Formalizing the proof of an intermediate-level

algebra theorem — An experiment

Antoine Chambert-Loir
Université Paris Cité, Institut de mathématiques de Jussieu — Paris Rive Gauche, 8 place
Aurélie Nemours, 75013 Paris
E-mail: antoine.chambert-loir@u-paris.fr
arXiv:2303.12404v1 [math.GR] 22 Mar 2023

Abstract. — Proof assistants are computer softwares that allow us to write mathematical
proofs so as to assess their correctness. In November 2021, I started the project of checking
the simplicity of the alternating groups within the Lean theorem prover and its mathlib
library. This text aims at reviewing this experiment.
Résumé. — Les assistants de preuves sont des logiciels qui permettent de rédiger des dé-
monstrations mathématiques et d’en garantir leur correction. En novembre 2021, j’ai débuté
un projet de vérification de la simplicité des groupes alternés au sein de l’assistant de preuve
Lean, et de sa librairie mathlib. Ce texte est un essai de compte rendu de cette expérience.

1. Introduction
Human mathematics is written in plain language, and we all know examples of
shortcomings that lead to “proofs” of wrong results. We also know for now more
than hundred years ago, notably by the works of P EANO (1889) or W HITEHEAD &
R USSELL (1927), that mathematics can be written using axiomatic systems, and,
at least in principle, in a rigid syntactic way, so as to avoid such problems, at least
if the chosen axiomatic system does not lead to contradiction. I write “in principle”,
because this rigid syntactic writing is extremly verbose: It took hundreds of pages
to Whitehead and Russell to prove that 1 + 1 = 2. One may find a pleasant, large
audience, account of this quest in the comic book D OXIADIS & PAPADIMITRIOU
(2009).
Since the 1950s, the development of computers led mathematicians to propose to
use their mechanical force to develop fully formalized proofs of the mathematical
corpus. Among such examples, let us mention N. G. De Bruijn’s Automath (1967),
A. Trybulec’s Mizar (1973), G. Huet’s team project Coq (1989), C. Coquand’s Agda
(1999) or L. de Moura’s Lean (2013). . .
In recent years, these softwares have allowed us to check delicate parts of the
mathematical corpus: Appel and Haken’s proof the Four color theorem (the regions
delimited by a finite planar graph can be colored in four colors so that any two
neighboring regions have different colors); Feit and Thompson’s proof of the Odd
order theorem (any finite group of odd cardinality is solvable), by G ONTHIER (2008)
and G ONTHIER et al (2013), (both in Coq); Hales’s proof of the Kepler conjecture
(the standard, “cubic close”, sphere packing is the densest one), by H ALES et al
(2017) (in HOL Light); following the challenge of S CHOLZE (2022), the proof of a
2 ANTOINE CHAMBERT-LOIR

delicate homological algebra result of Clausen and Scholze (in Lean, the so called
“Liquid tensor experiment”, 2022, by C OMMELIN and T OPAZ, with the help of many
more people involved); or Gromov’s proof of the h-principle and the sphere eversion
theorem by M ASSOT, VAN D OORN & N ASH (2022), also in Lean.
Actually, the latter results were not formalized in plain Lean, but were built
on the Lean mathematical library mathlib. Lead by a group of approximately
25 people, plus some 15 reviewers, this mathematical library is an ongrowing ef-
fort of roughly 300 people, with (as today) approximately 45 000 definitions and
110 000 mathematical statements (“theorems”) that cover many fields of mathe-
matics, such as additive combinatorics, complex variables, differential geometry
and Lebesgue integration. . . So that a collective effort is at all possible, the initial
authors of mathlib had to make careful architecture and design decisions, described
in (T HE MATHLIB C OMMUNITY, 2020). As Lean/mathlib is an open source project,
it is also relatively easy to install it on one’s own computer, and start joining this
collective effort. This is also facilitated by a comprehensive website and an online
discussion board where contributors share their problems and, remarkably gener-
ously, insights.
In November 2021, I embarked in checking in Lean/mathlib the proof that the
alternating group of a finite set of cardinality at least 5 is a simple group. While
this mathematical result is of a smaller scale, compared to the above-quoted accom-
plishements, it belongs to the classical (under)graduate mathematical corpus, and
I felt interesting to experiment the formalization process on a result of this inter-
mediate level. For reasons I will try to share, I chose a nonstandard way to do that,
that led me to unsuspected mathematical territories.
This text is a retrospective account of this journey.

I thank Javier Fresán for having inviting me to write this paper, a Spanish version
of which should appear in La Gaceta de la Real Sociedad Matemática Española. I
also thank him for his insightful comments, as well as those of Riccardo Brasca and
Patrick Massot, as well as Martin Liebeck and Raphaël Rouquier for their help. I
also thank the mathlib community for their enthusiasm in welcoming newcomers to
the game, and for the support they provide so generously.

2. Solvability, simplicity
Let us first recall the terms of the statement we have in mind.

Theorem 2.1. — Let n be an integer such that n Ê 5. The alternating group An is a

simple group.
(For n É 2, the group An is trivial; for n = 3, it is cyclic of order 3; for n = 4, it is a
nonabelian solvable group of order 12, its derived subgroup is an abelian subgroup
of index 3.)
In the Lean language, this theorem can be formulated as follows:
 AN EXPERIMENT IN FORMALIZATION OF AN INTERMEDIATE-LEVEL ALGEBRA THEOREM 3

L ISTING 1. Simplicity of the alternating groups of order at least five

theorem alternating_group.normal_subgroups {α : Type∗ }
[decidable_eq α] [fintype α]
(hα : 5 É fintype.card α)
{N : subgroup (alternating_group α)}
(hnN : N.normal) (ntN : nontrivial N) : N = >

The command theorem initiates a statement of a theorem, followed by the name

given to it, here alternating_group.normal_subgroups, and followed by a sequence
of arguments which are surrounded by various kinds of parentheses.
The first of these arguments, α, is declared as a type, the basic notion of “depen-
dent type theory”, the formal language of Lean: in this case, one may think of α as a
set. The next arguments impose that it is also finite, and hα is the assumption that
it has at least five members. The next three parameters are N, which is declared
as a subgroup of the alternating group on α, hnN which imposes that it is a normal
subgroup, and ntN that it is nontrivial (which, in the mathlib library, means that it
is not reduced to the unit element).
The conclusion of that theorem follows the colon symbol: N = >, meaning that
N is the full group of the alternating group. In the actual code, this 5-line text is
followed by the symbol := and the actual proof of this statement.
Every object of type theory is a type, and what Lean does is allowing the user
to write down new types, or members of those types. For example, in the example
above N is a member of the type subgroup (alternating_group α). Lean provides a
few basic means to define new types from older; for example, if α and β are types,
there is a type α → β which represents “functions” from α to β, in the sense that if
f : α → β is such a function and a : α (read: “a is a member of the type α”), then
f a is a member of the type β, with obvious rules regarding equality. Functions
of multiple arguments can be defined “à la Curry”: for example, if α, β and γ are
types, then f : α → β → γ maps a : α to f a : β → γ, which maps b : β to f
a b : γ, etc. Even the expression N = > of listing 1 designates a type, namely, the
type of proofs of equality between the two members N and > of the type subgroup
(alternating_group α), and the (omitted) code that follows constructs a member of
that type, that is, a proof of that statement: type theory puts mathematical struc-
tures and theorems on the same level.

