{"id":"080a0234-dc54-491d-88e0-b9a63ccfbc24","arxiv_id":"2505.05850","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For doubly-infinite tridiagonal non-Hermitian Hamiltonians, the singular values can be expressed as poles of a Green's function built from two matrix continued fractions.","lead":"Resonances in non-Hermitian quantum models are usually found as complex eigenvalues, which are hard to compute. The paper derives a semi-analytic continued-fraction formula for the real singular values of tridiagonal Bose-Hubbard-like Hamiltonians, recasting the search as a pair of matrix continued fractions.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Claim that singular values are poles of G(z) is false when a singular vector vanishes at the central site; a 3x3 non-Hermitian tridiagonal example gives a zero, not a pole, at a singular value.","rationale":"The reader identified the convergence of the two matrix continued fractions as the weakest assumption. That concern is real: Section 4.4 supplies only an expectation of convergence by analogy, and the author's own Ref. [6] provides a divergence benchmark. However, a more fundamental and independent problem undermines the central claim even when every continued fraction converges. Lemma 2 equates the singular values of H with the poles of G(z)=det F0(z), where F0 is the central block of the resolvent of the Hermitized block-tridiagonal matrix. The central block of a resolvent has a pole at an eigenvalue only if the associated eigenvector has a nonzero component in that block. When a singular vector vanishes at the central site, the eigenvalue appears as a regular point (or a zero) of G, not a pole. I verified this with a concrete non-Hermitian tridiagonal example of the paper's own form: H=[[0,1,0],[2,0,3],[0,4,0]] has singular values 0, sqrt(13), sqrt(17), but Eq. (43) yields G(z)=z^2/[(13-z^2)(17-z^2)], whose pole set is {±sqrt(13), ±sqrt(17)}; the singular value 0 is a zero, not a pole. This is not a borderline or measure-zero issue: the 3-mode PT-symmetric CBH model, which the paper explicitly discusses, exhibits the same phenomenon at γ=0. Therefore the central claim 'singular values can be sought as poles of G(z)' is false without an additional nonvanishing-central-component condition. The paper does not state such a condition, and its proof by analogy with Lemma 1 assumes it silently. This makes the main theorem internally inconsistent, not merely unproven in some limit. The convergence question remains a secondary obstacle: even a full convergence proof would not repair the pole/zero mismatch. For these reasons, the paper in its present form should be rejected; a revision would need to restrict the claim to singular values whose singular vectors have nonzero central projection (or provide a modified Green's function that captures all singular values) and then address the convergence of the continued fractions.","tokens_in":20915,"tokens_out":24933,"duration_ms":242219,"concrete_test":"Compute G(z) from Eq. (43) for the 3x3 non-Hermitian tridiagonal H=[[0,1,0],[2,0,3],[0,4,0]]. With M=N=1 and blocks from Eq. (35), the central Schur complement is diag((13-z^2)/z, (17-z^2)/z), so G(z)=z^2/[(13-z^2)(17-z^2)]. The singular values of H are 0, sqrt(13), sqrt(17); G has poles at ±sqrt(13), ±sqrt(17), but z=0 is a zero. If Lemma 2 were correct, z=0 would be a pole. Repeating the same computation for the 3-mode PT-symmetric CBH model (K=3, γ=0) gives G(z)=z^2/(4-z^2)^2, again missing the zero singular value. A second test: verify that adding an explicit invertibility condition at the central site (e.g., requiring the relevant singular vectors to have nonzero central component) restores the pole correspondence.","verdict_should_be":"REJECT","load_bearing_attack":"Lemma 2 (Section 4.4) states that the singular values of the original tridiagonal Hamiltonian are the poles of G(z)=det F0(z), with F0 defined by Eq. (43) as the central 2x2 block of the resolvent of the Hermitized block-tridiagonal matrix. This is only true for eigenvalues of the Hermitized matrix whose eigenvectors have nonzero projection onto the central two-dimensional site. No such condition is stated or proved. The failure is not exotic: for any tridiagonal H=[[0,a,0],[b,0,c],[0,d,0]] with a,b,c,d nonzero, H has eigenvalue 0 with eigenvector (c,0,-b), so the singular value 0 has right and left singular vectors vanishing at the central site. Concretely, take the non-Hermitian H=[[0,1,0],[2,0,3],[0,4,0]] (M=N=1 in Eq. (14), all diagonal a_k=0). Applying Eqs. (41)-(43) gives F0(z)=diag(z/(13-z^2), z/(17-z^2)), hence G(z)=z^2/[(13-z^2)(17-z^2)]. The singular values of H are 0, sqrt(13), sqrt(17); G has poles at ±sqrt(13) and ±sqrt(17), but z=0 is a double zero, not a pole. Thus the central claim of Lemma 2 is false as stated. The proof of Lemma 2, which says 'the idea is the same as in Lemma 1', silently assumes the matching-site component of every eigenvector is nonzero; that assumption can fail even in the motivating PT-symmetric Bose-Hubbard models (e.g., the 3-mode CBH model at γ=0). This is a logical gap, not merely a missing convergence proof.