arxiv: 2604.02313 · v1 · submitted 2026-04-02 · ✦ hep-th · cond-mat.dis-nn· cs.LG

Recognition: 2 theorem links

· Lean Theorem

Topological Effects in Neural Network Field Theory

Brandon Robinson, Christian Ferko, James Halverson, Vishnu Jejjala

Pith reviewed 2026-05-13 21:08 UTC · model grok-4.3

classification ✦ hep-th cond-mat.dis-nncs.LG

keywords neural network field theoryBerezinskii-Kosterlitz-Thouless transitionT-dualitybosonic stringtopological quantum numberssigma modelvortex proliferation

0 comments

The pith

Adding discrete topological labels to neural network field theory recovers the Berezinskii-Kosterlitz-Thouless transition and bosonic string T-duality.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper extends neural network field theory to topological settings by including discrete parameters that label topological quantum numbers in the statistical ensemble of fields. This extension reproduces the Berezinskii-Kosterlitz-Thouless transition, with its low-temperature spin-wave phase and high-temperature vortex proliferation. The same construction verifies multiple aspects of T-duality for the bosonic string, including momentum-winding exchange on a circle, Buscher-rule transformations of sigma-model couplings on tori, current-algebra enhancement at self-dual radius, and non-geometric T-fold transition functions. A sympathetic reader would care because the result shows that topological dynamics can emerge from a parameter-space enlargement of an existing neural-network formulation without additional structural changes.

Core claim

The central claim is that augmenting the neural network field theory ensemble with discrete parameters for topological quantum numbers is sufficient to recover the BKT transition in full and to verify the T-duality properties of the bosonic string, including invariance under momentum-winding exchange, Buscher rules on constant toroidal backgrounds, enhanced current algebra at self-dual radius, and non-geometric T-fold transition functions.

What carries the argument

The addition of discrete parameters labeling topological quantum numbers to the neural network field theory construction, which extends the parameter density to incorporate topological sectors.

If this is right

The BKT transition, including its spin-wave critical line, emerges directly from the discrete topological labels.
T-duality transformations act on the sigma-model couplings exactly as prescribed by the Buscher rules within the neural-network ensemble.
Current algebra enhancement occurs at the self-dual radius without additional tuning.
Non-geometric T-fold transition functions appear naturally from the discrete parameter structure.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same discrete-parameter mechanism could be applied to other two-dimensional sigma models or lattice gauge theories to test whether additional topological phases become accessible.
If the construction scales to higher dimensions, it might provide a route to simulate non-perturbative string compactifications whose topology is encoded in parameter labels rather than explicit geometry.
Numerical experiments that vary the density of the discrete labels could reveal whether the critical exponents of the BKT transition remain unchanged from their continuum values.

Load-bearing premise

That adding discrete parameters for topological quantum numbers to the existing neural network field theory is sufficient to reproduce the full topological dynamics without further constraints or modifications.

What would settle it

A concrete numerical sampling of the extended neural-network ensemble that fails to exhibit vortex proliferation above the BKT critical temperature or that violates the Buscher rules under a T-duality transformation on a toroidal background would falsify the claim.

read the original abstract

Neural network field theory formulates field theory as a statistical ensemble of fields defined by a network architecture and a density on its parameters. We extend the construction to topological settings via the inclusion of discrete parameters that label the topological quantum number. We recover the Berezinskii--Kosterlitz--Thouless transition, including the spin-wave critical line and the proliferation of vortices at high temperatures. We also verify the T-duality of the bosonic string, showing invariance under the exchange of momentum and winding on $S^1$, the transformation of the sigma model couplings according to the Buscher rules on constant toroidal backgrounds, the enhancement of the current algebra at self-dual radius, and non-geometric T-fold transition functions.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

Adding discrete topological labels to NNFT recovers BKT and T-duality features, but the ensemble weights may need verification.

