Bounded local defect leaves only finitely many block theories

← All digests

Family 203 · Algebra
Importance 5/5 · Provisional editorial rating
Sources consulted 2026-10-21

1. The question

A finite group can be enormous while a particular part of its modular representation theory has only a small local obstruction. Does bounding that obstruction leave only finitely many genuinely different representation theories?

Donovan’s conjecture makes this precise for blocks of finite group algebras. The local datum is a defect group, a finite group of prime-power order. The desired conclusion is finiteness up to Morita equivalence, meaning equivalence of entire module categories, not just equality of the number of irreducible representations.

Family 203 contains two manuscripts. M1, dated 24 September 2026, claims the conjecture over every fixed algebraically closed field of positive characteristic. M2, dated 25 September, claims an integral version over Witt vectors and then over each fixed complete discrete valuation ring of mixed characteristic with algebraically closed residue field. Both include nonabelian defect groups and the prime $2$.

2. Background and essential definitions

Blocks and their local size

Fix a prime $p$, a finite group $G$, and an algebraically closed field $K$ of characteristic $p$. The group algebra $KG$ consists of finite formal linear combinations of group elements. Its modules are representations of $G$ over $K$.

A primitive central idempotent $b$ selects an indecomposable algebra factor $KGb$, called a block. Every simple module belongs to one block. A defect group $D$ is a $p$-subgroup associated with the block, determined up to conjugacy. It measures local modular complexity. Defect zero is the simplest situation; larger defect groups allow more complicated extensions between representations.

The size of $D$ does not bound the size of $G$. That is why the conjecture is a genuine uniformity assertion: arbitrarily large ambient groups are allowed.

Equivalence of representation theories

Two $K$-algebras are $K$-linearly Morita equivalent when their module categories are equivalent by a functor respecting scalar multiplication. For example, $K$ and the matrix algebra $M_n(K)$ are Morita equivalent for every positive integer $n$. Their algebra dimensions differ, but the module theories differ only by matrix packaging.

A basic algebra removes these redundant matrix multiplicities. Over an algebraically closed field, it represents the same Morita class with one copy of each simple type and with simple modules of dimension one.

The Cartan matrix $(c_{ij})$ records how often the $i$-th simple module occurs among the composition factors of the projective cover of the $j$-th simple module. A projective cover is a minimal projective module mapping onto that simple module. For a basic algebra $A$,

$$\dim_K A=\sum_{i,j}c_{ij}.$$

Thus Cartan bounds control a concrete size associated with the representation theory, even though the original group algebra may be huge.

What the integral version remembers

A complete discrete valuation ring $R$ has a maximal ideal generated by one element, called a uniformizer, and permits convergent expansions in its powers. Mixed characteristic means that $R$ has characteristic zero while its residue field $R/\mathfrak m$ has characteristic $p$. The $p$-adic integers are the basic model, although their residue field is not algebraically closed.

Here the initial ring is $W(\overline{\mathbb F}_p)$, the Witt-vector ring of the algebraic closure of the prime field. An integral block $RGb$ retains information about modules and lattices over $R$, not merely their reductions modulo $\mathfrak m$. Consequently, integral Morita finiteness is stronger than finiteness of the reduced blocks.

3. Development history and previous partial results

The problem is attributed to Peter Donovan and appears in Alperin’s 1980 discussion of local representation theory. It builds on Brauer’s questions about bounding Cartan invariants in terms of defect groups.

Early progress often used the structure of particular group families. Scopes proved finiteness for symmetric-group blocks through combinatorial reductions. Kessar treated related double covers. Jost developed Morita reductions for general linear groups, and Hiss and Kessar extended the reduction methods to classical groups. These results already addressed substantial uniformity in rank and field size; the general theorem is not the first instance where those parameters can be unbounded.

A second line of work clarified what finiteness actually requires. Bounded Cartan matrices give bounded dimensions of basic algebras, but their multiplication constants could still vary through an infinite field. Hiss emphasized this rationality issue. Kessar separated Donovan finiteness into size control and control of Morita–Frobenius numbers, which measure how many coefficient-Frobenius iterations are needed to recover the same Morita class.

