{"id":"3046e9e2-5a94-43be-8560-82e20c39243b","arxiv_id":"1908.06157","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Minkowski's chain characterizes badly approximable forms by a smallest-error term bounded away from zero, singular forms by that term tending to zero, and badly approximable forms by existence of a basis of approximating integer vectors.","lead":"The authors use a 120-year-old sequence of integer matrices, the Minkowski chain, to give new if and only if tests for when a real linear form is badly approximable or singular. They also show that a Dirichlet-style approximation by a whole basis of integer vectors holds exactly for badly approximable forms.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 5's singular direction omits the coefficient-dependent scaling needed to make the Minkowski vector admissible; the proof is repairable but as written does not establish the result.","rationale":"The central equivalence in Theorem 3 rests on Lemma 5.3, whose 'straightforward calculation' the reader flagged; that volume lower bound is actually sound. The body {G_m<1} contains a cylinder over a neighborhood of the hyperplane x_l+L_α(x_1,...,x_n)=0, giving volume ≥V|β_l|/m with V depending only on α, and β_l is bounded independently of m by the standard basis. The main genuine soft spot is not there but in Theorem 5, where the proof fails to carry over the coefficient-dependent constant κ used in Theorem 3: without it, the vector obtained from the short lattice vector need not be admissible for the Minkowski minimum β1. This is repairable by rescaling t, and the remaining arguments in Theorems 3 and 6 are coherent once the small gaps (the misstated negation in Theorem 5 and the x1=0 case in Theorem 4) are corrected. Therefore the reader's CONDITIONAL verdict remains appropriate; no change is needed.","tokens_in":17353,"tokens_out":31965,"duration_ms":318800,"concrete_test":"Re-derive the singular direction of Theorem 5 with the corrected scaling: choose κ=1+Σ|α_j|, let m be any large integer, set t=m/(κε), and use the vector v supplied by λ1(t)<ε. Verify that the induced r satisfies ‖r‖∞≤m and |ξ(r)|<ε t^{-n}, so r is admissible for β1. Then check that the inequalities between (7.5) and (7.10) still yield |β1|<ε/t^n and |β_l|>c/(t^n ε^{1/n}). If all steps go through, the omission is cosmetic and Theorem 5 stands as stated.","verdict_should_be":"UNCHANGED","load_bearing_attack":"In the proof of Theorem 5 (singular implies |α_{k,1}| tends to 0), the text fixes m and sets t=m/ε. From λ1(t)<ε one gets a lattice vector v=(t^{-1}q_1,...,t^{-1}q_n,t^nξ) with |q_j|<ε t and |ξ|<ε t^{-n}. But the corresponding integral vector r=(q_1,...,q_n,p) has p=ξ-L_α(q), hence |p|≤|ξ|+Σ|α_j||q_j|<ε t^{-n}+(Σ|α_j|)ε t, which can exceed m when Σ|α_j|>1. The definition of β1 as the minimum over all r with ‖r‖∞≤m requires p to be bounded by m; without an extra constant κ depending on α, the vector r may not be admissible, so the bound |β1|<ε/t^n is unjustified. In Theorem 3 the same step is handled by setting m=⌈κtε⌉ with κ chosen so that max(|q_1|,...,|q_n|,|p|)≤m, but Theorem 5 omits this factor. This is a genuine but localized gap in the proof of the singular criterion; it does not affect the badly approximable criterion, whose proof contains the κ factor.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the Minkowski chain, a multidimensional generalization of the Hurwitz continued fraction chain, and applies it to Diophantine approximation. The authors prove that a real linear form L_α is badly approximable if and only if the first coordinate of the Minkowski chain is bounded away from zero (Theorem 3), and singular if and only if that coordinate tends to zero (Theorem 5). They also prove a Dirichlet-type theorem (Theorem 6) asserting that L_α is badly approximable exactly when for every Q there is a unimodular integral matrix with entries below Q whose image under the vector (α,1) has sup norm below c Q^{-n}. Along the way they give a proof of Minkowski's algebraic criterion (Theorem 2) and a consequence for Liouville numbers (Theorem 4). The proofs use Minkowski's second theorem on successive minima and a theory of reduced bases, with constants tracked explicitly in the central lemmas.","tokens_in":17601,"tokens_out":24295,"duration_ms":210564,"significance":"If the results are correct, the paper provides explicit and elegant arithmetic criteria for bad approximability and singularity, and a new sharp form of Dirichlet's theorem with a full basis of approximating