{"id":"570dd312-674a-417d-84cb-98ba8949a561","arxiv_id":"2412.08374","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"The authors introduce a 'simplex tensor network' RG algorithm for 3D symmetric boundary theories of 3+1D Dijkgraaf-Witten symTFTs and use it to numerically map the Z2 phase diagram, including the 3D Ising transition.","lead":"This paper develops a numerical renormalization group for 3D lattice models defined as boundaries of a 3+1D topological field theory (a symTFT). It tests the method on a Z2 symmetric example and maps the phase boundaries between a symmetry-broken phase and two symmetry-protected topological phases.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The implemented truncation is not shown to preserve Z2, so the nonzero ⟨gi⟩ on the exactly symmetric line y=1 (and in the Ising setup) is likely a truncation artifact; the 'any path through SB' conclusion and the local-order-parameter phase diagram are therefore not yet supported.","rationale":"The reader identified truncation fidelity as the weakest assumption; I agree that the approximation quality is central, and I sharpen this into a specific mechanism: the truncation is not constrained to the symmetric subspace, so it can explicitly break the very Z2 symmetry the algorithm claims to preserve. Because the initial tensors used for the order-parameter calculations are symmetric, the exact RG flow would keep all symmetry-odd expectation values exactly zero. The observed nonzero ⟨gi⟩ along y=1 and in the Ising setup, absent any described symmetry-breaking field or boundary condition, is therefore direct evidence that the truncated flow is not symmetry-preserving. This does not necessarily invalidate the exact RG construction, which is a genuine contribution, nor does it disprove the physical existence of a symmetry-breaking region between the two SPT phases; it does mean the numerical evidence in Figs. 7 and 8 cannot currently distinguish a physical SB phase from a truncation-induced artifact. Because this is checkable and potentially fixable, a conditional verdict is appropriate: the symmetry properties of the truncated tensors and the protocol for measuring ⟨gi⟩ must be documented and tested before the phase diagram and the symmetry-preservation claim can be accepted.","tokens_in":21263,"tokens_out":8084,"duration_ms":95598,"concrete_test":"Run the RG for T[x,y] with x=0,y=1 and, after every truncation step, test whether the tensor is Z2-equivariant, e.g. T(gi,gj,gk,gl) = T(−gi,−gj,−gk,−gl) on vertex indices with the corresponding action on edge and face indices. If it is not, recompute Fig. 7b after projecting every truncated tensor onto the symmetric subspace, e.g. by symmetrizing the initial guess and adding a penalty |B−P_Z2 B|^2 to Eq. (37). If the SB crescent along y=1 vanishes or changes materially, the 'any path through SB' claim is a truncation artifact. Independently, for the Ising check, add a small explicit field h·⟨gi⟩ to the partition function, compute ⟨gi⟩, and take h→0; verify that J_c is stable and that this controlled procedure, rather than truncation noise, is what locates J=0.22(4).","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central claim is that the numerical algorithm, including truncation, implements a symmetry-preserving RG flow. The exact re-triangulation map F in Eq. (25) is Z2-equivariant, so the exact flow of any symmetric initial tensor remains symmetric. But the truncation in §A.2 is unconstrained: the face HOSVD and the L-BFGS minimization of |E−F|^2 in Eq. (37) can select any of the two global-Z2-related minima, and no symmetrization step is described. This matters because the initial tensors used for the phase diagram are symmetric. In particular, for y=1 in Eq. (29), T[x,1] is a superposition of the symmetric 3-cochains βSPT0 and βSPT1, so the exact partition function is Z2-invariant and ⟨gi⟩=0 for every finite lattice and every exact RG step. The nonzero ⟨gi⟩ reported in Fig. 7b along this line, and the resulting claim that 'any path connecting the two Z2 SPT phases will go through an SB phase', therefore cannot arise from the exact flow; it must come from the truncation breaking the symmetry. The same issue applies to the 3D Ising check: TIsing[J] in Eq. (27) is Z2-symmetric, so ⟨gi⟩ is zero unless an explicit symmetry-breaking field or symmetry-breaking boundary condition is introduced, which the paper does not describe. Thus the local order parameter and the SB region may be artifacts of the approximate algorithm rather than properties of the boundary theories, directly undermining the advertised symmetry-preserving nature of the flow.