{"id":"1c25192d-719a-432f-a6d4-c73aa0010339","arxiv_id":"2412.12490","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Balanced pairs of imaginary on-site potentials can be arranged so the transfer matrix becomes SU(2)-like, producing real-energy delocalized states with an exact mobility edge.","lead":"A 1D chain with random gain-loss pairs, called dipolar disorder, is shown to host delocalized eigenstates with real energies even though purely imaginary disorder usually localizes everything. The mechanism is a transfer-matrix compactness condition that produces an exact mobility edge, and boundary conditions can tune how many such states appear.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The claimed exact mobility edge lacks a proof on the localized side: in the red region with E^2+V^2<4 the transfer matrices are elliptic, contradicting the paper's assertion that both have an eigenvalue >1.","rationale":"The concrete SU(2) construction for tr(MN)<=2 is self-contained and the PR scaling is credible evidence for delocalized states inside the blue region. The fragility is entirely on the complementary claim that the black line is an exact mobility edge. The paper's own text limits the localization argument to 'localization is expected' and supports it with an eigenvalue statement that is wrong for an open subset of the red region. This is not an attack on the result, but it means the sharp transition is not proven as written; it is conditional on the positivity check. The reader's weakest assumption already identified the missing positivity proof, and this stress test sharpens it by locating a concrete parameter regime where the stated reason cannot work. The verdict should remain CONDITIONAL, not stronger, because the central mechanism is sound and the missing piece is checkable.","tokens_in":14912,"tokens_out":11103,"duration_ms":106404,"concrete_test":"Compute the Lyapunov exponent lambda = lim_{N->infty} (1/N) log ||T_N|| for random binary products of M and N at a grid of (E,V) outside the black line, in particular E=0.5, V=1.5 and other points with 0<E^2+V^2<4, using N up to 10^6 and averaging over at least 1000 sequences. If lambda tends to zero at any such point, the claimed sharp mobility edge is invalid; if lambda is positive for all tested points, the localized-side claim is numerically supported and the paper's hyperbolic-eigenvalue justification should be corrected.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central result is a sharp mobility edge at tr(MN)=2. The delocalized side follows from the SU(2) construction; the localized side rests on the statement that in the red region both M and N have an eigenvalue larger than one, so localization is expected. That statement is false for an open part of the red region. Since the eigenvalues of M are (E^2+V^2-2 plus or minus sqrt((E^2+V^2-2)^2-4))/2, whenever 0 is less than r=E^2+V^2 less than 4 the discriminant is negative and both eigenvalues satisfy |lambda|=1; the matrices are elliptic, not hyperbolic. For example, E=0.5, V=1.5 has r=2.5 and 2|E|=1, so it lies outside the black line although no eigenvalue exceeds 1. The Furstenberg/noncompactness argument invoked for localization therefore does not follow from the stated eigenvalue property, and no separate proof is given that every energy with tr(MN)>2 has a positive Lyapunov exponent. The exactness of the mobility edge requires exactly that positivity, including in this elliptic red region.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript studies a 1D nearest-neighbor tight-binding chain with reciprocal hopping and a random binary imaginary on-site potential arranged in balanced ±iV dipoles. The disorder enters through the random orientation of each dipole, giving two transfer matrices M and N. The central claim is that for real energies satisfying tr(MN) ≤ 2, the two matrices can be simultaneously transformed by a fixed similarity transformation into SU(2) matrices, so every random product of them has a zero Lyapunov exponent, an infinite localization length, and delocalized eigenstates. This condition is shown to be equivalent to -4E^2 + E^4 + 2E^2V^2 + V^4 ≤ 0, yielding an exact mobility edge. Exact diagonalization and participation-ratio scaling are used to confirm delocalized states inside this region and localized states outside, with the number of delocalized real-energy states growing as a power law for generic twisted boundary conditions. The