A comprehensive theory of finite-bond-dimension effects in matrix product state approximations beyond the conformal-field-theory scaling picture

Finite-entanglement scaling is the statement that a matrix product state (MPS) of fixed bond dimension $\chi$ approximating a critical system behaves like the exact system with an effective relevant perturbation that cuts the correlation length off at $\xi(\chi)\propto\chi^{\kappa}$, with $\kappa$ fixed by the central charge of the underlying conformal field theory (CFT). Hodgkiss, Lootens, Ueda, Vanhecke and Verstraete (2026) computed the actual perturbations induced by finite $\chi$, by implicitly differentiating the fixed-point equations of boundary matrix product state (boundary-MPS) and corner transfer matrix renormalization group (CTMRG) contractions of two-dimensional tensor networks. For a single critical Ising model the induced perturbation converges to the spin operator, the most relevant operator of the Ising CFT, confirming the standard picture; but for two decoupled critical Ising chains the most relevant CFT perturbation captures only about two thirds of the induced perturbation, and subleading relevant interaction terms that entangle the two chains persist as $\chi$ grows. The open problem is to construct a comprehensive theory of finite-bond-dimension effects — predicting the operator content and $\chi$-dependence of the induced perturbation and the resulting scaling of correlation length, entanglement, and observables — valid both for critical systems beyond the single-CFT case (including tensor products of decoupled critical systems) and for gapped systems away from criticality, and reducing to the known CFT-based results in the single-CFT critical limit.

Background

Matrix product states (MPS) are variational wavefunctions for one-dimensional quantum spin chains in which the state is encoded in a repeating tensor carrying a virtual 'bond' index of dimension $\chi$; the bond dimension bounds the entanglement entropy across any cut of the chain by $\log\chi$, and the density matrix renormalization group (DMRG) and its infinite-size variant (iDMRG) variationally optimize such states. The same mathematics describes the leading eigenvector of the transfer matrix of a two-dimensional classical statistical model, whose partition function is a two-dimensional tensor network; in practice this network is contracted by boundary-MPS methods such as the corner transfer matrix renormalization group (CTMRG). Because a finite-$\chi$ state can carry only finite entanglement, its correlations decay exponentially with a finite correlation length $\xi$ even when the true system is critical; the Schmidt coefficients (eigenvalues of the cut density matrix) are the objects being truncated. MPS algorithms also tend to break the symmetries of a critical point, since symmetry-broken or dual descriptions can be more efficient — a phenomenon recently related to gauging dualities in the generalized Landau paradigm by Lootens, Delcamp and Verstraete.

For a critical chain the exact ground state is described at long distances by a conformal field theory (CFT), a scale-invariant field theory characterized by a central charge $c$ and by the scaling dimensions of its operators. A perturbation of the critical Hamiltonian by an operator of scaling dimension $x$ is called relevant when $x<2$ (in 1+1 dimensions): its coupling grows under the renormalization group (RG) and drives the system away from criticality, generating a finite correlation length. The standard theory of finite-entanglement scaling (FES), developed by Tagliacozzo et al., Pollmann et al., Pirvu et al. and Vanhecke et al., postulates that truncating $\chi$ acts as such a relevant perturbation: the optimal finite-$\chi$ MPS mimics the exact critical state cut off at scale $\xi(\chi)$, with the algebraic law $\xi(\chi)\propto\chi^{\kappa}$ where $\kappa$ depends only on the central charge,

$\kappa(c)=\frac{6}{c\,(1+\sqrt{12/c})}$

and the half-chain entanglement entropy satisfies $S=(c/6)\log\xi+\mathrm{const}$. This turns $\chi$ into a controllable scaling variable analogous to system size in finite-size scaling: critical exponents, the central charge, and other CFT data are extracted from the $\chi$-dependence of observables. Recent effective-theory work sharpened the picture: finite entanglement realizes a boundary CFT for the entanglement Hamiltonian — the logarithm of the reduced density matrix of a half-chain (Huang et al.) — and the interplay between the decreasing perturbation strength and the increasing infrared scale freezes the RG flow at a 'self-congruent point' (Schneider et al.). Throughout this framework the induced perturbation is taken to be, essentially, the most relevant operator of the underlying CFT — for the Ising transition, the spin (magnetic-field) operator $\sigma$.

