An invariant decomposition can involve many arbitrary choices of bases and complements. A character removes those choices by keeping the trace of each group action. Remarkably, over the complex numbers for a finite group, that smaller collection of numbers determines the representation up to equivalence. The proof connects averaging, Schur's lemma, and the trace of a projection.
Motivation
Suppose two matrix descriptions look different. Finding a simultaneous change of basis can require solving many equations. Traces give an easier first test: conjugate matrices have the same trace, so equivalent representations have identical traces at every group element. For a single arbitrary matrix, equal traces are far from enough to prove similarity. Why should an entire finite-group representation be determined by them?
The answer depends on complete reducibility. Every complex representation of a finite group is a direct sum of irreducibles, and traces add across direct sums. If the characters of inequivalent irreducibles behave like orthonormal vectors, taking an inner product extracts exactly the multiplicity of each constituent. This turns decomposition into arithmetic while keeping the structural reason for that arithmetic visible.
We restrict to finite groups and finite-dimensional complex representations throughout this chapter. Modular character theory is a different subject: if the characteristic divides the group order, averaging may fail and ordinary traces do not recover all extension data. No conclusion here should be transferred to that setting without further hypotheses.
Definitions
Definition 5.1
(Character, class function and inner product)
For a representation of , its character is . Trace is invariant under conjugation, so . A function constant on conjugacy classes is called a class function. The space of complex class functions has dimension equal to the number of conjugacy classes: the indicator functions of those classes form a basis.
Define a Hermitian inner product, linear in its first argument, by
One may sum over conjugacy classes instead, weighting each value by the size of its class. Omitting those weights changes the inner product and generally destroys orthogonality.
Finite group elements act with finite order. If has order , then , and its minimal polynomial divides . Over this polynomial has distinct roots, all roots of unity, so is diagonalizable and its eigenvalues have absolute value one. Therefore .
The character of is , because block diagonal traces add. The character of the tensor product, with action , is : the trace of a tensor product matrix is the product of the two traces, as one sees from diagonal entries in a product basis. The character of the dual representation is the complex conjugate character. At the identity, , giving a useful first consistency check.
Theorem/proposition
Theorem 5.2
(Character inner products count intertwiners)
For finite-dimensional complex representations of a finite group,
In particular, characters of inequivalent irreducibles are orthogonal and each irreducible character has norm one.
Theorem 5.3
(Multiplicity and completeness)
If represent all irreducible complex representations of , their characters form an orthonormal basis of class functions. For every representation , the multiplicity of in is . There are exactly as many irreducible types as conjugacy classes, and
The multiplicities are nonnegative integers, not merely numbers that happen to be close to integers in numerical calculations. The theorem also gives an irreducibility test: a nonzero representation with multiplicities has , so it is irreducible exactly when its character has squared norm one. The word representation matters, as the next example shows.
Counterexample mode
Norm one does not make a class function a character
Claim: every class function of norm one on is an irreducible character. Let take the values on the identity, the transpositions and the three-cycles. Then
yet , and for the standard character . Indeed , whose squared coefficients add to one. These coefficients are not integers, so is not even a difference of characters. The test is correct for characters of genuine representations: there the coefficients are nonnegative integers, and forces exactly one .
Proof sketch or proof idea
Proof
(Averaging on Hom, the regular representation and class sums)
Let , with group action . A fixed element is precisely an intertwiner. Averaging gives a projection
onto . The trace of a projection is its image dimension, since its minimal polynomial divides and its only eigenvalues are zero and one.
For fixed matrices on and on , the operator on has trace . To check this, use the matrix units sending the th source basis vector to the th target basis vector. The coefficient of in is , and summing over gives the product. Consequently
Schur's lemma now gives zero for inequivalent irreducibles and one for equivalent irreducibles. Maschke's theorem writes any representation as a sum of irreducibles, and additivity of characters extracts its multiplicities by the stated inner products. This already proves that equal characters imply equivalence, provided the list of irreducible types is exhaustive.
To see that there are finitely many types and find all of them in one representation, use the left regular representation . Every irreducible is a quotient of it: for a nonzero vector , the map has nonzero invariant image, hence is surjective. Maschke splits that quotient, so every irreducible occurs in the finite regular decomposition. Its character is at the identity and zero elsewhere, because left multiplication by a nonidentity group element fixes no basis element of . The multiplicity formula therefore gives multiplicity for . Taking dimensions yields the sum-of-squares identity.