vectors. The use of the Minkowski chain revives a classical algorithm and connects it to modern Diophantine approximation and to Dani's dynamical criteria. The proofs are constructive, and the reduction to the lattice Λ_t via successive minima and reduced bases is a clean and potentially influential technique. The paper also gives a self-contained treatment of the necessary geometry-of-numbers background, including a proof of the First Finiteness Theorem. These strengths make the paper a valuable contribution to the field, provided the localized gaps in the proofs of Theorems 5 and 4 are repaired as indicated below.","major_comments":[{"comment":"In the direction 'singular ⇒ |α_{k,1}|→0', the proof sets t = m/ε and obtains from λ1(t) < ε an integral vector r with |q_j| < εt = m for j = 1,...,n and |ξ(r)| < ε t^{-n}. It then claims |β1| < ε/t^n by comparing with r. However, for r to be admissible for the Minkowski algorithm at level m one needs ‖r‖∞ ≤ m. While |q_j| ≤ m is automatic, the bound on the last coordinate p is |p| ≤ |ξ(r)| + Σ |α_j| |q_j| < ε t^{-n} + (Σ |α_j|) m, which exceeds m when Σ |α_j| > 1. Thus r need not be admissible, and the inequality |β1| ≤ |ξ(r)| is not justified. This is precisely the issue that is correctly handled in the proof of Theorem 3 by taking m = ⌈κ t ε⌉ with a constant κ depending on α. The proof of Theorem 5 should incorporate the same κ; after that change the argument goes through with the same ε-dependence up to constants. Please repair this step.","section":"§7, proof of Theorem 5, first paragraph"},{"comment":"The negation of singularity is misstated. The text says that if α is not singular, 'there exists a c > 0 and a sequence {Q_j} tending to infinity such that for each j there are infinitely many q ∈ Z^n with ‖q‖∞ ≤ Q_j and ‖L_α(q)‖ ≥ c Q_j^{-n}.' This is not equivalent to the negation of the definition of singularity; it is much weaker, and the phrase 'infinitely many q' is neither necessary nor sufficient for the subsequent lower bound on β1. The correct negation is: there exists c > 0 and a sequence Q_j → ∞ such that for all q with 0 < ‖q‖∞ ≤ Q_j one has ‖L_α(q)‖ ≥ c Q_j^{-n}. With this correction, the rest of the proof, which follows the lines of Lemma 5.3, is sound.","section":"§7, proof of Theorem 5, converse"},{"comment":"The proof relies on the assertion 'A straightforward calculation shows that vol({x; G_m(x) < 1}) ≥ V |β_l|/m.' This volume bound is the only quantitative bridge between the Minkowski chain coefficients and the lattice Λ_t, and all constants in Lemma 5.3, and hence the quantitative forms of Theorems 3 and 5, depend on it. Since the paper promises proofs of all numbered lemmas, please supply the calculation or a precise reference. The bound is plausible (the body contains a slab of width proportional to |β_l|/m in the last coordinate for a positive-measure set of the first n coordinates), but the explicit dependence of V on α should be recorded.","section":"§5, Lemma 5.3"}],"minor_comments":[{"comment":"The assertion 'x_1 p^n + ... + x_n p q^{n-1} - y q^n is a non-zero integer since q ∤ x_1' fails when x_1 = 0. The conclusion of the lower bound remains true: if x_1 = 0, let j be the smallest index with x_j ≠ 0; after dividing by q^{j-1}, the expression is congruent to x_j p^{n-j+1} modulo q and hence is nonzero, giving an even stronger lower bound. Please adjust the argument to cover this case.","section":"§7, proof of Theorem 4"},{"comment":"The strict inequalities 0 < |α_{k,1}| < ... < |α_{k,n}| < 1 are asserted without proof. The sequential minimization gives |β_1| ≤ ... ≤ |β_l|, but strictness is not automatic; ties could occur even under the Q-linear independence assumption (for instance, two independent minimizers with opposite signs). Please either prove the strict inequalities or weaken them to non-strict where only non-strict inequalities are needed in the subsequent arguments.","section":"§2, around (2.4)"},{"comment":"There are several typographical errors: 'Fibonnaci' in Example (i), 'Minkowsi' in the Introduction, and garbled umlauts in the reference list (e.g., 'Bercksichtigung' for 'Berücksichtigung'). These should be corrected in the final version.","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":"The paper is within the scope of the journal and the main ideas are attractive. I have moderate confidence that the results are correct after the local repairs in the proofs of Theorems 5 and 4. The missing constant in the singular direction of Theorem 5 and the misnegation in its converse are the kind of issue that should have been caught by the authors; the volume calculation in Lemma 5.3 should be written out. I recommend major revision rather than rejection, and I expect that a careful revision will make the paper acceptable."