Executive Summary
A linear group is a subgroup with . Nothing is assumed about itself — it may be infinite, torsion, or finitely generated and wild. What is assumed is that the representation is finite-dimensional, and that alone is enough to run the machinery of §7 and §8 against .
The engine is a single counting device. If the algebra spanned by is all of , then an element of is determined by of its traces; so a semigroup with only distinct traces has at most elements. Everything on this page — Burnside's two theorems, and Schur's solution of the General Burnside Problem for linear groups — is that observation plus an induction on .
Overview
Let be a field and a -vector space with . A subgroup is a linear group (or matrix group). The inclusion makes a -module, so arrives equipped with a faithful representation of dimension — and, crucially, with a finite-dimensional image algebra , whatever the cardinality of .
That is the whole point of the subject. The group algebra of an infinite group is an unmanageable object — it is never semisimple — but its image in is a finite-dimensional algebra, and finite-dimensional algebras are completely understood by Wedderburn–Artin theory. Statements about are therefore traded for statements about a small algebra.
Burnside's two problems ask whether torsion alone forces finiteness. For arbitrary groups the answer is no, spectacularly; for linear groups the answer is yes, and the proof is the material of this page and of Schur's Theorem on Torsion Linear Groups.
Learning Objectives
- Define a linear group and identify the algebra it spans.
- State the General Burnside Problem and the bounded-exponent version , and give the status of each.
- Prove the Trace Lemma : when is absolutely irreducible over and .
- Prove Burnside's First Theorem and locate the point where destroys it.
- State Burnside's Second Theorem : a linear group is finite iff it has finitely many conjugacy classes.
- Use and to reduce the General Burnside Problem for linear groups to the abelian-by-finite case.
Definitions
Let be a field and a finite-dimensional -vector space. A linear group is a subgroup . Choosing a basis identifies with , , so a linear group is a group of invertible matrices. Two linear groups are regarded as the same if they are conjugate in .
- Torsion (periodic)
- Every has finite order. No uniform bound is asserted.
- Exponent
- for all , with least. A group of finite exponent is torsion; the converse fails.
- -group
- Every element has order a power of . Not assumed finite.
- Locally finite
- Every finitely generated subgroup is finite. Locally finite torsion; the converse is exactly .
- The -subspace of spanned by the elements of . Because is closed under multiplication it is a -subalgebra — the image of .
- Absolutely irreducible
- is absolutely irreducible over when is onto; over an algebraically closed field this is the same as irreducible, by Burnside's Theorem .
Throughout, k is an arbitrary field unless a characteristic hypothesis is stated, and G is not assumed finite or finitely generated unless said so.
Core Concepts
Traces carry more information than they look like they do
The trace form on is nondegenerate in every characteristic: if for all , take to read off . So a matrix is determined by its pairings against any -basis of .
If spans we may take that basis *inside *. Then each is pinned down by the scalars — each of which is again a trace of an element of . A small trace set therefore forces a small group.
Why bounded exponent bounds the traces
If then the eigenvalues of over are th roots of unity, of which there are at most . A trace is a sum of eigenvalues, so has at most elements — a bound depending only on and , not on . Feeding into the Trace Lemma gives .
What to do when the action is reducible
Absolute irreducibility is essential to the Trace Lemma and cannot simply be assumed. If has a proper nonzero -submodule, choose a basis adapted to it: every becomes block upper triangular,
The diagonal blocks give homomorphisms onto linear groups of smaller degree; induction on handles them.
and the induction on reduces everything to the kernel of , which consists of matrices . That kernel is abelian, and controlling it is the entire remaining difficulty.
Key Results
Let be a finitely generated torsion group. Must be finite?
Both hypotheses are needed: is finitely generated and infinite, and the group of all complex roots of unity is torsion and infinite. For abelian the answer is yes, since a finitely generated abelian torsion group is finite.
Let be a finitely generated group of finite exponent . Must be finite? Equivalently: is the free Burnside group on generators of exponent finite?
Lam calls this the Restricted Burnside Problem. Modern usage reserves that name for a different question — see Failure Modes and Common Mistakes below.
Let be a field and let be a subsemigroup such that is an absolutely irreducible module over the semigroup algebra . If the trace function has finite image of cardinality , then
Absolute irreducibility means the algebra map is surjective . Its image is spanned by the elements of , so we may pick forming a -basis of . Define
This is -linear. It is injective: if then for every by linearity, and taking to be the matrix units gives every entry . Hence is a linear isomorphism and in particular injective on the subset .
For each coordinate is the trace of the element (here closure of under multiplication is used), so it takes at most values. Therefore .