paper also discusses states with small imaginary energy and a D2 spectral symmetry.","tokens_in":15075,"tokens_out":13803,"duration_ms":137378,"significance":"If the central claim holds, this is a valuable analytically tractable example of an exact mobility edge in a genuinely random 1D non-Hermitian system, with a clear mechanism (emergent compactness) and experimentally accessible predictions. The SU(2) construction for the delocalized side is elegant, essentially self-contained, and not circular: the mobility edge is derived from a trace inequality rather than fitted, and the participation-ratio numerics independently support the delocalized phase. The main caveat is that the exactness of the mobility edge on the localized side is not proven in the present manuscript, and one of the stated justifications for it is false in an open region. The paper deserves a major revision rather than rejection, because the gap is repairable by adding a proof of positivity of the Lyapunov exponent outside the compact region or by reframing the exactness claim.","major_comments":[{"comment":"The exact mobility edge rests on the statement that in the red region 'both M and N have an eigenvalue larger than one, thus they do not belong to any compact subgroup of SL(2,C); therefore localization is expected.' This statement is false for an open part of the red region. From Eq. (21), tr(M)=E^2+V^2-2 and det(M)=1, so for 0<E^2+V^2<4 the eigenvalues of M are complex conjugates of modulus 1; the same holds for N. For example, E=0.5 and V=1.5 satisfy tr(MN)>2 (the mobility-edge polynomial evaluates to 5.25>0) but give E^2+V^2=2.5, so neither M nor N has an eigenvalue of modulus greater than 1. Consequently, Furstenberg's theorem cannot be invoked through the stated eigenvalue property, and the paper gives no proof that every random word with tr(MN)>2 has a positive Lyapunov exponent in this elliptic regime. Because the sharp mobility edge is a central claim, either this positivity must be proved (for instance by showing that the semigroup generated by M and N is non-compact and irreducible, or by an explicit lower bound on the Lyapunov exponent), or the exactness claim must be weakened.","section":"Emergent compactness, paragraph after Fig. 1(b)"},{"comment":"The simultaneous SU(2) construction is only worked out for (b*d-bd*)(-a*c+ac*)<0, and the equality case is dismissed by the sentence 'When the equal sign is taken, the matrix P is sufficient.' On the boundary tr(MN)=2, however, the product is zero for generic boundary points, so one of the off-diagonal entries of P^{-1}P* vanishes; the W construction in Eq. (14) then degenerates, and the appendix's case division maps this to the triangular Case 2, which is excluded by Condition 1. The manuscript therefore does not actually prove that the boundary of the blue region is delocalized, although the text explicitly asserts that the black line is included. A separate treatment of the boundary, or an explicit statement that the boundary is understood only as a limit, is needed for the claimed exact mobility edge.","section":"Appendix, Eq. (14)-(16) and the sentence before Eq. (5)"}],"minor_comments":[{"comment":"The stated condition is tr(MN) in [-2,2], but Condition 3 and Eq. (4) in the main text only impose tr(MN) ≤ 2; please clarify whether the lower bound is automatic for the dipolar model or is an additional requirement.","section":"Appendix, opening of the proof"},{"comment":"The square root in the definition of W has two branches; specify the chosen branch and explain why the resulting similarity transformation is single-valued.","section":"Eq. (14)"},{"comment":"The caption says the fraction of eigenstates is around 10^-4 for K=0 and pi, whereas the main text reports a ~10^-2 fraction of real energies and a ~10^-4 fraction of delocalized states; please make the caption and text consistent.","section":"Fig. 1(c) caption and main text"},{"comment":"The participation ratio is defined for normalized right eigenvectors; for a non-Hermitian Hamiltonian it would be helpful to state explicitly that no biorthogonal normalization is used and to justify this choice for the localization diagnostic.","section":"Eq. (9)"}],"recommendation":"major_revision","confidential_remarks":"For the editor: the delocalized side of the central claim is sound and the paper is well within the journal's scope, but the exact mobility edge is not fully established. The localized side rests