---
title: Local Gekeler Ratios & Ideal Class Monoids
url: https://www.emergentmind.com/topics/external-and-generated-graph-neural-network-enggnn
type: topic
---

# Local Gekeler Ratios & Ideal Class Monoids

Below is a self-contained account of the local Gekeler ratios based on the techniques of the paper “Calculating The Local Ideal Class Monoid and Gekeler Ratios.”  We work throughout with  
 A = 𝔽_q[T],  
 f(x) ∈ A[x] a monic irreducible of degree r,  
 R = A[x]/(f(x)),  
and a nonzero prime ideal 𝔭⊂A.  We explain

 1. the definition of the local Gekeler ratio  
 2. the algorithmic computation of v_𝔭(f) via local ideal class monoids  
 3. the global product ∏_𝔭 v_𝔭(f) and its role in counting Drinfeld modules  
 4. the main notation and key lemmas  
 5. a worked example  

Section 1.  Definition of the local Gekeler ratio  

Fix r and f as above and let 𝔭⊂A be a prime.  For each n≥1 we form the finite rings  
 A/𝔭ⁿ, Mat_r(A/𝔭ⁿ) = all r×r matrices over A/𝔭ⁿ,  
 SL_r(A/𝔭ⁿ) = invertible matrices of determinant 1.  

We write  
 N_n(f) =  ∣ { M∈Mat_r(A/𝔭ⁿ) ∣ charpoly(M)=f } ∣,  
and denote by |𝔭| the cardinality of the residue field A/𝔭.  Then the local Gekeler ratio is defined by  
(1) v_𝔭(f)  =  lim_{n→∞}  N_n(f) / ( |SL_r(A/𝔭ⁿ)| / |𝔭|^{n(r−1)} ).  

Explanation of each term:  
 • N_n(f) counts r×r matrices M over the finite ring A/𝔭ⁿ whose characteristic polynomial is exactly f(x) (viewed in (A/𝔭ⁿ)[x]).  
 • |SL_r(A/𝔭ⁿ)| is the size of the special linear group over A/𝔭ⁿ.  One shows |SL_r(A/𝔭ⁿ)| grows on the order of |𝔭|^{n(r²−1)}.  
 • The normalizing factor |𝔭|^{n(r−1)} accounts for the fact that monic polynomials of degree r impose r constraints but one determinant condition is redundant in SL_r.  

Convergence of the limit follows from standard “stabilization” arguments: for large n, the fibers of reduction Mod 𝔭ⁿ→Mod 𝔭^{n−1} become uniformly distributed, so the ratio stabilizes.  One of the aims of the paper is to identify this stable value with the size of a local ideal-class set.

Section 2.  Computing v_𝔭(f) via local ICM  

The key idea (pioneered in Gekeler’s work and made algorithmic in the paper) is that conjugacy classes of matrices M with charpoly f over A/𝔭ⁿ are in natural bijection, in the limit n→∞, with isomorphism classes of certain rank-r lattices in the semisimple algebra K⊗_A A_𝔭, where  
 K = Frac R = A[x]/(f).  

Concretely the ratio v_𝔭(f) equals the cardinality of the ideal-class monoid ICM(R_𝔭) of the completed local order R_𝔭 = R⊗_A A_𝔭.  We outline the steps:

 Algorithm for v_𝔭(f) :

 1. Factor 𝔭 in R.  Write  
    𝔭R = 𝔭₁^{e₁}·…·𝔭_k^{e_k}.  
    By Lemma 2.1 (spRpproduct) one checks  
      R_𝔭 ≅ ∏_{i=1}^k R_{𝔭_i},  
    each R_{𝔭_i} a complete local 𝐴_𝔭-order.  

 2. Compute all 𝔭-overorders S_𝔭 of R_𝔭.  Equivalently (Proposition 2.7, 2.8) one can compute the corresponding 𝔭-overorders S of the global order R and then localize.  Here an overorder is a ring R⊂S⊂K of finite index, and being a 𝔭-overorder means the index ideal [R:S] is a power of 𝔭.  

 3. For each local overorder S_𝔭 we compute the set of weak-equivalence classes W_{S_𝔭}(R_𝔭) of fractional R_𝔭-ideals whose multiplicator ring is S_𝔭.  Concretely one lifts to the global weak equivalence classes W_S(R) (computed by the algorithms of Stefano, see Remark 4.9) and then discards those that become trivial or coincide upon localization (Lemmas 3.2–3.6).  

 4. By Proposition 4.6 (specialized in Proposition 4.11) one shows that in the local setting every ideal class is represented by a unique weak class, so  
      ICM(R_𝔭)  =  ⨆_{S_𝔭}  W_{S_𝔭}(R_𝔭).  

 5. Finally set  
      v_𝔭(f)  =  |ICM(R_𝔭)|.  

Justification of convergence and correctness:  
• One shows (Theorem 5.1 in Gekeler, reproved in Section 5 of the paper) that the limit in (1) exists and equals the number of isomorphism classes of rank-r lattices Λ in K⊗_A A_𝔭 on which the A_𝔭-action has characteristic polynomial f.  Such Λ are exactly the R_𝔭-ideals up to principal equivalence.  
• Proposition 2.1–2.3 guarantee that Picard groups of the local overorders vanish, so the local ideal-class monoid is controlled purely by the weak equivalence classes W_{S_𝔭}(R_𝔭).  
• The steps in §2–§4 give a finite, explicit enumeration of all S_𝔭 and then of all W_{S_𝔭}(R_𝔭).  