Simple groups are those (nontrivial) groups whose only normal subgroups are
the two obvious examples, the full group and trivial group { e}. When a nontrivial
group G is not simple, it admits a normal subgroup H such that H 6= { e} and H 6= G;
then one can (try to) study G through its projection to the quotient group G/H, whose
kernel is H. When we restrict ourselves to finite groups, a full “dévissage” is possible
and a common metaphor presents finite simple groups as the “elementary particles”
of finite group theory. In this direction, a legendary theorem whose proof involved
hundreds of mathematicians and hundreds of mathematical papers written over
4 ANTOINE CHAMBERT-LOIR

a period of 50 years, is the classification of finite simple groups: All finite simple
groups appear in a list of groups of the following form:

– The cyclic group Z/ pZ, for some prime number p;

– The alternating group An , for some integer n Ê 5;
– Lists of finite groups “of Lie type”, related to linear algebra over finite fields,
whose easiest examples consist of the projective special linear groups PSL( n, F q )
over a finite field of cardinality q, assuming q Ê 4 if n = 2 (PSL(2, F2 ) and PSL(2, F3 )
are respectively isomorphic to S3 and A4 , hence are not simple);
– A list of 26 (so called “sporadic”) groups, related to exceptional combinatorial
geometries, such as the Matthieu groups M11 , M12 , M22 , M23 and M24 .

The difficult part of the classification of finite simple groups asserts that those
finite groups are the only simple groups, but we are only concerned here by the
easy part of the classification, that these groups are indeed simple.
The first ones, cyclic groups of prime order, are simple: it follows from Lagrange’s
theorem (the order of a subgroup divides the order of the group) that they have no
other subgroup than themselves and the trivial subgroup.
As an aside remark, let us note that the center Z(G) of a group G, the set of ele-
ments g ∈ G which commute with any other element of G, is also a normal subgroup.
Consequently, if G is simple, then either Z(G) = G, in which case G is commutative,
hence a cyclic group of prime order, or Z(G) = { e}. This explains why, from the second
item on, all groups of the above list have trivial center.
On the second item of that list come the alternating groups, which are the very
subject of this note, and whose simplicity is often established in algebra lectures
related to Galois theory and the solvability of algebraic equations (in one variable).
While Abel and Ruffini had proved that general algebraic equations of degree Ê 5
cannot be solved by radicals, Galois’s theorem refines that result by proving that
a given algebraic equation is solvable by radicals if and only if its Galois group is
solvable. The notion of a group of an equation was introduced by Galois, as well as
the notion of a normal subgroup and of solvable group, although he did not give a
name to these two concepts: the Galois group is the subgroup of the permutations of
the roots that preserve all algebraic relations with rational coefficients; and a finite
group G is solvable if it is trivial or if, by induction, it admits a nontrivial normal
subgroup H which is itself solvable and such that the quotient group G/H is com-
mutative. In modern terms, we say that a group G is solvable if its “derived series”
G, D(G), D(D(G)) . . . , the decreasing sequence of subgroups obtained by successively
taking commutator subgroups, eventually reaches the identity subgroup.
In that perspective, the Abel–Ruffini theorem boils down to the fact that a general
equation of degree n has Galois group the full symmetric group Sn , and that, for
n Ê 5, it is not solvable, itself a direct consequence of the following more precise
result.
 AN EXPERIMENT IN FORMALIZATION OF AN INTERMEDIATE-LEVEL ALGEBRA THEOREM 5

Proposition 2.2. — Let n be an integer such that n Ê 5. The commutator subgroup

of Sn is the alternating group An . The commutator subgroup of the alternating
group An is itself.
Proof. — Any commutator has signature 1, so that D(Sn ) ⊆ An . On the other hand,
a commutator of two transpositions (a b) and ( c d ) is trivial if they are equal or have
disjoint supports, but is equal to a 3-cycle otherwise, as the following computation
shows

(a b)( c a)(a b)( c a) = (a b c).

We see that any 3-cycle can we written as a commutator, so that D(Sn ) contains all
3-cycles, which are known to generate the alternating group An . (This works for
n Ê 3.)
To prove that D(An ) is An itself, we prove that the quotient group K = An /D(An )
is trivial. The group An is generated by 3-cycles g, so their images generate K. The
hypothesis n Ê 5 implies that all 3-cycles are conjugate in An ; consequently, they
all have the same image in K, say k, and K = 〈 k〉. Since the square of a 3-cycle
g = (a b c) is again a 3-cycle, namely g2 = (a c b), one has k = k2 , hence k = e and
K = { e}.

The relation with simplicity is that noncommutative solvable groups cannot be

simple. In fact, it is an elementary observation that the commutator subgroup D(G)
of any group G is a normal subgroup of G; if G is simple, then either D(G) = { e},
which means that G is commutative, or D(G) = G. So Galois’s theorem on alge-
braic equations of degree Ê 5 is often subsumed in mentioning that the alternating
group An is simple for n Ê 5, although the result which is needed is the easier
proposition given above.
It is sometimes written that Galois proved that simplicity theorem, although the
only explicit statement I could find in his works is the fact that the smallest possible
cardinality of a simple (noncommutative) finite group (he says “indecomposable”) is
5 · 4 · 3, but he does not state that it corresponds to the alternating group A5 . On
the other hand, group theorists of the 19th century, from Lagrange and Ruffini to
Jordan, gradually built the tools to understand Galois’s theorem in terms of the
simplicity of the alternating group.
There are many relatively easy proofs of the simplicity of An for n Ê 5, such as,
for example, the one given by (J ACOBSON, 1985, p. 247), but none of them looks as
being completely straightforward, in the sense that they do not tell why they work.
Moreover, some of them build on case disjunctions, or mental reasonings which,
although they are quite familiar to us, remain a bit awkward to specify exactly, to
the point that I am not even sure that our explanations suffice to our students.
So my initial idea was to find a proof that would be of a more systematic nature,
using arguments that are more prone to generalization. The principle of such a
proof, already hinted to in the book of W ILSON (2009), is given by the Iwasawa
criterion, to which I now turn.
6 ANTOINE CHAMBERT-LOIR

3. The Iwasawa criterion for simplicity

I WASAWA (1941) proposed a proof of the simplicity of the projective special linear
group PSL( n, F) of a field F of cardinality at least 4. Before that, this theorem was
limited to the case of finite fields (Dickson) or fields of characteristic 6= 2 (van der
Waerden). From that proof, the following geometric criterion can be extrated.

