gpt 6 astra ultra attempt on Math-432

Greenberg's Iwasawa-invariants conjecture

Definitions and mathematical statement

Let $F$ be a totally real number field and $p$ a prime. Its cyclotomic ${\mathbb{Z}}_p$-extension is the infinite extension $F_\infty/F$ with Galois group ${\mathbb{Z}}_p$ obtained from the cyclotomic tower. Let $F_n$ be its degree-$p^n$ layer and $A_n$ the $p$-primary subgroup of the ideal class group of $F_n$. The Iwasawa invariants $\mu,\lambda$ are the nonnegative integers in the eventual formula \[ \log_p|A_n|=\mu p^n+\lambda n+\nu\qquad(n\gg0), \] where $\nu$ is an integer. Greenberg's selected conjecture is $\mu=\lambda=0$ for every such $F,p$, equivalently $\sup_n|A_n|<\infty$. It asserts more than the vanishing of $\mu$ alone, and the total-reality hypothesis is essential to this formulation.

Formulation and concept references: [S1].

Short English statement

In a totally real field's cyclotomic tower, do the p-parts of the ideal class groups remain bounded in size?

Research attempt

GPT-6 Astra Ultra; 27 September 2026. Status: UNRESOLVED.

This attempt investigates what finite class-group data would actually need to certify. The target is bounded order, which requires both $\mu=0$ and $\lambda=0$. I derive exact tests inside the eventual range and then construct algebraic models separating the two growth mechanisms. The models are not asserted to arise from totally real number fields; their role is to expose what an arithmetic proof still has to exclude.

Finite differences after the onset of the formula.

Write $e_n=\log_p|A_n|$. If the Iwasawa formula holds at $n,n+1,n+2$, then, with $\Delta e_n=e_{n+1}-e_n$, \[ \Delta e_n=\mu(p-1)p^n+\lambda, \] and hence \[ \mu=\frac{\Delta e_{n+1}-\Delta e_n}{(p-1)^2p^n}, \qquad \lambda=\frac{p\Delta e_n-\Delta e_{n+1}}{p-1}. \tag{1} \] These are exact integer identities. They provide consistency checks for proposed eventual data, including divisibility and nonnegativity conditions. In particular, a single flat step $e_{n+1}=e_n$ in a range where the formula is known to hold forces $\mu=\lambda=0$, because both summands in $\mu(p-1)p^n+\lambda$ are nonnegative. It then forces all later orders to be constant.

The qualification about the range is indispensable. Given any cutoff $N$, consider abstract finite groups \[ B_n=\mathbb Z/p^{\max(0,n-N)}\mathbb Z \] with the natural reduction maps $B_{n+1}\to B_n$. Their orders are constant through level $N$, every transition map is surjective, and their inverse limit is $\mathbb Z_p$. For $n\ge N$ their exponents equal $n-N$, so the eventual invariants of their size sequence are $\mu=0,\lambda=1,\nu=-N$. Arbitrarily long initial stabilization and surjective transitions therefore do not imply eventual boundedness for a general inverse system. This is a countertest to that inference, not a construction of class groups of a number field.

Three explicit Iwasawa-algebra models.

Let $\Lambda=\mathbb Z_p[[T]]$ and $\omega_n=(1+T)^{p^n}-1$. For a $\Lambda$-module $M$, the quotient $M/\omega_n M$ is its coinvariant module for the corresponding open subgroup of the procyclic group. The following computations isolate different possible sources of growth without invoking a classification of arithmetic Iwasawa modules.

First take $M_\mu=\Lambda/(p)$. In characteristic $p$, repeated Frobenius gives $(1+T)^{p^n}-1=T^{p^n}$. Thus \[ M_\mu/\omega_nM_\mu\cong \mathbb F_p[[T]]/(T^{p^n}),\qquad \log_p|M_\mu/\omega_nM_\mu|=p^n. \tag{2} \] This is the pure $\mu=1$ growth pattern. The module is torsion over $\Lambda$, but torsion over this two-dimensional ring does not mean finite as an abelian group.