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a semi-analytic method for locating the singular values of non-Hermitian tridiagonal (Jacobi-like) Hamiltonians of Bose-Hubbard type. After reviewing the scalar continued-fraction Green's function for one-sided tridiagonal problems, the author block-tridiagonalizes the Hermitized operator [0 H; H† 0], introduces two 2×2 matrix continued fractions in Eqs. (41)–(42), and states in Lemma 2 that the Green's function is G(z)=det F0(z) with F0 defined by Eq. (43). The abstract and introduction claim that the singular values of H can be sought as poles of this Green's function. The algebraic derivation of the Schur-complement/F0 formula is standard, but the paper supplies no proof of convergence in the doubly infinite limit and, more importantly, the pole claim is false as stated when a singular vector has zero projection on the central block.","tokens_in":21260,"tokens_out":7558,"duration_ms":81833,"significance":"If the pole statement were correct, the paper would offer a parameter-free, semi-analytic route to singular values of large or doubly infinite non-Hermitian tridiagonal models, extending the one-sided result of Ref. [6] and giving a practical tool for the cited Bose-Hubbard-like models. The manuscript is self-contained in its algebraic manipulations and contains no fitted parameters, which is a genuine strength. However, the advertised central claim is not merely missing a proof: it fails for a simple 3×3 example embedded in the paper's own class of models. The useful algebraic core can likely be repaired, but the present version does not establish the stated result.","major_comments":[{"comment":"The statement that the singular values of H are poles of G(z)=det F0(z) is false when a singular vector vanishes at the central site. For H=[[0,1,0],[2,0,3],[0,4,0]], which is of the form (14) with M=N=1 and all a_k=0, the recurrences (41)-(43) give F0(z)=diag(z/(13-z^2), z/(17-z^2)), hence G(z)=z^2/[(13-z^2)(17-z^2)]. The singular values of H are 0, sqrt(13), sqrt(17); G has poles at ±sqrt(13) and ±sqrt(17), but z=0 is a double zero, not a pole. The singular vector for σ=0 is (3,0,-2), whose central component vanishes, so the Hermitized eigenvector has zero projection on the central 2×2 block. The proof of Lemma 2, which says that 'the idea is the same as in Lemma 1', silently assumes this projection is nonzero. The same phenomenon already occurs in the scalar Lemma 1, e.g. for H=[[0,1,0],[1,0,1],[0,1,0]], whose zero-energy eigenvector (1,0,-1) has f0(z)=z/(2-z^2) with a zero, not a pole, at z=0. Lemma 2 therefore needs an explicit nonvanishing condition on the central projection, or a reformulation in terms of zeros/poles of the Schur complement, together with a treatment of the exceptional eigenvectors.","section":"Section 4.4, Lemma 2 and Eq. (43)"},{"comment":"Convergence of the two matrix continued fractions in the limit M,N→∞ is load-bearing but is never established. The text only says that convergence 'may be expected' by analogy with the one-sided case of Ref. [6] and that the sparse 2×2 blocks make convergence 'quick'. No proof is given for the doubly infinite two-sided setting, and no numerical test is reported. The one-sided proof in Ref. [6] does not automatically cover the asymmetric two-sided case with non-Hermitian off-diagonal elements b_k, c_{k+1}. Since G(z) is defined through these infinite recurrences, the paper must either prove convergence under assumption (15) or explicitly restrict the claims to finite M,N.","section":"Section 4.4, Eqs. (41)-(43)"},{"comment":"The paper does not state the invertibility and regularity assumptions under which the matrix recurrences (38), (41), and (42) are well defined. The scalar assumption (22) is formulated for the f_j, but the analogous non-vanishing conditions for the 2×2 matrices F_j, and for the denominators appearing in Eqs. (41)-(43), are missing. As the counterexample above shows, the denominators can vanish at a singular value, and the limiting formula then produces a zero rather than