Theorem 3.1. — Let a group G act on a set X, and assume that we are given, for
every x ∈ X, a subgroup T( x) of G, such that the following properties hold:
– For every x ∈ X, the group T( x) is commutative;
– For every g ∈ G and every x ∈ X, one has T( g · x) = gT( x) g−1 ;
– The groups T( x) generate G.
If, moreover, the action of G on X is quasiprimitive, then any normal subgroup N
of G that acts nontrivially on X contains the commutator subgroup D(G) of G.
An action of a group G on a set X is said to be quasiprimitive if any normal sub-
group of G which acts nontrivially on X acts transitively. This property may look
obscure, but it appears naturally in the framework of primitive actions, a classic
theme of 19th century group theory which remained very important in finite group
theory but seems to have disappeared from the algebra package we offer to under-
graduate students. Let us define it in terms of partitions of X (sets of nonempty
disjoint subsets of X whose union is X):

Definition 3.2. — A transitive action of a group G on a set X is primitive if there

are exactly two partitions of X which are invariant under G, the coarse partition {X}
and the discrete partition consisting of all singletons.
In particular, this definition implies that the set X has at least two elements.
If H is a normal subgroup of X, then the partition of X in orbits of H is an invariant
partition; consequently, if the action of X is primitive, then either H acts trivially,
or it acts transitively: for every x, x0 ∈ X, there exist h ∈ H such that h · x = x0 . This
shows that primitive actions are quasiprimitive.
Higher transitivity conditions give important examples of primitive actions.

Lemma 3.3. — Let us assume that the action of G is 2-fold transitive: X has at least
two elements and for any two pairs ( x, y) and ( x0 , y0 ) of distinct elements of X, there
exists g ∈ G such that g · x = x0 and g · y = y0 . Then this action is primitive.
The proof is elementary: consider an element B of a partition Σ of X which has at
least two elements x, y and let us show that B = X. Let z ∈ X be such that z 6= x, y.
By the 2-fold transitivity condition applied to ( x, y) and ( x, z), there exists g ∈ G such
that g · x = x and g · y = z. The set g · B belongs to Σ but has a common point with B,
namely x, so that g · B = B. In particular, z ∈ B. This proves that B = X.
We just observed that members of a G-invariant partition are subsets B of X such
that either g · B ∩ B = ; or g · B = B; in the traditional terminology of group theory,
they are called blocks, and blocks which are neither empty, nor singletons, nor the
full sets are called blocks of imprimitivity. Conversely, if B is a nonempty block and
 AN EXPERIMENT IN FORMALIZATION OF AN INTERMEDIATE-LEVEL ALGEBRA THEOREM 7

if the action is transitive, then the set of all g · B, for g ∈ G, gives a G-invariant
partition of X.
As an example of a transitive, but not primitive action, one may consider the
action of S4 on the set of pairs of elements of {1, 2, 3, 4}: in this case, there are
nontrivial blocks, such as B = {{1, 2}, {3, 4}}. In fact, we will have to meet this example
later, and some variants of it.
The terminology “primitive” comes from Galois, in the language of equations: as
explained by (N EUMANN, 2006, p. 390), when the Galois group G of an irreducible
polynomial equation f ( x) = 0 acts on its roots, there are m blocks of size n if and
only if there is an auxiliary equation of degree m the adjunction of one root of which
allows f to be factored as f 1 f 2 , where f 1 has degree n.
The Lean definitions follow these descriptions, see listing 2, with a few adjust-
ments to follow the general mathlib conventions.

L ISTING 2. Blocks, primitive actions

variables (G : Type ) {X : Type∗ } [has_smul G X]
∗

/-- A trivial block is a subsingleton or > (it is not necessarily a

...block)-/
def is_trivial_block (B : set X) := B.subsingleton ∨ B = >

/-- A block is a set which is either fixed or moved to a disjoint subset

-/
def is_block (B : set X) := (set.range (λ g : G, g ·
B)).pairwise_disjoint id

/-- An action is preprimitive if it is pretransitive and

the only blocks are the trivial ones -/
class is_preprimitive
extends is_pretransitive G X : Prop :=
(has_trivial_blocks’ : ∀ {B : set X}, (is_block G B) → is_trivial_block
B)

First of all, definitions are always given under very minimal hypotheses, one idea
being that they could serve in more general contexts than the ones that are gener-
ally considered, so as to avoid the need for infinite variations of otherwise identical
proofs. Another principle to have definitions as general as possible is that chang-
ing a definition later on requires to adjust all theorems that refer to it, a painful
and long task. In our case, “actions” of a type G on another type X just presumes a
map G → X → X embodied in the predicate has_smul G X, and then denoted by the
symbol ·, not even requiring that G has an inner multiplicative structure! It is remi-
niscent of the “groups with operators” introduced in the first chapter of (B OURBAKI,
1998) with a similar intention.
Then a “subset” B of X (something called set X) is a block if and only if the sets
g · B, for g in X, are pairwise equal or disjoint. The (possibly) obscure definition
makes use of mathlib’s general predicate set.pairwise_disjoint.
8 ANTOINE CHAMBERT-LOIR

Trivial blocks are detected by the predicate is_trivial_block, defined as either

“subsingletons” (the empty set or a singleton, with definition “∀ x, y ∈ B, x = y”) or
the full set >.
Another mathlib indosyncracy that appears in the definitions above is the con-
cept of “pretransitive” actions, meaning “transitive but possibly empty”. Again, the
idea is to defer the non-emptiness hypotheses to the statements that actually and
explicitly need them. We thus define an action to be preprimitive if it is pretransi-
tive and if all blocks are trivial.
In what follows, it will be important to use the following equivalent characteri-
zation of primitive actions. (Recall that the fixator G x of an element x in X is the
subgroup of G consisting of all g ∈ G such that g · x = x.)