For odd $p$, take $M_\lambda=\Lambda/(T-p)\cong\mathbb Z_p$. Evaluation at $T=p$ is well-defined because the power series converges $p$-adically. Its coinvariants have order \[ p^{v_p((1+p)^{p^n}-1)}=p^{n+1}. \tag{3} \] For completeness, if $v_p(u)\ge1$ and $p$ is odd, the binomial expansion of $(1+u)^p-1$ has first term $pu$ of valuation $v_p(u)+1$, while every other term has strictly larger valuation. Induction starting at $u=p$ proves $3$. For $p=2$, use $\Lambda/(T-4)$ instead. If $v_2(u)\ge2$, then $2u+u^2$ has valuation $v_2(u)+1$, giving order $2^{n+2}$. These models have $\mu=0$ but $\lambda=1$; proving only the first invariant vanishes leaves a real logical possibility of unbounded growth.

Finally let $M_0=\mathbb Z/p^a\mathbb Z$ with $T$ acting by zero. Then $\omega_n$ acts by zero for every $n$, so all coinvariants have order $p^a$. This is the bounded pattern. Direct sums multiply the orders, hence add the exponents in $2$--$3$. In particular the formal sequence $p^n+n+1+a$ occurs from an explicit torsion module and has both positive invariants. Formula $1$ recovers them exactly.

What class-group ranks can and cannot show.

The groups in $3$ are cyclic: their $p$-ranks equal one at every level while their orders tend to infinity. Consequently bounded $\mathbb F_p$-dimension of $A_n/pA_n$ is insufficient for Greenberg's claim. Likewise identifying an inverse limit as a finitely generated $\mathbb Z_p$-module is not enough; a nonzero free $\mathbb Z_p$ summand is infinite and is compatible with the second model. An argument using generators needs an independent uniform bound on their orders, not merely a uniform bound on their number.

It would also be unjustified to identify $A_n$ with coinvariants of an inverse-limit module without a control theorem. Norm maps, ramified primes, and the discrepancy between a finite layer and its image in a limit must be accounted for. The computations above are deliberately stated for actual module quotients, so that no unproved arithmetic control assertion is hidden in them.

A concrete certificate that would suffice.

For a given pair $(F,p)$, one possible proof route would establish an explicit index $n_0$ from which the Iwasawa formula applies and verify one equality $|A_{n_0+1}|=|A_{n_0}|$. Equation $1$ then finishes that input. A different route could prove a norm-control statement with uniformly bounded kernels and cokernels together with finiteness of the relevant inverse limit. Neither route is supplied by a plateau alone. The first gap here is a general arithmetic stabilization or control theorem for every totally real $F$ and every prime $p$.

Source check, verification, and outcome.

The recent paper [S1] treats specified real biquadratic families and conditions, particularly at the prime two; its stated family results do not remove the universal field and prime quantifiers. Exact finite checks verify the valuations in $3$, the characteristic-$p$ binomial identity in $2$, and the recovery formulas $1$ on the explicit models. No actual number-field class groups were computed in this attempt. The outcome is a rigorous diagnostic for growth and stabilization, including a demonstration that bounded rank and $\mu=0$ do not by themselves answer the question. The stated Greenberg conjecture remains UNRESOLVED.

Sources

• S1 Mohamed Mahmoud Chems-Eddin et al., "Greenberg's conjecture and Iwasawa module of Real biquadratic fields I," arXiv:2503.15727 $2025$; companion published version in J. Number Theory $2025$. https://arxiv.org/html/2503.15727v2. Formulation source.
• E1 Mohamed Mahmoud Chems-Eddin et al., Greenberg's conjecture and Iwasawa module of Real biquadratic fields I, arXiv:2503.15727v2, Theorem 1.4 and its field and prime hypotheses. https://arxiv.org/html/2503.15727v2. Version-specific scope check of [S1]. Abstract module examples here are not computations of the paper's number fields. Consulted 27 September 2026.