a pole. A precise statement of the regularity domain of z, including what happens at points where intermediate F_j are singular, is needed for Lemma 2 to be meaningful.","section":"Section 4.3 and Eq. (39)"}],"minor_comments":[{"comment":"The sentence 'the convergence may be expected quick here' is grammatically awkward and unsupported; it should be rewritten as a precise quantitative claim or removed.","section":"Section 4.4"},{"comment":"The regularity assumption (22) is written for scalar continued fractions; the corresponding matrix invertibility conditions for the F_j in Eqs. (41)-(42) should be stated explicitly.","section":"Section 3.2, Eq. (22)"},{"comment":"G(z) is called a Green's function but is defined as the determinant of a 2×2 block of the resolvent of the Hermitized operator; the terminology should be defined and distinguished from the full resolvent.","section":"Section 4.4, Eq. (43)"},{"comment":"There is a typo in the footnote: 'atytempts' should be 'attempts'.","section":"Footnote 11"},{"comment":"Reference [7] is a Wikipedia page; a standard textbook reference for singular values would be more appropriate.","section":"References"},{"comment":"The paper contains no numerical illustration of Eq. (43), even for a small finite matrix; a finite-dimensional example that confirms the pole locations, or that explicitly exhibits the problematic zero case, would greatly improve the presentation.","section":"General"}],"recommendation":"major_revision","confidential_remarks":"The counterexample in the first major comment is decisive: the advertised pole claim is false without an additional nonvanishing condition. I recommend major revision rather than rejection because the algebraic Schur-complement framework is sound and the statement can likely be repaired by adding the missing projection condition and restricting or proving convergence in the infinite limit. However, if the authors cannot either prove the corrected pole characterization or clearly delimit its validity, the paper's main advertised result should be withdrawn."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take on arXiv:2505.05850. The paper does something worthwhile: it carries the author's earlier continued-fraction treatment of complex tridiagonal Hamiltonians over to the doubly-infinite Bose-Hubbard-like case, then uses the Hermitization trick of Pushnitski and Stampach to turn the singular-value problem into a block-tridiagonal Hermitian eigenvalue problem, and writes the Green's function as a pair of matrix continued fractions. The algebra in Lemmas 1-2 is standard and checks out as formal computation. The matrix version of the continued fraction for the 2x2 blocks looks like the genuinely new piece.\n\nThe central claim, however, is wrong as stated. Lemma 2 asserts that singular values are poles of G(z)=det F0(z). That is only true when the corresponding eigenvector of the Hermitized operator has a nonzero projection on the central two-dimensional site. When that projection is zero, the resolvent block F0(z) is regular there, and G(z) shows a zero, not a pole. The stress-test example is a concrete instance: H=[[0,1,0],[2,0,3],[0,4,0]] gives F0(z)=diag(z/(13-z^2), z/(17-z^2)), so G has poles at ±sqrt(13), ±sqrt(17), but the singular value 0 is a double zero. This is not a contrived corner case; the same structure occurs in the 3-mode CBH model at gamma=0. The proof of Lemma 2 simply says the idea is the same as in Lemma 1, but Lemma 1's scalar argument implicitly assumes psi_0 != 0. In the matrix case the analogous assumption is a nonzero two-component block, and it is not stated.\n\nOther soft spots are less severe: convergence of the infinite matrix continued fractions is only argued by analogy with Ref. [6], and there are no numerical examples at all. Those are addressable. The pole/zero issue is load-bearing.\n\nNet: this is a useful formalism to know about, but not a finished method. I would send it to a referee — the flaw is specific and fixable, and the paper deserves serious scrutiny rather than desk rejection — but the referee should be asked to check the matching-site assumption and, if it fails, to say how the missing singular values are recovered.","headline":"Lemma 2 overstates its result: poles of G(z) only capture singular values whose eigenvectors have a nonzero central-block component, and the presented 3x3 example shows a missing singular value as a zero instead.","tokens_in":21814,"tokens_out":6042,"would_cite":false,"duration_ms":53377,"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 claims that the singular values