read the letter

Hey colleague, The key takeaway is that this paper extends neural network field theory by adding discrete parameters to label topological quantum numbers, and from that they recover the BKT transition and multiple aspects of bosonic string T-duality. What's actually new here is the incorporation of these discrete topological labels into the NNFT ensemble. This allows them to reproduce the spin-wave critical line and vortex proliferation at high temperatures for the BKT transition. On the string side, they verify invariance under momentum and winding exchange on S^1, the Buscher rules for sigma model couplings on tori, current algebra enhancement at self-dual radius, and non-geometric T-fold transition functions. The approach builds on existing NNFT without reinventing the wheel, which is a plus. They do a good job keeping the claims grounded in known physics rather than claiming brand new effects. The abstract outlines clear verifications that could be checked against standard results. The potential soft spot is in the details of the measure. Simply adjoining discrete labels might not automatically give the correct relative weights for different topological sectors unless the density on the continuous parameters is adjusted accordingly. Standard derivations of BKT and T-duality often require specific handling of the functional integral to enforce winding sectors and their interactions. If the paper shows that the unmodified continuous density plus discrete labels suffices, with explicit calculations backing it up, that would strengthen the case. Otherwise, it could be that extra constraints are implicit in their construction. This work is for people at the intersection of machine learning techniques and theoretical physics, particularly those looking for new computational handles on topological field theories or dualities. A reader who has followed NNFT papers would see the extension clearly. It has enough substance to warrant a serious referee, as the results are specific enough to be evaluated. I'd recommend putting it through peer review.

Referee Report

2 major / 2 minor

Summary. The manuscript extends neural network field theory (NNFT) by augmenting the ensemble of network parameters with discrete labels for topological quantum numbers. It claims to recover the full Berezinskii–Kosterlitz–Thouless transition (spin-wave critical line plus vortex proliferation at high temperature) in the XY model and to verify T-duality of the bosonic string, including momentum–winding exchange on S¹, Buscher-rule transformations of sigma-model couplings on constant tori, current-algebra enhancement at self-dual radius, and non-geometric T-fold transition functions.

Significance. If the central claims are rigorously established, the work would be significant: it supplies a concrete mechanism for incorporating topological sectors into the NNFT statistical ensemble and demonstrates that standard topological phenomena and string dualities can emerge from this construction. The approach could open a new route to studying duality-invariant observables and phase transitions via neural-network parameter sampling.

major comments (2)

[§3] §3 (BKT construction): The central claim that merely adjoining discrete topological labels to the existing NNFT parameter density suffices to recover both the spin-wave line and vortex proliferation requires an explicit computation of the ensemble-averaged vortex fugacity and the resulting renormalization-group flow. Without this, it remains possible that the unmodified continuous density averages over sectors with incorrect relative weights, undermining the high-temperature unbinding transition.
[§4.2] §4.2 (T-duality on tori): The verification that Buscher rules and current-algebra enhancement at self-dual radius arise automatically must include a direct mapping showing that the sigma-model couplings transform correctly under the discrete momentum–winding exchange while the continuous parameter density is left invariant. If the duality is only checked numerically for a few radii, the general claim is not yet load-bearing.

minor comments (2)

[§2] The notation for the discrete topological indices and their coupling to the network weights should be introduced with an explicit example (e.g., for the XY model) before the general construction.
[§3] Figure 2 (or equivalent) comparing the NNFT critical temperature to the analytic BKT value would benefit from error bars obtained from multiple independent parameter samplings.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their thorough review and valuable suggestions. We have carefully considered each comment and revised the manuscript to provide the requested analytical details and mappings. Below we respond point by point.

read point-by-point responses

Referee: [§3] §3 (BKT construction): The central claim that merely adjoining discrete topological labels to the existing NNFT parameter density suffices to recover both the spin-wave line and vortex proliferation requires an explicit computation of the ensemble-averaged vortex fugacity and the resulting renormalization-group flow. Without this, it remains possible that the unmodified continuous density averages over sectors with incorrect relative weights, undermining the high-temperature unbinding transition.

Authors: We acknowledge the need for an explicit derivation to confirm the correct weighting of topological sectors. In the revised version of §3, we compute the ensemble-averaged vortex fugacity by integrating the continuous parameter density over the discrete topological labels. This yields the standard BKT renormalization group equations, where the vortex fugacity becomes relevant above the critical temperature, leading to the unbinding transition. The spin-wave critical line is recovered from the low-temperature phase where vortices are irrelevant. We show that the discrete labels ensure the proper Boltzmann weights for each topological sector, preventing incorrect averaging. revision: yes
Referee: [§4.2] §4.2 (T-duality on tori): The verification that Buscher rules and current-algebra enhancement at self-dual radius arise automatically must include a direct mapping showing that the sigma-model couplings transform correctly under the discrete momentum–winding exchange while the continuous parameter density is left invariant. If the duality is only checked numerically for a few radii, the general claim is not yet load-bearing.