Let be a field of characteristic and let be a subgroup of finite exponent . Assume (a vacuous condition when ). Then
No finite-generation hypothesis is needed. Thus the Bounded Burnside Problem has an affirmative answer for linear groups whenever the characteristic does not divide the exponent.
Enlarging to its algebraic closure changes neither nor , so assume . Induct on .
**Base .** Here and every satisfies , an equation with at most roots in a field. So .
Irreducible case. Suppose acts irreducibly on . Since is algebraically closed, is then absolutely irreducible over by Burnside's Theorem . Each satisfies , so its eigenvalues lie among the th roots of unity — at most of them — and , a sum of such, takes at most values. The Trace Lemma gives .
Reducible case. Choose a basis adapted to a proper nonzero submodule, so that each has the block form , and let be the group of blocks that occur, a linear group in of exponent dividing . By induction .
The map , , is a group homomorphism; its kernel consists of the matrices in . For such an element, induction on the exponent gives , so . As , the scalar is invertible in and . The homomorphism is therefore injective and
Let be an infinite field of characteristic and put
Then is abelian of exponent and is infinite. Here and the last step of the proof collapses: for every , so the kernel is everything.
A linear group over a field is finite iff it has only finitely many conjugacy classes.
One direction is trivial. Conversely, assume has finitely many conjugacy classes; again enlarge to (conjugacy classes are an invariant of as an abstract group, so nothing changes) and induct on . The trace is a class function, so is finite.
If acts irreducibly, the Trace Lemma applies directly and is finite. If not, use the block form . Each is a homomorphic image of , hence has finitely many conjugacy classes, hence is finite by induction.
Let , so . By below is abelian, so for every ; thus and every element of has a finite -conjugacy class. Since has only finitely many classes altogether, is covered by finitely many finite classes and is finite. Then .
For the block sizes fixed above,
so the kernel of is isomorphic to a subgroup of the additive group of matrices — abelian, and of exponent when . This one computation is what both and Schur's Theorem lean on.
Let be a finitely generated torsion group containing an abelian subgroup of finite index. Then is finite. (That is, has an affirmative answer for abelian-by-finite groups.)
Write and choose a finite generating set of containing these coset representatives and closed under inversion. For write with and .
Let be the subgroup generated by the finitely many . It is a finitely generated abelian torsion group, hence finite.
Claim: every word in the generators lies in . Induct on word length. Given with , write ; then because is abelian and . Hence is a union of finite sets.
Let be a field and a finitely generated torsion subgroup. Then has finite exponent.
The entries of finitely many generators and their inverses generate a subfield finitely generated over the prime field , so we may assume itself is finitely generated over . Choose a transcendence basis and set ; then is algebraic and finitely generated, hence finite, say . Viewing as a -space of dimension embeds in .
For let be its minimal polynomial, of degree at most . Since has finite order, divides , so all roots of are roots of unity and the coefficients of are algebraic over .
**Case .** The coefficients are algebraic integers lying in ; but is purely transcendental over , so its algebraic elements lie in , and . Each coefficient is an elementary symmetric function of at most roots of unity, hence bounded in absolute value by . Finitely many integer polynomials of bounded degree and bounded coefficients: finitely many possibilities for .
**Case .** The same argument places the coefficients in the algebraic closure of inside , which is ; so and there are again only finitely many polynomials of degree at most .
Finally determines the order of : it is the least with . So only finitely many orders occur and their lowest common multiple is an exponent for .
They are the two halves of Schur's Theorem. Lemma upgrades torsion to bounded exponent, which is what needs; Lemma repairs the one case cannot handle, namely , where the kernel of survives but is abelian of finite index. The assembly is carried out in Schur's Theorem on Torsion Linear Groups.
Proof Techniques and Method
How these proofs work, and which move to reuse.
Extend the field first
Finiteness of , its exponent, and its conjugacy classes are all unchanged by . Enlarging to an algebraically closed field converts irreducible into absolutely irreducible via Burnside's Theorem, which is what the Trace Lemma requires.
Dichotomy plus induction on n
Either the action is irreducible — apply the Trace Lemma and stop — or it is not, in which case a block triangular basis produces two linear groups of strictly smaller degree. Every theorem in this section has this shape.
Handle the unipotent kernel
The kernel of is always abelian . Kill it when ; when you cannot, exploit that it is abelian of finite index and invoke .
The pattern is worth naming: bound an invariant on the irreducible pieces, then glue along a controlled abelian kernel. It recurs verbatim in the proof of the Lie–Kolchin–Suprunenko theorem and in the construction of the unipotent radical.
Worked Example
Groups of exponent 2: the bound is true but wasteful
Let and let have exponent . Theorem predicts . The truth is far smaller. Every satisfies , and a group of exponent is abelian, since gives .
