An AI-discovered smooth random fast dynamo on $\mathbb{T}^3$

Abstract: We construct a random, time-dependent divergence-free velocity field on $\mathbb{T}<sup>3$---refreshing iid on finite time blocks and obeying deterministic $C<sup>\infty_{t,x}$ bounds---that exhibits fast dynamo behavior. That is, for every fixed, sufficiently small resistivity, the almost sure exponential growth rate of the magnetic field solving the associated linear resistive induction equation is at least $1/2$; the exceptional null set may depend on the resistivity. We in fact get a time-uniform lower bound---with a random prefactor obeying a uniform-in-$κ$ inverse moment bound. The argument relies on a particular algebraic structure in Fourier space of the induction equation solution operator that allows us to propagate expected growth of the logarithmic size of three specially chosen Fourier modes. This allows us to reduce to a simple recursion, avoiding the complicated infinite-dimensional dynamics typical to the dynamo problem. The central proof idea was generated autonomously by ChatGPT 5.6 Sol Ultra; the manuscript was written (and verified) by the author.

• The paper constructs the first smooth random fast dynamo on T³, using three randomly translated sine shears to produce almost sure magnetic-energy growth of at least e^{t/2}.
• Its proof combines an exact tridiagonal Fourier-mode structure with Jensen’s formula and a martingale estimate, yielding a time-uniform lower bound with controlled random fluctuations for resistivities up to (4π²)⁻¹ log(π/e).
• The result closes the smooth random case while leaving deterministic smooth fast dynamos, broader non-degeneracy criteria, and the optimal growth rate as important open problems.

Overview

This paper constructs a random, time-dependent, divergence-free velocity field $U$ on the three-dimensional torus $\mathbb{T}^3$ that acts as a fast dynamo: for every fixed resistivity $\kappa$ in a positive interval extending down to zero, the solution of the linear resistive induction equation exhibits almost sure exponential growth at rate at least $1/2$ (2608.20105). The velocity field is refreshingly simple—it cycles through three sine shears on unit time blocks with iid random spatial translations—and obeys deterministic $C^\infty_{t,x}$ bounds. The result is notable both as the first smooth random fast dynamo and because the central proof idea was generated autonomously by ChatGPT 5.6 Sol Ultra, with the author subsequently rewriting and verifying the argument.

Background and context

The kinematic dynamo problem asks for divergence-free velocity fields $u$ under which solutions of

$$\partial_t b^\kappa - \kappa \Delta b^\kappa + u \cdot \nabla b^\kappa - b^\kappa \cdot \nabla u = 0$$

grow exponentially in $L^2$, with growth rate uniform in $\kappa \in [0,\kappa_0]$. Rigorous fast dynamos are listed among Arnold's problems. Prior constructions on flat spaces—on $\mathbb{R}^3$, on $\mathbb{T}^3$0 in limsup form, and fully on $\mathbb{T}^3$1—all operate at Lipschitz spatial regularity, which is the critical regularity for the problem (2608.20105). Smooth examples were previously limited to slow dynamos (growth rate vanishing as $\mathbb{T}^3$2) or subsequential fast dynamos, where the rate stays bounded away from zero only along a sequence of resistivities. The present work supplies the missing smooth case, albeit in a random rather than deterministic formulation.

The paper also situates itself relative to the theory of uniform-in-diffusivity exponential mixing, which shares structural similarities; there, analogous "almost sure for each diffusivity" statements arise, and the author conjectures that a general theorem—any sufficiently non-degenerate random velocity field is a fast dynamo—may be attainable, though this paper does not provide one.

The construction and main theorem

The velocity field is built from three shears

$\mathbb{T}^3$3

activated cyclically on unit time blocks via a smooth bump $\mathbb{T}^3$4 supported on $\mathbb{T}^3$5, each block shifted by an iid uniform random translation $\mathbb{T}^3$6. The field is smooth in $\mathbb{T}^3$7 with pathwise derivative bounds independent of the noise realization.

Main result. For all $\mathbb{T}^3$8, there exists an almost surely finite random constant $\mathbb{T}^3$9 such that for all $\kappa$0,

$\kappa$1

and moreover $\kappa$2 uniformly over the full resistivity interval.

