Motivation
A representation describes how vectors move. A linear functional measures a vector, so its transformation law should preserve the measurement when both the vector and functional are moved together. This requirement explains the inverse in the dual action. Without that inverse, familiar matrix formulas reverse multiplication and fail to define the intended representation.
Pairings then turn questions about vectors into questions about maps. An invariant bilinear form gives an equivariant map into a dual representation; a perfect form makes that map an isomorphism. This supplies a concrete way to recognize self-duality and connects matrix equations, invariant tensors, and Schur's lemma.
Throughout, is a finite group, is a field, and representations are finite-dimensional left -modules. The constructions do not require to be invertible in . We use the tensor universal property and intertwiner equations.
Definitions
Definition 4.2.1
(Dual representation)
For a representation , its vector-space dual is . Define the dual representation by
The field on the right carries the trivial action. This definition makes evaluation invariant: . If is the matrix of in a basis of , the matrix in the dual basis is . The transpose alone is not the dual action.
Definition 4.2.2
(Tensor and Hom representations)
The tensor representation has diagonal action . The Hom representation has underlying vector space and action
Its fixed subspace is , the space of intertwiners. Taking all linear maps before taking fixed points is essential: the Hom representation contains maps that are not equivariant.
Definition 4.2.3
(Invariant, nondegenerate, symmetric and alternating forms)
A bilinear form is invariant if for every . Its left radical consists of with for every ; its right radical is defined with the two slots reversed. The form is nondegenerate, or perfect, if both radicals are zero. It is symmetric if , and alternating if for every . Alternating implies by expanding . The converse needs characteristic different from two.
These are bilinear forms. A complex Hermitian form is conjugate-linear in one slot and belongs to a different construction. A representation can preserve a positive Hermitian form while admitting no nonzero invariant complex bilinear form.
Theorem/proposition
Theorem 4.2.4
(Duality preserves exactness and irreducibility)
The dual action and diagonal tensor action define representations. For an exact sequence of finite-dimensional representations,
precomposition gives an exact sequence
The image of is the annihilator . The evaluation map is an equivariant isomorphism. A nonzero is irreducible if and only if is irreducible.
Theorem 4.2.5
(Invariant tensors and invariant pairings)
There is an equivariant isomorphism
Consequently . Sending to identifies invariant bilinear forms on with . Such a form is nondegenerate exactly when is an isomorphism. In particular, admits a perfect invariant bilinear form exactly when .
Concept lensAlgebraic
Invariant forms as a matrix equation
Choose a basis and write with Gram matrix . The functional has coefficient column , so has matrix . Invariance of is the equation
and it suffices to impose it for generators of . Transposing and multiplying on the left by turns it into , which is exactly the intertwiner equation from to in the dual basis. The correspondence of the theorem is therefore a finite linear system in the entries of , and perfectness is the condition .
The last statement of the theorem becomes much sharper when is irreducible.
Proposition 4.2.6
(The symmetry of an irreducible self-dual representation)
Suppose additionally that is algebraically closed and . If is nonzero, irreducible, and isomorphic to , its invariant bilinear forms form a one-dimensional space. Every nonzero such form is nondegenerate, and all nonzero forms are symmetric or all are alternating. No semisimplicity hypothesis is needed.
Proof sketch or proof idea
Establishing the actions before using their matrices
Proof
(Duality preserves exactness and irreducibility)
For ,
The identity acts identically, and each acts linearly with inverse . For the tensor action, is bilinear, so the universal property gives a linear operator. Its composites agree with the required group multiplication on elementary tensors, which span the tensor product. Thus the action is well defined on sums as well as elementary tensors.
Write a functional as a column of coefficients , so that its value on a coordinate column is . Then
This derives the dual matrix. Merely transposing would give , with the factors reversed. The inverse reverses the order a second time and restores the left action.
If is equivariant, then is equivariant: evaluated at , both and equal . For an invariant subspace , its annihilator is invariant because . The dual exact sequence now follows concretely. The map is injective since is onto; its image consists exactly of functionals vanishing on , by quotient factorization. Every functional on extends to by extending a basis, so is onto.
That last extension need not be an invariant functional. Surjectivity is a statement about the underlying vector spaces; the maps in the sequence are equivariant. It does not assert that taking fixed vectors preserves surjections.
