Rigorous origin of the finite-bond-dimension scaling law $\xi(D)\sim D^{\kappa}$ in tensor network renormalization

Tensor network renormalization (TNR) coarse-grains two-dimensional critical lattice models at finite bond dimension $D$, and the truncation acts numerically as an emergent relevant perturbation of the conformal fixed point, generating a correlation length $\xi(D)$ that appears to follow the same power law as matrix product state finite-entanglement scaling, $\xi(D)\sim D^{\kappa}$ with $\kappa=6/(c(1+\sqrt{12/c}))$, where $c$ is the central charge. No rigorous or model-independent derivation of this law exists — not for TNR, and not even for the one-dimensional matrix product state case, where the exponent rests on unverified assumptions about the entanglement spectrum and where the naive conformal-field-theory perturbation structure is known to fail for decoupled copies. The problem is to establish (or refute) this scaling law with a controlled derivation of the exponent.

Background

Two-dimensional classical lattice models at criticality can be written exactly as contracted networks of local tensors, each tensor carrying the Boltzmann weights of one site. Coarse-graining such a network while truncating every intermediate tensor to bond dimension $D$ (the number of values kept per virtual index) defines real-space renormalization on the network: the tensor renormalization group (TRG) of Levin and Nave performs the local truncation, and tensor network renormalization (TNR) of Evenbly and Vidal additionally removes short-range entanglement between steps, so that the coarse-grained tensor is expected to flow toward a universal fixed-point tensor encoding the infrared theory. At a critical point the infrared theory is a two-dimensional conformal field theory (CFT), characterized by its central charge $c$ (a number measuring the density of low-energy degrees of freedom) and by the scaling dimensions and operator product expansion (OPE) coefficients of its local operators, the latter being the constants that govern how a product of two local operators decomposes into a sum of local operators; extracting this conformal data from fixed-point tensors and their transfer matrices is a major application of TNR-type algorithms.

Because every tensor is truncated to bond dimension $D$, a finite-$D$ calculation cannot represent exact scale invariance: repeated coarse-graining eventually generates a finite correlation length $\xi(D)$, or equivalently a gap in the spectrum of the associated one-dimensional quantum problem. The one-dimensional analogue is understood semi-quantitatively. A matrix product state (MPS; a one-dimensional tensor network whose bond dimension limits its entanglement) with bond dimension $D$ approximating a critical chain has a finite correlation length obeying the finite-entanglement scaling law $\xi(D)\sim D^{\kappa}$ with $\kappa=6/(c(1+\sqrt{12/c}))$ (for example $\kappa\approx 2.034$ for the Ising CFT with $c=1/2$ and $\kappa\approx 1.344$ for $c=1$). This exponent was derived heuristically by Tagliacozzo et al. and by Pollmann et al., and simplified by Pirvu et al.; the crucial input is the Calabrese-Lefevre description of the continuum decay of the entanglement spectrum (the ordered Schmidt coefficients across a bipartition of the critical chain). The derivation assumes analyticity properties of the entanglement spectrum in the complex replica plane that have never been verified, as shown by Gliozzi and Tagliacozzo, and the law is therefore often called a scaling hypothesis, following Vanhecke et al. The same finite-$D$ scaling had earlier been observed in the corner transfer matrix renormalization group (CTMRG), the environment-based coarse-graining method introduced by Nishino, Okunishi, and Kikuchi.

Ueda and Oshikawa, in the PhD thesis and the paper of Refs. 1 and 2, made the TNR version quantitative. Analyzing finite-size spectra of the renormalized tensor for the critical Ising and three-state Potts models, they showed that finite $D$ acts as an emergent relevant perturbation (an operator whose coupling grows under renormalization) of the CFT fixed-point Hamiltonian — within the $\mathbb{Z}_2$-symmetric sector of the Ising model this is the thermal, i.e. energy-density, operator — producing a crossover from critical to massive behavior at system sizes $L\sim\xi(D)$, and that $\xi(D)$ follows the MPS exponent within numerical accuracy. The thesis argues heuristically that the same scaling should hold across MPS, CTMRG, and TNR, because a renormalized TNR tensor can be contracted into a corner transfer matrix (the object evolved by CTMRG) and the half-system entanglement spectrum is reproduced by contracting four renormalized tensors; but it states explicitly that no mathematical proof of the scaling of $\xi(D)$ in TNR exists at this point, and it presents the relations underlying this shared-exponent argument as suggestive rather than established.