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper constructs a tensor-network representation of 2+1D symmetric theories as boundaries of a 3+1D Dijkgraaf-Witten symTFT, generalizing the \"strange correlator\" construction. Each tetrahedron carries a simplex tensor with vertex, edge, and face indices. The authors derive exact RG equations from re-triangulation invariance of the DW partition function (Eqs. (14)–(25)) and then introduce a numerical truncation scheme: face bonds are truncated by HOSVD, while edge bonds are truncated by L-BFGS minimization of the squared error (Eq. (37)). The algorithm is applied to Z2-symmetric boundary theories, interpolating linearly among three fixed-point tensors (βSPT0, βSPT1, βSB). The paper reports the 3D Ising critical coupling J=0.22(4) and uses local (⟨gi⟩) and non-local membrane (M) order parameters to map a phase diagram containing an SB phase and two SPT phases, including a complex-x continuation claiming that any path between the SPT phases passes through the SB phase.","tokens_in":21639,"tokens_out":5092,"duration_ms":51604,"significance":"The conceptual framework is attractive: deriving the RG flow from the topological invariance of the symTFT partition function is a clean and potentially powerful idea, and the simplex tensor network is a novel representation of 3D boundary states. The exact re-triangulation map F in Eq. (25) is explicitly constructed and is Z2-equivariant, which is a genuine strength. The paper is transparent about numerical limitations, stating that the gradient-descent truncation is sensitive to initial guesses and showing fluctuations in the figures. If the approximate flow can be made to respect the symmetry, the method would be a useful new tool for exploring 2+1D symmetric phases, including SPT transitions. However, the numerical results as presented do not yet establish the advertised symmetry-preserving flow, and the central phase-diagram claims rest on truncation behavior that is not controlled.","major_comments":[{"comment":"The truncation procedure is not Z2-equivariant, so the nonzero ⟨gi⟩ on the symmetric line y=1 is likely a truncation artifact. For T[x,1] in Eq. (29), the exact initial state is a superposition of the two Z2-symmetric cochains βSPT0 and βSPT1; hence the exact partition function is Z2-invariant and ⟨gi⟩=0 on every finite lattice and after every exact RG step. Because the HOSVD and L-BFGS steps in Eqs. (37)–(38) are unconstrained and no symmetrization is described, the broken symmetry observed in Fig. 7b must originate from the truncation, not from the physical theory. The same issue applies to the 3D Ising check in Eq. (27): TIsing[J] is symmetric, so ⟨gi⟩ vanishes identically unless an explicit symmetry-breaking field is introduced, which the paper does not describe. The local order parameter therefore does not yet provide evidence for an SB phase or for the claim that any path connecting the two SPT phases crosses an SB phase. The authors should impose the symmetry on the truncated tensors (or quantify the symmetry-breaking error and demonstrate it vanishes with increasing bond dimension) and rerun the phase diagram.","section":"§A.2, §4.2, Eq. (29), Fig. 7"},{"comment":"The central claim of a symmetry-preserving RG flow is only demonstrated for the exact map F in Eq. (25). After truncation, the algorithm is not shown to preserve the Z2 symmetry, and the acknowledged sensitivity of the L-BFGS step to initial guesses and hyperparameters (Sec. 5) means the flow—and hence the phase boundaries in Fig. 7—are not controlled. A convergence study with larger bond dimensions (e.g., edge/face dimension d=3) or with a symmetrized truncation is necessary to establish that the reported phase diagram is a property of the boundary theories rather than of the approximation. Without such a check, the numerical RG results cannot be taken as evidence for the symmetry-preserving nature of the flow.","section":"§3.3, §A.2, Sec. 5"},{"comment":"The conclusion 'any path connecting the two Z2 SPT phases will go through an SB phase' is drawn from the complex-x continuation along the specific line y=1 in Eq. (29). This is only a one-parameter family in theory space; it does not rule out other paths that avoid the SB region without additional symmetry arguments. The statement should be weakened to apply to the interpolating family studied, or supported by a general argument (e.g., based on the 2-form symmetry and the known classification of Z2 SPT phases). Even within the family, the conclusion depends on the broken-symmetry truncation issue raised above.","section":"§4.2, Fig. 7b"}],"minor_comments":[{"comment":"There is a typo: 'generatd' should be 'generated'.","section":"§3.3"},{"comment":"The word 'repect' should be 'respect'.","section":"§4.2"},{"comment":"The word 'generlization' should be 'generalization'.","section":"Sec. 5"},{"comment":"The definition of the membrane operator appears to have a typo: the product of three identical factors D_ijll' D_ijll' D_ijll' should presumably involve distinct simplices (ijll', ikll', jkll'), as described in the text and Figure 8(a). Please clarify.","section":"Eq. (31)"},{"comment":"The reported critical coupling J=0.22(4) has a large error bar, but the paper does not explain how the uncertainty was estimated. Please state the estimation method (e.g., number of runs, statistical fluctuation) so the reader can judge the agreement with the Monte Carlo