Lemma 3.4. — The action of G on X is primitive if and only if it is transitive and if

for every x ∈ X, its fixator G x is a maximal(1) subgroup of G.
More generally, one can show that for any x ∈ X, the mapping H 7→ H · x induces an
order preserving bijection from the lattice of subgroups H of G such that G x ⊆ H ⊂ G
to the lattice of blocks B in X that contain x. We copied in listing 3 the Lean defi-
nition of this order preserving bijection (in fact, its inverse): it takes the form of an
“order equivalence” of types, as indicated by the symbols 'o. The first type, { B :
set X // a ∈ B ∧ is_block G B }, is the type of all B : set X (basically, subsets
of X) satisfying the properties that a ∈ B and is_block G B, the latter type encod-
ing that B is a block for the action of G on X (which could be left implicit, because
B being of type set X, this type is known). Lean is capable to guess by itself that
this type inherits the ordering relation given by inclusion on set X. The second
type, set.Ici (stabilizer G a), designates, in the lattice subgroup G, the subset
of those subgroups containing stabilizer G a. This “order equivalence” consists of
two functions, to_fun and inv_fun, proofs (left_inv and right_inv) that they are
inverse of each other, and a proof (map_rel_iff) that they respect the order. Then
comes the definition of the function to_fun, which maps such a B, together with the
witnesses ha : a ∈ B and hB : is_block G B, to stabilizer G B, accompanied with
stabilizer_of_block hB ha. As one can guess, the former designates the stabilizer
of B in G, together with the additional information that it contains stabilizer G a.
That information is provided by the function stabilizer_of_block : is_block G
B → a ∈ B → stabilizer G a É stabilizer G B whose code, of course, had been
given earlier in the source. The inverse function inv_fun maps H : subgroup G to-
gether with hH : stablizer G a É H to mul_action.orbit H a which represents the
orbit of a under the action of the subgroup H, together with the relevant proofs that
this set contains a and is a block. Then come three proofs, left_inv and right_inv
stating that the two preceding functions are inverse of each other, while map_rel_iff
states that they respect the order relation. In the listing showed there, we replaced

(1)
Recall that a subgroup H of G is maximal if H 6= G and if any subgroup H0 of G containing H is H
or G.
 AN EXPERIMENT IN FORMALIZATION OF AN INTERMEDIATE-LEVEL ALGEBRA THEOREM 9

these three proofs by ...; the codes of the first two ones are 2-line long, that of the
third one is 17-line long.

L ISTING 3. Order equivalence between blocks containing a point and sub-

groups contining its stabilizer
variables {G: Type∗ } [group G] {X : Type∗ } [mul_action G X]
/-- Order equivalence between blocks in X containing a point a
and subgroups of G containing the stabilizer of a
(Wielandt, Finite Permutation Groups, th. 7.5)-/
def stabilizer_block_equiv [htGX : is_pretransitive G X] (a : X) :
{ B : set X // a ∈ B ∧ is_block G B } 'o set.Ici (stabilizer G a) := {
to_fun := λ 〈B, ha, hB〉, 〈stabilizer G B, stabilizer_of_block hB ha〉,
inv_fun := λ 〈H, hH〉, 〈mul_action.orbit H a,
mul_action.mem_orbit_self a, is_block_of_suborbit hH〉,
left_inv := ...,
right_inv := ...,
map_rel_iff’ := ...,
end }

3.5. — We end this section with a proof of the Iwasawa criterion (theorem 3.1).
Fix a point a ∈ X. We first prove that the subgroup 〈N, T(a)〉 generated by N
and T(a) is equal to G. By assumption, N acts transitively on X. Since N is normal,
the hypothesis that the action is quasiprimitive implies that for every b ∈ X, there
exists n ∈ N such that n · a = b. Since nT(a) n−1 = T( b), this implies that 〈N, T(a)〉
contains T( b). Since b is arbitrary, the subgroup 〈N, T(a)〉 contains the subgroup
generated by all T( x), for x ∈ X, which, by assumption, is G.
The subgroup N is normal; the desired conclusion that it is contains the de-
rived subgroup of G is equivalent to the commutativity of the quotient G/N. Since
〈N, T(a)〉 = G, the composition T(a) → G → G/N is surjective; since T(a) is commuta-
tive, we conclude that G/N is commutative, as we wished to.

4. Normal subgroups of symmetric and alternating groups

In this section, we consider an integer n; we generally assume that n Ê 5.
The symmetric group Sn acts not only on the set X = {1, . . . , n}, but also on the
sets X[k] of k-element subsets of X, for any integer k such that 0 É k É n. The action
is trivial if k = 0 or k = n, because then X[k] is reduced to a single element, but it
is faithful otherwise: any element g 6= e acts nontrivially. The following proposition
asserts that this action is moreover primitive, unless n = 2 k.

Proposition 4.1. — Let k and n be integers such that 0 < k < n − k < n. If 4 É n,
then the actions of An and Sn on X[k] are primitive.
Given this primitivity result, the approach of Iwasawa allows us to understand
the normal subgroups of the symmetric and alternating groups. We will only need
to use the cases k = 2, k = 3 and k = 4.
10 ANTOINE CHAMBERT-LOIR

4.2. — Let us first consider the case k = 2. For any 2-element subset x = {a, b} of X,
let us consider the subgroup T( x) generated by the transposition (a b): it is commu-
tative of order 2; the relation ( g · a g · b) = g(a b) g−1 implies that these subgroups sat-
isfy the relation T( g · x) = gT( x) g−1 ; and since Sn is generated by all transpositions,
they generate the symmetric group. Consequently, Iwasawa’s criterion implies that
if this action is primitive, then any normal subgroup N of Sn such that N 6= { e} con-
tains D(Sn ), which as we have seen, is equal to An . Since Sn /An has order 2, the
only subgroups of Sn that contain An are An and Sn .
What about the primitivity assumption? Note that the action of Sn on X[2] is not
2-fold transitive, because one cannot map {1, 2} and {1, 3} to the sets {1, 2} and {3, 4}.
Let us observe that it is nevertherless primitive; here, we will use that 2 < n −2, that
is, n > 4. (W ILSON, 2009, §2.5.1) shows that the fixator of any element of X[2] is a
maximal subgroup, and we will discuss this in greater generality in the next section,
but let me tell right now the following proof as explained to me by G. Chenevier.
Let B be an imprimitivity block of X[2] , and let {a, b} be a pair in B.
First assume that B contains another pair of the form {a, c}. Consider g ∈ G
such that g · a = c and g · b = a; then B and g · B share the element {a, c}, so that
g · B = B; consequently, B contains the pair { g · a, g · c} = { c, g · c}, hence all pairs of the
form { c, d }. Redoing the argument from {a, c} and { c, d } we deduce that B contains
any pair, hence B = X[2] .
Assume then that B contains a pair { c, d } which is disjoint from {a, b}. Since n Ê 5,
we may consider a fifth element e in X; let us prove that { c, e} ∈ B. Indeed, there
exists g ∈ Sn which maps a to a, b to b, c to c and d to e, hence {a, b} to itself, and
{ c, d } to { c, e}; then B and g · B have {a, b} in common, so that g · B = B and { c, e} ∈ B.
In particular, B contains two pairs { c, d } and { c, e} whose supports are not disjoint
and the first part of the argument implies that B = X[2] .
We thus obtain the following result (also a consequence of theorem 2.1).

