, C. Han, E. Huffman, J. S. Hofmann, Y.-C. He, "Uncovering conformal symmetry in the 3d Ising transition: State-operator correspondence from a quantum fuzzy sphere regularization", Physical Review X 13, 021009 (2023). DOI 10.1103/PhysRevX.13.021009; arXiv:2210.13482, https://arxiv.org/abs/2210.13482
17. Atsushi Ueda, Sander De Meyer, Adwait Naravane, Victor Vanthilt, Frank Verstraete, "Global Tensor Network Renormalization for 2D Quantum systems: A new window to probe universal data from thermal transitions", arXiv:2508.05406 (2025), https://arxiv.org/abs/2508.05406
18. Shinichiro Akiyama, Raghav G. Jha, Jun Maeda, Yuya Tanizaki, Judah Unmuth-Yockey, "Tensor renormalization group approach to critical phenomena via symmetry-twisted partition functions", Physical Review D 113, 074502 (2026). DOI 10.1103/ywhr-lt6w; arXiv:2601.02681, https://arxiv.org/abs/2601.02681
19. Victor Vanthilt, Adwait Naravane, Chenqi Meng, Atsushi Ueda, "A Practical Introduction to Tensor Network Renormalization with TNRKit.jl", SciPost Phys. Codeb. 77 (2026). DOI 10.21468/SciPostPhysCodeb.77; arXiv:2604.06922, https://arxiv.org/abs/2604.06922
20. Shinichiro Akiyama, "Renormalization group on tensor networks", Proceedings of Science 518, 003 (2026). DOI 10.22323/1.518.0003; arXiv:2603.02741, https://arxiv.org/abs/2603.02741