Finally let be the number of conjugacy classes. Orthogonality gives . The center has dimension , since a central group-algebra element has coefficients constant on conjugacy classes, and each class sum is central. Each central element acts as a scalar on every irreducible by Schur's lemma. The resulting linear map is injective: if the element acts as zero on all irreducibles, it acts as zero on their direct sum decomposition of the regular representation, and hence is itself zero by applying it to the identity basis element. Thus . Equality proves the basis assertion and establishes that the character table has no missing rows.
Proof X-ray
Which ingredient does which job
The proof uses three ingredients for three different conclusions. Division by in is where the complex setting enters: it makes the image of a projection, so a dimension becomes a trace. The identity turns that trace into a sum of character values, and , proved from finite order in the definitions, turns the sum into the Hermitian inner product. Irreducibility enters only through Schur's lemma, which pins the dimension to zero or one. Completeness is a separate step: orthogonality alone gives , and the reverse inequality needs the class sums in the center, which act by scalars and are detected faithfully on the regular representation.
Worked examples
Worked example 5.4
(The full character table of S3)
The conjugacy classes are the identity, the three transpositions, and the two three-cycles. The trivial representation has character and the sign representation has character . The permutation representation on three letters has character , counting fixed letters. Removing its constant line gives the standard character . Its squared norm is , proving irreducibility independently of the invariant-line calculation. The degrees satisfy , and there are three conjugacy classes, so these three rows are exhaustive.
| Representation | Identity (size 1) | Transposition (size 3) | Three-cycle (size 2) |
|---|---|---|---|
| Trivial | 1 | 1 | 1 |
| Sign | 1 | -1 | 1 |
| Standard | 2 | 0 | -1 |
Worked example 5.5
(A tensor square decomposed without choosing invariant bases)
Let be the standard representation. Its tensor square has character by pointwise multiplication. Inner products with the three irreducible rows are , , and . Therefore . The dimensions add to four as required. The alternating square is one-dimensional with character , which is sign; hence the symmetric square is . The character calculation finds types and multiplicities, though it does not itself supply explicit embeddings of the summands.
Constructing the tensor summands explicitly
The preceding calculation proves an isomorphism type. To use a particular tensor, we also need actual subspaces and maps. Work in
Our convention is . The generators and act by
These four formulas suffice to check equivariance: generate , and commuting with their actions implies commuting with every product and inverse. On the tensor square use the diagonal action .
The swap commutes with this action. Since our field is , the operators and are complementary projections. Their images are the symmetric and alternating tensor subspaces. We realize as the symmetric subspace; its usual quotient definition is isomorphic to this realization by symmetrization. Put
Then form a basis of the symmetric subspace, and spans the alternating subspace. Substitution in the generator formulas gives and , so is the sign representation. No character table is needed for this identification.
The trivial line and the second standard summand
Define
For example , and . Likewise , and . Consequently
Thus is trivial, and , is an equivariant isomorphism from onto . For injectivity, the coefficient of is and its coefficient is ; both vanish only when .
We must also prove that these subspaces form a direct sum. The bilinear form is preserved by coordinate permutations, so its associated linear contraction
is equivariant with trivial target. Here , , ; hence and . On the three-dimensional symmetric subspace, is nonzero and its kernel therefore has dimension two. That kernel equals . Since is outside the kernel, the symmetric subspace is . Including the alternating subspace proves the concrete decomposition
The contraction is complex bilinear, with no conjugates; the argument uses neither positivity nor orthonormality of .
Proof X-ray
From a multiplicity to an embedded summand
The character calculation predicts the types and dimensions. The vectors locate them. Matching the generator actions proves the type of each embedded subspace; the swap and contraction prove the sum is direct and exhaustive. Merely finding four nonzero vectors, or matching their total dimension, would leave those logical obligations open. If an irreducible occurs repeatedly, its individual copies can admit different embeddings; characters determine their total multiplicity without selecting a preferred copy.
Reading the components of a tensor
For an arbitrary tensor
write . Contraction gives ; the alternating projection gives . Comparing the remaining and coefficients yields
Each component map is equivariant because it projects along the other invariant summands. Thus is invariant and transforms by sign, whereas transforms by the same two-dimensional matrices as the coefficients of .
Worked example 5.6
(A simple tensor can have three nonzero components)
Taking and gives
Expanding the right-hand side cancels its , and coefficients, leaving . Being a simple tensor does not place it inside a single irreducible summand.
