A representation describes a group through linear transformations. The defining equation is short, but its consequences depend on three pieces of data: the group, the scalar field, and the vector space. Keeping those data visible prevents two common mistakes: assuming every invariant subspace has an invariant complement, and assuming an irreducible action has only scalar commuting operators over every field.
Motivation
Suppose a group permutes a finite collection of objects. Place one basis vector at each object and let every permutation move the basis vectors accordingly. The resulting matrices encode the action linearly. Their entries can be considered over the real numbers, the complex numbers, or a finite field. Although the permutations are unchanged, decompositions of the resulting vector space can differ dramatically with the field.
Linear algebra suggests simplifying simultaneous transformations by finding invariant subspaces and splitting the space into smaller pieces. A complement chosen merely as a vector-space complement need not be invariant. If we project onto a chosen invariant subspace, conjugating the projection by group elements gives many other projections. Averaging them may produce a projection that commutes with the entire action. Whether that average exists depends on being able to divide by the order of the group.
Modules make this strategy systematic. The group algebra packages every linear combination of group operations into one ring, so an invariant subspace becomes a submodule. Direct-sum decomposition then means decomposition into simpler modules. Maschke's theorem explains when this process always succeeds for finite-dimensional representations; Schur's lemma describes maps between the simplest pieces.
Definitions
Definition 4.1
(Finite-dimensional representation)
Let be a group and a field. A finite-dimensional representation is a homomorphism , where is a finite-dimensional -vector space. Thus and . The zero-dimensional representation is allowed, but it will not be called irreducible.
Definition 4.2
(Group algebra and modules)
The group algebra consists of finite formal sums with coefficients in , with multiplication defined by distributivity and the group product. A representation gives an -module by . Conversely an -module is a representation after restricting its scalar action to group elements. These constructions recover one another.
Definition 4.3
(Invariant subspaces, irreducibility and complete reducibility)
An invariant subspace satisfies for every . In fact equality holds, since also acts. A nonzero representation is irreducible if its only invariant subspaces are zero and itself. A representation is completely reducible if it is a direct sum of irreducible subrepresentations. In this chapter all decompositions are finite because the vector spaces are finite-dimensional.
Definition 4.4
(Intertwiners and equivalence)
An intertwiner is linear and satisfies for every . Write for the vector space of such maps and when source and target agree. Equivalent representations are connected by an invertible intertwiner; a change of basis is a particular way to describe the same action, not a new representation type.
Theorem/proposition
Theorem 4.5
(Maschke's theorem)
Let be finite and assume its order is nonzero in , equivalently does not divide . Every invariant subspace of a finite-dimensional -representation has an invariant complement. Consequently every finite-dimensional representation is completely reducible.
Theorem 4.6
(Schur's lemma and its scalar form)
For irreducible representations over any field, a nonzero intertwiner is an isomorphism. Thus is a division ring. If is algebraically closed and is finite-dimensional, then .
The scalar conclusion is stronger than the division-ring conclusion and needs its field hypothesis. Neither part of Schur's lemma needs finite. Maschke's averaging does require a finite group and an invertible group order. These hypotheses belong to different arguments and should not be merged into a single vague assumption that the setting is “nice.”
Concept lensGeometric
Complete reducibility through an invariant inner product
Over , averaging can be applied to an inner product instead of a projection. Starting with a positive definite Hermitian inner product, set . The sum remains positive definite and is invariant by reindexing, so a finite complex representation can be made unitary. The orthogonal complement of an invariant subspace is then invariant: if and , then . This describes the complements of Maschke's theorem geometrically over , but it does not replace the ring-theoretic proof over arbitrary fields: positive definite Hermitian geometry is additional structure special to that scalar setting.
Proof sketch or proof idea
Proof
(Averaging a projection)
Let be invariant. Ordinary finite-dimensional linear algebra gives a linear projection restricting to the identity on . It need not commute with . Define
Each summand maps into because is invariant. For , it sends to , so also restricts to the identity on . Conjugation by permutes the terms by , so commutes with every group element. Since its image lies in and its restriction to is the identity, .
Every decomposes as . The first term lies in , the second in , and the intersection is zero. Equivariance makes the kernel invariant. This proves the complement assertion. If is nonzero, choose a nonzero invariant subspace of least dimension; it is irreducible. Split it off and induct on dimension to obtain a direct sum of irreducibles. Termination follows from strictly decreasing finite dimension, not from an assumption that all modules have finite composition length.
Proof X-ray
What averaging repairs and what it preserves
Conjugation makes the projection vary, but it preserves its image containment in the invariant subspace and its identity restriction there. The average therefore retains those two properties while gaining equivariance by permutation of the summands. The hypothesis on is used exactly once: the sum of the conjugates restricts to on , and only division by turns this into the identity. Averaging arbitrary idempotents would not guarantee an idempotent average: the common image and common restriction are what close this proof.
Schur's lemma needs neither averaging nor any hypothesis on the group order.
Proof
(Kernels, images and eigenvalues)
For Schur's lemma, the kernel and image of an intertwiner are invariant subspaces. A nonzero map between irreducibles has nonzero image, hence surjective, and a kernel smaller than the entire source, hence zero. Thus it is an isomorphism. A nonzero equivariant endomorphism has an equivariant inverse: rearrange after multiplying by . This proves the division-ring assertion.
If is algebraically closed and finite-dimensional and nonzero, any linear endomorphism has an eigenvalue . Then is equivariant and has a nonzero kernel. Schur's first part forces it to be zero. Therefore . Finite dimension supplies an eigenvalue through the characteristic polynomial, and algebraic closure ensures that polynomial has a root in the scalar field.