on a false eigenvalue assertion in an elliptic region, and the boundary of the compact region is not covered by the appendix proof. These issues are repairable within the manuscript's scope by adding a proof of positivity of the Lyapunov exponent outside the compact region or by softening the exactness claim to a numerically supported one. I therefore recommend major revision rather than rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThis one is worth your time. The new element is a third way to get 1D delocalization in a disordered non-Hermitian chain: a binary imaginary potential in balanced gain–loss pairs (dipolar disorder), with a two-site transfer-matrix block that can be simultaneously conjugated into SU(2) when tr(MN)≤2. The SU(2) compactness then gives a zero Lyapunov exponent, and the participation-ratio numerics independently confirm delocalized real-energy states in the predicted region. The appendix proof of the SU(2) condition is clear for the open region, and the mobility-edge curve follows from the trace inequality, not from a fit. That is the paper's real asset.\n\nThe soft spots are on the other side of the claimed exact mobility edge. The paper says that outside the SU(2) region both M and N have an eigenvalue larger than one, so localization follows from Furstenberg. That statement is false for the part of the red region with E^2+V^2<2: there the matrices are elliptic and their eigenvalues lie on the unit circle (take E=0, V=0.5). A random product of non-commuting elliptic matrices can still have a positive Lyapunov exponent, so localization is plausible, but the paper does not prove it for every point with tr(MN)>2. For an 'exact' mobility edge, that missing proof is the main referee issue. The boundary tr(MN)=2 itself is also handled only in a limiting sense, since the similarity transformation can degenerate there. A separate but honest caveat: the delocalized states are subextensive (Nd grows like L^α, α<1), so the fraction vanishes in the thermodynamic limit; the SM states this, though the abstract's phrasing is easy to misread.\n\nA stress-test note flagged the same localized-side gap; its specific discriminant claim is off for E^2+V^2>2, but the underlying objection stands.\n\nVerdict: this deserves a serious referee. The mechanism is novel, the delocalized side is rigorously grounded, and the numerics are consistent. The authors should either prove positivity of the Lyapunov exponent outside the compact region or soften the claim from 'exact mobility edge' to a well-supported transition with a proven delocalized region. I would bring it to a reading group and would cite the compactness condition; I just would not repeat the word 'exact' until the other side is closed.","headline":"A novel SU(2)-compactness mechanism gives a solid delocalized side in a non-Hermitian disordered chain, but the claimed exact mobility edge is missing a proof of localization outside, and one supporting statement is demonstrably false.","tokens_in":15633,"tokens_out":13549,"would_cite":true,"duration_ms":104375,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["82B44","81Q12"],"pacs":["72.15.Rn"],"model":"deepseek-v4-flash","headline":"Paired gain and loss sites can delocalize a 1D disordered chain, exact mobility edge found","keywords":["non-Hermitian localization","Anderson localization","mobility edge","transfer matrix","SU(2) compactness","imaginary potential disorder","Lyapunov exponent","participation ratio"],"falsifier":"Numerically compute the Lyapunov exponent $\\lambda_L = \\lim_{N\\to\\infty} (1/N) \\log ||T_{\\text{tot}}^{(N)}||$ for a long random word of $M$ and $N$ at energies just outside the curve $\\mathrm{tr}(MN) = 2$ for fixed $V$. If $\\lambda_L$ vanishes, or if the participation ratio of real-energy eigenstates scales linearly with $L$ in that region, the claimed mobility edge is wrong. Alternatively, at exactly $\\mathrm{tr}(MN) = 2$, check whether the $W$-rescaling exists: if $(b^*d - bd^*)(-a^*c + ac^*) = 0$ forces $W$ to degenerate and no simultaneous SU(2) conjugation exists, the boundary is not exact.","tokens_in":1928,"feed_emoji":"⚛️","tokens_out":2909,"duration_ms":66977,"temperature":0.7,"pith_summary":"The paper asks whether purely on-site gain and loss, without non-reciprocal hopping or quasiperiodic structure, can delocalize eigenstates in a one-dimensional disordered chain. It answers yes, provided the random imaginary potential is arranged in