Proposition 4.3. — For n Ê 5, the normal subgroups of Sn are { e}, An and Sn .

4.4. — We now pass to k = 3. For any 3-element set x = {a, b, c} in X, we consider

the alternating group T( x) of these three elements, viewed as a subgroup of An ; it
is the subgroup generated by the 3-cycle (a b c). As above, the relations T( g · x) =
gT( x) g−1 hold, and these subgroups generate the alternating group. Assuming that
the action of An on X[3] is primitive, we deduce from Iwasawa’s criterion that any
normal subgroup of An either is trivial, or contains D(An ); in other words, An is
simple.
We shall see in the next section that the primitivity condition holds for n 6= 6 (and
why it does not for n = 6), so the case n = 6 requires another argument.

4.5. — For this, let us consider k = 4. For any 4-element set x = {a, b, c, d } in X,
let us consider Klein’s Vierergruppe V( x) in the alternating group of these four
elements, viewed as a subgroup of An . It is commutative of order 4, and con-
sists of the identity and of the three “double transpositions” (a b)( c d ), (a c)( b d ) and
 AN EXPERIMENT IN FORMALIZATION OF AN INTERMEDIATE-LEVEL ALGEBRA THEOREM 11

(a d )( b c). This is already an intrinsic definition of V( x) (permutations with sup-

port in x whose cycle type is either empty or (2, 2)); it can also be defined as the
derived subgroup of the alternating group on these four elements. Consequently,
the relations V( g · x) = gV( x) g−1 hold. Let us show that these subgroups V( x) gen-
erate An ; the argument will use that n Ê 5. We start from the remark that An
consists of permutations which are products of an even number of transpositions. If
two successive transpositions in such a product have disjoint supports, they belong
to some V( x). Otherwise, if their supports share an element a, say (a b)(a c), then
using that n Ê 5, we can insert a cancelling product ( d e)( d e), so that (a b)( d e) and
( d e)(a c) belong to subgroups of the form V( x).
Applying Iwasawa’s criterion, this construction shows that the alternating
group An is simple as soon as the action of An on X[4] is primitive.

4.6. — We note that a variant of these arguments leads to a reasonably simple

proof of the simplicity of A5 . Indeed, by taking complements, the action of An on X[k]
is isomorphic to the action on X[n−k] . When n = 5, the case k = 4 is reduced to the
case k = 1 and it suffices to prove that the action of A5 on X is primitive. To that
aim, it suffices to observe that the action of A5 is 2-transitive. (In fact, it is even
3-transitive.)

5. Primitivity and maximal subgroups

To conclude the proof of theorem 2.1, it remains to explain the proof of propo-
sition 4.1 The fixator of the element {1, . . . , k} of X[k] is the intersection of An with
the subgroup Sk × Sn−k associated with the partition of {1, . . . , n} in {1, . . . , k} and
{ k + 1, . . . , n}. Since the action of An on X[k] is transitive, lemma 3.4 reduces us to
prove that this subgroup is a maximal subgroup of An .
In this way, we note that the hypothesis n 6= 2 k is really necessary for this propo-
sition: the subgroup Sn × Sn of S2n is not maximal, it is a subgroup of index 2 of
the stabilizer of the partition {{1, . . . , n}, { n + 1, . . . , 2 n}}, a group also described as the
wreath product Sn o (Z/2Z).

5.1. — In §4.2, we saw an elementary proof of proposition 4.1 for k = 2 and n Ê 5,

and it seems likely that an elementary proof exists for any k. R. Rouquier gave
me one that works for k = 3 and n Ê 7. However, I want to describe another ap-
proach, explained to me by M. Liebeck, that I think emphasizes the status of that
proposition within finite group theory.
One of the first treatises on group theory is that of J ORDAN (1870). Then groups
were “permutation groups”, permuting letters, (henceforth algebraic expressions
on these letters), or, since the connection with the Galois theory of equations was
explicit, roots of a polynomial equation. It had been observed that the symmetric
group on n letters is n-fold transitive — almost by definition, given two systems of
distinct elements in {1, . . . , n}, x1 , . . . , xn and y1 , . . . , yn , say, there is a permutation g
such that g · x i = yi for all i , and g is even unique. Only slightly less obvious was the
12 ANTOINE CHAMBERT-LOIR

fact that the alternating group on n letters is ( n − 2)-fold transitive: given distinct
systems x1 , . . . , xn−2 and y1 , . . . , yn−2 , there are two permutations g such that g ·
x i = yi for all i , one is even and the other is of the form (a b) g, where a, b are
the two elements of {1, . . . , n} not in { y1 , . . . , yn−2 }. It had also been observed that
beyond these two cases, a permutation subgroup on n letters has to act much less
transitively and 19th century mathematicians proved many theorems that aimed at
quantifying this limit. For example, Mathieu had proved that unless it contains the
alternating group, a subgroup of Sn isn’t n/2-fold transitive, while J ORDAN (1872)
proved that it isn’t m-fold transitive if n − m is a prime number > 2.
As explained in C AMERON (1981), once the classification of simple finite groups
had been achieved, it could be checked on the list that a 6-fold transitive subgroup
of Sn must be symmetric or alternating.
Parallel to the classification is the understanding of all maximal subgroups of a
given finite simple group. In the case of the alternating group, an explicit list has
been provided independently by M. O’Nan and L. Scott. As remarked by C AMERON
(1981), this question is closely related to the description of all subgroups of the
symmetric group Sn which act primitively on {1, . . . , n}.
This classification theorem takes the given form: Let G be a strict subgroup of An
or Sn ; then G is conjugate to a subgroup of one of six types of which the first three
take the form:
(a) A product Sm × Sn−m , where 0 < m < n — the intransitive case.
(b) The “wreath product” Sm o S p , where n = pm, namely the subgroup generated
by the product of p symmetric groups acting on p disjoint sets of m letters (isomor-
phic to Sm × · · · × Sm ), and a permutation that permutes cyclically these p sets —
the imprimitive case;
(c) An affine group of an F p -vector space of dimension d , where n = p d is the
power of a prime number.
It applies in particular to maximal subgroups, and L IEBECK et al (1987) established
the converse assertion, deciding which of the groups of this list are maximal. That
case (a) is maximal when m 6= n − m is exactly the statement of proposition 4.1.
However, when n = 2 m, case (a) is not maximal but case (b) gives the corresponding
maximal case. For n = 4, for example, the subgroup given by (b) has order 8, hence
is a 2-sylow subgroup of S4 , while the subgroup S2 × S2 has order 4.
Cases of the form (c) were of particular interest to Galois, who proved that they
appear for the Galois groups of irreducible equations of prime degree which are
solvable by radicals. In other words, solvable and transitive subgroups of S p can
be viewed, up to conjugacy, as a group of permutations of the form x 7→ ax + b on F p ,
for a ∈ F×p and b ∈ F p . Since the identity is the only permutation of that form that
fixes two elements, Galois obtains that an irreducible equation of prime degree is
solvable by radicals if and only if any of its roots can be expressed rationally by any
two of them.
Galois also defined primitive algebraic equations which correspond exactly to the
case where the Galois group acts primitively on their roots. In the solvable case,
 AN EXPERIMENT IN FORMALIZATION OF AN INTERMEDIATE-LEVEL ALGEBRA THEOREM 13