Counterexample mode
Maschke fails when the characteristic divides the group order
Claim: every invariant subspace of a finite-dimensional representation of a finite group has an invariant complement, whatever the field. Take and , and let act on by left multiplication. The augmentation is an equivariant surjection onto the trivial representation, with invariant kernel . A complement would be an invariant line, on which acts by a scalar of square one, that is by ; so it would be spanned by a fixed vector. But is fixed only when , so the only fixed line is the kernel itself. In the basis the generator acts by , the modular example below with . The argument generalizes whenever divides : a splitting of sends to a left-invariant element , whose augmentation contradicts . The repair is exactly the hypothesis of Maschke's theorem. This does not claim that every representation in that characteristic is nonsplit, only that the universal complement property fails.
Worked examples
Worked example 4.7
(The standard representation of the symmetric group)
Let permute coordinates of . The line is invariant, and so is . They intersect trivially and have dimensions one and two, giving . To prove irreducible, suppose an invariant line exists. A vector on it must be an eigenvector for the transposition with eigenvalue or . In the negative case it is a multiple of , whose image under is not in the same line. In the positive case it is a multiple of ; swapping the last two coordinates again leaves that line only for the zero vector. Both possibilities contradict invariance. Since a proper nonzero subspace of a two-dimensional space is a line, is irreducible.
Worked example 4.8
(An irreducible real representation with nonscalar endomorphisms)
Let a generator of act on by rotation through . The rotation has no real eigenvalue, so it has no invariant real line and the representation is irreducible. The quarter-turn matrix commutes with every plane rotation and is not a real scalar. In fact the commuting matrices are , a copy of as a real division algebra. This does not contradict Schur's lemma: it illustrates its division-ring form. After extending scalars to , the rotation has two eigenlines with eigenvalues and , and is reducible there.
Worked example 4.9
(A modular representation with an invariant line but no complement)
Let have characteristic and let act on by . Since with , we have , so this defines a representation. The line is invariant. If it had an invariant complementary line, the induced action on that line would have an eigenvector independent of . But , so there is no such vector. The matrix is not diagonalizable even though the group is finite. The failure occurs exactly where division by in the averaging formula is impossible.
Common mistakes
Common mistake
(Abelian does not force one-dimensional irreducibles over every field)
“Abelian group” does not imply “every irreducible representation is one-dimensional” over every field. Over an algebraically closed field, for a finite-dimensional irreducible representation of an abelian group, each group action commutes with the representation and hence is scalar by Schur. Every line is then invariant, forcing dimension one. The real rotation example shows precisely why the field hypothesis is necessary.
Common mistake
(Commuting with one element is not enough)
A matrix commuting with one selected group element need not be an intertwiner for the whole representation unless that element generates the group. Likewise, a subspace stable under one matrix need not be stable under every group matrix. A generating set suffices, provided stability under inverses is justified; for an invertible operator preserving a finite-dimensional subspace, restriction is injective and hence surjective, which supplies inverse stability.
Common mistake
(An invariant complement exists but is not unique)
Complete reducibility means an invariant complement exists; it does not say a given vector-space complement is invariant or that the complement is unique. The average constructs one from an initial choice. Different choices may give different complements: for the trivial action on , every line is invariant, so when is infinite the line has infinitely many invariant complements. Decomposition into isomorphism types will be studied by characters, which avoid depending on a particular choice of invariant basis.
Summary
A finite group representation is a module over its group algebra. Maschke's theorem uses an invertible group order to average an arbitrary projection into an equivariant one. Schur's lemma follows from kernel and image invariance; the scalar version additionally needs finite dimension and an algebraically closed field. These two theorems separate questions of existence of invariant decompositions from questions about maps between irreducible pieces.
Exercises
Exercise 4.10
(Let be finite and algebraically closed with characteristic not dividing . Prove that every irreducible representation of an abelian is one-dimensional.)
FoundationProof
Solution(3)
- 1
Each group element acts by an intertwiner
For all , , so .
Because: is abelian.
- 2
Apply the scalar form of Schur's lemma
Key step
Each equals for some .
Because: is algebraically closed and is finite-dimensional and irreducible.
- 3
Every line is invariant
∎Scalars preserve every line, and irreducibility forces the nonzero space to have dimension one. The characteristic assumption was not used: this argument needs only algebraic closure and finite dimension.
Exercise 4.11
(For a complex representation of a finite group, prove that is a projection onto the fixed subspace with invariant kernel.)
CoreProof
Solution(3)
- 1
The image consists of fixed vectors
Reindexing gives and likewise for every . The first identity puts the image in .
Because: Left and right multiplication by permute .
- 2
Fixed vectors are fixed by
If , every summand equals , so .
- 3
Conclude
∎The image is exactly , and because is the identity on its image. If , then , so the kernel is an invariant complement. No basis of fixed vectors has to be guessed in advance.
Exercise 4.12
(In the modular example with , list the invariant lines over and explain why the representation is reducible but not completely reducible.)
Foundation
Solution(3)
- 1
List the lines
Over the nonzero vectors of are , and , so there are exactly three lines.
- 2
Apply the generator
Key step
, and . The last two lines are interchanged, so only is invariant.
- 3
Conclude
The proper invariant subspace makes the representation reducible. A decomposition into irreducibles in dimension two would need two invariant lines, so none exists.
Exercise 4.13
(If and are inequivalent irreducible complex representations, determine . If they are equivalent, determine its dimension.)
Core
Solution(3)
- 1
The inequivalent case
By Schur's lemma a nonzero intertwiner would be an isomorphism, so every intertwiner is zero and .
- 2
Reduce the equivalent case to endomorphisms
Key step
Choose an isomorphism . For any intertwiner , the endomorphism lies in , so .
Because: The scalar form of Schur's lemma applies over .
- 3
Conclude
is one-dimensional. The chosen is not canonical, but its existence determines the dimension.