Is a continuous Frobenius structure or artifact?

Object. In characteristic p, the Frobenius x ↦ x^q is an endomorphism only for q a power of p — checked over 𝔽₂₅, the only q < 40 with (x+y)^q = x^q + y^q are 5, 25 and 29 ≡ 5 (mod 24). The family is discrete, and its discreteness is what makes Frobenius a time step: one generator, integer powers, closed orbits of integer length.

Over an idempotent semiring — characteristic one, where x + x = x — the map Fr_λ(x) = x^λ is an automorphism for every real λ > 0, with Fr_a ∘ Fr_b = Fr_{ab}. Tested on 20,000 random instances, no failures. The discrete family has become a one-parameter group.

Presenting problem (the friction). This dissolves an apparent type error and may be hiding a real one.

The type error it dissolves: over a finite field a Frobenius graph is a single correspondence, so intersecting it with the diagonal gives a number; over the rationals the scaling flow is a one-parameter family, so the same intersection gives a function of the parameter, which looks like a category mismatch. If a one-parameter family is simply what Frobenius is in characteristic one, there is no mismatch — the flow is the Frobenius, and the function of the parameter is the explicit formula.

The friction is that idempotency is an extremely strong hypothesis and it is doing all the work. In an idempotent semiring the binomial coefficients that obstruct (x+y)^λ = x^λ + y^λ in characteristic p are not small; they are absent, because addition is max. So the continuity of the family may be reporting a fact about the limit p → 1 of a geometry, or it may be reporting that max-plus arithmetic has no binomial theorem to obstruct, which would be true and uninformative.

Differential. 1. Genuine: characteristic one is the limiting geometry, the continuous Frobenius family is its Frobenius, and the parameter is a real analogue of the integer exponent, with the flow inheriting the role of the time step. 2. Degenerate: the automorphism property is a triviality of idempotent addition, the continuous parameter is not a deformation of anything, and the analogy with p-power Frobenius survives only at the level of notation. 3. Genuine but mis-parameterized: the correct object is not ℝ₊^× acting by exponents but some completion of the p-power family in which the discrete structure survives as a distinguished subgroup, which would predict that ℝ₊^× carries extra data currently invisible. 4. Both, at different places: the continuous family is real at the archimedean place, where there is no residue characteristic to be a power of, and degenerate at the finite places, where the discreteness is the whole content.

Discriminating tests. Ask whether the continuous family has any property that fails for a generic one-parameter automorphism group of an idempotent semiring — if not, branch 2. Look for a deformation argument connecting p-power Frobenius to Fr_λ with p as a parameter, not merely an analogy between endpoints; branch 1 requires one and branch 3 predicts it exists with a discrete subgroup surviving. Check whether the archimedean local term's distinguished status — it is the only place whose denominator requires a finite part, with the universal constant γ + log 2π — correlates with branch 4.

Stakes. Whether "the flow is already the Frobenius" is a structural identification or a re-description. A great deal is currently being hung on it.

---

First data point (2026-09-12), and it moves the case toward branch 2.

The discriminating test in the first line above — does the continuous family have any property a generic one-parameter automorphism group of an idempotent semiring lacks — was run in the cheapest possible form: test arbitrary monotone bijections φ, not just power maps, against both semiring operations.

Result, 20,000 random pairs each:

Every monotone bijection preserves the addition, exactly. The additive half of the automorphism property is a fact about max being order-preserving, and nothing else — it does not distinguish exponentiation from any other monotone map, and it is not evidence for a Frobenius structure. The "20,000 tests, no failures" figure was measuring this half.

What survives is the multiplicative half, and there Fr_λ is distinguished for a reason that is visible once stated: multiplication is addition of exponents, so preserving it means λ(a+b) = λa + λb, i.e. ℝ-linearity on the value group. The one-parameter family is the scaling action on a real vector space.

That is not nothing, but it is much less than a Frobenius. A Frobenius over a finite field is a distinguished generator with integer powers and closed orbits of integer length; scaling on a real line has no distinguished generator at all, which is precisely what makes the parameter continuous. The identification "the flow is the Frobenius" therefore rests on the flow being the scaling action, and the analogy is carried by the value group rather than by any semiring structure.

What this leaves. Branch 2 is now the default and branches 1 and 3 need a positive argument that was not being asked for before: exhibit a property of Fr_λ that a generic ℝ-linear scaling on a value group lacks, or exhibit the deformation with p as a continuous parameter. Branch 4 is untouched and is the one worth pushing on, since the archimedean place is exactly where there is no residue characteristic for an exponent to be a power of.

---

Second data point (2026-09-12): a third instance of the same one-parameter scaling, in a setting nobody calls a Frobenius.

Renormalization has since been placed as Birkhoff factorization in a loop group, a local-to-global splitting whose one-parameter subgroup is the renormalization group flow — and noted as the same shape as the scaling flow read as a characteristic-one Frobenius.

This is a third appearance of one structure: the scaling action on the value group (R11's surviving content), the scaling flow on the adele class space, and now the RG flow. The question is what a third instance is evidence for.

It cuts toward branch 2. Nobody calls the renormalization group a Frobenius, and nothing about it suggests a distinguished generator with integer powers and closed orbits. What it has in common with the other two is that it is a one-parameter group acting by rescaling, which is the generic structure R11 already isolated as all that survives on the idempotent side. Three instances of a generic structure is evidence that the structure is generic.

The identification therefore needs to name what is Frobenius-like about the flow other than being a one-parameter rescaling, since the RG has that and is not a Frobenius. The best candidate remains the one in branch 4: at a finite place the orbit lengths are log p, forced by valuations being integers and nothing being put in by hand, which the RG has no analogue of. If that is the whole content, the identification holds at the finite places for a reason that has nothing to do with characteristic one, and characteristic one contributes only the continuity of the parameter — which R11 showed carries no information on the additive side.