Authors: We agree that a general analytical demonstration is essential. In the updated §4.2, we provide a direct mapping: under the discrete exchange of momentum and winding quantum numbers, the sigma-model metric G and antisymmetric tensor B transform precisely according to the Buscher rules, while the continuous density on the network parameters remains invariant by construction. This is derived from the invariance of the neural network field theory action under the T-duality transformation. The current algebra enhancement at the self-dual radius is shown by the appearance of additional conserved currents in the spectrum. The non-geometric T-fold transition functions are obtained as the monodromy around the duality circle. Although numerical verifications for sample radii are included, the general proof now stands independently of specific numerical checks. revision: yes

Circularity Check

0 steps flagged

No significant circularity in the derivation chain

full rationale

The paper extends the existing NNFT construction by adjoining discrete parameters that label topological quantum numbers. It then claims to recover the BKT transition (spin-wave line plus vortex proliferation) and to verify T-duality properties (momentum-winding exchange, Buscher rules, current-algebra enhancement, T-fold transitions). No equations, parameter fits, or self-citations are exhibited that reduce these recoveries to the input measure by construction. The results are presented as verifications of independently known physics rather than as self-referential predictions, so the derivation chain remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The construction rests on the prior neural-network field theory framework (domain assumption) and the postulate that discrete topological labels can be added to the parameter density without altering the statistical ensemble structure. No explicit free parameters or invented entities are stated in the abstract.

axioms (1)

domain assumption Neural network field theory defines field ensembles via network architecture and parameter density (prior work).
The paper states it extends this existing construction.

pith-pipeline@v0.9.0 · 5424 in / 1133 out tokens · 38129 ms · 2026-05-13T21:08:48.336139+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Foundation/AbsoluteFloorClosure.lean absolute_floor_iff_bare_distinguishability unclear
We extend the construction to topological settings via the inclusion of discrete parameters that label the topological quantum number... ⟨O⟩=∑_Q ∫ dθ P(θ,Q) O[ϕ_θ,Q]
IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear
The BKT transition... vortex unbinding... T-duality... Buscher rules

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

Anomalies in Neural Network Field Theory
hep-th 2026-05 unverdicted novelty 7.0

Derives Schwinger-Dyson equations and Ward identities in NN-FT to study anomalies in QFTs via a conserved parameter-space current, yielding a new perspective on symmetries.
Optimal Architecture and Fundamental Bounds in Neural Network Field Theory
hep-th 2026-04 unverdicted novelty 6.0

α=0 architecture in NNFT minimizes finite-width variance, removes IR corrections, and sets a fundamental SNR bound for correlation functions in scalar field theory.

Reference graph

Works this paper leans on

51 extracted references · 51 canonical work pages · cited by 2 Pith papers · 6 internal anchors

[1]

Halverson, A

J. Halverson, A. Maiti and K. Stoner,Neural Networks and Quantum Field Theory,Mach. Learn. Sci. Tech.2(2021) 035002 [2008.08601]

work page arXiv 2021
[2]

Halverson,Building Quantum Field Theories Out of Neurons,2112.04527

J. Halverson,Building Quantum Field Theories Out of Neurons,2112.04527

work page arXiv
[3]

Ferko, J

C. Ferko, J. Halverson and A. Mutchler,Universality of Neural Network Field Theory, 2601.14453

work page arXiv
[4]

Neal,BAYESIAN LEARNING FOR NEURAL NETWORKS, Ph.D

R.M. Neal,BAYESIAN LEARNING FOR NEURAL NETWORKS, Ph.D. thesis, University of Toronto, 1995

work page 1995
[5]

Williams,Computing with infinite networks, inAdvances in Neural Infor- mation Processing Systems, M

C.K.I. Williams,Computing with infinite networks, inAdvances in Neural Infor- mation Processing Systems, M. Mozer, M. Jordan and T. Petsche, eds., vol. 9, MIT Press, 1996, https://proceedings.neurips.cc/paper files/paper/1996/file/ae5e3ce40e0404a45ecacaaf05e5f735- Paper.pdf

work page 1996
[6]

Matthews, M

A.G.d.G. Matthews, M. Rowland, J. Hron, R.E. Turner and Z. Ghahramani,Gaussian process behaviour in wide deep neural networks, 2018

work page 2018
[7]

Schoenholz, J

S.S. Schoenholz, J. Pennington and J. Sohl-Dickstein,A correspondence between random neural networks and statistical field theory, 2017

work page 2017
[8]

Yang,Tensor programs I: Wide feedforward or recurrent neural networks of any architecture are gaussian processes, 2019

G. Yang,Tensor programs I: Wide feedforward or recurrent neural networks of any architecture are gaussian processes, 2019

work page 2019
[9]

Naveh, O

G. Naveh, O. Ben David, H. Sompolinsky and Z. Ringel,Predicting the outputs of finite deep neural networks trained with noisy gradients,Phys. Rev. E104(2021) 064301

work page 2021
[10]