of a doubly infinite non-Hermitian tridiagonal Hamiltonian are the poles of a Green's function built from a pair of matrix continued fractions.","keywords":["non-Hermitian Hamiltonians","singular values","matrix continued fractions","Green's functions","Bose-Hubbard model","tridiagonal matrices","resonances","exceptional points"],"falsifier":"Take a finite Bose-Hubbard-like matrix $H[M,N]$ with diagonals satisfying $|a_k| \\sim |k|^p$ for large $|k|$ and constant off-diagonals, compute the poles of $G(z)$ from recurrences (41)-(43) at increasing $M,N$, and compare with the singular values of a direct SVD of the same finite matrix; if the pole sets do not converge to the singular values as $M,N$ grow, or if the recurrences blow up before convergence, Lemma 2 is falsified.","tokens_in":20681,"feed_emoji":"⚛️","tokens_out":5409,"duration_ms":55680,"temperature":0.7,"pith_summary":"The paper tackles the hard problem of locating complex resonance energies of non-Hermitian quantum Hamiltonians. It argues that for tridiagonal Hamiltonians with diagonals growing at both infinities, a cheaper target works: the real singular values of H, which can be recovered as poles of an auxiliary Green's function. The Green's function is assembled from two independent matrix continued fractions, one running up the index ladder and one running down. If this construction is correct, resonance physics in Bose-Hubbard-like many-boson models becomes accessible to a semi-analytic computation that avoids direct complex diagonalization.","feed_headline":"Two continued fractions locate resonance singular values","feed_subtitle":"Non-Hermitian Bose-Hubbard-like models get a Hermitized Green's function whose poles give the real singular values.","key_machinery":"The load-bearing object is the pair of two-by-two matrix continued fractions in recurrences (41) and (42), together with the central block $F_0(z)$ of Eq. (43). The recurrences are the block-tridiagonal analogues of the scalar continued fractions that define Green's functions for ordinary Jacobi matrices; they solve the factorization of $H - \\sigma$ into $U F L$. Their convergence, expected under the two-sided growth condition (15), is what makes $G(z)$ well defined. The sparse structure of the 2x2 blocks ($A_k$ has zeros on the diagonal, $B_k$ and $C_k$ have a single nonzero entry) is invoked as the reason convergence should be quick.","core_discovery":"The central claim is Lemma 2: after Hermitization by embedding H and H† into a self-adjoint block operator and permuting basis states, the Hamiltonian becomes block-tridiagonal with 2x2 blocks. Factoring $H - \\sigma = U F L$ and solving the block recurrences yields $F_0(z) = [A_0 - z - C_0 F_{-1}(z) B_{-1} - B_0 F_1(z) C_1]^{-1}$, where $F_1$ and $F_{-1}$ are the limits of the recursions $F_k = 1/(A_k - z - B_k F_{k+1} C_{k+1})$ and $F_{-j} = 1/(A_{-j} - z - C_{-j} F_{-j-1} B_{-j-1})$. The Green's function $G(z) = \\det F_0(z)$ then has poles exactly at the singular values $\\sigma_n$ of $H$. The paper presents this as a generalization of the one-sided continued-fraction method, needed because Bose-Hubbard-like Hamiltonians are doubly infinite with growth at both ends.","pith_inferences":["Inference: if the two-sided convergence is proved rigorously, the construction gives a deterministic, parameter-free way to compute singular values of non-Hermitian many-body Hamiltonians, complementing stochastic and Krylov-based approaches.","Inference: the fixed-point basis of the one-sided convergence proof in the companion work suggests the two-sided recurrences may also be stable under small random perturbations, which would make the method usable for ensemble studies of non-Hermitian random tridiagonal matrices.","Inference: a numerical scaling test, fixing the off-diagonal structure and varying the growth rate of $a_k$ while monitoring the truncation depth needed for a target pole accuracy, would turn the expected convergence into a quantitative criterion.","Inference: in PT-symmetric $M=N$ models, tracking where poles of $G(z)$ coalesce could locate higher-order exceptional points, connecting singular-value poles to the EP-unfolding phenomenology."],"forward_implications":["Resonance energies for non-Hermitian tridiagonal models can be located without finding complex eigenvalues: one computes real singular values as poles of $G(z)$.","The method removes the one-sided limitation of earlier continued-fraction treatments and covers doubly infinite Bose-Hubbard-like Hamiltonians with two-sided diagonal growth.","Because the 2x2 blocks are sparse, the matrix continued