value.","section":"Sec. 4.1"},{"comment":"The figures would benefit from explicit color bars and a description of the plotted quantity (e.g., whether the expectation value is normalized). In particular, Figure 7b's axes (real and imaginary parts of x) and the meaning of the blue/yellow regions should be stated in the caption.","section":"Fig. 7 and Fig. 8"}],"recommendation":"major_revision","confidential_remarks":"The manuscript's core construction is appealing and the exact re-triangulation relations are clean. My main concern is that the advertised 'symmetry-preserving RG flow' is broken by the truncation step, and the reported SB phase and the 'any path' conclusion depend on this truncation in a way that is not controlled. The authors are clearly aware of the truncation's sensitivity, but they have not shown that a symmetrized or higher-bond-dimension truncation yields the same phase diagram. I would request the editor to ask for a direct test of symmetry preservation (e.g., measure the Z2 charge of the truncated tensors) or a comparison with a symmetrized algorithm. The paper seems within scope for SciPost Physics, but the numerical evidence needs to be made robust before publication."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here's my take. The genuinely new contribution is the simplex tensor network ansatz for boundary states of a 3+1D DW symTFT, together with an exact RG map F derived from re-triangulation and the cocycle condition. That part is clean: Eq. (25) is a precise prescription for combining tensors, and the invariance argument is sound. The paper also gives a nice construction of a non-local membrane order parameter built from fixed-point tensors, and the Z2 demonstration does reproduce the 3D Ising transition at J=0.22(4), which matches the Monte Carlo value within error bars. That is a meaningful proof of principle.\n\nThe soft spot is the numerical truncation. Bond dimensions are capped at 2, the edge truncation uses gradient descent on |E-F|^2, and the authors themselves note sensitivity to initial guesses and hyperparameters, with visible fluctuations. The stress-test point is sharper: nothing in the truncation enforces the Z2 symmetry that the exact flow preserves. For T[x,1], the initial tensor is symmetric, so the exact partition function has <gi>=0 everywhere; any nonzero value on the y=1 line in Fig. 7b must be a truncation artifact. That directly undermines the claim that 'any path connecting the two Z2 SPT phases goes through an SB phase,' and it casts doubt on the SB region in the phase diagram. The same logic applies to the Ising check, though the agreement with the known critical coupling suggests the artifact may be mild there.\n\nSo the exact formalism is solid and worth building on, but the central numerical claim about the SPT-SPT transition is not currently supported. The paper would need a symmetrized truncation, or at least a systematic study of the symmetry-breaking tendency of the current one, plus code and data to be credible. I would still send it to a serious referee, because the framework is novel and the exact part is on firm ground, but I would expect heavy revision.","headline":"New simplex tensor network RG framework with exact re-triangulation equations, but the Z2 phase-diagram claims rest on a truncation that likely breaks the symmetry it is meant to preserve.","tokens_in":22157,"tokens_out":3141,"would_cite":true,"duration_ms":32688,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper proposes a symmetry-preserving numerical renormalization group for 2+1D symmetric theories expressed as boundary states of a 3+1D Dijkgraaf-Witten symTFT, and demonstrates it on the Z2 phase diagram, mapping the Ising…","keywords":["symmetry topological field theory","Dijkgraaf-Witten model","simplex tensor network","renormalization group","3D Ising transition","symmetry-protected topological phases","higher-form symmetry"],"falsifier":"A concrete check: rerun the same RG flow from the same initial tensors with a different truncation rule (for example, keep three bond states instead of two or use a different optimizer) and see whether the magnetization transition remains at $J=0.22(4)$ and whether the membrane order parameter still separates the two SPT phases; a shift larger than the quoted uncertainty would mean the truncation, not the re-triangulation invariance, is what fixes the reported phase boundaries.","tokens_in":21055,"feed_emoji":"🕸️","tokens_out":10724,"duration_ms":101259,"temperature":0.7,"pith_summary":"This paper develops a numerical renormalization group for 2+1D symmetric lattice theories by treating them as boundary conditions of a 3+1D Dijkgraaf-Witten topological bulk theory. The boundary is triangulated into tetrahedra, each carrying a tensor with vertex, edge, and face indices, forming what the authors call a simplex tensor network state. The RG step exploits re-triangulation invariance of the bulk partition function: combining neighboring tetrahedra through a cocycle move produces new tensors exactly, and bond