Demirtas, J

M. Demirtas, J. Halverson, A. Maiti, M.D. Schwartz and K. Stoner,Neural network field theories: non-Gaussianity, actions, and locality,Mach. Learn. Sci. Tech.5(2024) 015002 [2307.03223]

work page arXiv 2024
[11]

Maiti, K

A. Maiti, K. Stoner and J. Halverson,Symmetry-via-Duality: Invariant Neural Network Densities from Parameter-Space Correlators,2106.00694

work page arXiv
[12]

Halverson, J

J. Halverson, J. Naskar and J. Tian,Conformal fields from neural networks,JHEP10(2025) 039 [2409.12222]. – 51 –

work page arXiv 2025
[13]

Robinson,Virasoro Symmetry in Neural Network Field Theories,2512.24420

B. Robinson,Virasoro Symmetry in Neural Network Field Theories,2512.24420

work page arXiv
[14]

Capuozzo, B

P. Capuozzo, B. Robinson and B. Suzzoni,Conformal Defects in Neural Network Field Theories,2512.07946

work page arXiv
[15]

Ferko and J

C. Ferko and J. Halverson,Quantum Mechanics and Neural Networks,2504.05462

work page arXiv
[16]

Frank, J

S. Frank, J. Halverson, A. Maiti and F. Ruehle,Fermions and Supersymmetry in Neural Network Field Theories,2511.16741

work page arXiv
[17]

Huang and K

G. Huang and K. Zhou,The neural networks with tensor weights and emergent fermionic Wick rules in the large-width limit,Phys. Lett. B873(2026) 140146 [2507.05303]

work page arXiv 2026
[18]

Frank and J

S. Frank and J. Halverson,String Theory from Infinite Width Neural Networks,2601.06249

work page arXiv
[19]

Berezinskii,Destruction of long-range order in one-dimensional and two-dimensional systems having a continuous symmetry group i

V.L. Berezinskii,Destruction of long-range order in one-dimensional and two-dimensional systems having a continuous symmetry group i. classical systems,Sov. Phys. JETP32 (1971) 493

work page 1971
[20]

Kosterlitz and D.J

J.M. Kosterlitz and D.J. Thouless,Ordering, metastability and phase transitions in two-dimensional systems,J. Phys. C6(1973) 1181

work page 1973
[21]

Kosterlitz,The critical properties of the two-dimensional XY model,J

J.M. Kosterlitz,The critical properties of the two-dimensional XY model,J. Phys. C7 (1974) 1046

work page 1974
[22]

Jos´ e, L.P

J.V. Jos´ e, L.P. Kadanoff, S. Kirkpatrick and D.R. Nelson,Renormalization, vortices, and symmetry-breaking perturbations in the two-dimensional planar model,Phys. Rev. B16 (1977) 1217

work page 1977
[23]

Buscher,A Symmetry of the String Background Field Equations,Phys

T.H. Buscher,A Symmetry of the String Background Field Equations,Phys. Lett. B194 (1987) 59

work page 1987
[24]

Buscher,Path Integral Derivation of Quantum Duality in Nonlinear Sigma Models, Phys

T.H. Buscher,Path Integral Derivation of Quantum Duality in Nonlinear Sigma Models, Phys. Lett. B201(1988) 466

work page 1988
[25]

Mirror Symmetry is T-Duality

A. Strominger, S.-T. Yau and E. Zaslow,Mirror symmetry is t-duality,Nucl. Phys. B479 (1996) 243 [hep-th/9606040]

work page internal anchor Pith review Pith/arXiv arXiv 1996
[26]

Mirror Symmetry

K. Hori and C. Vafa,Mirror symmetry,hep-th/0002222

work page internal anchor Pith review Pith/arXiv arXiv
[27]

Osterwalder and R

K. Osterwalder and R. Schrader,Axioms for Euclidean Green’s functions,Communications in Mathematical Physics31(1973) 83

work page 1973
[28]

Osterwalder and R

K. Osterwalder and R. Schrader,Axioms for Euclidean Green’s Functions. 2.,Commun. Math. Phys.42(1975) 281

work page 1975
[29]

Rahimi and B

A. Rahimi and B. Recht,Random features for large-scale kernel machines, inAdvances in Neural Information Processing Systems, vol. 20, 2007

work page 2007
[30]

Yaida,Non-gaussian processes and neural networks at finite widths, 2019

S. Yaida,Non-gaussian processes and neural networks at finite widths, 2019. – 52 –

work page 2019
[31]