he proved that the degree has to be a power p n of a prime number p and, up to an

enumeration of the vector space Fnp , the Galois group G is a subgroup of the group
of permutations of the form x 7→ A x + b, for A ∈ GL( n, F p ) and b ∈ Fnp , that contains
all translations x 7→ x + b. Moreover, the subgroup G0 of GL( n, F p ) consisting of
all elements of G of the form x 7→ A x has no nontrivial invariant subspace. The
interested reader shall find more details on this fascinating story in chapter 14
of C OX (2012).

5.2. — But let us go back to the promised proof of proposition 4.1. Let G be a
subgroup of An strictly containing (Sk × Sn−k ) ∩ An , where 0 < k < n and n 6= 2 k.
We need to prove that G coincides with An . By symmetry, we may assume that
k < n − k. The case k = 1 is easy. Indeed, the action of An on {1, . . . , n} is ( n − 2)-fold
transitive, hence it is 2-fold transitive, because n Ê 4, hence it is primitive. We now
assume that 2 É k; then n Ê 5.
A theorem of (J ORDAN, 1870, Note C to §398, page 664) asserts that a primitive
subgroup of Sn that contains a cycle of prime order p is at least ( n − p + 1)-fold
transitive. When p = 2, we get that this subgroup is ( n − 1)-fold transitive, hence it
has to be the whole Sn , while when p = 3, it is ( n − 2)-fold transitive, and it is not
too difficult to deduce that it contains An . Since 1 É k < n − k < n and n Ê 5, we have
n − k Ê 3 and our subgroup G contains a 3-cycle. To conclude, it remains to establish
that it acts primitively on {1, . . . , n}.
One first proves that G acts transitively on {1, . . . , n}. In fact, G contains the
subgroups Sk and Sn−k ; in particular, it acts transitively on the elements of each
subset {1, . . . , k} and { k + 1, . . . , n}, hence it has at most two orbits. But since it strictly
contains (Sk × Sn−k ) ∩ An , it cannot leave {1, . . . , k} and { k + 1, . . . , n} invariant.
Arguing as for transitivity, G acts k-fold transitively on {1, . . . , k} and ( n − k)-fold
transitively on { k+1, . . . , n}; since 2 É k < n− k, it acts in particular 2-fold transitively,
hence primitively, on both of these sets.
We consider imprimitivity blocks B for the action of G, assuming that they have
at least two elements and are distinct from {1, . . . , n}.
First observe that B cannot contain { k + 1, . . . , n}, because its translates g · B, for
g ∈ B such that g · B 6= B, would have to be contained in {1, . . . , k}, which is impossible
since k < n − k. In particular, B meets { k + 1, . . . , n} in at most one element. If it
is disjoint from { k + 1, . . . , n}, it is contained in {1, . . . , k}. Since G acts primitively
on {1, . . . , k}, one then has B = {1, . . . , k}. Consider an element g of G which does not
stabilize {1, . . . , k}. Then g · B is a block distinct from B, hence disjoint, so that g · B
is a block contained in { k + 1, . . . , n}. By primitivity, g · B = { k + 1, . . . , n}, contradicting
the beginning of the proof.
In particular, there are elements a ∈ {1, . . . , k} ∩ B and b ∈ { k + 1, . . . , n} ∩ B. To
conclude the proof by a contradiction, it suffices to establish that B contains
{ k + 1, . . . , n}. So let c ∈ { k + 1, . . . , n} and consider an element g ∈ G that fixes {1, . . . , k}
such that g · b = c. Then g · B and B both contain a, hence g · B = B, hence c ∈ B, as
was to be shown.
14 ANTOINE CHAMBERT-LOIR

6. Intermezzo: conjugacy classes in symmetric groups

At the end, the proof of the simplicity theorem of the alternating group A6 re-
quired a discussion of the Klein subgroup of A4 . When we discuss this group be-
tween colleagues, possibly in class, the proof fact that it is indeed a subgroup usually
boils down to a mere: “one checks that. . . ”.
Of course, such an argument is not sufficient for the computer, and I spent some
time trying to imagine how we should prove such facts in the computer. The proof I
resorted to happened to be fun, nevertheless slightly sophisticated.
Let X = {a, b, c, d } be a set with four elements, and let V be the subset of SX
consisting of the identity and of all double transpositions. In order to prove that V
is a subgroup of SX , I prove that V is the only 2-sylow subgroup of AX . The proof
runs as follows, in which we consider an arbitrary 2-sylow subgroup S of AX .
– Since S4 has cardinality 4! = 24, the alternating group A4 has cardinality 12,
and S has cardinality 4.
– The order of any element g of S divides 4; since its entries thus divide 4, the
cycle type of g belongs to (), (2), (2, 2) or (4). Since the second and last cases give odd
permutations, we have g ∈ V;
– Now the number of permutation of a given cycle type in a symmetric group can
be computed explicitly, more on this below, and the computation shows that V has
4 elements;
– Since S and V both have 4 elements, and S ⊆ V, this proves V = S, as claimed.
The computation of the number of permutations of a given cycle type in the
symmetric group SX is by itself a classic and important result in combinatorics
of finite permutation groups. We return to the general case of a finite set X, let
n be its cardinality, and consider a partition π of the integer n; let us write m i
for the number of parts equal to i . A permutation of cycle type π takes the form
(a 1 )(a 2 ) . . . (a m1 )( b 1 b01 )( b 2 b02 ) . . . ( b m2 b0m2 ) . . . : m 1 cycles of length 1, m 2 cycles of
length 2, etc. In order to compute the number of permutations of cycle type π, we
just have to fill the letters with distinct elements of X, which apparently makes
for n! permutations. However, for each cycle of length i , only the cyclic ordering of
the elements matters, so we have to divide the result by i m i . Moreover, the order
Q
in which we write the m i cycles of given length i does not matter, so the result
Q
needs to be further divided by m i !. Finally, the number of permutations of cycle
Q mi Q
type π is n!/ i m i !.
There is however a more conceptual, and more precise, way to prove this formula.
Fix any permutation g which has cycle type π. Since the number we wish to com-
pute is the cardinality of the orbit of g under the conjugation action, it suffices to
prove that the cardinality of the centralizer Z g of g is equal to i m i m i !.
Q Q

