{"id":"11aee997-6f6f-4bf7-a198-ebc528ce6411","arxiv_id":"2505.23936","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"A smooth flow on T^3 is built so that the induction equation grows magnetic energy exponentially at rate at least 1/4, for any prescribed countable set of diffusivities accumulating at zero.","lead":"Keefer Rowan constructs a smooth, divergence-free velocity field on the three-torus that makes magnetic fields grow exponentially for a dense set of arbitrarily small diffusivities, a property called a subsequentially fast dynamo. It is the first compact-domain example of dynamo growth with a rate uniform in diffusivity along a sequence, sitting between slow and fast dynamos.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection identified: the no-common-bad-subspace computation in Corollary 3.7 checks out, and the central claim is supported.","rationale":"The reader's conditional verdict is due to typographical errors in Proposition 2.3 and Eq. (4.2), not to a flaw in the main construction. My stress-test of the most load-bearing algebraic condition, the no-common-bad-subspace verification, shows that it is correct; the transversality of the two bad eigenspaces is stable under small diffusion. The proof of Theorem 1.5 does rely on being able to make the control interval in Proposition 2.2 arbitrarily long, and the proposition as printed only states existence of some time 2n. This is a genuine formal gap in the write-up, but the proof of Proposition 2.2 supplies exactly the stronger statement needed, because the liminf bound implies the desired inequality for all sufficiently large n. Thus no substantive correctness risk to the central claim was identified. The manuscript still needs the minor corrections flagged by the reader, plus a one-line strengthening or clarification of Proposition 2.2, so the verdict remains conditional rather than accept or reject.","tokens_in":17780,"tokens_out":43135,"duration_ms":417227,"concrete_test":"Perform a symbolic computation of the two-dimensional non-growing eigenspaces of the matrices in Lemma 3.10 at kappa=0 for lambda=R and lambda=-R, and verify their intersection is exactly span{e_z}; additionally verify that for the chosen R the dominant eigenvalues exceed e. This directly confirms the no-common-bad-subspace condition used in Corollary 3.7.","verdict_should_be":"UNCHANGED","load_bearing_attack":"I stress-tested the reader's weakest assumption, the no-common-bad-subspace condition behind Corollary 3.7. For the matrices A(W_R) and A(W_{-R}) in Lemma 3.10, the non-growing eigenspaces at kappa=0 are span{(b-S)/(2alpha), i, 0; e_z} and span{(S-b)/(2alpha), i, 0; e_z}, where b=beta R and S=sqrt(b^2+4alpha^2); their intersection is exactly span{e_z}. Hence every nonzero v with e_z dot v = 0 has an expanding component under at least one matrix. Because the eigenvalues are simple and the matrix elements depend continuously on kappa, this transversality persists on a small interval [0,kappa0], so Corollary 3.7 is sound. The only issue I found is formal: Proposition 2.2 states existence of some time 2n, while the proof of Theorem 1.5 needs to choose 2n arbitrarily large. The proof of Proposition 2.2 does support this, since liminf (1/n) log|A^n v| > 1 gives the estimate for all sufficiently large n, so this is a fixable strengthening rather than a threat to the central claim.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper constructs, for any countable collection of diffusivities in [0,κ0], a smooth divergence-free velocity field u on R_+ × T^3 such that the solution of the induction equation with initial data b0 = sin(x)e_z has limsup growth rate at least 1/4 for each diffusivity. The proof combines finite-time control of Fourier modes via explicit flows W_R and W_{-R}, averaging over random translations, and an induction that visits each diffusivity infinitely often. It also includes a unique continuation estimate ensuring nontriviality, and uniform C^∞ estimates on the velocity field.","tokens_in":17979,"tokens_out":16477,"duration_ms":146347,"significance":"If correct, this is the first subsequentially fast dynamo on T^3, giving a positive answer to a weakened form of Arnold's problem. The construction is self-contained: the Fourier matrix elements are computed explicitly in terms of Bessel functions (Lemmas 3.9–3.11), the no-common-bad-subspace condition needed for uniform growth is verified directly, and all estimates are uniform in the chosen diffusivity sequence. The adaptive choice of the flow is a legitimate existence mechanism rather than circular reasoning. The main quantitative claim rests on a finite-time perturbation argument, which correctly avoids the singular infinite-time κ→0 limit.","major_comments":[{"comment":"The proposition as stated guarantees the