Evaluation is injective because functionals separate vectors, and is onto by finite dimension. Also
so it is equivariant. If is irreducible and were a nonzero proper invariant subspace of , its annihilator in would be invariant and have dimension , strictly between zero and . This contradicts irreducibility. Applying the same argument to and using proves the converse.
Turning tensors into maps
Proof
(Tensors correspond to linear maps)
The rule defining is bilinear in , so it induces a linear map. Given a basis and dual basis , define
The equality gives . Conversely, gives . These tensors span, so the maps are inverses. Although the inverse was written using a basis, uniqueness of the inverse to the basis-free proves its independence of that choice.
For equivariance, evaluate at :
Hence taking fixed subspaces gives exactly the stated intertwiner correspondence. For , the tensor is basis independent and invariant because it corresponds to the identity map. Its summands need not individually be fixed.
Turning forms into dual maps
Proof
(Invariant forms correspond to maps into the dual)
Bilinearity makes linear. Invariance is equivalent to
This proves both directions of the correspondence. Its kernel is the left radical. If that kernel is zero, equal finite dimensions make onto; for a nonzero , some functional in its image does not vanish at , so the right radical is also zero. Conversely a nondegenerate form has zero kernel and therefore gives an isomorphism.
Proof
(Symmetry type of an irreducible self-dual representation)
For an irreducible self-dual , choose an equivariant isomorphism . Schur's scalar lemma gives , and every nonzero member is an isomorphism. If is a nonzero invariant form, its transpose is invariant too, so . Transposing twice yields . In characteristic different from two, gives symmetry; gives and hence alternation. A nonzero form cannot have both types in that characteristic. This argument uses irreducibility and algebraic closure, not Maschke averaging.
Proof X-ray
Where each hypothesis enters the symmetry argument
The proof uses its hypotheses one at a time. Self-duality supplies a single isomorphism , so invariant forms exist at all. Irreducibility and algebraic closure enter together and only once: through , the space is identified with , which Schur's lemma makes equal to . The key move is that transposition is a linear map of that one-dimensional space whose square is the identity, so it can only be multiplication by . The characteristic hypothesis enters last: makes the two cases exclusive and turns into . Dropping algebraic closure enlarges the space so that transposition can have both eigenvalues on it, as the counterexample under Common mistakes shows; dropping irreducibility allows degenerate nonzero forms, as in the exercises below.
Worked examples
Worked example 4.2.7
(A one-dimensional representation need not be self-dual)
Over , let a generator of act on by a primitive cube root . Its dual acts by , so and are inequivalent. Any bilinear form has . Invariance forces , hence . Thus has no perfect invariant bilinear form. Nevertheless the Hermitian form is invariant because .
The evaluation pairing between and is perfect and invariant, and is trivial. There is no contradiction: a perfect pairing between two different representations is not a self-pairing of .
Worked example 4.2.8
(The S3 plane and a characteristic-dependent radical)
Let permute coordinates in and let . The dot product is invariant. In the basis , , its restriction has Gram matrix
It is perfect when , even in characteristic two. In characteristic three, solving gives . The radical is therefore . This line belongs to precisely because and is fixed by every permutation. The ambient dot product is still perfect on ; its restriction to an invariant plane has become degenerate. A restriction must be checked rather than assumed perfect.
Worked example 4.2.9
(Self-duality survives a nonsplit modular extension)
Let and let the generator of act on by
Here and direct multiplication gives . Thus is invariant and perfect. It is alternating because , and also symmetric, as can occur in characteristic two.
The invariant line satisfies . It has no invariant complement. Indeed a complementary line contains with ; invariance would put in that same line, contradicting complementarity. Therefore does not imply complete reducibility. Orthogonality fails to split this extension because the restricted form on is zero.
Common mistakes
The tensor product in this chapter is over , with simultaneous group action. It is not a tensor product of two left modules over . The latter expression lacks the right-module input needed for balancing over a generally noncommutative group algebra.
Counterexample mode
Algebraic closure cannot be dropped from the symmetry proposition
Claim: over any field of characteristic different from two, an irreducible self-dual representation has a one-dimensional space of invariant bilinear forms. Over , let a generator of act on by rotation through :
Its characteristic polynomial has no real root, so there is no invariant line and the representation is irreducible. Since , the equation is equivalent to , that is to , whose solutions are . The invariant forms therefore make up a two-dimensional space. The dot product is perfect, so ; yet gives an invariant form that is neither symmetric nor alternating, because . The repair is the hypothesis used in the proof: Schur's lemma gives only over an algebraically closed field, whereas here .
