WorksMathematics
Galois's lost proof on
Abstract
In his famous ``testamentary letter'' Galois formulated several results without proof. Among them he claimed — here we express it in modern terms — that the projective special linear group has a subgroup of index only for . Nowadays this result is seen as a corollary of the complete enumeration of all subgroups of obtained by Gierster in 1881. In this paper we attempt to give a proof that uses only methods available to Galois.
Dedicated to the memory of Peter M. Neumann (1940–2020)
Introduction
The famous “testamentary letter” written by Évariste Galois on the eve of his fatal duel contains a wealth of mathematical results. In the first part of the letter Galois summarizes his theory on the solvability of polynomial equations what was the subject of his “First Memoir” submitted to the French Academy on January 16, 1831. (For a modern treatment of this work see [4].) After that almost a quarter of the letter (namely, the bottom part of page 3, and the whole page 4) is devoted to the study of the Galois group of the modular equation arising in the theory of elliptic functions. In modern terms this group is the projective general linear group acting on the roots of the modular equation indexed by the elements of the projective line over the -element field ( a prime number), i.e., . The action is given by fractional linear transformations , where . In particular, and .
Galois first notices that the group has a normal subgroup of index 2, namely, the subgroup consisting of those fractional linear transformations, where is a quadratic residue modulo (in other words, a nonzero square in the -element field). Today this normal subgroup is called the projective special linear group . For and the degree of the modular equation is 3 and 4, respectively, so the equation is solvable by radicals, and the group is solvable. So we will assume that .
Galois has two substantial results concerning . We will call them theorems, although they are not designated so in the running text of the testamentary letter.
Theorem 1. Let be a prime. The projective special linear group is a simple group.
Theorem 2. Let be a prime. The projective special linear group has a subgroup of index only when , 7, and 11.
This research was supported by the National Research, Development and Innovation Office of Hungary (NKFIH), grant no. 153681.
Concerning the first theorem he simply says “But it is easy to see that it is not further decomposable properly unless or .” (See [14 p. 89]. The English translations of Galois’s texts are taken from this edition of Galois’s works by Peter M. Neumann.) We have no clue what kind of easy proof he had in mind.
I have always been admiring Galois for the second theorem. Because of the three exceptional cases it is absolutely non-trivial. Until recently, I have been happy to see this result as a consequence of the complete classification of subgroups of (where is a prime power), see, e.g., [10 Kapitel II, §8]. But this is based on much later developments in group theory. So how could Galois prove this result? Again, he does not say anything about the proof. After outlining the examples for he states boldly: “In all rigour this reduction is not possible in the higher cases.” (See [14, p. 91].) However, analysing his constructions for the small primes may provide hints for a possible proof applying only tools already available to him. In this note we try to give such a proof in Section 3`.
Before doing that we make some historical remarks in § 2.
Finally, in § 4, we relate Galois’s examples for the cases to the rotation group of the tetrahedron, the cube, and the icosahedron, respectively.
Historical remarks
Except for some scattered calculations with fractional linear transformations, Galois’s manuscripts do not contain any explanation for his claims about . The examples he gave for the cases may give some hints how he looked at these subgroups. By describing a pairing of the permuted elements (for details see the next section), he is obviously indicating that these subgroups are imprimitive permutation groups. One finds the definition of imprimitivity in his handwriting on a torn piece of paper among the fragments in Dossier 15 at the Library of the French Academy (see [14 p. 289]). Our suggested reconstruction of Galois’s proof is based on this concept.
One might wonder why Galois wrote “In all rigour” introducing the statement about the nonexistence of subgroups of index for . On the one hand, he tacitly corrects an earlier error. Namely, in the short announcement of his results on the theory of equations (see [14 p. 53]) that was published in Férussac’s Bulletin in April, 1830, he erroneously stated that such a reduction (i.e., the existence of such a subgroup) is not possible even for . On the other hand, he is now absolutely certain that the result in this new version is correct, as he writes on the last page of the testamentary letter: “Often in my life I have risked advancing propositions of which I was not certain. But all that I have written here has been in my head for almost a year and it is not in my interest to make a mistake so that one could suspect me of having announced theorems of which I did not have the complete proof”. This sentence also suggests that he had not written down the proof of this result. The manuscript of his “Second Memoir” (see [14 Chapter V]) is unfinished, it does not reach the point of our interest, leaving room for guesswork.
Enrico Betti [1] was the first to publish a proof of Galois’s result (Theorem 2 above) in 1853. However, as it is sometimes the case with Betti’s proofs (cf. [13]), his argument is not complete. In fact, he gives a sufficient condition for the existence of a subgroup of index in , and checks that this is not satisfied by primes greater than 11. Nevertheless, his intuition was right. The necessary condition we will derive in § 3 is quite similar to Betti’s sufficient condition. Here is the English translation (with my explanations in square brackets) of his theorem:
Theorem 3. (Betti [1 p. 88]) When the odd powers of a primitive root of the prime number , for which , satisfy either one or the other of the two congruences
then the group [= PSL] is the sum [i.e., union] of derived groups similar to joint letters [i.e., cosets of a subgroup].