existence of some time 2n with the stated growth, but the proof of Theorem 1.5 requires that 2n can be chosen arbitrarily large: the induction step explicitly says 'choose R∈2N large enough' so that after subtracting the time spent in Propositions 2.3 and 2.4, the ratio R/t_n is at least 1/4. This quantifier is absent from the statement of Proposition 2.2. The proof of Proposition 2.2 does support the stronger formulation, because liminf_{n→∞} (1/n) log |(A_1^κ)^n v| > 1 implies the estimate |(A_1^κ)^n v| ≥ e^n holds for all sufficiently large n; the proposition should be restated with this quantifier.","section":"§2, Proposition 2.2 and proof of Theorem 1.5"}],"minor_comments":[{"comment":"The final factor in the lower bound should be ||b(0,·)||^2_{L^2_x} rather than ||b(0,·)||_{L^2_x}; as printed the inequality is false in general, e.g., for b0 = sin(x)e_z on the unit torus since ||b0||_{L^2} < 1. The unique-continuation conclusion is unaffected because the bound remains strictly positive, but the statement and Eq. (4.2) need the square.","section":"§4, Proposition 2.3 and Eq. (4.2)"},{"comment":"The proof writes the product pT Vλ,0 ... pT Uλ,0, but the matrix displayed (with -iλαβ in the (1,2) entry) is the product with V_{-λ}; replace Vλ by V_{-λ} in the proof text and in the subsequent 'it suffices to show' sentence for consistency.","section":"§3.1, proof of Lemma 3.10"},{"comment":"In the sentence 'A fast dynamo is then a velocity field in which the fastest exponential growth rate as uniformly bounded away from 0 for all sufficiently small initial data,' the phrase 'initial data' should presumably be 'diffusivities' or 'uniformly in κ'.","section":"§1, Definition 1.3"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is a strong, self-contained construction. The only substantive issue is the missing quantifier in Proposition 2.2, which is load-bearing for the induction in Theorem 1.5 but is easily fixed because the existing proof establishes the stronger statement. The other errors are typographical and do not threaten the central claim. I expect that a revision addressing these points will be ready for acceptance."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"I read Rowan's paper on the subsequentially fast dynamo. The main result is genuinely new: a smooth divergence-free flow on the compact torus that gives exponential magnetic growth at a uniform rate along a prescribed sequence of diffusivities tending to zero. That sits between slow and fast dynamos, and the definition is introduced cleanly. The previous fast dynamo construction on R^3 does not transfer to the compact setting, so this is not a small variation on known work.\n\nWhat the paper does well: the averaging reduction is transparent, the two-flow control construction is explicit, and the spectral computation behind Corollary 3.7 checks out. I verified the no-common-bad-subspace condition at kappa=0: the non-growing eigenspaces of the two matrices intersect only in span{e_z}, so every divergence-free Fourier datum has an expanding component under at least one matrix. The continuity argument to positive kappa is sound. The proof is self-contained, and the adaptive control mechanism is a legitimate existence argument, not circular. The author also honestly flags that the construction gives growth only along lacunary times and does not resolve the full fast dynamo problem.\n\nThe soft spots are real but minor. Proposition 2.3 and Eq. (4.2) drop the square on the initial L^2 norm; as printed, the lower bound is false for the paper's own initial data, though the Gronwall argument one line earlier gives the correct inequality. Lemma 3.10 has a sign/subscript slip in the matrix elements: the off-diagonal sign convention should match the V_lambda versus V_{-lambda} definition. And Proposition 2.2 states existence of some time 2n, while Theorem 1.5 needs times that can be chosen arbitrarily large; the proof actually supplies that, since the liminf estimate gives the bound for all sufficiently large n, so this is a fixable strengthening rather than a flaw.\n\nNone of this threatens the central claim. The typos are of the kind that survive a first draft but need correction before this goes to print. The citation pattern is fine: the paper engages the relevant physics and PDE literature, and the one self-citation is to a related transport-noise result, not to a load-bearing ingredient.