minimal 'dipolar' pairs $(+iV, -iV)$ of random orientation. For real eigenvalues inside a sharp mobility edge given by $-4E^2 + E^4 + 2E^2V^2 + V^4 = 0$, the two-letter transfer matrices $M$ and $N$ can be simultaneously similarity-transformed into compact SU(2) matrices, so the Lyapunov exponent vanishes and the localization length diverges. Outside this region localization is expected. The fraction of delocalized states depends on twisted boundary conditions and grows as a power law with system size for generic twist angles.","feed_headline":"Paired gain and loss sites delocalize a 1D disordered chain","feed_subtitle":"Dipolar disorder makes transfer matrices compact, giving zero Lyapunov exponent and infinite localization length.","key_machinery":"The machinery is the pair of dipolar transfer matrices $M$ and $N$, each a two-site block with imaginary potential of opposite sign, together with the condition $\\mathrm{tr}(MN) \\le 2$. Because $M = N^*$ and $\\mathrm{tr}(M) \\in [-2,2]$, both have unit-modulus eigenvalues; a normalized eigenvector matrix $P$ diagonalizes $M$, and if $\\mathrm{tr}(MN) \\le 2$ the quantity $(b^*d - bd^*)(-a^*c + ac^*) \\le 0$, so a further rescaling $W$ makes $U = W^{-1}P^{-1}P^*W$ an SU(2) matrix. Thus $PW$ simultaneously conjugates $M$ and $N$ into SU(2), and compactness of the group forces a random product to have zero Lyapunov exponent.","core_discovery":"The central claim is that emergent compactness, not non-reciprocity or quasiperiodicity, is enough to defeat standard one-dimensional localization in a non-Hermitian chain with random imaginary potential disorder. When the disorder is made dipolar -- each disorder block containing one gain and one loss site -- the elementary transfer matrices $M = T_+ T_-$ and $N = T_- T_+$ are complex conjugates with real trace. For any real energy with $\\mathrm{tr}(MN) \\le 2$, the paper constructs a similarity transformation (using a diagonalizing matrix $P$ and a rescaling $W$) that maps both $M$ and $N$ into SU(2) simultaneously. Since SU(2) is compact, a random product of such matrices cannot grow, giving zero Lyapunov exponent and infinite localization length. The equality $\\mathrm{tr}(MN) = 2$ yields the exact mobility edge $-4E^2 + E^4 + 2E^2V^2 + V^4 = 0$, matching participation-ratio numerics.","pith_inferences":["The same emergent-compactness mechanism could apply to other two-letter transfer-matrix alphabets of conjugate pairs with real trace, such as non-Hermitian hopping phases or modified disorder distributions, whenever $\\mathrm{tr}(MN) \\le 2$ holds.","Because the scaling exponent $\\alpha$ is less than 1, the fraction of delocalized states vanishes in the thermodynamic limit even though their number diverges; whether this affects transport in the infinite-size limit is an open question the paper does not settle.","The strong boundary-condition sensitivity hints that the real-energy density can be engineered by spectral twist, which could be probed experimentally in photonic waveguide arrays with controlled gain and loss.","The spectrum's D2 symmetry, proven via the trace being a polynomial in $E^2$, may generalize to other purely imaginary disorder ensembles and yield paired real energies beyond the dipolar construction."],"forward_implications":["All real-energy eigenstates in the SU(2) region are delocalized, with participation ratio scaling linearly with system size in finite-size numerics.","The mobility edge is exactly the curve $-4E^2 + E^4 + 2E^2V^2 + V^4 = 0$, separating delocalized from localized real eigenstates.","For generic twist angle $K$, the number of delocalized real eigenstates grows as $L^\\alpha$ with $\\alpha$ between 0.2 and 0.4, so delocalized states survive in the thermodynamic limit; for periodic and anti-periodic boundary conditions the count remains small.","Most complex-energy eigenstates are localized, but eigenvalues with sufficiently small imaginary part show delocalized scaling up to numerically accessible sizes.","Tuning the twisted boundary condition $K$ controls how many real energies and delocalized states appear in the spectrum."],"supporting_citations":[{"why":"Establishes the benchmark that one-dimensional potential disorder localizes the entire spectrum, which this model is designed to evade.","marker":"[1]"},{"why":"Supplies the earlier non-Hermitian delocalization route via non-reciprocal hopping, the alternative this paper avoids.","marker":"[9]"},{"why":"Provides