Let GG be a finite group and kk a field. Consider finite-dimensional of GG over kk, equivalently finite-dimensional kGkG-modules.

Theorem (Maschke)

Assume that G|G| is invertible in kk (equivalently char(k)G\mathrm{char}(k)\nmid |G|). Then every finite-dimensional kk-representation VV of GG is semisimple: for every GG- WVW\subseteq V, there exists a GG-stable subspace UVU\subseteq V such that

V  =  WUV \;=\; W \oplus U

as GG-representations (equivalently as kGkG-modules). In particular, every representation is .

In module language: every finite-dimensional kGkG-module is a .

Standard averaging proof (key construction)

Let WVW\subseteq V be GG-stable. Choose any kk-linear projection p:VWp:V\to W (not necessarily GG-equivariant). Define the averaged map

pG(v)  =  1GgGgp(g1v).p_G(v) \;=\; \frac{1}{|G|}\sum_{g\in G} g\cdot p(g^{-1}\cdot v).

Then:

  • pGp_G is GG-equivariant, i.e. a VWV\to W,
  • pGW=idWp_G|_W = \mathrm{id}_W, so pGp_G is a GG-equivariant projection,
  • therefore ker(pG)\ker(p_G) is GG-stable and gives the complement U=ker(pG)U=\ker(p_G).

This averaging step is exactly where the condition 1Gk\frac{1}{|G|}\in k is used.

Remarks and consequences
Examples
  1. A “good characteristic” decomposition for S3S_3 over C\mathbb C. Let S3S_3 act on C3\mathbb C^3 by permuting coordinates (a permutation representation). The 1-dimensional subspace
    W={(a,a,a):aC}W = \{(a,a,a):a\in\mathbb C\}
    is S3S_3-stable (the trivial subrepresentation). Maschke guarantees an invariant complement, and indeed one is
    U={(x1,x2,x3)C3:x1+x2+x3=0},U=\{(x_1,x_2,x_3)\in\mathbb C^3 : x_1+x_2+x_3=0\},
    giving C3WU\mathbb C^3 \cong W \oplus U, where UU is the 2-dimensional standard irreducible.
  1. A “good characteristic” example where not all irreducibles are 1D. Take G=C3=gG=C_3=\langle g\rangle over k=F2k=\mathbb F_2. Since char(F2)=23\mathrm{char}(\mathbb F_2)=2\nmid 3, Maschke applies: every F2[C3]\mathbb F_2[C_3]-module is semisimple. In particular, F2[C3]\mathbb F_2[C_3] (the regular representation) decomposes as a direct sum of irreducibles, but over F2\mathbb F_2 there is a 2-dimensional irreducible factor (because x2+x+1x^2+x+1 is irreducible over F2\mathbb F_2). So semisimple does not mean “splits into 1-dimensional pieces”; it means “splits into irreducibles over the given field.”
  1. Failure in “bad characteristic”: CpC_p over Fp\mathbb F_p. Let G=Cp=gG=C_p=\langle g\rangle and k=Fpk=\mathbb F_p, so char(k)=pG\mathrm{char}(k)=p\mid |G|. Consider the 2-dimensional representation V=k2V=k^2 where gg acts by the unipotent matrix
    ρ(g)=(1101).\rho(g)=\begin{pmatrix}1&1\\0&1\end{pmatrix}.
    The line W=e1W=\langle e_1\rangle is GG-stable, but there is no GG-stable complement line: the only eigenvectors of ρ(g)\rho(g) lie in e1\langle e_1\rangle, so no other 1-dimensional subspace is preserved. Hence VV is not a direct sum of subrepresentations, and Maschke’s conclusion fails.