truncation keeps the computation finite. For a $\\mathbb{Z}_2$ symmetric theory, the authors interpolate between three topological fixed-point boundaries (two symmetry-protected phases and a symmetry-broken phase), and they locate the phase transitions with a local order parameter $\\langle g_i\\rangle$ and a non-local membrane operator. The magnetization transition is found at $J=0.22(4)$, matching the known 3D Ising critical coupling, and the membrane operator cleanly distinguishes the two SPT phases; a sympathetic reader would care because this provides a symmetry-preserving numerical tool for searching 2+1D gapless fixed points and CFTs.","feed_headline":"New tensor RG locates 3D Ising transition at J=0.22(4)","feed_subtitle":"Flow preserves Z2 symmetry and a membrane operator separates the two SPT phases.","key_machinery":"The central object is the simplex tensor network: a tetrahedron tensor $T^E_F(V)$ with four vertex indices, six edge indices, and four face indices, contracted over shared edge and face indices to build the boundary state. The central move is re-triangulation of the boundary 3-manifold with the bulk 4-simplex cocycle $\\alpha$; the exact coarse-graining function $F(\\alpha,A,B)$ defined in Eq. (25) combines two tetrahedra by summing a shared face and merging the remaining indices into composite ones, and it is applied three times per RG round to remove edge, face, and body centers of a $2\\times2\\times2$ cube. The fixed points of the flow are topological boundary states described by 3-cochains $\\beta$ solving the Frobenius condition $\\alpha_{01234}=\\prod_{i=0}^4 \\beta^{(-1)^i}(g_0,\\dots,\\hat g_i,\\dots,g_4)$; in the $\\mathbb{Z}_2$ case this yields the trivial SPT, twisted SPT, and symmetry-breaking fixed-point tensors whose interpolations parametrize the phase space.","core_discovery":"The paper claims that re-triangulation invariance of a 3+1D Dijkgraaf-Witten partition function, together with a controlled bond truncation, defines a correct symmetry-preserving RG flow on 2+1D boundary theories. In this construction, the boundary state $\\Omega$ is a product of tetrahedron tensors $T^E_F(V)$ whose vertex, edge, and face indices encode locality and entanglement; the bulk ground state $|\\Psi\\rangle$ is the path integral of the DW model with a bulk vertex $S$ connected to each boundary vertex. The identity that carries the argument is the exact map $T'_{ijlm}=\\alpha_{ijklm}\\sum_{f_{klm}} A_{iklm}B_{jklm}$ with composite indices $e_{ij}=(g_k,e_{ik},e_{jk})$, $f_{ijl}=(e_{kl},f_{ikl},f_{jkl})$, and $f_{ijm}=(e_{km},f_{ikm},f_{jkm})$, denoted $F(\\alpha,A,B)$, which combines two tetrahedra into one while leaving the strange-correlator partition function invariant. Iterating this map over the edge, face, and body centers of a $2\\times2\\times2$ cube gives one RG round; truncating all edge and face bonds to dimension at most two (faces by a higher-order singular value decomposition, edges by a gradient-descent minimization of the contracted squared error) makes the flow numerical. Running the flow on the linear interpolation $T[x,y]$ of the three $\\mathbb{Z}_2$ fixed-point boundaries, the paper finds that $\\langle g_i\\rangle$ detects the symmetry-broken region with the Ising transition at $J=0.22(4)$, and that expectation values of the fixed-point membrane operator $\\hat M$ grow with the number of inserted operators and separate the two SPT phases.","pith_inferences":["A direct test the authors did not run is to repeat the flow with a better edge truncation, for example one that exploits the $\\mathbb{Z}_2$ charge sectors of the edge bonds; we infer that this would remove much of the fluctuation seen in figures 7 and 8 and could sharpen the reported phase boundaries.","If the truncation is the only obstruction, higher bond dimensions should reveal the tricritical point discussed by the paper: we infer that the current $D\\le2$ limit smears the purported tricritical region, and a $D=3$ calculation is a concrete way to check.","The impurity construction for order parameters suggests a general recipe: inserting any topological fixed-point boundary tensor as an impurity should detect the corresponding higher-form symmetry in any interpolated boundary, which could provide a numerical probe for deconfined quantum critical points beyond the $\\mathbb{Z}_2$ example considered here."],"forward_implications":["Because the flow is constructed from the bulk topological theory, the same re-triangulation machinery applies to any finite discrete group $G$ with a 4-cocycle, so phase diagrams for $\\mathbb{Z}_N$ or $\\mathbb{Z}_N\\times\\mathbb{Z}_M$ symmetric boundary theories become numerically accessible.","In the $\\mathbb{Z}_2$ example, the fixed point approached at $J=0.22(4)$ realizes the 3D Ising transition, so the workflow can be used to search for other symmetry-enriched gapless boundaries by choosing different interpolating paths among fixed-point tensors.","The membrane operator built from the nontrivial SPT tensor provides a finite-size non-local