Because is separable in characteristic , each is diagonalisable with eigenvalues in , and a commuting family of diagonalisable operators is simultaneously diagonalisable. In a suitable basis therefore sits inside the diagonal matrices with entries :
For : the sharp bound is , Burnside's bound is . The theorem is a finiteness statement, not a counting statement.
The bound is attained by the full group of diagonal sign matrices, so nothing better is possible in general.
Checking the Trace Lemma on the quaternion group
Take , , and let be the quaternion group in its faithful -dimensional representation, generated by
This representation is irreducible, hence absolutely irreducible over . The traces are , , and for the six elements of order ; so . The Trace Lemma gives , and indeed .
Comparison and Classification
| Exponent | Answer | Due to |
|---|---|---|
| yes — the group is abelian | elementary | |
| yes | Burnside (1902) | |
| yes | Sanov (1940) | |
| yes | M. Hall (1958) | |
| open | — | |
| odd, | no | Novikov–Adjan (1968) |
| odd, | no | Adjan (1975) |
| large even | no | Ivanov, Lysënok (1990s) |
| any , linear with | yes, with a bound | Burnside |
| Needs f.g. | Needs bounded exponent | Needs char condition | Gives explicit bound | |
|---|---|---|---|---|
| Trace Lemma | no | no | no | yes |
| Burnside I | no | yes | yes | yes |
| Burnside II | no | no | no | no |
| Schur | yes | no | no | no |
Hypotheses of the three finiteness theorems for linear groups
Relationship Map
For a subgroup the implications run as follows; the dashed steps are exactly the theorems of this section.
- Torsion linear group — , every element of finite order
- and finitely generated
- bounded exponent, by
- finite, by Schur
- and of exponent with
- finite of order , by
- no finite generation needed
- and nothing else
- locally finite, by
- may be infinite:
- and finitely generated
Applications and Industry Use
Applications here means where this structure is used — inside mathematics and in the engineering and computing disciplines that consume it.
Deciding finiteness of a matrix group
Algorithms in GAP and Magma that test whether a matrix group given by generators over a number field or finite field is finite descend from these trace arguments: compute the trace set, or the order of the elements, and bound the group.
Bieberbach and point groups
Point groups of crystals are finite subgroups of . Finiteness of the torsion subgroups involved, and Jordan-type bounds on their index over an abelian normal subgroup, are what make the classification of space groups a finite computation.
Obstructions to linearity
Because a finitely generated torsion linear group is finite, Golod's infinite finitely generated -groups admit no faithful finite-dimensional representation over any field. This is the standard first proof that a group is not linear.
Finite matrix groups over finite fields
Groups of exponent prime to inside are automatically finite and of bounded order; this underwrites the search space arguments used when matrix groups are deployed as key spaces or automorphism groups of codes.
The honest summary is that this material is a bridge. Its consumers are the structure theory of infinite groups, the theory of algebraic groups, and the algorithms that implement both; it is rarely applied outside mathematics in its own right.
Failure Modes and Common Mistakes
- Do not read as a sharp count; for exponent the true bound is .
- Do not assume the eigenvalue count is exact when : in characteristic the polynomial has repeated roots and fewer than distinct ones — harmless for the bound, fatal for the kernel argument.
- Do not confuse linear group with algebraic group. A linear group is any abstract subgroup of , with no Zariski-closedness assumed.
- Do not expect to give a bound: it is a pure finiteness statement, with no function of and the number of classes extractable from the proof as given.
Historical Notes and Lessons Learned
- 1878Jordan's theoremJordan proves that a finite subgroup of GL_n(C) has an abelian normal subgroup of index bounded by a function of n alone — the first general finiteness constraint on linear groups.
- 1902Burnside asksBurnside poses the General Burnside Problem and settles exponent 3. His trace method for linear groups appears at the same time.
- 1911SchurSchur proves that a finitely generated torsion subgroup of GL_n(C) is finite, settling the General Burnside Problem for linear groups in characteristic zero.
- 1940–1958Small exponentsSanov settles exponent 4 and M. Hall exponent 6; both proofs are combinatorial and give no method for general N.
- 1964Golod–ShafarevichGolod constructs, for each prime p, an infinite finitely generated p-group. The General Burnside Problem is answered negatively, and the examples are necessarily non-linear.
- 1968Novikov–AdjanThe free Burnside group of exponent N is shown infinite for odd N at least 4381, later improved by Adjan to odd N at least 665; the even case follows in the 1990s through work of Ivanov and Lysënok.