Section 3.  The global product and Drinfeld modules  

In the theory of Drinfeld modules of rank r over finite fields one is led to count isogeny classes with fixed characteristic polynomial f.  Gekeler’s mass formula expresses the weighted size of that isogeny class as a product of local densities:  

(2) Mass_f  :=  ∑_{[ϕ] with charpoly(ϕ)=f}  1/|Aut(ϕ)|  
           =  C · ∏_{𝔭⊂A}  v_𝔭(f),  

where C is an explicit global factor (volume of the global adele group, etc.) and each v_𝔭(f) is exactly the local ratio (1).  Thus once all local ICM(R_𝔭) are known, plugging into (2) gives the full weighted count of Drinfeld modules with charpoly f.

Section 4.  Notation, conventions, and key lemmas  

Notation  

 • A = 𝔽_q[T], R = A[x]/(f(x)), K = Frac(R).  
 • For a prime 𝔭⊂A, let A_𝔭, R_𝔭 be the completions, and R_{(𝔭)} the localization.  
 • The factorization 𝔭R = ∏𝔭_i^{e_i} induces R_𝔭 ≅ ∏_i R_{𝔭_i} (Lemma 2.1).  
 • If S→T is an extension of orders, S is a 𝔭-overorder of R if the index ideal [R:S] =ord(R/S) is a power of 𝔭 (Def 2.4).  

Picard triviality  

 • For each local factor R_{𝔭_i}, Pic(R_{𝔭_i})=1 (Lemma 2.3).  Hence Pic(R_𝔭)=1 (Cor 2.4).  

Weak equivalence  

 • Two fractional R_𝔭-ideals I,J are weakly equivalent (Def 3.3) exactly when 1∈(I∶J)(J∶I) (Proposition 3.1).  
 • These classes lift from global weak classes W_S(R) by localization and intersect–test (Lemmas 3.2, 3.4).  

Ideal class monoid  

 • ICM(R_𝔭) = all fractional ideals modulo principal ones.  By Proposition 4.11 one shows  
       ICM(R_𝔭) = ⨆_{S_𝔭} W_{S_𝔭}(R_𝔭),  
   a finite disjoint union over the finitely many local overorders S_𝔭.  

Section 5.  Example  

We illustrate with a simple rank-2 example.  Let q=3,  
  A = 𝔽_3[T], f(x)=x² − T, r=2.  
We take the prime 𝔭=(T).  Then in R = 𝔽_3[T,x]/(x²−T) one has  
  𝔭R = (T, x²−T) = (𝔭₁)²,  
a single prime of ramification index 2.  Thus R_𝔭 is a local 𝐴_𝔭-order in the quadratic extension K_𝔭 = 𝐹₃((T^{1/2})).  

Step 1.  Overorders of R_𝔭  
 • The maximal order in K_𝔭 is 𝒪 = 𝐹₃[[T^{1/2}]].  Its 𝔭-index in R_𝔭 is Tℭ, so 𝒪 is the unique nontrivial 𝔭-overorder.  
 • Thus the only local overorders are S₁=R_𝔭 and S₂=𝒪.  

Step 2.  Weak classes W_{S_i}(R_𝔭)  
 • Since Pic(S_i)=1, each weak class is just the set of R_𝔭-ideals I with (I∶I)=S_i, modulo multiplication by an element of K_𝔭^×.  
 • One checks easily there is exactly one class for S₁ (namely I=R_𝔭 itself) and one class for S₂ (namely I=𝒪).  

Hence ICM(R_𝔭) has cardinality 2, and so  
  v_𝔭(f) = |ICM(R_𝔭)| = 2.  

For every other prime 𝔮≠(T), the polynomial f mod 𝔮 remains square‐free of degree 2, so R ⊗ A_𝔮 is a product of two unramified DVR’s and by Lemma 4.12 its ICM is trivial of size 1.  Consequently the global product is  
  ∏_{𝔭⊂A} v_𝔭(f) = 2·1·1·…  = 2,  
and so the weighted size of the isogeny class of Drinfeld modules with characteristic polynomial x²−T is proportional to 2.  

---

This completes the exposition.  In practice one implements Steps 1–4 above by:  
 • factoring 𝔭 in R,  
 • enumerating 𝔭-overorders via the “order-ideal” criterion (Prop 2.7–2.8),  
 • computing global weak classes W_S(R) by the algorithms of Stefano (adapted to inseparable f if needed),  
 • localizing and testing equivalences by Lemmas 3.4–3.6,  
 • forming the disjoint union of W_{S_𝔭}(R_𝔭) to get ICM(R_𝔭).  

Finally v_𝔭(f)=|ICM(R_𝔭)| and ∏_𝔭 v_𝔭(f) enters Gekeler’s mass formula for Drinfeld modules.

Source: https://www.emergentmind.com/topics/external-and-generated-graph-neural-network-enggnn