Two features deserve emphasis. First, the bound is time-uniform: it holds for all $\kappa$3, not merely along a subsequence or in limsup, with fluctuations controlled by a prefactor having a uniform inverse moment bound. Second, the exceptional null set may depend on $\kappa$4. Consequently, the proof does not yield a single deterministic velocity field that is a fast dynamo for all $\kappa$5 simultaneously—that would require intersecting uncountably many full-probability events. Fubini's theorem does give asymptotic exponential growth for almost every $\kappa$6 in the interval, but the author is explicit that a genuinely deterministic smooth fast dynamo remains open.

Proof mechanism

The argument exploits two algebraic facts, combined to reduce the infinite-dimensional induction dynamics to a scalar recursion.

Tridiagonal Fourier structure. Because the velocity field is a translated shear, the unit-time solution operator has Fourier coefficients of the form $\kappa$7. For the specific shear $\kappa$8, a direct computation shows that a single pulse maps one distinguished Fourier mode to exactly three modes—a tridiagonal structure on selected rows. Concretely, the mode combination $\kappa$9 after one pulse equals

$1/2$0

where $1/2$1 is the random translation difference and the coefficients satisfy $1/2$2 precisely when $1/2$3—this constraint fixes the admissible resistivity range.

Jensen's formula. A complex-analytic lemma (Jensen's formula, not Jensen's inequality) gives that for $1/2$4 with $1/2$5 uniform,

$1/2$6

with an exponential moment bound on the centered fluctuation for all $1/2$7. Crucially, the lower bound involves only the extremal coefficients; intermediate coefficients would incur a degree-dependent loss, which is why the tridiagonal structure is essential rather than incidental.

Cycling through the three coordinate-permuted shears yields a recursion $1/2$8 for $1/2$9 along a carefully chosen sequence of modes. A martingale argument then shows the cumulative noise fluctuations are asymptotically dominated by the drift: there exists a random variable $C^\infty_{t,x}$0 with $C^\infty_{t,x}$1 such that $C^\infty_{t,x}$2 for all $C^\infty_{t,x}$3. Since $C^\infty_{t,x}$4 by Plancherel, and Grönwall interpolation handles times between integers, the all-time lower bound follows.

The reduction to a scalar recursion is what makes the proof tractable; it deliberately avoids the infinite-dimensional dynamics that typically complicate dynamo analysis.

AI involvement

The paper documents its provenance in detail. The core idea—the tridiagonal structure plus Jensen's formula combination—was produced autonomously by ChatGPT 5.6 Sol Ultra under a multi-agent search prompt adapted from prior work. The original AI-generated manuscript contained minor errors, phrased results via top Lyapunov exponents using a backward-in-time adjoint argument, and was judged inadequate in exposition. Through iterative revision—including the author's suggestion of the forward-in-time argument and the all-time bound, and ChatGPT's drafting of the exponential martingale lemma—the author ultimately rewrote the manuscript from scratch, simplifying the velocity fields to directly related coordinate permutations and streamlining the iteration. The final text was hand-verified, with ChatGPT used for explanation, copy-editing, and checking.

The author draws a pointed methodological conclusion: frontier models are effective at example-finding but produce proofs that are unclear and unrefined without substantial human prodding—once a viable strategy emerges, the models do not attempt to simplify or isolate the essential ingredients.

Limitations and open questions

Several caveats are stated plainly in the paper:

• No deterministic fast dynamo. The null set in the almost sure statement depends on $C^\infty_{t,x}$5, so the construction does not produce a single smooth deterministic fast dynamo on $C^\infty_{t,x}$6—arguably the sharpest remaining version of Arnold's problem.
• Resistivity interval not all of $C^\infty_{t,x}$7 for arbitrary $C^\infty_{t,x}$8. Growth is established only up to $C^\infty_{t,x}$9, a small fixed threshold dictated by the requirement $u$0.
• Bespoke structure. The proof relies on an exact tridiagonal algebraic coincidence specific to these shears; whether a general class of random fields satisfying a non-degeneracy condition constitutes fast dynamos remains open.
• Rate optimality. The lower bound of $u$1 is not shown to be sharp, and no upper bound on the growth rate is provided.

Conclusion

The paper establishes that smooth random velocity fields on $u$2 can exhibit fast dynamo behavior with a quantitative, time-uniform, moment-controlled lower bound on magnetic energy growth, closing the gap between Lipschitz and smooth regularity in the random setting. Its proof technique—propagating expected log-growth of selected Fourier modes through a tridiagonal recursion disciplined by Jensen's formula—offers a template that may extend beyond this specific construction. Equally significant is its documented demonstration that autonomous AI systems can supply the pivotal mathematical idea for a publishable result, while underscoring that human refinement remains indispensable for producing rigorous, readable mathematics.