Later work sharpens the picture without closing it. Schneider et al. constructed an effective field theory for the renormalization flow of finite-$D$ MPS, finding a renormalization-group self-congruent point (a point at which the emergent relevant coupling stops flowing because its lattice-scale strength decreases as the induced correlation length grows) and confirming that the finite-entanglement scaling hypothesis remains valid within this framework, while reiterating that the theoretical origin of the scaling remains unsolved. Hodgkiss et al. developed an implicit-differentiation method that identifies the effective Hamiltonian whose ground state has the largest fidelity with the finite-$D$ MPS: for a single critical Ising chain the conventional picture, in which the single most relevant perturbation dominates, is confirmed, but for two decoupled critical Ising chains subleading relevant perturbations contribute non-negligibly, which leads the authors to call for a reconsideration of the standard finite-entanglement scaling and for the development of a complete theory of finite-bond-dimension effects. Chan and Chen gave an operational definition of the crossover scale separating the finite-size and finite-entanglement regimes in classical tensor-network calculations and used it to extract conformal data without a unique fixed-point tensor, but without deriving the exponent. Mortier et al. showed, for two-dimensional metals with non-chiral Fermi surfaces, that the scaling exponent is not a function of the central charge alone. A solution must therefore explain the mechanism by which truncation generates the emergent relevant perturbation and the universality and domain of validity of the resulting exponent.

Problem Statement

Determine with mathematical rigor — or by a controlled analytic derivation with explicitly stated and independently justified assumptions — the finite-bond-dimension scaling of tensor network renormalization for two-dimensional critical lattice models whose infrared limit is a conformal field theory of central charge $c$. Prove (or disprove) that TNR-type coarse-graining — the family of real-space tensor network renormalization schemes in the sense of the Evenbly-Vidal TNR described in the Background, including variants such as the Loop-TNR used in the numerical studies of Refs. 1 and 2, in which every intermediate tensor is truncated to bond dimension $D$ at each coarse-graining step — induces in the renormalized tensor an emergent relevant perturbation of the conformal fixed point, and that the induced correlation length satisfies $\xi(D)\sim D^{\kappa}$ with the exponent $\kappa=6/(c(1+\sqrt{12/c}))$; or, if this exponent is not correct in general, determine the correct asymptotic law (exponent, logarithmic corrections) and delimit the domain of validity of the original formula — including the role of subleading relevant perturbations, and the classes, such as decoupled copies or Fermi-surface systems, for which the central charge alone is known not to fix the scaling and the correct law must be determined in place of the original formula.

The claim to be settled combines three separately refutable sub-claims: the mechanism claim, that truncation to bond dimension $D$ induces in the renormalized tensor an emergent relevant perturbation of the conformal fixed point; the exponent claim, that the induced correlation length obeys $\xi(D)\sim D^{\kappa}$ with $\kappa=6/(c(1+\sqrt{12/c}))$; and the scope claim, that both hold throughout the TNR-type family above. A controlled derivation establishing all three resolves the problem; so does an explicit member of the class for which $\xi(D)$ does not scale as $D^{\kappa}$, together with the correct asymptotic law and validity boundary for that member. Establishing or refuting the mechanism claim alone, while the exponent claim stands, is a partial result and must be reported as such.

Numerical evidence or consistency checks alone, without a controlled derivation, do not constitute a solution.