Common mistake
(A nonzero invariant form need not be perfect)
Self-duality requires an invertible solution of . A nonzero solution may be singular when is reducible: for the diagonal action of in the exercises, is such a solution. Likewise a representation and its dual always have the same dimension, but need not be isomorphic.
Common mistake
(A perfect form does not automatically provide complements)
For an invariant and an invariant form, is invariant, because . Even if the form on is perfect, can be nonzero; the modular example has . An orthogonal decomposition requires the restricted form on to be nondegenerate, which is a different condition from nondegeneracy on the whole space.
Common mistake
(Duality does not conjugate coefficients)
Complex conjugate coefficients are not part of algebraic duality. For finite complex representations one obtains , but the last equality uses finite-order eigenvalues. The defining formula remains precomposition by the inverse, over every field.
Summary
Evaluation determines the inverse dual action. Duality reverses exact sequences while preserving their equivariant maps, and finite-dimensional double duality preserves the original representation. Tensor–Hom correspondence turns invariant tensors into intertwiners. Invariant forms become maps to the dual, making nondegeneracy an isomorphism test. Their behavior on subspaces still requires separate attention, especially in modular characteristic.
Exercises
Exercise 4.2.10
(A generator of acts on by the matrix below. Compute its dual matrix and verify that the dual matrix has cube .)
Foundation
Solution(2)
- 1
Show that over every field
The trace of is and its determinant is , so Cayley–Hamilton gives and hence . This holds in every characteristic, including three; it does not assert that the action is diagonalizable there. Consequently
- 2
Transpose the inverse
Exercise 4.2.11
(Without computing a change-of-basis matrix, explain why does not depend on the basis and is fixed by .)
Core
Solution(3)
- 1
Identify the tensor under
For every , . Every basis therefore produces a tensor whose image is .
- 2
Deduce basis independence
Key step
Two bases give tensors with the same image , so the tensors are equal.
Because: is defined without a basis and is injective.
- 3
Deduce invariance
The identity is fixed by the Hom action: . For the tensor , equivariance gives , and injectivity gives .
Exercise 4.2.12
(Suppose and acts on by . Find every invariant bilinear form and decide which are perfect.)
Core
Solution(3)
- 1
Write the invariance equation for the generator
For ,
Because: Invariance under a generator implies invariance under the whole group.
- 2
Solve the linear system
Key step
The equation gives and , hence , while and are arbitrary.
Because: is invertible in .
- 3
Decide perfectness
The form is perfect exactly when . In particular is a nonzero degenerate invariant form; irreducibility is what excludes this behavior in the symmetry proposition.
Exercise 4.2.13
(In the characteristic-three example, prove that descends to a perfect invariant form on , and compute the value of the descended form on the class of paired with itself.)
CoreProof
Solution(4)
- 1
The form is well defined on classes
If lie in the radical , then , so changing either representative leaves the value unchanged.
Because: The form is symmetric, so the radical line is killed in either slot.
- 2
Invariance descends
Every permutation fixes , so acts on the quotient, and .
- 3
The descended form is perfect
Key step
If a class pairs to zero with every class, its representative pairs to zero with all of . It lies in the radical, so the class is zero.
- 4
Compute the self-pairing
∎The vector is not a multiple of , so its class spans the one-dimensional quotient. Its self-pairing is in characteristic three.
Exercise 4.2.14
(Over , let the two standard generators of act by the matrices and below. Prove irreducibility and exhibit a perfect invariant alternating form. Can a nonzero invariant symmetric bilinear form exist?)
Synthesis
Solution(4)
- 1
Check the defining relations
The matrices satisfy and , so . They therefore give the stated action.
- 2
Prove irreducibility
Key step
A proper nonzero invariant subspace would be a line. Since has the distinct eigenvalues , that line must be or , but and , so interchanges these lines.
Because: An invariant line must be an eigenline of .
- 3
Exhibit an invariant alternating form
Let . For a two-by-two matrix , direct expansion gives . Both generators have determinant one, so defines an invariant alternating form, perfect because .
- 4
Rule out symmetric forms
The invariant forms are the multiples of , all alternating. A form that is both symmetric and alternating satisfies , so it is zero in characteristic zero. Hence the only invariant symmetric bilinear form is zero.
Because: The symmetry proposition applies: is algebraically closed and the representation is irreducible and self-dual.