Worked example 5.7
(Counting orbits with a permutation character)
Let act on a finite set , and let have basis indexed by . Its character at counts the points fixed by . An invariant vector has coefficients constant on every orbit, so the fixed subspace has dimension equal to the number of orbits. The trivial-character multiplicity formula gives
For the action of on ordered pairs of letters, fixed-pair counts are the squares of the fixed-letter counts: nine at the identity, one at a transposition, and zero at a three-cycle. Their weighted average is . The two orbits are pairs with equal entries and pairs with distinct entries.
Each column of a character table describes a conjugacy class, and the table carries consistency checks: degrees are positive integers, irreducible rows have norm one, distinct rows are orthogonal with class-size weights, and the squared degrees add to the group order. Passing them is evidence for a proposed table; proving that the rows are characters still requires constructing representations or another argument.
For instance, on an abelian group every conjugacy class is a singleton. The number of irreducibles equals the group order, and the sum-of-squares formula forces all their degrees to be one. For a cyclic group of order , a generator acts in a one-dimensional representation by an th root of unity, producing all characters. Their orthogonality reduces to the finite geometric-series identity. Thus the character table of a cyclic group is the discrete Fourier matrix, with its rows normalized by the group-order inner product.
Common mistakes
Common mistake
(An integer-valued class function need not be a character)
An arbitrary integer-valued class function need not be a character. Nonnegative integral multiplicities in the irreducible basis are necessary and sufficient. Negative multiplicities describe a virtual character, which can be useful algebraically but is not the character of an actual representation. A negative value at a particular group element is harmless: the standard character equals minus one at a three-cycle.
Common mistake
(Traces at one generator do not classify)
Characters classify complex representations of a finite group because complete reducibility and orthogonality hold. Equal traces of one group generator do not generally suffice when the group is not cyclic, and even in a cyclic group the full eigenvalue data may not be determined by one trace. Use equality of the whole character, or a justified complete table calculation.
Common mistake
(Dimension does not decide irreducibility)
Do not infer irreducibility from dimension alone. A two-dimensional representation can be a direct sum of two lines. Compute its norm or analyze invariant subspaces. Likewise, the sum-of-squares identity describes the exhaustive irreducible list, not the arbitrary summands of one selected representation.
Summary
Averaging on a Hom space turns character inner products into dimensions of intertwiner spaces. Schur's lemma supplies orthogonality, and Maschke supplies the decomposition that makes multiplicity extraction possible. The regular representation contains every irreducible and yields the sum of squared degrees. Central class sums show that irreducible characters span all class functions. These facts make character tables a complete computational language for finite complex representations.
Exercises
Quick check
(Decompose the regular representation into the three types in the displayed table.)
Solution · Answer
The regular character is . Its inner products with trivial, sign, and standard are one, one, and two. Thus , of dimension .
Quick check
(Determine whether the class function with values on the three classes is a character. Give a reason stronger than looking at a negative value.)
Solution · Answer
The inner product with sign is . A genuine representation cannot contain a negative number of sign constituents. Therefore this function is not a character. The negative value at a three-cycle was not itself an obstruction; the standard character already has that feature.
Quick check
(Write all irreducible characters of and verify orthogonality for the characters sending to and to .)
Solution · Answer
For , define . These are four one-dimensional representations, exhaustive by the class-count theorem. The inner product of and is . More generally a nontrivial geometric sum gives zero for distinct rows and one for equal rows.
Quick check
(Let have character . Prove that the dimension of its fixed subspace is and apply it to the tensor-square example.)
Solution · Answer
The averaging operator projects onto . Its trace is the image dimension and also the average of the traces. For , the average is , agreeing with its single trivial summand.
Quick check
(In the explicit tensor decomposition, express in the basis . Verify by expansion that its alternating component is zero.)
Solution · Answer
The coefficients are , , , . Thus ; expanding gives . Symmetry also forces the alternating component to vanish, since the symmetric and alternating subspaces intersect only in zero over .
Quick check
(Let act diagonally on . Prove , using the constructed summands rather than multiplying six four-by-four matrices.)
Solution · Answer
Averaging fixes and kills : the signs sum to zero. It also kills , since its image there lies in . Indeed a fixed vector in has all three coordinates equal and sum zero, so it vanishes over . Therefore .
Quick check
(Explain which step of this explicit decomposition fails if the same displayed tensor formulas are interpreted over a field of characteristic three. Exhibit a nonzero intersection of the proposed trivial and standard summands.)
Solution · Answer
Division by three or six is unavailable. More structurally, in characteristic three, so the nonzero vector lies in both the proposed trivial line and . Also , so contraction no longer separates the trivial line from its kernel. These are actual failures of the direct-sum proof. The complex character table and its averaging formulas were never asserted in this characteristic.