Motivation
Induction assembles copies of a representation along cosets. Restricting that induced representation to a second subgroup does not generally leave those copies separate: the second subgroup moves them in orbits. Each orbit has its own stabilizer and its own transported action on the original vector space. Double cosets are the bookkeeping that keeps these three pieces together.
This produces more than a restriction formula. For complex representations, Frobenius reciprocity converts the decomposition into an exact count of self-intertwiners. That count tells us when an induced representation is irreducible. Examples will explain both outcomes: a nontrivial character of induces irreducibly to , while a character of a transposition subgroup does not.
Let be finite, , and a finite-dimensional representation of over a field . The sided tensor construction of induction and Frobenius reciprocity are proved prior background. The decomposition below works in arbitrary characteristic; only the later character and irreducibility statements specialize to .
Definitions
Definition 14.1.1
(Double cosets and transported stabilizer actions)
The double coset of is . Choose one representative from each member of and put
The vector space becomes an -representation by , using the original -action on the right. This makes sense precisely because . It is a representation since conjugation is a homomorphism.
The subgroup acts on the left coset set by left multiplication. Its orbit through is the collection of cosets in , and its stabilizer is . Thus double cosets classify orbits, while left cosets within each double coset classify the separate copies of . Confusing these two levels gives the wrong number or size of summands.
Write . The group algebra is a right -module by right multiplication; the balancing relation is . Restriction to keeps the same vector space and only the -action. All direct sums in this finite-group setting are finite.
Theorems
Theorem 14.1.2
(Mackey's restriction decomposition)
There is an isomorphism of -representations
For the chosen representatives, the map from the summand is . No division by a group order occurs, so this assertion is valid over every field. The identification uses choices of representatives; its target restriction is intrinsically defined.
Theorem 14.1.3
(Complex induced irreducibility criterion)
Let be irreducible over , and in this statement take . Then
The identity double coset contributes one. The induced representation is irreducible exactly when every other summand is zero. In particular, if is normal, it is irreducible exactly when is not isomorphic to any conjugate with .
Proof sketch or proof idea
Split the group algebra by support
Proof
(Support decomposition and coset normal form)
Distinct double cosets partition the basis of its group algebra. Therefore, as a left-, right- bimodule,
Each summand is preserved by those two actions because its support is closed under left multiplication by and right multiplication by . Tensoring over distributes over this finite direct sum. One can see the distribution directly: send to the tuple of , and use the projections onto the support subspaces to define its inverse. Balancing is respected because these projections commute with the right -action.
It remains to identify the piece . Define
It is balanced: for ,
It is plainly -equivariant, and it is onto since any equals . For injectivity, choose representatives of the left cosets . The cosets are precisely the distinct left -cosets inside : equality is equivalent to . The source consequently has the direct-sum normal form , and the target has . The map sends each copy identically onto its matching copy, proving injectivity without an averaging argument. Combining the pieces proves the formula.
Proof X-ray
The stabilizer explains the conjugated action
An element stabilizing the coset satisfies with . Moving it through the tensor sign therefore makes it act on as that conjugate element of . The twist is forced by balancing; it is not an optional convention inserted after the orbit decomposition. The coset normal form then proves that no vectors were lost or duplicated.
From a restriction formula to the number of self-maps
Proof
(Counting self-intertwiners)
Now work over , take , and write for the character of . For characters of a finite group , use the linear-first inner product . The character inner product counts intertwiner dimensions by the previously proved orthogonality theorem.
Frobenius reciprocity and the decomposition give
The last use of reciprocity follows also by conjugating the usual formula, since the inner product is conjugate symmetric. Each term is a nonnegative integer, the dimension of the stated intertwining space. For , and the term is by complex Schur.
Maschke decomposes the nonzero induced module as . Its endomorphism dimension is . That number is one exactly when there is one irreducible summand with multiplicity one. This proves the criterion. Notice the logical separation: the double-coset isomorphism did not need semisimplicity, but this conversion from an endomorphism count to irreducibility did.
The particularly transparent normal-subgroup case
Proof
(The inertia subgroup count)
If is normal, every , and the double cosets are the ordinary cosets of . The restriction is therefore the sum of all transported conjugates . Schur says the intertwiner term is one when and zero otherwise. Define the inertia subgroup
Conjugation acts on isomorphism classes, so this is a subgroup. It contains : for , the original action of gives an isomorphism from to . Thus the endomorphism dimension is , and induced irreducibility is equivalent to .
An abelian normal subgroup can have many different characters, but some are fixed by conjugation. Normality alone, or one-dimensionality of the inducing representation, does not ensure irreducibility. What matters is whether a conjugate type returns before leaving the subgroup's own coset.
Worked examples
Worked example 14.1.4
(A character of induces the standard representation)
Let , , and let have character , where is a nonreal cube root of one. A transposition conjugates to , so . The normal-subgroup criterion shows is irreducible of dimension two. Its character is two at the identity, zero at transpositions, and at three-cycles: it is the standard representation. In contrast, the trivial character of is fixed by conjugation, and its induction is the sum of the trivial and sign representations of .