\n\nThis paper deserves a serious referee. I would send it out, with a request that the typos be fixed and the time-of-existence point in Proposition 2.2 clarified. I would also bring it to our reading group; the construction is clever and the averaging idea is reusable.","headline":"A serious constructive PDE paper that likely delivers the first subsequentially fast dynamo on T^3; the main proof is coherent and the only problems are typos that should be fixed before publication.","tokens_in":18551,"tokens_out":1557,"would_cite":true,"duration_ms":17227,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["35Q35","76W05","37N10","35B40"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper constructs a smooth divergence-free flow on a three-torus that amplifies magnetic fields exponentially along any prescribed countable sequence of vanishing diffusivities, the first subsequentially fast dynamo on this geometry.","keywords":["kinematic dynamo","subsequentially fast dynamo","fast dynamo","induction equation","magnetohydrodynamics","Fourier mode averaging","renewing flows","three-torus"],"falsifier":"Evaluate the matrices $A_1^\\kappa$ and $A_2^\\kappa$ of Corollary 3.7 numerically from the Bessel-function entries in Lemma 3.9 and Lemma 3.10 across $\\kappa\\in[0,\\kappa_0]$; the uniform-growth claim fails exactly if a nonzero vector $v$ with $e_z\\cdot v=0$ falls in the intersection of the non-growing eigenspaces of both matrices. A simulation of the induction equation with $b_0=\\sin(x)e_z$ under either control flow would then show a unit-wavenumber mode whose Fourier mass does not grow.","tokens_in":17530,"feed_emoji":"🧲","tokens_out":14932,"duration_ms":140675,"temperature":0.7,"pith_summary":"This paper constructs a smooth, divergence-free velocity field on the three-torus whose magnetic energy grows exponentially along any prescribed sequence of vanishing diffusivities, a behaviour the author calls a subsequentially fast dynamo. The main theorem states that for any countable list of diffusivities $(\\kappa_j)\\subset[0,\\kappa_0]$, one flow $u$ makes the solution of the induction equation with initial field $b_0=\\sin(x)e_z$ satisfy $\\limsup_{t\\to\\infty}\\max_{|k|=1}\\frac{1}{t}\\log|\\widehat{b^{\\kappa_j}}(t,k)|^2\\ge\\frac{1}{4}$ for every $j$, so the dynamo rate obeys $\\gamma(u,\\kappa_j)\\ge\\frac{1}{4}$. If the construction is correct, this is the first subsequentially fast dynamo on $\\mathbb{T}^3$, intermediate between a slow dynamo and a genuinely fast dynamo: the rate is uniformly positive along the chosen diffusivities, but not for all diffusivities near zero. The paper works in the kinematic setting, choosing the velocity field by hand and asking whether the linear induction equation amplifies a fixed seed field despite diffusion.","feed_headline":"A smooth torus flow amplifies fields for vanishing diffusivities","feed_subtitle":"It works for any countable list of diffusivities, with exponential growth rate at least 1/4.","key_machinery":"The load-bearing mechanism is the averaged single-mode identity of Corollary 3.4: for any advecting flow $u$, averaging the solution operator over uniform translations $\\tau_y$ of the flow diagonalizes the Fourier dynamics, $\\int \\widehat{T^{\\tau_y u,\\kappa}_{s,t}b}(k)\\,dy=(T^{u,\\kappa}_{s,t}(k,k))\\,\\widehat{b}(k)$, so the evolution of each Fourier coefficient becomes multiplication by a $3\\times3$ matrix on $\\mathbb{C}^3$. The paper then exhibits two smooth compactly supported control flows $W_R$ and $W_{-R}$ on a time interval of length two whose single-mode matrices $A_1^\\kappa,A_2^\\kappa$ each have an eigenvalue exceeding $e$ and share no common non-expanding eigenspace for vectors perpendicular to $e_z$, the subspace forced by the divergence-free condition. Corollary 3.7 transfers this spectral property from $\\kappa=0$ to all $\\kappa\\in[0,\\kappa_0]$ by treating diffusion as a regular perturbation of the finite-time matrix. Concatenating translated copies of the selected control flow makes the averaged Fourier mass grow like $|(A_i^\\kappa)^n v|\\ge e^n$, and a measure argument passes that growth to a concrete choice of translations; two auxiliary propositions guarantee that a nonzero seed can never vanish in finite time and that any Fourier mass can be moved onto a unit-wavenumber mode.","core_discovery":"The central claim is Theorem 1.5: there is a positive $\\kappa_0$ such that, for any countable collection of diffusivities $(\\kappa_j)\\subset[0,\\kappa_0]$, one can build a smooth divergence-free flow $u$ on $\\mathbb{R}_+\\times\\mathbb{T}^3$ with the following property. With the fixed initial data $b_0=\\sin(x)e_z$, the solution $b^{\\kappa_j}$ of the induction equation $\\partial_t b-\\kappa\\Delta b+u\\cdot\\nabla b-b\\cdot\\nabla u=0$ satisfies $\\limsup_{t\\to\\infty}\\max_{|k|=1}\\frac{1}{t}\\log|\\widehat{b^{\\kappa_j}}(t,k)|^2\\ge\\frac{1}{4}$ for every $j$, and therefore $\\gamma(u,\\kappa_j)\\ge\\frac{1}{4}$. The flow satisfies uniform regularity estimates that do not depend on the sequence. Taking the diffusivity list to be $\\mathbb{Q}\\cap[0,\\kappa_0]$ yields a subsequentially fast dynamo, meaning $\\limsup_{\\kappa\\to0}\\gamma(u,\\kappa)>0$, with