The Unbalanced Group Theorem

Overview

The classification of finite simple groups (CFSG) is a monumental achievement in the history of mathematics. A group is simple if it is nontrivial and has no normal subgroups other than itself and the identity subgroup. Every finite group can be broken into simple factors by a composition series. Indeed, the Jordan–Hölder theorem says that the resulting factors, counted with multiplicity, do not depend on the series. Classifying these factors does not tell us every way to assemble a finite group, but it tells us which building blocks are available. The CFSG catalogs every finite simple group:

The proof grew into a collective enterprise spanning much of the twentieth century. Richard Brauer made centralizers of elements of order two central to the recognition of simple groups. Walter Feit and John Thompson’s odd-order theorem, published in 1963, showed that every nonabelian finite simple group has such an element [11]. Michio Suzuki and Thompson made powerful tools for recovering a group from its subgroups, while Daniel Gorenstein gave the effort a guiding program and an organizing framework. Michael Aschbacher, Richard Lyons, Ronald Solomon, and many others supplied the structural and recognition theorems that carried that program forward. Solomon’s historical account [25] describes both the ideas and the extraordinary collaboration behind them.

What is now referred to as the first-generation proof is made up of an extensive web of papers, whose main framework was already established by the early 1980s. Gorenstein’s 1983 volume [16] records much of the relevant work. Work toward completion and verification continued. For example, the substantial remaining case of quasithin groups was resolved by Aschbacher and Stephen Smith in work published in 2004. Aschbacher’s account [3] explains the distinction between the early announcements and the work needed to secure the proof. Beginning in the early 1980s, Gorenstein, Lyons, and Solomon also undertook a second-generation project, seeking to reorganize and simplify large parts of the proof into a systematic series of volumes. These efforts reflect the scale of the achievement.

CFSG should be thought of as a problem of identification. Suppose there were a finite simple group of smallest order missing from the “known list”. All the simple groups arising as sections of its proper subgroups would already be known. One then examines those subgroups, compares the information they provide, and proves that the supposedly new group must itself be one of the known groups. To this end, group theorists have made great use of centralizers of involutions: the subgroups consisting of all elements that commute with a given element of order two. Such centralizers are proper subgroups of a nonabelian simple group, but their structure can constrain the whole group.

The Unbalanced Group Theorem (UGT) is one of the key results that makes this “local-to-global” strategy work. Roughly speaking, if even one involution centralizer has a nontrivial normal subgroup of odd order, then the ambient simple group belongs to particular familiar families.

To make this precise, we recall some definitions. Throughout, all groups in this paper are finite. Let $G$ be a group. An involution is an element of order two. The centralizer of a subset $A$ of $G$ is

We denote this centralizer by $C_G(t)$ when $A=\{t\}$. The odd core $O(G)=O_{2^{\prime}}(G)$ is the largest normal subgroup of $G$ of odd order. It exists because the product of two normal subgroups of odd order is again a normal subgroup of odd order. The subscript $2^{\prime}$ means “prime to two.”

Normality depends on the ambient group. In particular, a subgroup can be normal in $C_G(t)$ without being normal in $G$. Although a nonabelian simple group $G$ has no nontrivial normal subgroup of odd order, an involution centralizer $C_G(t)$ may have one. The unbalanced-group problem asks what this tells us about $G$.

Such a simple group is called unbalanced. Generally, a finite group $G$ is unbalanced if $O(C_G(t))\nleq O(G)$ for some involution $t\in G$. For a nonabelian simple group $G$, the Feit–Thompson theorem [11] implies $G$ has even order and hence $O(G)=1$, so the two conditions agree.

The Unbalanced Group Theorem answers this question under the inductive hypothesis used in the classification program, which we now make precise. A section of a group $X$ is a quotient $A/N$, where $A\leq X$ and $N\trianglelefteq A$. Let $\mathcal{K}$ be the catalog of known nonabelian finite simple groups. Namely, they are the alternating groups, the simple groups of Lie type, and the twenty-six sporadic groups. A $\mathcal{K}$-group is a group whose nonabelian simple sections belong to this catalog. A group is $\mathcal{K}$-proper if every proper subgroup is a $\mathcal{K}$-group. This condition is natural in a classification proof. If $G$ is a nonabelian simple group of least order outside the catalog, every nonabelian simple section of a proper subgroup has smaller order. Consequently $G$ is $\mathcal{K}$-proper.

Groups of Lie type are the finite groups constructed from algebraic groups over finite fields, together with their twisted analogs. The groups $\operatorname{PSL}_n(q)$ are examples; they are the determinant-one matrices over the finite field of $q$ elements modulo the scalar matrices. Similarly, $\operatorname{PGL}_n(q)$ is the group of all invertible matrices modulo the nonzero scalar matrices. We say that a simple group is of odd Lie type if it is isomorphic to a simple group of Lie type in odd characteristic. There are small coincidences such as $A_5\cong\operatorname{PSL}_2(5)$. We now state the simple-group case of the Unbalanced Group Theorem.