Charles Hermite [8], [9] was intrigued to find explicit equations of degree that are arise as reductions of the modular equation of degree in the cases . (See also [7].)
The first correct proof of Theorem 2 can be found in Camille Jordan’s Traité des substitutions et des équations algébriques published in 1870. He works in the matrix group and using linear algebraic methods he shows that this group cannot have any subgroup of index , whenever . His general reasoning requires , so he has to add some ad hoc argument to deal with the case . (See [11 pp. 347–351].)
In his doctoral thesis [6] Felix Klein’s student, Joseph Gierster determined all subgroups of . Galois’s theorem immediately follows from this result. More generally, subgroups of the projective special linear groups over arbitrary finite fields of elements, have been described by Dickson [3] (see also [10]).
The proof
For preparation we first prove a simple property of the group of linear functions over , the -element field (where is a prime number).
Lemma. Any subgroup in the group of linear functions over the -element field whose order is not divisible by fixes an element.
Proof. Let be a subgroup in the group of linear functions , where , , and assume that does not divide . If , , then has order , so such a linear function cannot belong to . If , then has a unique fixed point . We have to show that if with and with both belong to , then they have the same fixed point, i.e., . Indeed, we have , and similarly , hence , where . Here , thus by the observation above we must have , that is , as it was required. □
Now let’s turn to the proof of Galois’s theorem. The constructions in the exceptional cases will be given in the next section.
Proof of Theorem 2. Let and let be a subgroup of index . We are going to show that . We have , hence . The order of the stabilizer divides both the order of and the order of . The latter is the group of linear functions of order , so . On the other hand, the index of in is at most the number of permuted elements (“letters” in Galois’s terminology), that is . Thus we obtain and , hence is transitive. From the Lemma above we infer that fixes some . Let be the translation . Then the conjugate subgroup fixes and . So, replacing by , we may assume without loss of generality that fixes as well, so .
Now we define an equivalence relation on by setting if . This relation is clearly -invariant, since it follows that , whenever and , so . We have seen that the equivalence class of is . If are different from and , then for any square we can apply the transformation , which is now an element of the subgroup , and obtain . By the transitivity of it follows that every class of has two elements, hence we will talk about partners of elements (linked letters — “lettres conjointes” — in Galois’s manuscript). The -invariance of means that is an imprimitive permutation group.
Following Galois, let us denote by the partner of . First we consider the case when is a square. Then implies , hence is the partner of , so , . (It can happen only for primes .) So , and then for every square , we have . Now the partners of squares are squares themselves, hence the partner of any non-square must be a non-square. If for non-squares and , then is a square, and it follows that by the same reasoning as above. Summarizing, we have obtained that if the partner of is a square, then the partner of every is .
Now we look at the case when is not a square. Then for every nonzero square we have that , so the partner of every nonzero square is the non-square .
Now knowing the equivalence relation (depending on , of course) we have to check whether the subgroup of consisting of those fractional linear transformations that preserve is indeed a transitive subgroup, as has to be. So must contain a permutation that maps to . Since preserves , we get that maps to . Thus the fractional linear transformation must have the form
with an such that is a square. Note that this must preserve the relation .
In the first case, when and the partner of every is , we must have for every that , i.e.,
hence . This equation can hold for at most two elements , therefore we must have , hence in this case.
In the second case, when the partner of every nonzero square is with a fixed non-square , we must have for every square that for some square . It means that either
or
That is, for every square either or holds. Each of these quadratic equations can have at most two roots in , so we obtain that the number of non-zero squares in is at most five, thus . □
Examples in the cases
We have seen that if a subgroup of index exists in then it is imprimitive, namely, it preserves a partition of into pairs. If we give an appropriate partition then it is a matter of checking if the set of permutations in that preserve the given partition is indeed a subgroup of index (or, equivalently, of order ). In fact, Galois gives the suitable partitions:
for : ();
for : ();
for : ().
(In the case it is not given explicitly, but since the pairs and are mentioned, the remaining two elements trivially yield the missing pair.)
In the last two cases he also writes down the elements of the subgroup in the form
where is a pair, and has the same quadratic character as . (Formally, it should have the same quadratic character as , as it was corrected in later editions of Galois’s works, but in the examples given for and it does not matter. Furthermore, the meaning of and must be clarified. Namely, these should be and , respectively, with being a square.)
Our discussion in Section 3 yields that if we choose as one class of the partition, then the other classes have either the form (and this can happen only for ) or for all nonzero squares with a fixed non-square . In the latter case every square must satisfy one of the quadratic equations or , furthermore, . This necessary condition allows the values or for , and or for . If some value of works then so does as well, because the transformation interchanges the two partitions, so if the subgroup preserving the partition corresponding to has the required order, then conjugating it by yields a subgroup preserving the partition corresponding to . So in each case () there is essentially one suitable partition, namely, the one found by Galois. (That is, all subgroups of index in are conjugate in the larger group .)