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Worth engaging. The new results are real: Theorems 3, 5, and 6 give Minkowski-chain criteria for badly approximable and singular forms and a genuine basis-level Dirichlet variant. The paper is also a useful revival of Minkowski's old algorithm, with a self-contained treatment and tracked constants. The central arguments for Theorems 3 and 6 look sound to me; the lattice machinery (reduced bases, First Finiteness Theorem) is applied coherently.\n\nThe soft spots are all in the proofs of Theorems 4 and 5. Theorem 4's proof is wrong as written: it claims that for a Liouville α, for all x with 0<||x||∞≤Q, ||L(α^n,...,α)(x)|| ≥ (1−ε)Q^{−n}. But for x=(0,...,0,1), the form value is α itself, which can be much smaller than Q^{−n} because α is Liouville. That stronger claim is false. The theorem itself is still true, since the single vector (0,...,0,1) already violates bad approximability, but the proof needs to be replaced, not patched.\n\nTheorem 5 has two localized issues. First, the negation of 'singular' is misstated: it says there are infinitely many q with ||L(q)|| ≥ c Q^{−n}, but the later minimum bound requires the inequality to hold for all nonzero q up to Q. Second, in the singular→|α_{k,1}|→0 direction, the text sets t=m/ε and then treats a lattice vector as admissible for the Minkowski index m without ensuring |p|≤m; the missing κ factor from Theorem 3 is exactly what is needed. The stress-test note on this is correct. Both are easily repairable.\n\nSo the paper is not in publishable condition as-is, but the flaws are not fatal to the main claims. The novelty and the central criteria justify sending it to a careful referee, who should ask for corrections to these two proofs. I would cite the paper once the corrections are in.","headline":"Solid, well-written paper with genuinely new Minkowski-chain criteria for badly approximable and singular forms; the central arguments hold up, but the proofs of Theorems 4 and 5 have localized gaps that need fixing before publication.","tokens_in":18150,"tokens_out":7836,"would_cite":true,"duration_ms":75632,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["11J13","11J70","11H06"],"pacs":[],"model":"deepseek-v4-flash","headline":"The Minkowski chain separates badly approximable forms from singular ones, and yields a Dirichlet-type theorem that produces a whole basis of approximating integer vectors.","keywords":["Minkowski chain","Hurwitz chain","badly approximable forms","singular linear forms","Dirichlet approximation","successive minima","reduced bases","Liouville numbers"],"falsifier":"Run the Minkowski chain algorithm on a concrete badly approximable form, for instance $\\alpha=(\\theta^2,\\theta)$ with $\\theta=2\\cos(2\\pi/7)$, and compute $|\\alpha_{k,1}|$ for large $k$: Theorem 3 predicts a positive lower bound (indeed the tuples are finite), so any computed value below that bound contradicts the theorem. Conversely, for a Liouville number such as $\\lambda=\\sum_{m\\ge1}10^{-m!}$ and a fixed $n$, Theorems 3 and 5 predict $|\\alpha_{k,1}|\\to0$; a direct chain computation that stabilizes away from zero would refute the claimed dichotomy.","tokens_in":17134,"feed_emoji":"🔗","tokens_out":9410,"duration_ms":84219,"temperature":0.7,"pith_summary":"The paper claims that a classical multidimensional continued-fraction algorithm, the Minkowski chain, exactly detects two contrasting Diophantine behaviours of a real linear form $L_\\alpha(x)=\\alpha_1x_1+\\cdots+\\alpha_nx_n$. After pushing $\\alpha$ through the chain and reading the first coordinate $\\alpha_{k,1}$ of the transformed tuple, the form is badly approximable precisely when $|\\alpha_{k,1}|$ stays bounded away from zero, and singular precisely when $|\\alpha_{k,1}|\\to 0$. The same machinery gives a new variant of Dirichlet's approximation theorem: a linear form is badly approximable if and only if, at every scale $Q$, there is a unimodular integer matrix $A$ with $\\|A\\|_\\infty<Q$ whose row errors satisfy $\\|A(\\alpha_1,\\dots,\\alpha_n,1)^\\top\\|_\\infty<cQ^{-n}$, meaning one gets a whole basis of approximating vectors rather than a single one. The significance is that these are discrete, algorithmic criteria: they turn a neglected classical algorithm into