Remark 1.3. The version of UGT stated above has been formalized in Lean in the Axiom Math CFSG repository. Our project website provides a link to the formal proof.

Theorem 1.2 gives an important reduction in the classification program. Suppose that $G$ were a nonabelian simple group of smallest order outside the known catalog. Its proper subgroups satisfy the required inductive hypothesis. If $O(C_G(t)) \ne1$ for some involution $t$, the theorem would identify $G$ with a group in the catalog, a contradiction. Therefore, any such counterexample must satisfy

UGT therefore removes all unbalanced groups from the remaining classification problem.

The theorem also has an important connection with the $B$-theorem. Roughly speaking, the $B$-theorem ensures that the simple pieces visible after removing the odd core of an involution centralizer can be realized by suitable subgroups of the centralizer itself. More precisely, when $O(G) = 1$, every quasisimple component of $C_G(t)/O(C_G(t))$ is the image of a quasisimple component of $C_G(t)$. We define components in Section 2. This lifting property allows one to study actual subgroups of $G$, compare them across different centralizers, and apply recognition theorems to identify $G$.

For a smallest counterexample to CFSG, the displayed consequence of UGT makes this lifting property immediate. Namely, there is no odd core to remove. The corresponding $B$-property for its proper subgroups and their sections follows from independently established results for $K$-groups. Therefore, the simple-group form of UGT supplies the ambient reduction needed here without requiring an independent proof of the unrestricted $B$-theorem as an additional input. For the historical relationship between the two theorems, see [13 §1], [16 Chapter 3], and [5 §§1.7–1.8].

The restriction to involutions in $G$ distinguishes our statement from some other formulations of UGT.

Remark 1.4. More general versions of UGT treat involutions acting from outside a simple normal subgroup. Theorem 1.2 is the simple-group form required for the classification induction.

For alternating groups, three fixed points of an involution account for the nontrivial odd core.

Example 1.5. As noted above, the groups $A_{4a+3}$, for $a \ge1$, are unbalanced. Indeed, for $A_7$, let

The involution $t$ fixes precisely the three points 5, 6, 7. Every element of $C_{A_7}(t)$ permutes these fixed points, so conjugation preserves $U$, the alternating group on those three points. Therefore, $U$ is normal in $C_{A_7}(t)$ with order three, and consequently

Hence $A_7$ is unbalanced, even though simplicity prevents $U$ from being normal in $A_7$ itself.

The same argument works in $A_{4a+3}$. Take an involution $t$ consisting of $2a$ disjoint transpositions and fixing the remaining three points. The even number of transpositions ensures that $t \in A_{4a+3}$. The subgroup of order three generated by a 3-cycle on the fixed points is normal in $C_{A_{4a+3}}(t)$, so this centralizer has nontrivial odd core.

An obstruction to an accepted route to UGT

Our work on formalizing CFSG requires making the hypotheses and dependencies of its many theorems explicit. In this effort, we have assembled a comprehensive errata list. Most of the errors are straightforward to correct: typographical errors, omitted hypotheses, and similar local issues. The difficulty considered here requires a change to an argument.

The first-generation proof of UGT is not contained in a single paper. Gilman–Solomon’s Theorem A [13 p. 59] proves the Unbalanced Group Conjecture for a group whose proper sections satisfy it, subject to Hypothesis S [13 p. 58], a list of standard-component problems, each asking that a group with a standard component of a specified type be identified. These problems were settled in separate papers over the following years, with the $\operatorname{PSL}_3(4)$ case completed by Cheng and Held in 1985 [10 pp. 147–148]. No single reference, as far as we know, records the point at which Hypothesis S was fully settled. Gorenstein’s 1983 volume [16] develops the unbalanced-group argument; the historical overview in [17 pp. 40–41] describes the collective work on the related $B_2$-property. The pattern, a result announced modulo components completed later, is discussed in [3]. Walter’s 1986 memoir [27] gives a treatment organized around the signalizer functor method, and is recognized by Aschbacher [4 §4, p. 10] as a route to UGT and the $B$-conjecture. It is the treatment examined here.

Unexpectedly, our formalization found explicit counterexamples to a closure assertion in Walter’s proof of UGT [27], leaving a key step of that argument unjustified. We explain the obstruction and develop an alternative argument for the simple-group case using Solomon’s 1976 theory of maximal $2$-components.