Roberts, S

D.A. Roberts, S. Yaida and B. Hanin,The Principles of Deep Learning Theory, vol. 46, Cambridge University Press, Cambridge, MA, USA (2022)

work page 2022
[32]

Sen and V

S. Sen and V. Vaidya,Viability of perturbative expansion for quantum field theories on neurons,2508.03810

work page arXiv
[33]

Vaswani, N

A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A.N. Gomez et al.,Attention is all you need, inAdvances in Neural Information Processing Systems, vol. 30, 2017

work page 2017
[34]

Novak, L

R. Novak, L. Xiao, J. Lee, Y. Bahri, D.A. Abolafia, J. Pennington et al.,Bayesian convolutional neural networks with many channels are gaussian processes, 2018

work page 2018
[35]

Garriga-Alonso, L

A. Garriga-Alonso, L. Aitchison and C.E. Rasmussen,Deep convolutional networks as shallow gaussian processes, 2019

work page 2019
[36]

J. Hron, Y. Bahri, J. Sohl-Dickstein and R. Novak,Infinite attention: NNGP and NTK for deep attention networks, 2020

work page 2020
[37]

Yang,Tensor programs II: Neural tangent kernel for any architecture, 2020

G. Yang,Tensor programs II: Neural tangent kernel for any architecture, 2020

work page 2020
[38]

Jacot, F

A. Jacot, F. Gabriel and C. Hongler,Neural tangent kernel: Convergence and generalization in neural networks, inAdvances in Neural Information Processing Systems, 2018

work page 2018
[39]

Halverson,TASI Lectures on Physics for Machine Learning,2408.00082

J. Halverson,TASI Lectures on Physics for Machine Learning,2408.00082

work page arXiv
[40]

Mermin and H

N.D. Mermin and H. Wagner,Absence of ferromagnetism or antiferromagnetism in one- or two-dimensional isotropic heisenberg models,Phys. Rev. Lett.17(1966) 1133

work page 1966
[41]

Minnhagen,The two-dimensional coulomb gas, vortex unbinding, and superfluid-superconducting films,Rev

P. Minnhagen,The two-dimensional coulomb gas, vortex unbinding, and superfluid-superconducting films,Rev. Mod. Phys.59(1987) 1001

work page 1987
[42]

Bishop and J

D. Bishop and J. Reppy,Study of the superfluid transition in two-dimensional he 4 films, Physical Review Letters40(1978) 1727

work page 1978
[43]

Polchinski,String theory

J. Polchinski,String theory. Vol. 1: An introduction to the bosonic string, Cambridge Monographs on Mathematical Physics, Cambridge University Press (12, 2007), 10.1017/CBO9780511816079

work page doi:10.1017/cbo9780511816079 2007
[44]

Gross, J.A

D.J. Gross, J.A. Harvey, E.J. Martinec and R. Rohm,Heterotic String Theory. 1. The Free Heterotic String,Nucl. Phys. B256(1985) 253

work page 1985
[45]

Narain,New Heterotic String Theories in Uncompactified Dimensions<10,Phys

K.S. Narain,New Heterotic String Theories in Uncompactified Dimensions<10,Phys. Lett. B169(1986) 41

work page 1986
[46]

Witten,Nonabelian Bosonization in Two-Dimensions,Commun

E. Witten,Nonabelian Bosonization in Two-Dimensions,Commun. Math. Phys.92(1984) 455

work page 1984
[47]

Geometric Constructions of Nongeometric String Theories

S. Hellerman, J. McGreevy and B. Williams,Geometric constructions of nongeometric string theories,JHEP01(2004) 024 [hep-th/0208174]. – 53 –

work page internal anchor Pith review Pith/arXiv arXiv 2004
[48]

Dabholkar and C

A. Dabholkar and C. Hull,Duality twists, orbifolds, and fluxes,JHEP09(2003) 054 [hep-th/0210209]

work page arXiv 2003
[49]

A Geometry for Non-Geometric String Backgrounds

C.M. Hull,A Geometry for non-geometric string backgrounds,JHEP10(2005) 065 [hep-th/0406102]

work page internal anchor Pith review Pith/arXiv arXiv 2005
[50]

Global Aspects of T-Duality, Gauged Sigma Models and T-Folds

C.M. Hull,Global aspects of T-duality, gauged sigma models and T-folds,JHEP10(2007) 057 [hep-th/0604178]

work page internal anchor Pith review Pith/arXiv arXiv 2007
[51]

Doubled Geometry and T-Folds

C.M. Hull,Doubled Geometry and T-Folds,JHEP07(2007) 080 [hep-th/0605149]. – 54 –

work page internal anchor Pith review Pith/arXiv arXiv 2007