a sharp test that unifies the one-dimensional continued-fraction theory with the geometry of numbers in several variables.","feed_headline":"A chain algorithm separates badly approximable from singular forms","feed_subtitle":"The one-coordinate test also yields a Dirichlet theorem that outputs a whole basis of integer vectors.","key_machinery":"The Minkowski chain is the lexicographic algorithm that, for each $m$, chooses an $\\ell\\times\\ell$ nonsingular integer matrix $A_m$ whose rows minimize $\\|A_m(\\alpha_1,\\dots,\\alpha_n,1)^\\top\\|_\\infty$ row by row, taking the subsequence of distinct matrices as $B_k$. Its coordinate $\\alpha_{k,1}=\\beta_1/\\beta_\\ell$ is the quantity whose size separates badly approximable from singular behaviour. The proof machinery also uses the one-parameter determinant-one lattice $\\Lambda_t\\subset\\mathbb{R}^\\ell$, the sup-norm successive minima $\\lambda_1(t)$, and a reduced-basis theorem that bounds the product $\\lambda_1\\cdots\\lambda_\\ell$ from both sides. The load-bearing bridge is Lemma 5.3, which converts bad approximability into $|\\beta_1|\\gg m^{-n}$ and $|\\beta_\\ell|\\ll m^{-n}$; the conversion rests on a volume estimate for the auxiliary body $\\{G_m<1\\}$, whose volume is claimed to be at least $V|\\beta_\\ell|/m$.","core_discovery":"At the centre is Theorem 6, with Theorems 3 and 5 giving the two halves of the same dichotomy. For $\\alpha=(\\alpha_1,\\dots,\\alpha_n)$ with $1,\\alpha_1,\\dots,\\alpha_n$ linearly independent over $\\mathbb{Q}$, the Minkowski chain produces matrices $B_k$ and tuples $B_k(\\alpha)=(\\alpha_{k,1},\\dots,\\alpha_{k,n})$ with $0<|\\alpha_{k,1}|<\\cdots<|\\alpha_{k,n}|<1$. The paper proves that $L_\\alpha$ is badly approximable exactly when $\\inf_k|\\alpha_{k,1}|>0$, and singular exactly when $\\lim_{k\\to\\infty}|\\alpha_{k,1}|=0$. The proof links the chain coefficients $\\beta_1,\\dots,\\beta_\\ell$ (with $\\ell=n+1$) to the successive minima of the lattice $\\Lambda_t$ of determinant one: bad approximability forces the power-law spacing $cm^{-n}<|\\beta_1|<\\cdots<|\\beta_\\ell|<Cm^{-n}$, while failure of bad approximability makes $|\\beta_1/\\beta_\\ell|=|\\alpha_{k,1}|$ arbitrarily small for some $k$. Theorem 6 upgrades the usual one-vector Dirichlet bound to a full basis statement, and its converse shows that, when the form is not badly approximable, every unimodular matrix with $\\|A\\|_\\infty<Q$ has row error larger than $cQ^{-n}$. For $n=1$ this recovers the classical fact that an irrational is badly approximable iff its partial quotients are bounded.","pith_inferences":["The same reduced-basis argument should extend to systems of several linear forms, since the proof never uses special features of a single form beyond the lattice $\\Lambda_t$; the paper itself notes this as a natural next step.","Because the Minkowski chain is an explicit algorithm, Theorem 3 suggests a numerical test: truncate the chain and monitor $|\\alpha_{k,1}|$; a visible positive floor would be evidence of bad approximability, whereas repeated dips toward zero would indicate singularity.","Theorem 6 can be read as saying that bad approximability is equivalent to a uniform basis-valued Dirichlet condition; an interesting open question is the optimal size of the constant $c$ and whether the full-basis conclusion can be quantified effectively from a finite prefix of the chain."],"forward_implications":["For $n=1$, Theorem 3 together with the chain-to-continued-fraction dictionary recovers the classical fact that an irrational is badly approximable iff its partial quotients are bounded.","For every Liouville number $\\alpha$ and every fixed $n$, the form $L_{(\\alpha^n,\\dots,\\alpha)}$ is not badly approximable, so its Minkowski-chain coordinate $|\\alpha_{k,1}|$ tends to zero as $k\\to\\infty$.","Whenever $L_\\alpha$ is badly approximable, Corollary 3 produces infinitely many unimodular matrices $A\\in\\mathrm{GL}(\\ell,\\mathbb{Z})$ with $\\|A(\\alpha_1,\\dots,\\alpha_n,1)^\\top\\|_\\infty<c\\|A\\|_\\infty^{-n}$.","Minkowski's original finite-set criterion for algebraic numbers is proved with the same machinery: $\\alpha$ is algebraic of degree $\\ell$ exactly when the chain tuples form a finite set.","The lattice form of the criterion connects the algebraic chain to bounded trajectories in the space of unimodular lattices, giving a geometric certificate for bad approximability and singularity."],"supporting_citations":[{"why":"Supplies Minkowski's original chain construction and the