Worked example 14.1.5
(A transposition subgroup produces two double cosets)
Take , and . The double cosets are and , of sizes two and four. Since , their outside intersection subgroup is trivial. For either one-dimensional character of , the identity contribution to the endomorphism formula is one and the outside contribution is . The induction has endomorphism dimension two, so it is not irreducible. It is for the trivial character and for the nontrivial one. The restriction formula reads , of dimension three: one double-coset piece accounts for two ordinary cosets at once.
Worked example 14.1.6
(Proving the three-dimensional standard representation of irreducible)
Let stabilize the point four in , and induce its trivial representation. The resulting permutation representation is . On the four-point coset set, has the orbit and the orbit . Hence there are two double cosets; the second stabilizer is isomorphic to . Each term of the endomorphism formula for the trivial representation is one, so the endomorphism dimension is two. Transitivity makes the fixed subspace exactly the line of constant vectors. Its invariant complement is the coordinate-sum-zero space, of dimension three. If that complement were reducible, semisimplicity would make the total endomorphism dimension at least three. It is therefore irreducible. This obtains the result without assuming the character table of .
Concept lensStructural
Double cosets are orbits, not extra representation data
The action of on supplies an orbit and a stabilizer. The inducing space supplies a fiber above one coset, and balancing supplies its stabilizer action. Inducing that fiber back across its orbit is exactly one Mackey summand. The algebraic support decomposition and the orbit–stabilizer picture describe the same construction, with the subgroup intersection telling us how much of the original action remains visible.
Common mistakes
Common mistake
(Summing over left cosets instead of double cosets)
A double-coset representative is not one representative of every left coset. Replacing the sum by a sum over all left cosets would count some entire induced orbit pieces several times. The twist uses , and its domain is the intersection ; omitting either condition can make the claimed action undefined.
Common mistake
(Using the irreducibility criterion outside the complex semisimple setting)
The complex irreducibility criterion uses both Schur and semisimplicity. The arbitrary-field decomposition does not permit importing complex character averages into a characteristic where their denominators vanish. Nor is an induced module generally irreducible merely because its starting module is irreducible.
Counterexample mode
A nontrivial character of a normal subgroup can induce reducibly
Claim: if is a proper normal subgroup and is a nontrivial irreducible representation of , then is irreducible. Take of order four, and with . Since is abelian, , so and the endomorphism dimension is . Explicitly, in the basis , the generator acts by , which has eigenvalues and , so the induction is the sum of two lines. The repair is the criterion itself: nontriviality is irrelevant, and what is needed is .
Summary
Restriction moves the induction cosets in subgroup orbits. Double cosets split those orbits, intersections give their stabilizers, and tensor balancing forces the transported actions. Over the complex numbers, reciprocity turns that decomposition into a nonnegative self-intertwiner count, giving an exact criterion for induced irreducibility and concrete new character calculations.
Exercises
Quick check
(Prove directly from orbits. Use it to check the dimensions in Mackey's formula for a representation of dimension .)
Apply orbit–stabilizer to the action of on .
Solution
The orbit through has size by orbit–stabilizer. The orbits partition the cosets, giving the sum. The corresponding induced summand has dimension , so summing gives , the original induced dimension. No characteristic assumption enters this count.
Quick check
(Let be the order-eight dihedral group, , and for . Determine which inductions are irreducible and describe the reducible ones.)
Compute the conjugate character and use the normal-subgroup form of the criterion.
Solution
Conjugation by sends to . Since is normal of index two, the inductions are irreducible exactly for . Those two inductions are isomorphic: on both have character and on reflections both vanish. For , the two one-dimensional extensions have , ; for , they have , . Each pair occurs once by Frobenius reciprocity, and its dimensions fill the two-dimensional induction, proving the stated decompositions.
Quick check
(In let . Prove that a nontrivial complex character of induces an irreducible three-dimensional representation of , and compute its character.)
Find the inertia subgroup of each nontrivial character of .
Solution
The subgroup is normal and isomorphic to . Its three nontrivial characters each have a different order-two kernel. Conjugation by a three-cycle cyclically permutes the three nonidentity double transpositions, hence also these three kernels and characters. The inertia subgroup of any nontrivial character is exactly , so induction is irreducible of dimension three. Its restriction to is the sum of the three nontrivial characters: the identity value is three, and at any nonidentity element the values are one, minus one, minus one, summing to minus one. Outside the induced character is zero because normality prevents any coset fixed by such an element. This accounts for all elements of .
Quick check
(Over , take and . Explain why Mackey's decomposition is still valid, but the regular -module is not a direct sum of simple submodules. Prove its fixed line has no -invariant complement.)
Evaluate an arbitrary -map to the trivial module on .
Solution
The double-coset proof uses only tensor balancing and coset bases, so it still applies. For restriction to the trivial subgroup it simply gives one-dimensional vector-space summands, with no assertion that they are -submodules. In the regular module the fixed line is spanned by . Any -map to that trivial line is determined by the image of one: every basis group element must have the same image. It sends to times that image, which is zero in . A projection onto the fixed line would have to fix its nonzero vector , impossible. Thus the line has no invariant complement and the regular module is not semisimple.