References

1. Atsushi Ueda, "Renormalization group flow and fixed-point in tensor network representations", PhD thesis (2024), arXiv:2401.18068, doi:10.48550/arXiv.2401.18068, https://arxiv.org/abs/2401.18068
2. Atsushi Ueda and Masaki Oshikawa, "Finite-size and finite bond dimension effects of tensor network renormalization", Phys. Rev. B 108, 024413 (2023), doi:10.1103/PhysRevB.108.024413, https://arxiv.org/abs/2302.06632
3. Atsushi Ueda and Masahito Yamazaki, "Fixed-point tensor is a four-point function", arXiv:2307.02523 (2023), https://arxiv.org/abs/2307.02523
4. L. Tagliacozzo, Thiago R. de Oliveira, S. Iblisdir, and J. I. Latorre, "Scaling of entanglement support for matrix product states", Phys. Rev. B 78, 024410 (2008), doi:10.1103/PhysRevB.78.024410, https://arxiv.org/abs/0712.1976
5. Frank Pollmann, Subroto Mukerjee, Ari M. Turner, and Joel E. Moore, "Theory of Finite-Entanglement Scaling at One-Dimensional Quantum Critical Points", Phys. Rev. Lett. 102, 255701 (2009), doi:10.1103/PhysRevLett.102.255701, https://doi.org/10.1103/PhysRevLett.102.255701
6. B. Pirvu, G. Vidal, F. Verstraete, and L. Tagliacozzo, "Matrix product states for critical spin chains: Finite-size versus finite-entanglement scaling", Phys. Rev. B 86, 075117 (2012), doi:10.1103/PhysRevB.86.075117, https://arxiv.org/abs/1204.3934
7. Pasquale Calabrese and Alexandre Lefevre, "Entanglement spectrum in one-dimensional systems", Phys. Rev. A 78, 032329 (2008), doi:10.1103/PhysRevA.78.032329, https://doi.org/10.1103/PhysRevA.78.032329
8. Ferdinando Gliozzi and Luca Tagliacozzo, "Entanglement entropy and the complex plane of replicas", J. Stat. Mech. (2010) P01002, doi:10.1088/1742-5468/2010/01/P01002, https://doi.org/10.1088/1742-5468/2010/01/P01002
9. Bram Vanhecke, Jutho Haegeman, Karel Van Acoleyen, Laurens Vanderstraeten, and Frank Verstraete, "Scaling Hypothesis for Matrix Product States", Phys. Rev. Lett. 123, 250604 (2019), doi:10.1103/PhysRevLett.123.250604, https://doi.org/10.1103/PhysRevLett.123.250604
10. Jan T. Schneider, Atsushi Ueda, Yifan Liu, Andreas M. Läuchli, Masaki Oshikawa, and Luca Tagliacozzo, "Self-congruent point in critical matrix product states: An effective field theory for finite-entanglement scaling", SciPost Phys. 18, 142 (2025), doi:10.21468/SciPostPhys.18.4.142, https://arxiv.org/abs/2411.03954
11. Luke Hodgkiss, Laurens Lootens, Atsushi Ueda, Bram Vanhecke, and Frank Verstraete, "On the origin of finite entanglement scaling", arXiv:2607.15124 (2026), https://arxiv.org/abs/2607.15124
12. Sing-Hong Chan and Pochung Chen, "Extracting conformal data from finite-size tensor-network flow in critical two-dimensional classical models", Phys. Rev. Research (2026), doi:10.1103/sp3b-cnyg, https://arxiv.org/abs/2604.15749
13. Gong Cheng, Lin Chen, Zheng-Cheng Gu, and Ling-Yan Hung, "Precision Reconstruction of Rational Conformal Field Theory from Exact Fixed-Point Tensor Network", Phys. Rev. X 15, 011073 (2025), doi:10.1103/PhysRevX.15.011073, https://arxiv.org/abs/2311.18005
14. Nikolay Ebel, Tom Kennedy, and Slava Rychkov, "Transfer matrix and lattice dilatation operator for high-quality fixed points in tensor network renormalization group", Phys. Rev. B 112, 104424 (2025), doi:10.1103/PhysRevB.112.104424, https://arxiv.org/abs/2409.13012
15. Michael Levin and Cody P. Nave, "Tensor Renormalization Group Approach to Two-Dimensional Classical Lattice Models", Phys. Rev. Lett. 99, 120601 (2007), doi:10.1103/PhysRevLett.99.120601, https://doi.org/10.1103/PhysRevLett.99.120601
16. G. Evenbly and G. Vidal, "Tensor Network Renormalization", Phys. Rev. Lett. 115, 180405 (2015), doi:10.1103/PhysRevLett.115.180405, https://doi.org/10.1103/PhysRevLett.115.180405
17. T. Nishino, K. Okunishi, and M. Kikuchi, "Numerical renormalization group at criticality", Phys. Lett. A 213, 69-72 (1996), doi:10.1016/0375-9601(96)00128-4, https://doi.org/10.1016/0375-9601(96)00128-4
18. Hiroshi Ueda, Kouichi Okunishi, and Tomotoshi Nishino, "Doubling of entanglement spectrum in tensor renormalization group", Phys. Rev. B 89, 075116 (2014), doi:10.1103/PhysRevB.89.075116, https://doi.org/10.1103/PhysRevB.89.075116
19. Quinten Mortier, Ming-Hao Li, Jutho Haegeman, and Nick Bultinck, "Finite-Entanglement Scaling of 2D Metals", Phys. Rev. Lett. 131, 266202 (2023), doi:10.1103/PhysRevLett.131.266202, https://arxiv.org/abs/2305.02091