gpt 6 astra ultra attempt on Math-431

Colmezs Conjecture on Faltings Heights of CM Abelian Varieties

Definitions and mathematical statement

A CM field $E$ is a totally imaginary quadratic extension of a totally real field; write $[E:{\mathbb{Q}}]=2g$. A CM type $\Phi$ chooses one embedding from each complex-conjugate pair. For an abelian variety with full-ring complex multiplication of type $(\mathcal O_E,\Phi)$, use the stable Faltings height normalized as in Howard, Sections 3.2--3.3, [E1]: the normalized arithmetic degree of the line of invariant top differential forms, with the archimedean norm given there.

For $G_{{\mathbb{Q}}}=\operatorname{Gal}(\overline{{\mathbb{Q}}}/{\mathbb{Q}})$ define $A_\Phi(\sigma)=|\Phi\cap\sigma\Phi|$, and average this function over the finite $G_{{\mathbb{Q}}}$-orbit of $\Phi$, obtaining a class function $A_\Phi^0$. Expand it in irreducible Artin characters: \[ A_\Phi^0=\sum_\chi m_\Phi(\chi)\chi. \] With $L(s,\chi)$ excluding archimedean factors and $f_\chi$ its Artin conductor, the conjecture is \[ h^{\rm Falt}(E,\Phi)= -\sum_\chi m_\Phi(\chi) \left(\frac{L'(0,\chi)}{L(0,\chi)}+\frac12\log f_\chi\right). \] This is an identity for each CM type, not merely an average. The full-ring hypothesis and height normalization matter: heights for nonmaximal endomorphism orders require correction terms and are not silently included in this formula.

Formulation and concept references: [S1], [E1].

Short English statement

For each complex multiplication type, does the Faltings height equal the precise expression built from Artin L-functions and their conductors?

Research attempt

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

The averaged Colmez theorem is a natural starting point, but the target is an identity for every type. I examine exactly when symmetry converts the averaged identity into the individual one, compute a small case where that inference is insufficient, and check the normalization in dimension one. The calculations concern the stated full-ring stable height; no correction for nonmaximal CM orders is suppressed.

Residuals and the information lost by averaging.

Let $C(E,\Phi)$ denote the right side of the conjectured formula and put \[ \Delta(\Phi)=h^{\rm Falt}(E,\Phi)-C(E,\Phi). \] The averaged theorem, in the normalization of [E1], says \[ \sum_{\Phi}\Delta(\Phi)=0, \tag{1} \] where the sum runs over all $2^g$ CM types. This invokes the established averaged theorem; it is not a new proof of it. Both terms in $\Delta$ are invariant under the Galois action on types. On the analytic side, the class-averaged functions agree on an orbit. On the geometric side, the stable Faltings height is invariant under conjugating the CM abelian variety, as used in the CM height formalism of [E1].

Suppose there are orbits $O_1,\ldots,O_r$, of sizes $w_1,\ldots,w_r$, and let $\delta_i$ be the residual on $O_i$. Then all that $1$ says is \[ w_1\delta_1+\cdots+w_r\delta_r=0. \tag{2} \] If the action is transitive, $r=1$ and $\delta_1=0$, giving the individual conjecture for that field. If $r>1$, equation $2$ alone leaves an $(r-1)$-dimensional space of possible real residual vectors. In particular, neither the existence of a total average nor Galois invariance gives each individual equality without more information.

Two quartic orbit calculations.

For a cyclic quartic CM field, label embeddings by $\mathbb Z/4$ with complex conjugation translation by $2$. A CM type chooses one point from each pair $\{0,2\}$ and $\{1,3\}$. The four choices are \[ \{0,1\},\quad\{1,2\},\quad\{2,3\},\quad\{3,0\}. \] Translation acts transitively, so $2$ proves the individual equality from the averaged theorem in this case. The argument depends on the action on types, not merely on the degree of the field.

For a biquadratic CM field, write its Galois group as $\{0,a,b,c\}$ with $c=a+b$ the complex conjugation. The conjugate pairs are $\{0,c\}$ and $\{a,b\}$. There are again four types, but now their orbits are \[ \{\{0,a\},\{b,c\}\},\qquad \{\{0,b\},\{a,c\}\}. \tag{3} \] Thus averaging yields $2\delta_1+2\delta_2=0$, compatible algebraically with any residuals $(u,-u)$. This does not suggest that the known abelian cases fail: those have additional input. It proves only that the single averaging equation and orbit symmetry are not by themselves a proof even in this small example. The distinction is important when trying to export an averaged theorem to arbitrary CM fields.

An exact character restriction.