- 1989–1991ZelmanovThe Restricted Burnside Problem in its modern form is answered affirmatively: there are only finitely many finite d-generator groups of any given exponent. Fields Medal, 1994.
The methodological lesson is sharp. Every positive result here is obtained by representing the group and counting inside a finite-dimensional algebra; every negative result is obtained by building a group that admits no such representation. The two Burnside problems are, in effect, questions about the limits of linearity.
Quick Reference
| Reference | Statement | Key hypothesis |
|---|---|---|
| General Burnside Problem | f.g. torsion | |
| Bounded Burnside Problem | f.g. of exponent | |
| Trace Lemma, | absolutely irreducible semigroup | |
| exponent , | ||
| finite finitely many classes | linear | |
| block kernel is abelian | block triangular form | |
| abelian-by-finite f.g. torsion is finite | abelian subgroup of finite index | |
| f.g. torsion linear bounded exponent | finitely generated |
Frequently Asked Questions
Why does the Trace Lemma insist on absolute irreducibility rather than irreducibility?
Because the proof needs to be all of , so that a basis of can be chosen inside and so that the nondegenerate trace form detects every matrix. Over a non-closed field an irreducible action may span a much smaller algebra — the rotation group acting on spans a -dimensional algebra — and then the injectivity argument fails outright. Over an algebraically closed field the two notions coincide, by Burnside's Theorem .
Where exactly does Burnside's First Theorem break when the characteristic divides the exponent?
Only in the reducible case, and only at the very last step. Everything up to and including the bound on the diagonal blocks survives. What fails is the injectivity of : an element of the kernel satisfies , which forces only when is invertible in . When the kernel can be an infinite elementary abelian -group, as shows.
Is the bound ever close to sharp?
No, and it is not meant to be. For exponent the correct bound is against a prediction of . The value of the theorem is that a bound exists depending only on and , so that the class of such groups is uniformly finite — a statement of the same character as Jordan's theorem, and equally non-quantitative in practice.
Do these results say anything about non-linear groups?
They say what such groups must avoid. Since a finitely generated torsion linear group is finite , any infinite finitely generated torsion group — Golod's -groups, the free Burnside groups of large exponent — has no faithful finite-dimensional representation over any field whatsoever. Non-linearity is often proved exactly this way, or via Mal'cev's theorem that finitely generated linear groups are residually finite.
Why is the semigroup version of the Trace Lemma worth stating?
Because inverses are never used: the proof only needs the products to lie back in . Stating it for semigroups makes it directly applicable to multiplicative sets of matrices arising as images of monoids, and to the intermediate sets produced inside inductions, without a separate argument.
Does Burnside's Second Theorem give a bound on ?
Not as proved. The argument shows is finite and is a finite union of finite classes, but the class sizes are not controlled by and the number of classes in any explicit way in this proof. Contrast , which is completely explicit.
References
- T. Y. Lam, A First Course in Noncommutative Rings, Graduate Texts in Mathematics 131, Springer-Verlag, 1991, §9 (pp. 149–162).
- B. A. F. Wehrfritz, Infinite Linear Groups, Ergebnisse der Mathematik 76, Springer-Verlag, 1973.
- S. I. Adjan, The Burnside Problem and Identities in Groups, Ergebnisse der Mathematik 95, Springer-Verlag, 1979.
- M. Vaughan-Lee, The Restricted Burnside Problem, 2nd edition, London Mathematical Society Monographs, Oxford University Press, 1993.
- E. S. Golod, “On nil-algebras and finitely approximable p-groups”, Izvestiya Akademii Nauk SSSR, Seriya Matematicheskaya 28 (1964), 273–276.
- C. W. Curtis and I. Reiner, Representation Theory of Finite Groups and Associative Algebras, Wiley-Interscience, 1962.
AI Suggested Questions
- Work through Golod's construction of an infinite finitely generated p-group and identify precisely where linearity would fail.
- How does Mal'cev's residual finiteness theorem for finitely generated linear groups compare with Schur's theorem as a non-linearity criterion?
- State Jordan's theorem with an explicit bound on the index, and explain how it strengthens Burnside's First Theorem in characteristic zero.
- What is known about the Bounded Burnside Problem for exponent 5, and why do the Novikov-Adjan methods not reach it?
- Explain how Zelmanov's solution of the Restricted Burnside Problem uses Lie algebra methods and the Hall-Higman reduction.
- Can the bound in the Trace Lemma be improved if G is assumed to be a group rather than a semigroup?
- Give an example of an infinite linear group with finitely many conjugacy classes over a field that is not algebraically closed, or prove none exists.
- How do algorithms in GAP or Magma decide finiteness of a matrix group over a number field, and what is their complexity?