positive dynamo rate on a dense set of diffusivities in an interval about zero. Growth is proved only along a lacunary sequence of times for each diffusivity, so the same construction does not produce a true fast dynamo, and the paper states that a genuinely fast dynamo on $\\mathbb{T}^3$ would require a different approach.","pith_inferences":["The explicit Bessel-function matrices allow a direct numerical check of the no-common-bad-subspace condition: if some $v\\perp e_z$ were contracted by both $A_1^\\kappa$ and $A_2^\\kappa$ for a $\\kappa\\in[0,\\kappa_0]$, the uniform-growth claim would collapse, and scanning $R$ numerically could also indicate whether the rate $1/4$ is optimal.","The translation-averaging identity is a general finite-dimensional reduction for renewing flows; the same two-control strategy might produce subsequentially fast dynamos for other linear transport problems whenever a pair of controls with disjoint non-growing subspaces exists.","Because the built flow is assembled from time-compactly supported translated pieces, it is intrinsically time-dependent and non-stationary; whether an autonomous or stationary smooth flow on $\\mathbb{T}^3$ can be subsequentially fast is a question the paper does not address."],"forward_implications":["Taking the diffusivity sequence to be the rationals in $[0,\\kappa_0]$ gives a subsequentially fast dynamo: $\\limsup_{\\kappa\\to0}\\gamma(u,\\kappa)\\ge1/4>0$, and the dynamo rate is positive on a dense set of diffusivities near zero.","The same flow works for any countable, preassigned list of diffusivities; the flow depends on the list, but the uniform regularity estimates quantifying the flow do not.","The growth rate is independent of which diffusivity is being visited, so the method gives a uniform-in-$j$ lower bound of $1/4$ for every element of the chosen sequence.","Growth is only guaranteed at the end of each visit to a diffusivity, not at all large times, so the construction does not upgrade to a true fast dynamo; the paper explicitly leaves that as a different problem."],"supporting_citations":[{"why":"Supplies the renewing-flow framework and the family of Chapter 11.4 advecting flows from which the control flows $U_\\lambda,V_\\lambda,W_\\lambda$ are drawn.","marker":"[CG95]"},{"why":"Provides the fast-dynamo candidate that $W_\\lambda$ is modeled on and whose single-mode matrix computation seeds the spectral controls of Proposition 3.6.","marker":"[Ota93]"},{"why":"Supplies the double-exponential advection-diffusion lower bound that Proposition 2.3 adapts to show the magnetic field cannot vanish in finite time.","marker":"[MD18]"},{"why":"Cited for the parabolic unique-continuation principle underlying the no-vanishing guarantee that lets the construction revisit each diffusivity indefinitely.","marker":"[Poo96]"}],"fun_headline_variants":["Subsequential fast dynamo on the 3-torus","Flow on torus gives fast dynamo for chosen diffusivities","Exponential growth for a sequence of vanishing diffusivities","A dynamo between slow and fast: subsequential","Fast dynamo on a subsequence of diffusivities"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The argument rests on the computed fact that for every diffusivity in $[0,\\kappa_0]$, no nonzero divergence-free Fourier datum at unit wavenumber is left unstretched by both of the two designed translation-averaged control flows; if that computation failed, the uniform exponential growth mechanism would collapse.","fun_headline_variants_meta":{"raw":{"variants":["Subsequential fast dynamo on the 3-torus","Flow on torus gives fast dynamo for chosen diffusivities","Exponential growth for a sequence of vanishing diffusivities","A dynamo between slow and fast: subsequential","Fast dynamo on a subsequence of diffusivities"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000574,"raw_usage":{"total_tokens":2698,"prompt_tokens":919,"completion_tokens":1779,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":535,"completion_tokens_details":{"reasoning_tokens":1710}},"tokens_in":535,"tokens_out":1779,"duration_ms":12851,"temperature":1.0,"reasoning_tokens":1710,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T12:40:49.138912+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Evaluate the matrices $A_1^\\kappa$ and $A_2^\\kappa$ of Corollary 3.7 numerically from the Bessel-function entries in Lemma 3.9 and Lemma 3.10 across $\\kappa\\in[0,\\kappa_0]$; the uniform-growth claim fails exactly if a nonzero vector $v$ with $e_z\\cdot v=0$ falls in the intersection of the non-growing eigenspaces of both matrices. A simulation of the induction equation with $b_0=\\sin(x)e_z$ under either control flow would then show a unit-wavenumber mode whose Fourier mass does not grow.","supporting_citations":[],"review_version":1}