fractions are expected to converge quickly, making the approach competitive with standard singular-value algorithms.","The same construction carries over to odd numbers of bosons and to asymmetric cut-offs $M \\neq N$, since the recurrences treat the two sides independently.","Onset of divergence of the continued fractions is tied to a physical regime change, typically toward a continuous spectrum."],"supporting_citations":[{"why":"Supplies the Hermitization idea: singular values of H are eigenvalues of the block operator built from H and H†.","marker":"[5]"},{"why":"Provides the one-sided matrix-continued-fraction convergence proof that the present two-sided construction extends by analogy.","marker":"[6]"},{"why":"Introduces the non-Hermitian Bose-Hubbard model whose doubly infinite tridiagonal structure motivates the whole paper.","marker":"[3]"},{"why":"Gives the original single-continued-fraction Green's function method that the paper generalizes to two continued fractions.","marker":"[8]"},{"why":"Supplies the matrix-continued-fraction and block-Lanczos technique underlying the block recurrences.","marker":"[16]"},{"why":"Offers an earlier matrix-continued-fraction application with slow convergence, used as a contrast for the expected speed here.","marker":"[33]"},{"why":"Provides phenomenological motivation for doubly infinite matrix models with two-sided growth.","marker":"[30]"}],"fun_headline_variants":["Double continued fractions locate complex-energy poles","Hermitized trick yields Green's function resonance poles","Matrix fractions solve non-Hermitian Bose-Hubbard spectra","Two-sided fractions expose singular values of H","Continued-fraction Green's function spots resonances"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole construction assumes that the two matrix continued fractions in recurrences (41) and (42) converge as M and N tend to infinity under the two-sided growth condition (15); the paper offers only an analogy-based expectation, not a proof or numerical demonstration.","fun_headline_variants_meta":{"raw":{"variants":["Double continued fractions locate complex-energy poles","Hermitized trick yields Green's function resonance poles","Matrix fractions solve non-Hermitian Bose-Hubbard spectra","Two-sided fractions expose singular values of H","Continued-fraction Green's function spots resonances"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000172,"raw_usage":{"total_tokens":1251,"prompt_tokens":898,"completion_tokens":353,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":514,"completion_tokens_details":{"reasoning_tokens":281}},"tokens_in":514,"tokens_out":353,"duration_ms":4639,"temperature":1.0,"reasoning_tokens":281,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T22:54:05.435330+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a finite Bose-Hubbard-like matrix $H[M,N]$ with diagonals satisfying $|a_k| \\sim |k|^p$ for large $|k|$ and constant off-diagonals, compute the poles of $G(z)$ from recurrences (41)-(43) at increasing $M,N$, and compare with the singular values of a direct SVD of the same finite matrix; if the pole sets do not converge to the singular values as $M,N$ grow, or if the recurrences blow up before convergence, Lemma 2 is falsified.","supporting_citations":[{"cited_title":"Pushnitski, F","cited_arxiv_id":null,"evidence_quote":"Supplies the Hermitization idea: singular values of H are eigenvalues of the block operator built from H and H†."},{"cited_title":"Complex tridiagonal quantum Hamiltonians and matrix continued fractions","cited_arxiv_id":"2504.16424","evidence_quote":"Provides the one-sided matrix-continued-fraction convergence proof that the present two-sided construction extends by analogy."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Introduces the non-Hermitian Bose-Hubbard model whose doubly infinite tridiagonal structure motivates the whole paper."},{"cited_title":"Singh, S","cited_arxiv_id":null,"evidence_quote":"Gives the original single-continued-fraction Green's function method that the paper generalizes to two continued fractions."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the matrix-continued-fraction and block-Lanczos technique underlying the block recurrences."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Offers an earlier matrix-continued-fraction application with slow convergence, used as a contrast for the expected speed here."},{"cited_title":"Garmon, G","cited_arxiv_id":null,"evidence_quote":"Provides phenomenological motivation for doubly infinite matrix models with two-sided growth."}],"review_version":1}