We outline the proof route where the difficulty arises. The starting point is a nonabelian simple $\mathcal{K}$-proper group $G$ with an involution $t$ such that $O(C_G(t)) \ne1$. The goal is to show that $G$ is in one of the families in Theorem 1.2. The reductions either identify $G$ with one of the required groups or lead to a configuration involving two commuting involutions and a $2$-component of one of their centralizers. A $2$-component is a perfect subnormal subgroup whose quotient by its odd core is quasisimple; after further quotienting by the center, one obtains a nonabelian simple group (see § 2).

The task is to find a $2$-component whose structure and position in $G$ allow a recognition theorem to identify $G$. To do this, the proof compares $2$-components in different involution centralizers and selects one that is maximal with respect to these comparisons. Results of Thompson–Burgoyne and Aschbacher, together with the alternating-group analyses, handle many of the possibilities. The remaining cases require the work of Foote, Gilman–Solomon, Griess–Solomon, and Harris [2, 5, 9, 12, 13, 22, 23].

The difficulty arises in the comparison stage. The argument works with components whose associated simple quotients belong to a specified family. A later step requires a quotient obtained by taking an involution centralizer to remain in that family. The closure assertion used to justify this can fail. Namely, an involutory automorphism can have a centralizer with a component whose simple quotient lies outside the chosen family. Therefore, the inference placing the resulting component back in the family is unjustified.

This assertion is in John Walter’s 1986 memoir [27], in the preparation for Part II, Proposition 5.9(ii). The proposition appears on p. 145, with its proof on pp. 145–148. The passage on p. 136 following (5.2) introduces the family $\mathcal{L}_0$ consisting of the groups $\operatorname{PSL}_2(q)$ for odd prime powers $q$ other than $9$, Fermat primes, and Mersenne primes. A Fermat prime has the form $2^{2^n}+1$ for an integer $n \ge0$, and a Mersenne prime has the form $2^n-1$ for an integer $n \ge2$.

The obstruction comes from field involutions, automorphisms induced by the order-two automorphism of a quadratic extension of finite fields. For $\operatorname{PSL}_2(r^2)$, this automorphism raises each matrix entry to its $r$th power. The following proposition shows how a group in $\mathcal{L}_0$ can have a field-involution centralizer whose component lies outside $\mathcal{L}_0$.

Proof. Put $k=\mathbb{F}_r$ and $F=\mathbb{F}_{r^2}$, the fields with $r$ and $r^2$ elements. A projective matrix $[A] \in\operatorname{PGL}_2(F)$ is fixed by $\sigma$ precisely when $A^\sigma=\lambda A$ for a scalar $\lambda$. Divide $A$ by any nonzero entry $a$. Since $a^\sigma=\lambda a$, the matrix $A/a$ is fixed entrywise, and has entries in $k$. Hence $C_{\operatorname{PGL}_2(F)}(\sigma)=\operatorname{PGL}_2(k)$.

Every element of $k^\times$ is a square in $F^\times$. For example, if $g$ generates $F^\times$, then $k^\times=\langle g^{r+1}\rangle$, and $r+1$ is even. Therefore, every projective matrix over $k$ has square determinant over $F$, and the entire fixed group lies in $\operatorname{PSL}_2(F)$. This proves the centralizer assertion.

The subgroup $\operatorname{PSL}_2(r)$ is nonabelian simple, perfect, and normal of index two in $\operatorname{PGL}_2(r)$. Any odd normal subgroup of the latter maps trivially to this quotient of order two, hence lies in $\operatorname{PSL}_2(r)$ and is trivial. The derived subgroup is $\operatorname{PSL}_2(r)$, so this is its unique component and its unique 2-component. □

Theorem 1.7 shows why the proposed closure fails. Start with $S=\operatorname{PSL}_2(49)$, which belongs to $\mathcal{L}_0$. The number 49 is an odd prime power, is not 9, and is not a prime. The field involution fixes a subgroup isomorphic to $\operatorname{PGL}_2(7)$. The simple component of that subgroup is $\operatorname{PSL}_2(7)$, which does not belong to $\mathcal{L}_0$, because 7 is a Mersenne prime. Therefore, taking the component of an involution centralizer can carry us outside the chosen family. The same calculation gives the following three examples; the middle column records the simple component of the field-involution centralizer.

Here the involution is an automorphism of $S$. It can be realized as an element of the larger group $S\mathbin{\rtimes}\langle\sigma\rangle$; it is not an involution belonging to $S$ itself.

We explain where the closure assertion enters Walter’s proof. The argument compares components in different involution centralizers and needs to show that a newly encountered component still belongs to a specified family. The type of a component is the isomorphism type of its nonabelian simple quotient, after removing its odd core and center. Knowing that two components have the same type transfers membership in a family only if one already knows that this common type belongs to that family.