For specific defect groups, the manuscripts cite results for abelian $2$-groups, several extraspecial groups, Suzuki $2$-groups, and quaternion defect groups. Eaton, Eisele and Livesey developed an integral finiteness criterion combining bounded defect, bounded Cartan data and bounded integral Morita–Frobenius number. Their work also established integral finiteness for abelian $2$-groups.

Passing from simple-group components to arbitrary groups introduces another difficulty: extensions carry multiplication data. Külshammer’s crossed-product methods and Eisele’s integral work supplied a framework for retaining these data. Farrell and Kessar obtained uniform rationality bounds for quasisimple blocks, but such bounds alone do not determine how the blocks fit into arbitrary extensions.

The two new manuscripts claim to supply the missing uniform estimates and compatible extension comparisons, then to strengthen the comparison over the integral coefficient ring itself.

4. What the new manuscripts claim

M1: finiteness over a fixed field

Fix $p$, an algebraically closed field $K$ of characteristic $p$, and an integer $M\geq1$. M1 asserts that

$$\{KGb:\ G\text{ finite},\ |D_b|\leq M\} \big/\text{Morita equivalence over }K$$

is a finite set. Here $D_b$ denotes a defect group of the block. Neither the order of $G$ nor the dimensions of its simple representations are bounded in advance.

Fixing a particular defect group and bounding defect-group order give equivalent finiteness formulations, because only finitely many groups have order at most $M$. The field is fixed before the blocks vary.

The resulting finite list also bounds Cartan entries, Loewy lengths of basic algebras, and extension-space dimensions between simple modules in each fixed degree. The Loewy length counts how many successive powers of the radical are required to reach zero.

M2: finiteness over fixed integral coefficients

M2 first asserts the analogous finiteness of $\mathcal O$-linear Morita classes for

$$\mathcal O=W(\overline{\mathbb F}_p).$$

It then extends the conclusion to each fixed complete discrete valuation ring $R$ of characteristic zero with algebraically closed residue field of characteristic $p$. Ramified rings are included: the uniformizer need not be $p$. No assumption that the fraction field splits all the finite groups is imposed.

M2 depends on M1’s field theorem for Cartan bounds and on its specified integral comparison constructions. M1 is independent of the integral conclusion. This order matters: reduction of M2 later recovers a field statement, but that recovery is not the premise used to prove M1.

Both papers also use coefficient and block-comparison material from the manuscript in family 202; this should not be confused with deriving Donovan finiteness from the numerical Alperin weight equality alone.

5. Why the advance matters

Importance: 5/5, provisional. A general Donovan theorem would establish a central finiteness principle for modular representation theory: a bounded local defect permits only finitely many module-category types, even as the ambient finite group grows without bound.

This is stronger than bounding a list of numerical invariants separately. A finite Morita list controls all categorical information preserved by those equivalences, including systems of extension groups. The integral version retains additional arithmetic information that disappears after passing to a field.

The proposed proof also separates two common sources of uncontrolled variation: the size of a mathematical object and the coefficients needed to define it. This distinction has a broader lesson. A family of small multiplication tables can still be infinite if their entries wander through an infinite field.

The rating concerns the scope and consequences of the claimed finiteness, not an explicit classification. Neither manuscript’s headline theorem supplies a short catalogue for every prescribed defect group.

6. The main idea

Bound both size and arithmetic variation

Over $\overline{\mathbb F}_p$, every individual finite multiplication table is defined over some finite field. The difficulty is bounding that field uniformly. If both the dimension and the finite field of definition are bounded, there are only finitely many possible tables.

M1 first attacks Cartan bounds for quasisimple groups, the building blocks closely related to finite simple groups. In groups of Lie type, it constructs projective characters using geometric objects on spaces of pairs of flags. A flag is a nested sequence of vector subspaces. The estimates count projective multiplicities rather than raw vector-space dimensions, so large finite tori do not automatically destroy the bounds.