What has to be done is to show that the given fractional linear transformations form a subgroup. Their number is indeed , since in we can choose arbitrarily (in ways), then it determines its partner uniquely, and finally the quadratic character of should be the same as the one of , so we have choices for . Galois probably verified it by calculating compositions of such transformations. There are only very few traces of this in the extant manuscripts in the holdings of the Library of the Institut de France [5].
We give a concise verification using some geometric ideas. This approach is admittedly anachronistic, since for Galois a group always meant a group of permutations of the roots of an equation. The study of symmetry groups of geometric objects was a later development in group theory (see [15]).
Our goal is to exhibit a subgroup of order in for the primes , , and . So these subgroups should have order 12, 24, and 60, respectively. These are the orders of the group of rotational symmetries of the Platonic solids. In each case we will choose a -element orbit of the symmetry group and label its points by the elements of the projective line appropriately, so that the permutations induced by the rotations on the chosen orbit will correspond to elements of . The orbit we take in the case of the regular tetrahedron will consist of the six midpoints of the edges, for the cube and the icosahedron we take the set of vertices. The suitable labelings are shown in the figures below.
Figure 1.

Figure 2.
Figure 3.
To save calculations, it is enough to check that the permutations corresponding to a pair of generators of the rotation group are indeed fractional linear transformations. In the table below we give two rotations for each case. The axis of one of these rotations goes through two (opposite) points (labelled by 0 and ) of the orbit, and the order of this rotation is . The other rotation does not fix any point in the chosen orbit, its axis is , with the points and are marked in the figures. It is easy to see that the two given rotations generate a transitive permutation group on the orbit of size , and since the stabilizer of a point contains a rotation of order , the subgroup generated by the two rotations is indeed the full rotation group of order of the Platonic solid.
| polyhedron | axis | angle | permutation | function | |
| 5 | tetrahedron | ||||
| 7 | cube | ||||
| 11 | icosahedron | ||||
Table 1.
A different approach was taken by John H. Conway [2], who considered Galois’s partitions as products of transpositions for , etc.), and showed that each of them has exactly conjugates under the action of , so this set of involutions can be considered as the required objects on which the the group () acts.
In another related work, Bertram Kostant [12 p. 364] mentions that the set of vertices of the icosahedron may be identified with the projective line over the 11-element field in such a fashion that there exists a partition into six pairs of elements whose stabilizing subgroup of is . Unfortunately, he does not give the explicit identification in this paper, neither does he mention that this pairing can be found in the last letter of Galois.
Acknowledgements
In October, 2020 I have exchanged a few e-mails with Peter M. Neumann on this subject. His kind comments were helpful for my project. The nice figures of the three polyhedra were produced by Géza Kós.
References
- [1]Enrico Betti, Sopra l’abbassamento delle equazioni modulari delle funzioni ellittiche, Annali di Scienze matematiche e fisiche 4 (1853), 81–100.
- [2]John Horton Conway, Three lectures on exceptional groups, in: Finite Simple Groups (eds. M. B. Powell and G. Higman), Academic Press, London and New York, 1971. 215–247.
- [3]Leonard Eugene Dickson, Linear groups with an exposition of the Galois field theory, Teubner, Leipzig, 1901.
- [4]Harold M. Edwards, Galois for 21st-century readers, Notices of the AMS 59 (2012), 912–923.DOI
- [5]Évariste Galois, Manuscripts in the holdings of the Institut de France, http://www.bibliotheque-institutdefrance.fr/numerisation
- [6]Joseph Gierster, Die Untergruppen der Galois’schen Gruppe der Modulargleichungen für den Fall eines primzahligen Transformationsgrades, Mathematische Annalen 18 (1881), 319–365.
- [7]Catherine Goldstein, Charles Hermite’s stroll through the Galois fields, Revue d’histoire de mathématiques 17 (2011), 211–270.
- [8]Charles Hermite, Sur la théorie des équations modulaires, Comptes rendus hebdomadaires des séances des l’Académie des sciences 48 (1859), 940–947, 1079–1084, 1096–1102; 49 (1859), 16–24, 110–118, 141–144.
- [9]Œuvres de Charles Hermite, Tome II, Gauthier–Villars, Paris, 1908.
- [10]Bertram Huppert, Endliche Gruppen I, Springer-Verlag, Berlin–Heidelberg–New York–Tokyo, 1967.DOI
- [11]Camille Jordan, Traité des substitutions et des équations algébriques, Gauthier–Villars, Paris, 1870.
- [12]Bertram Kostant, The graph of the truncated icosahedron and the last letter of Galois, Notices of the AMS 42 (1995), 959–968.DOI
- [13]Pasquale Mammone, Sur l’apport d’Enrico Betti en théorie de Galois, Bolletino di Storia delle Scienze Matematiche 9 (1989), 143–169.
- [14]Peter M. Neumann, The mathematical writings of Évariste Galois, European Mathematical Society, Zürich, 2011.DOI
- [15]Hans Wussing, Die Genesis des abstrakten Gruppenbegriffes, VEB Deutscher Verlag der Wissenschaften, Berlin, 1969.