order parameter for the 2-form symmetry, allowing the two $\\mathbb{Z}_2$ SPT phases to be distinguished without computing non-contractible loop expectation values.","The exact RG map $F(\\alpha,A,B)$ together with the three-step cube coarse-graining defines a tensor-network renormalization scheme for a new class of 3D quantum states, which the paper suggests could be used for time evolution or variational ground-state studies.","For purely imaginary interpolation parameters the magnetization map shows that any path joining the two $\\mathbb{Z}_2$ SPT phases crosses a symmetry-broken region, pinning down the topology of the phase diagram."],"supporting_citations":[{"why":"supplies the 3+1D strange-correlator construction of boundary partition functions and the exact RG operator whose numerical implementation this paper develops.","marker":"[1]"},{"why":"introduces strange correlators for mapping topological to conformal field theories, the conceptual template for boundary states as tensor networks.","marker":"[16]"},{"why":"provides the higher-order singular value decomposition used to truncate face bonds and the 3D Ising partition-function reference for critical coupling.","marker":"[27]"},{"why":"defines the Dijkgraaf-Witten state-sum TQFT whose 4-simplex cocycles are the bulk input for the RG equations.","marker":"[32]"},{"why":"gives the group-cohomology classification of SPT phases and DW cocycles used to identify the fixed-point boundary tensors.","marker":"[33]"},{"why":"provides the braiding-statistics characterization of Z2 SPT phases that distinguishes the two symmetric fixed-point boundaries.","marker":"[38]"},{"why":"studies the same three-phase Z2 boundary diagram and the purported tricritical point that the paper's interpolation is compared with.","marker":"[39]"},{"why":"supplies the high-precision Monte Carlo value of the 3D Ising critical coupling against which J=0.22(4) is benchmarked.","marker":"[41]"}],"fun_headline_variants":["Simplex tensor RG maps 3D Ising transition at J=0.22(4)","Symmetry-preserving tensor RG pinpoints 3D transition at J=0.22(4)","Tensor RG for symTFT boundary finds Ising at J=0.22(4)","Membrane operator separates SPT phases in new tensor RG"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that truncating every edge and face bond to dimension at most two, with the paper's gradient-descent choice of truncated tensors, still reproduces the phase diagram of the exact untruncated flow; the authors themselves note this truncation is sensitive to initial guesses and hyperparameters and leaves visible fluctuations in figures 7 and 8.","fun_headline_variants_meta":{"raw":{"variants":["Simplex tensor RG maps 3D Ising transition at J=0.22(4)","Symmetry-preserving tensor RG pinpoints 3D transition at J=0.22(4)","Tensor RG for symTFT boundary finds Ising at J=0.22(4)","Membrane operator separates SPT phases in new tensor RG"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001062,"raw_usage":{"total_tokens":4548,"prompt_tokens":1132,"completion_tokens":3416,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":748,"completion_tokens_details":{"reasoning_tokens":3325}},"tokens_in":748,"tokens_out":3416,"duration_ms":25237,"temperature":1.0,"reasoning_tokens":3325,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T17:51:57.608993+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A concrete check: rerun the same RG flow from the same initial tensors with a different truncation rule (for example, keep three bond states instead of two or use a different optimizer) and see whether the magnetization transition remains at $J=0.22(4)$ and whether the membrane order parameter still separates the two SPT phases; a shift larger than the quoted uncertainty would mean the truncation, not the re-triangulation invariance, is what fixes the reported phase boundaries.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"supplies the 3+1D strange-correlator construction of boundary partition functions and the exact RG operator whose numerical implementation this paper develops."},{"cited_title":"Vanhove, M","cited_arxiv_id":null,"evidence_quote":"introduces strange correlators for mapping topological to conformal field theories, the conceptual template for boundary states as tensor networks."},{"cited_title":"Dijkgraaf and E","cited_arxiv_id":null,"evidence_quote":"defines the Dijkgraaf-Witten state-sum TQFT whose 4-simplex cocycles are the bulk input for the RG equations."},{"cited_title":"Chen, Z.-C","cited_arxiv_id":null,"evidence_quote":"gives the group-cohomology classification of SPT phases and DW cocycles used to identify the fixed-point boundary tensors."},{"cited_title":"Levin and Z.-C","cited_arxiv_id":null,"evidence_quote":"provides the braiding-statistics characterization of Z2 SPT phases that distinguishes the two symmetric fixed-point boundaries."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"supplies the high-precision Monte Carlo value of the 3D Ising critical coupling against which J=0.22(4) is benchmarked."}],"review_version":1}