Rank requires separate treatment. Combinatorial moves on character labels transport cones of projective characters through classical groups. A cone here consists of nonnegative linear combinations of the relevant projective characters. Most moves preserve the needed data exactly after normalization; controlled errors in the remaining moves are summable. Determinant and volume estimates then bound the columns encoding projective multiplicities.

These moves need not themselves be Morita equivalences. Their immediate purpose is to transport a numerical estimate. Special endpoint configurations are handled by modular Harish–Chandra induction, with bounds depending on a weight parameter rather than the potentially unbounded rank.

Keep the multiplication in group extensions

Suppose a larger algebra is assembled from a base algebra $B$ and a finite group $Q$. Choose symbols $u_q$ for the components. A crossed product has rules such as

$$u_q b=\alpha_q(b)u_q,\qquad u_q u_t=a_{q,t}u_{qt}.$$

The maps $\alpha_q$ describe actions, while the units $a_{q,t}$ record corrections in multiplication. Knowing only the outer action is insufficient: different correction factors may produce different algebras.

M1 constructs Frobenius comparisons that retain these factors and the actions of Sylow $p$-subgroups. Combined with structural reductions, integral base finiteness and descent, this yields a finite list over the algebraic closure of the prime field. Scalar extension then gives the statement for every fixed algebraically closed $K$.

Lift exact control to the Witt ring

The integral problem cannot be solved by counting fixed points over a finite field. The Witt ring $W(\mathbb F_q)$ is infinite. Furthermore, errors that vanish after reduction can survive as principal units, units congruent to $1$ modulo the maximal ideal.

M2 refines the comparison so that its particular scalar discrepancies lie in the Teichmüller subgroup, the distinguished multiplicative lifts of nonzero residue-field elements. This restriction permits a cohomological normalization for $p$-group actions; arbitrary principal-unit errors would not permit the same argument.

A rigidity argument then controls the finitely many possible labelled gradings of a fixed total order, retaining the chosen identification of its identity component. Finally, the construction removes a potentially large normal subgroup of order prime to $p$, while preserving these controlled restrictions. An existing integral finiteness criterion completes the finite list. Base change of the Morita bimodules transfers it to a fixed ring $R$.

7. Scope, limitations, and open questions

The conclusions include arbitrary finite groups, arbitrary defect groups, nonprincipal blocks and all primes. The coefficient field or ring is fixed in each finiteness statement. A single classification simultaneously ranging over all coefficient rings is not asserted.

Morita equivalence does not mean algebra isomorphism. The matrix-algebra example explains why infinitely many algebra presentations can represent one categorical type. Likewise, the theorem does not say that only finitely many ambient finite groups realize a given defect group.

The bounds and finite lists should not be read as an efficient algorithm for finding the block associated with an arbitrary input group. Explicit catalogues, sharp bounds and stronger kinds of equivalence remain separate questions.

The integral argument depends on precise compatibility of actions and multiplication factors. Describing it merely as “lift the field theorem” would omit the principal new obstacle addressed by M2.

8. Sources and further reading

Scopes, Cartan matrices and Morita equivalence for blocks of the symmetric groups, Journal of Algebra 142 (1991), 441–455, is an early finiteness result. Kessar, A remark on Donovan’s conjecture, Archiv der Mathematik 82 (2004), 391–394, explains the size-versus-rationality separation.

Eaton, Eisele and Livesey, Donovan’s conjecture, blocks with abelian defect groups and discrete valuation rings, Mathematische Zeitschrift 295 (2020), 249–264, supplies the integral finiteness criterion used by the new work. Farrell and Kessar, Rationality of blocks of quasi-simple finite groups, Representation Theory 23 (2019), 325–349, provides the quasisimple rationality input.

Source manuscripts

• M1. OpenAI, Donovan’s Conjecture over Algebraically Closed Fields (September 24, 2026).
• M2. OpenAI, Integral Donovan Finiteness over Witt Vectors (September 25, 2026).

Sources consulted on 21 October 2026 at repository revision 1581d4c. Download the manuscript citations.