Hodgkiss et al. (2026) tested this premise directly. They compute the perturbation of the defining tensor $T$ of a two-dimensional tensor network that optimally improves the finite-$\chi$ boundary-MPS approximation — the natural cost functional being a Schatten 4-norm of the corner-transfer-matrix environment, a CTM analogue of the variance — using a new implicit-differentiation solver for the nonlinear fixed-point equations of the contraction (a method of independent interest for projected entangled pair state (PEPS) optimization, developed algorithmically in the companion paper by Burgelman et al.). For the single critical 2D Ising model they confirm the textbook answer: the induced perturbation converges to $\sigma$, with overlap $0.998\to0.999999$ for $\chi=8\to64$. But for two decoupled critical Ising chains — a tensor product of two critical systems, total central charge $c=1$, represented by a single MPS — the magnetic-field perturbation $\sum_i\sigma^z_i+\tau^z_i$ carries only about two thirds of the induced perturbation even at large $\chi$, and the remainder is dominated by subleading relevant interaction terms that entangle the two chains and persist as $\chi$ grows. The mechanism is kinematic: the Schmidt coefficients of a product state are products $\lambda_i\lambda_j$, so MPS algorithms profit from weakly entangling the chains and discarding small products — an effect absent from the CFT reasoning, in which the most relevant perturbations do not couple the chains. The authors conclude that finite-entanglement scaling is 'not just a simple application of the standard scaling theory of CFT' and call for a complete theory of MPS scaling, including gapped systems where no CFT description exists.

Problem Statement

Construct a comprehensive theory of finite-bond-dimension effects in MPS and boundary-MPS approximations — a theory that goes beyond the standard CFT-based scaling theory and is valid both at and away from criticality. From the data of the model being approximated (its defining tensor or Hamiltonian) and the bond dimension $\chi$, the theory should predict:

1. The effective perturbation induced by finite $\chi$: its operator content and its dependence on $\chi$ — which relevant, subleading, symmetry-breaking, and entangling perturbations the optimal finite-$\chi$ approximation effectively adds to the model, and with what weights.
2. The resulting scaling laws: the induced correlation length $\xi(\chi)$, the bipartite entanglement, and the corrections to observables and to spectra (for example transfer-matrix gaps) as functions of $\chi$.
3. In the regime of a single critical system described by one CFT, reproduction of the established results as limiting cases — induced perturbation equal to the most relevant CFT operator, $\xi(\chi)\propto\chi^{\kappa(c)}$, and $S=(c/6)\log\xi$.
4. In the regime where that picture demonstrably fails — the minimal case being a tensor product of critical systems such as two decoupled critical Ising chains, where the most relevant CFT perturbation carries only about 2/3 of the induced perturbation and subleading entangling relevant perturbations persist as $\chi$ grows — an explanation of which perturbations survive, why, and what scaling laws (effective exponents, effective central charge, crossover functions) replace or refine the single-CFT ones.
5. Away from criticality, in gapped phases where no CFT description exists, a characterization of the finite-$\chi$ biases of correlation lengths, entanglement, and observables.

Items 1 and 2 are the core of the theory and are to be delivered as a single framework valid in both the critical tensor-product regime (item 4) and the gapped regime (item 5); item 3 is a limiting-case consistency requirement on that framework, not an independent deliverable.

These components are facets of one scientific objective — a predictive account of what bond-dimension truncation does to the approximated model, at and away from criticality — and are to be delivered within one coherent framework, in the spirit of the source paper's conclusion that 'the current results should serve as a basis for developing a complete theory of MPS scaling' (Hodgkiss et al., arXiv:2607.15124).