algebraic-number criterion that the paper revives and generalizes.","marker":"[17]"},{"why":"Provides the one-dimensional Hurwitz chain of Farey pairs, which Theorem 1 matches to the Minkowski chain for $n=1$.","marker":"[11]"},{"why":"Gives the Davenport-Schmidt improvement of Dirichlet's theorem, whose non-improvement for Liouville numbers is used in the proof of Theorem 4.","marker":"[7]"},{"why":"Supplies the dynamical criterion for badly approximable and singular systems, from which the paper's Lemma 7.1 is extracted.","marker":"[4]"},{"why":"Is the standard source for the definition of badly approximable forms, Dirichlet's theorem, and the continued-fraction facts used in the $n=1$ case.","marker":"[22]"},{"why":"Provides Minkowski's theorem on successive minima, the central geometric input behind Lemma 5.3.","marker":"[19]"},{"why":"Contains the theory of reduced bases, including the comparison lemma and the First Finiteness Theorem used in the proofs of Theorems 3, 5, and 6.","marker":"[25]"}],"fun_headline_variants":["Minkowski chain separates badly approximable from singular forms","Chain gives full basis for badly approximable forms","Successive minima decide bad approximability via chain","Dirichlet theorem upgraded to basis for badly approximable forms","Badly approximable iff chain coefficients stay away from zero"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof depends on a stated but deferred volume calculation for the auxiliary norm body $\\{G_m<1\\}$; every later inequality inherits its constants from that estimate, so if the true volume is smaller than claimed, the lattice argument would not establish the characterization.","fun_headline_variants_meta":{"raw":{"variants":["Minkowski chain separates badly approximable from singular forms","Chain gives full basis for badly approximable forms","Successive minima decide bad approximability via chain","Dirichlet theorem upgraded to basis for badly approximable forms","Badly approximable iff chain coefficients stay away from zero"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000434,"raw_usage":{"total_tokens":2232,"prompt_tokens":990,"completion_tokens":1242,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":606,"completion_tokens_details":{"reasoning_tokens":1165}},"tokens_in":606,"tokens_out":1242,"duration_ms":8996,"temperature":1.0,"reasoning_tokens":1165,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T12:56:02.586413+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the Minkowski chain algorithm on a concrete badly approximable form, for instance $\\alpha=(\\theta^2,\\theta)$ with $\\theta=2\\cos(2\\pi/7)$, and compute $|\\alpha_{k,1}|$ for large $k$: Theorem 3 predicts a positive lower bound (indeed the tuples are finite), so any computed value below that bound contradicts the theorem. Conversely, for a Liouville number such as $\\lambda=\\sum_{m\\ge1}10^{-m!}$ and a fixed $n$, Theorems 3 and 5 predict $|\\alpha_{k,1}|\\to0$; a direct chain computation that stabilizes away from zero would refute the claimed dichotomy.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies Minkowski's original chain construction and the algebraic-number criterion that the paper revives and generalizes."},{"cited_title":"Ann., 14, 1894, 417436; and Oeuvres, tome II, pp","cited_arxiv_id":null,"evidence_quote":"Provides the one-dimensional Hurwitz chain of Farey pairs, which Theorem 1 matches to the Minkowski chain for $n=1$."},{"cited_title":"& Schmidt, W","cited_arxiv_id":null,"evidence_quote":"Gives the Davenport-Schmidt improvement of Dirichlet's theorem, whose non-improvement for Liouville numbers is used in the proof of Theorem 4."},{"cited_title":"G., Divergent trajectories of flows on homogeneous spaces and Diophantine approximation","cited_arxiv_id":null,"evidence_quote":"Supplies the dynamical criterion for badly approximable and singular systems, from which the paper's Lemma 7.1 is extracted."},{"cited_title":"M., Diophantine approximation","cited_arxiv_id":null,"evidence_quote":"Is the standard source for the definition of badly approximable forms, Dirichlet's theorem, and the continued-fraction facts used in the $n=1$ case."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides Minkowski's theorem on successive minima, the central geometric input behind Lemma 5.3."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Contains the theory of reduced bases, including the comparison lemma and the First Finiteness Theorem used in the proofs of Theorems 3, 5, and 6."}],"review_version":1}