the Lie-group background distinguishing non-compact SL(2,C) from compact SU(2).","marker":"[55]"},{"why":"Gives the relation between Lyapunov exponent, localization length, and Anderson transition diagnostics used throughout.","marker":"[57]"},{"why":"Supplies the theorem that random products in a non-compact group generally yield a positive Lyapunov exponent, the obstruction overcome by compactness.","marker":"[62]"},{"why":"States the explicit analytical mobility edge $-4E^2 + E^4 + 2E^2V^2 + V^4 = 0$.","marker":"[64]"},{"why":"Provides the transfer-matrix mapping with twisted boundary conditions used to define the spectrum and count real energies.","marker":"[65]"},{"why":"Supplies the participation-ratio diagnostic and large-PR delocalization criterion used for numerical verification.","marker":"[42]"}],"fun_headline_variants":["Paired gain and loss sites banish localization in 1D","Emergent SU(2) compactness gives infinite localization length","Exact mobility edge from dipolar non-Hermitian disorder","Dipolar disorder produces compact transfer matrices","Gain-loss pairing unlocks delocalization in 1D"],"cache_read_input_tokens":17792,"weakest_assumption_plain":"The sharp phase boundary assumes that every energy outside the SU(2) region indeed gives a positive Lyapunov exponent for the random $M/N$ product; the paper cites a standard random-product theorem for this but does not prove it for this specific two-letter alphabet, and on the boundary $\\mathrm{tr}(MN) = 2$ the constructed similarity transformation via $W$ can degenerate, so the exactness of the boundary is not fully pinned down by the written proof.","fun_headline_variants_meta":{"raw":{"variants":["Paired gain and loss sites banish localization in 1D","Emergent SU(2) compactness gives infinite localization length","Exact mobility edge from dipolar non-Hermitian disorder","Dipolar disorder produces compact transfer matrices","Gain-loss pairing unlocks delocalization in 1D"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000752,"raw_usage":{"total_tokens":3305,"prompt_tokens":862,"completion_tokens":2443,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":478,"completion_tokens_details":{"reasoning_tokens":2361}},"tokens_in":478,"tokens_out":2443,"duration_ms":18402,"temperature":1.0,"reasoning_tokens":2361,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T14:03:09.150127+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Numerically compute the Lyapunov exponent $\\lambda_L = \\lim_{N\\to\\infty} (1/N) \\log ||T_{\\text{tot}}^{(N)}||$ for a long random word of $M$ and $N$ at energies just outside the curve $\\mathrm{tr}(MN) = 2$ for fixed $V$. If $\\lambda_L$ vanishes, or if the participation ratio of real-energy eigenstates scales linearly with $L$ in that region, the claimed mobility edge is wrong. Alternatively, at exactly $\\mathrm{tr}(MN) = 2$, check whether the $W$-rescaling exists: if $(b^*d - bd^*)(-a^*c + ac^*) = 0$ forces $W$ to degenerate and no simultaneous SU(2) conjugation exists, the boundary is not exact.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the Lie-group background distinguishing non-compact SL(2,C) from compact SU(2)."},{"cited_title":"Hatano and D","cited_arxiv_id":null,"evidence_quote":"Supplies the earlier non-Hermitian delocalization route via non-reciprocal hopping, the alternative this paper avoids."},{"cited_title":"Furstenberg, Noncommuting random products, Trans- actions of the American Mathematical Society 108, 377 (1963)","cited_arxiv_id":null,"evidence_quote":"Supplies the theorem that random products in a non-compact group generally yield a positive Lyapunov exponent, the obstruction overcome by compactness."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"States the explicit analytical mobility edge $-4E^2 + E^4 + 2E^2V^2 + V^4 = 0$."},{"cited_title":"Kohmoto, Localization problem and mapping of one- dimensional wave equations in random and quasiperiodic media, Physical Review B 34, 5043 (1986)","cited_arxiv_id":null,"evidence_quote":"Provides the transfer-matrix mapping with twisted boundary conditions used to define the spectrum and count real energies."},{"cited_title":"Longhi, Topological phase transition in non-hermitian quasicrystals, Physical review letters 122, 237601 (2019)","cited_arxiv_id":null,"evidence_quote":"Supplies the participation-ratio diagnostic and large-PR delocalization criterion used for numerical verification."}],"review_version":1}