Work in a finite Galois CM extension containing the normal closure of $E$, and let $c$ denote central complex conjugation. For any type, $c\Phi$ is its complement among the $2g$ embeddings. Therefore \[ A_\Phi(\sigma)+A_\Phi(c\sigma)=g. \tag{4} \] The same identity holds after class averaging. Taking the inner product with the trivial character gives coefficient $g/2$. If $\chi$ is a nontrivial irreducible character on which $c$ acts as $+1$, substitution $\sigma\mapsto c\sigma$ in the character inner product shows that the coefficient of $\chi$ in $A_\Phi^0$ is zero: the two contributions add to the inner product of the constant function $g$ with $\chi$. Since $c$ is central of order two, Schur's lemma makes its action on every irreducible representation scalar $+1$ or $-1$. Consequently \[ A_\Phi^0=\frac g2\,\mathbf1+ \sum_{\chi(c)=-\chi(1)}m_\Phi(\chi)\chi. \tag{5} \] Only odd characters remain beyond the trivial contribution. This checks which analytic terms can appear, but it does not calculate the height.

There is also positivity on the representation side. If $v$ is the indicator vector of $\Phi$ in the permutation representation on embeddings, then $A_\Phi(\sigma)=\langle v,\sigma v\rangle$ with the usual real-valued interpretation. Decomposing into irreducibles and averaging conjugates gives nonnegative character coefficients: each coefficient is a squared isotypic norm divided by the irreducible dimension. Such positivity does not say that $\Delta(\Phi)$ is nonnegative. The logarithmic derivatives can have either relevant sign, and the height-minus-expression residual has no positivity theorem provided here. Thus one cannot turn $1$ into individual vanishing by an unjustified "sum of nonnegative errors" argument.

The elliptic normalization check.

When $g=1$, there are two types and they form one Galois orbit. For the imaginary quadratic character $\chi_D$, \[ A_\Phi^0=(\mathbf1+\chi_D)/2. \] The trivial character has conductor one and its finite $L$-function is $\zeta(s)$, with $\zeta'(0)/\zeta(0)=\log(2\pi)$. Since $f_{\chi_D}=|D|$, the formula in the question specializes to \[ h^{\rm Falt}(E,\Phi)= -\frac12\frac{L'(0,\chi_D)}{L(0,\chi_D)} -\frac14\log|D|-\frac12\log(2\pi). \tag{6} \] This agrees with the degree-one specialization of the averaged formula in [E1]. Omitting the trivial character would lose the final constant; inserting completed $L$-functions without compensating archimedean terms would change it. Equation $6$ is used as a normalization test, not as a numerical measurement of a CM elliptic curve's height.

Source scope, verification, and first gap.

The 2026 Maillot--Rossler manuscript [S1, E2] proposes a refinement relating the conjecture to reciprocity laws. Its inspected abstract is not an announcement of a proof of the full individual identity. The known averaged theorem is the input explicitly invoked here. Exact enumeration verifies the two quartic orbit partitions, $4$, and the trivial-character coefficient. It computes neither Faltings heights nor Artin logarithmic derivatives in higher degree.

The strongest conclusion of this approach is individual equality when the action on all types is transitive, together with the character and normalization checks. In general, the first gap is to eliminate the nonzero orbit residual vectors allowed by $2$, using information beyond the single averaged relation. No such additional relation or positivity theorem has been proved here. The full individual Colmez conjecture remains UNRESOLVED.

Sources

• S1 Maillot and Rossler, The Conjecture of Colmez and Reciprocity Laws for Modular Forms, abstract. https://arxiv.org/abs/2603.28536. Formulation source.
• E1 Benjamin Howard, On the averaged Colmez conjecture, arXiv:1811.00428v2 $2018$, Sections 3.2–3.3, Conjecture 3.3.1. https://arxiv.org/html/1811.00428v2. Added primary source: Definitions and normalization of the Faltings height, the CM-type class function, and the individual Colmez identity. Consulted 24 September 2026.
• E2 Vincent Maillot and Damian Roessler, The conjecture of Colmez and reciprocity laws for modular forms, arXiv:2603.28536. https://arxiv.org/abs/2603.28536. Recent-paper scope check: the inspected abstract proposes a refinement, rather than asserting a proof of all individual cases. Consulted 27 September 2026.
• E3 Benjamin Howard, On the averaged Colmez conjecture, arXiv:1811.00428v2, Theorem A and Section 3.3. https://arxiv.org/html/1811.00428v2. Additional use of inherited [E1]: the averaged theorem, its normalization, and its distinction from the individual conjecture. Consulted 27 September 2026.