If h ∈ Z g , then hgh−1 = g, so that the cycles of hgh−1 are those of g. In other

words, Z g acts by conjugation on the set of cycles of g, respecting their lengths.
This gives a group morphism ϕ : Z g → i Sm i .
Q
This morphism is surjective. In fact, one can even show that ϕ has a section.
Indeed, fix, for each cycle c of g, an element a c in c; then, for any permutation σ
 AN EXPERIMENT IN FORMALIZATION OF AN INTERMEDIATE-LEVEL ALGEBRA THEOREM 15

of the set of cycles of g which preserves their lengths, there is a unique element h σ
of Z g such that h σ (a c ) = a σ(c) for all c, and the map σ 7→ h σ is a group morphism.
Now, the kernel of ϕ is the subgroup of all elements h ∈ Z g such that hch−1 = c for
all cycles c of g. Necessarily, h stabilizes the support of each such c, so it maps a c
to some power iterate of a c under g; fix k c ∈ Z (modulo the cardinality n c of the
support of c) such that h(a c ) = g k c (a c ) = c k c (a c ); using the fact that c is a cycle, it
follows that h acts like c k c on the support of c. Finally, we see that h is the product
of these powers c k c . In other words, ker(ϕ) is a product of cyclic groups, c (Z/ k c Z),
Q
which we rewrite as i (Z/ i Z)m i , since m i is the number of cycles c such that n c = i .
Q
In particular, the order of ker(ϕ) is equal to i m i .
Q
Finally, Card(Z g ) = Card(Im(ϕ)) Card(ker(ϕ)) = i i m i i m i !, as was to be shown.
Q Q

7. Simplicity of classical groups

7.1. — The simplicity criterion is not explicitly stated by I WASAWA (1941), but
it is directly proved and applied in the case of the projective special linear
group PSL( n, F) acting on the projective space Pn−1 (F) of lines in Fn . Unless F
has 2 or 3 elements, a linear algebra argument shows that this action is 2-fold
transitive, hence primitive.
For every line ` ∈ Pn−1 (F), consider the subgroup T(`) of transvections with re-
spect to `, namely the elements g ∈ SL( n, F) such that the range of g −id is contained
in `. Using that SL( n, F) is generated by transvections, we see that they give rise to
a datum as in Iwasawa’s criterion. Consequently, any normal subgroup of SL( n, F)
which acts nontrivially on Pn−1 (F) contains the commutator subgroup of SL( n, F),
which is known to be SL( n, F) itself. Finally, the only elements of SL( n, F) which
act trivially on Pn−1 (F) are the homotheties, and they form the center of SL( n, F), a
finite subgroup isomorphic to the set of nth roots of unity in F. As a consequence,
the quotient PSL( n, F) = SL( n, F)/Z(SL( n, F)) is a simple group.

7.2. — This reasoning can also be applied for other cases of geometric groups. In
his paper, Iwasawa himself indicates that the same method works for the symplectic
group PSp(2 n, F) (“complex projective groups” in the earlier terminology) acting on
the projective space P2n−1 (F). Iwasawa does not explicitly consider the notion of a
primitive action in his paper: his arguments are only spelt out for a 2-fold transitive
action. However, he mentions in a footnote that while the action of the symplectic
group on P2n−1 (F) is not 2-fold transitive, it is quasiprimitive, and this suffices for
his proof. On the other hand, K ING (1981) established that the stabilizers of this
action are maximal subgroups, so that this action is even primitive.
In fact, it seems that the simplicity of the appropriate groups of geometric trans-
formations can all be established in this way.
I find it remarkable how much this method, that relates the simplicity of a group
with the structure of its maximal subgroups, is absolutely in par with the point of
view of Jordan and early group theorists!
16 ANTOINE CHAMBERT-LOIR

8. Remarks regarding the formalized proof and the formalization process

As recalled in the introduction, implementation of mathematical proofs in com-
puters is not a very recent activity, but the Lean/mathlib movement puts us at a
crossroad in so that it makes an indefinitely-extending library of formalized proofs
conceivable. Built on the experiment described in this note, I would like to risk my-
self to enumerating some remarks about this prospect, the hopes and fears it may
rise.
Formalization of mathematical proofs has many goals.
Some of its proponents raise the idea that it will make us truly certain of the
validity of the new mathematical theorems we prove. The idea here is that the
traditional peer review seems to reach its limits, both for mathematical and socio-
logical reasons.
Some papers are simply so complicated that nobody can reasonably claim to have
checked their validity with absolute certainty. This was the case for Hales’s proof of
the Kepler conjecture, before he, leading a team of 21 mathematicians, formalized
that proof. In some sense, this is still the case for the classification of finite simple
groups, whose size and technicality makes it inacessible to most of the mathemati-
cal community.
On the other hand, most research papers are of an apparently smaller size, but
the sociology of the fields evolved. The increasing importance of research grants for
the funding of research, if not for obtaining permanent academic positions, led us
to a stage where the collectivity wants their papers published more quickly that it
can assert their validity, if not just read. As a consequence, papers are reviewed too
quickly, their publication is conditioned to preliminary opinions, leading to all imag-
inable biases, and new journals are created to host this ever growing mathematical
litterature.
If we could check ourselves the validity of our proofs within formalization soft-
ware, and deliver it at the same time we submit a paper, it is likely that this paper
could have been written in a different way: not just to quickly convince a referee
that the proofs are true, but spending more time than we presently do to explain
the statements, their interest, their context, the path that lead to their proof, as
well as aiming at a possibly larger audience.
For this to happen, we need a huge archive of mathematical proofs written in a
common language, with common definitions. The experience of the Bourbaki books
suggests that something is possible, but it also reminds that not all mathematicians
will be willing to comply to the mathematical writing style of other.
If the style of Bourbaki has been sometimes defined as too abstract, it is nothing
in comparison with that of mathlib. Indeed, in order to avoid repeating proofs, the
authors of that library make a permanent effort to put its definitions and state-
ments in a (natural, but possibly frightening) generality. Linear algebra starts
with a discussion of semimodules over monoids, so that the relevant part applies
to more exotic contexts, such as (max, +)-algebras. (Bourbaki made a similar step
when they defined “groups with operators”, but their notion does not seem to be
 AN EXPERIMENT IN FORMALIZATION OF AN INTERMEDIATE-LEVEL ALGEBRA THEOREM 17