References

1. Luke Hodgkiss, Laurens Lootens, Atsushi Ueda, Bram Vanhecke, Frank Verstraete, "On the origin of finite entanglement scaling" (2026), arXiv:2607.15124, https://arxiv.org/abs/2607.15124
2. Lander Burgelman, Anna Francuz, Paul Brehmer, Lukas Devos, Jutho Haegeman, Frank Verstraete, Bram Vanhecke, "Implicit differentiation of tensor network algorithms" (2026), arXiv:2607.15030, https://arxiv.org/abs/2607.15030
3. Jan T. Schneider, Atsushi Ueda, Yifan Liu, Andreas M. Läuchli, Masaki Oshikawa, 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://doi.org/10.21468/SciPostPhys.18.4.142, arXiv:2411.03954, https://arxiv.org/abs/2411.03954
4. Rui-Zhen Huang, Long Zhang, Andreas M. Läuchli, Jutho Haegeman, Frank Verstraete, Laurens Vanderstraeten, "Emergent conformal boundaries from finite-entanglement scaling in matrix product states", Phys. Rev. Lett. 132, 086503 (2024), doi:10.1103/PhysRevLett.132.086503, https://doi.org/10.1103/PhysRevLett.132.086503, arXiv:2306.08163, https://arxiv.org/abs/2306.08163
5. Atsushi Ueda, Masaki Oshikawa, "Finite-size and finite bond dimension effects of tensor network renormalization" (2023), arXiv:2302.06632, https://arxiv.org/abs/2302.06632
6. Quinten Mortier, Ming-Hao Li, Jutho Haegeman, Nick Bultinck, "Finite-Entanglement Scaling of 2D Metals", Phys. Rev. Lett. 131, 266202 (2023), doi:10.1103/PhysRevLett.131.266202, https://doi.org/10.1103/PhysRevLett.131.266202, arXiv:2305.02091, https://arxiv.org/abs/2305.02091
7. B. Pirvu, G. Vidal, F. Verstraete, L. Tagliacozzo, "Matrix product states for critical spin chains: finite-size scaling versus finite-entanglement scaling", Phys. Rev. B 86, 075117 (2012), doi:10.1103/PhysRevB.86.075117, https://doi.org/10.1103/PhysRevB.86.075117, arXiv:1204.3934, https://arxiv.org/abs/1204.3934
8. Frank Pollmann, Subroto Mukerjee, Ari M. Turner, 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, arXiv:0812.2903, https://arxiv.org/abs/0812.2903
9. L. Tagliacozzo, T. R. de Oliveira, S. Iblisdir, J. I. Latorre, "Scaling of entanglement support for matrix product states", Phys. Rev. B 78, 024410 (2008), doi:10.1103/PhysRevB.78.024410, https://doi.org/10.1103/PhysRevB.78.024410, arXiv:0712.1976, https://arxiv.org/abs/0712.1976
10. Bram Vanhecke, Jutho Haegeman, Karel Van Acoleyen, Laurens Vanderstraeten, 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
11. Laurens Lootens, Clement Delcamp, Frank Verstraete, "Entanglement and the density matrix renormalisation group in the generalised Landau paradigm", Nature Physics 21, 1657 (2025), doi:10.1038/s41567-025-02961-2, https://doi.org/10.1038/s41567-025-02961-2, arXiv:2408.06334, https://arxiv.org/abs/2408.06334
12. Atsushi Ueda, Lander Burgelman, Luca Tagliacozzo, Laurens Vanderstraeten, "Interface roughening in the 3-D Ising model with tensor networks" (2026), arXiv:2601.07829, https://arxiv.org/abs/2601.07829
13. Sing-Hong Chan, Pochung Chen, "Extracting conformal data from finite-size tensor-network flow in critical two-dimensional classical models" (2026), arXiv:2604.15749, https://arxiv.org/abs/2604.15749