commonly used.) In complex analysis, the characterization of analytic functions

as functions which are differentiable in the complex sense is proved using the
Kurzweil–Henstock integral, because that allows us to avoid any Lebesgue inte-
grability assumption on the derivative.
One of the difficulties by working with many simultaneous group actions is that
type theory, the inner language of Lean, does not allow the many abuses of lan-
guage that we do while doing mathematics, without even thinking about it. Take,
for example, a group G acting on a set X, a subset A of X, and a point a ∈ X A.
Then we can consider the fixator GA of A in X, and its action on X A, then the
fixator GA,a of a in GA , and its action on X (A ∪ {a}), which — obviously — coin-
cides with the action of the fixator of A ∪ {a} on its complement. However, these
actions look sufficiently different to Lean, syntaxically, and it is not able to identify
automatically. The suggestion I received on the Zulip discussion blackboard was
that I should not even try to identify them, but that it was sufficient to relate them
through equivariant maps. If groups G and H act on X and Y, and ϕ : G → H is a
morphism of groups, then a ϕ-equivariant map from X → Y is just a map f : X → Y
such that f ( g · x) = ϕ( g) · f ( x) for all g ∈ G and x ∈ X. Then several basic results allow
us to transfer primitivity or transitivity properties from the action of G on X to the
action of H on Y, or vice versa. This is an example of an elementary definition,
with basic companion results, that we probably wouldn’t dare introducing explicitly
in a standard mathematical discussion — probably too trivial for specialists, but
already too obscure for beginners. Learning to appreciate the relevance of introduc-
ing such abstract concepts takes some time, requires a community of knowledgeable
mathematicians, as well as the will to follow their point of view.
On the other hand, as the myth of the Babel tower should remind us, leaning
towards some ever-expanding generality comes at a high risk, that the full edifice
collapses. My own experiment has been agreeable enough to me to sincerely wish
that the community skillfully avoids that risk.

References
N. B OURBAKI (1998), Elements of Mathematics. Algebra I. Chapters 1–3. Transl.
from the French. Softcover Edition of the 2nd Printing 1989., Berlin: Springer,
softcover edition of the 2nd printing 1989 edition.
P. J. C AMERON (1981), “Finite Permutation Groups and Finite Simple Groups”.
Bulletin of the London Mathematical Society, 13 (1), pp. 1–22.
D. A. C OX (2012), Galois Theory, Pure and Applied Mathematics, John Wiley &
Sons, Hoboken, N.J, 2nd ed edition.
A. D OXIADIS & C. H. PAPADIMITRIOU (2009), Logicomix—An Epic Search for
Truth, Bloomsbury Press, New York.
G. G ONTHIER (2008), “Formal proof—the four-color theorem”. Notices of the Amer-
ican Mathematical Society, 55 (11), pp. 1382–1393.
18 ANTOINE CHAMBERT-LOIR

G. G ONTHIER, A. A SPERTI, J. AVIGAD & e. AL . (2013), “A machine-checked proof of

the odd order theorem”. Interactive Theorem Proving, Lecture Notes in Comput.
Sci. 7998, pp. 163–179, Springer, Heidelberg.
T. H ALES, M. A DAMS, G. B AUER, T. D. D ANG, J. H ARRISON, L. T. H OANG,
C. K ALISZYK, V. M AGRON, S. M C L AUGHLIN, T. T. N GUYEN, Q. T. N GUYEN,
T. N IPKOW, S. O BUA, J. P LESO, J. R UTE, A. S OLOVYEV, T. H. A. T A, N. T. T RAN,
T. D. T RIEU, J. U RBAN, K. V U & R. Z UMKELLER (2017), “A formal proof of the
Kepler conjecture”. Forum of Mathematics. Pi, 5, pp. e2, 29.
K. I WASAWA (1941), “über die Einfachheit der speziellen projektiven Gruppen”.
Proceedings of the Imperial Academy, 17 (3), pp. 57–59.
N. J ACOBSON (1985), Basic Algebra. I, W. H. Freeman and Company, New York,
second edition.
C. J ORDAN (1870), Traité des substitutions et des équations algébriques, Gauthier-
Villars, Paris.
C. J ORDAN (1872), “Sur la limite de transitivité des groupes non alternés”. Bulletin
de la Société Mathématique de France, 1, pp. 40–71.
O. K ING (1981), “On some maximal subgroups of the classical groups”. Journal of
Algebra, 68 (1), pp. 109–120.
M. W. L IEBECK, C. E. P RAEGER & J. S AXL (1987), “A classification of the maximal
subgroups of the finite alternating and symmetric groups”. Journal of Algebra,
111 (2), pp. 365–383.
P. M ASSOT, F. VAN D OORN & O. N ASH (2022), “Formalising the h-principle and
sphere eversion”. arXiv:2210.07746.
P. M. N EUMANN (2006), “The Concept of Primitivity in Group Theory and the
Second Memoir of Galois”. Archive for History of Exact Sciences, 60 (4), pp. 379–
429.
G. P EANO (1889), Arithmetices principia nova methoda exposita, Fratelli Bocca
Editrice, Torino.
P. S CHOLZE (2022), “Liquid Tensor Experiment”. Experimental Mathematics,
31 (2), pp. 349–354.
T HE MATHLIB C OMMUNITY (2020), “The lean mathematical library”. Proceedings
of the 9th ACM SIGPLAN International Conference on Certified Programs and
Proofs, CPP 2020, pp. 367–381, Association for Computing Machinery, New York,
NY, USA.
A. N. W HITEHEAD & B. R USSELL (1927), Principia Mathematica, Cambridge Uni-
versity Press, Cambridge [Eng.] New York, 2d ed edition.
R. A. W ILSON (2009), The Finite Simple Groups, Graduate Texts in Mathemat-
ics 251, Springer London, London.