Group Rings of Infinite Groups Are Never Semisimple

Q: Does the proof need k to be commutative or semisimple?

No. The only hypothesis is k ≠ 0, which is used once, to know that the augmentation ideal is a proper left ideal. Everything else happens in the group: the invariance f = f and the finiteness of supports. This is unusually clean — most results about kG depend heavily on k.

Q: Where does the argument break if G is finite?

At the last step only. For finite G the element f with f = f for all does exist — it is a multiple of the group sum G — and it is an idempotent precisely when |G| is invertible, giving |G|^-1G. So the finite case is where the argument turns into Maschke's construction rather than a contradiction.

Q: Can kG still be a nice ring when G is infinite?

Yes. For G = Z and k a field, kG is a principal ideal domain, hence noetherian, hereditary and J-semisimple. For G polycyclic-by-finite, kG is noetherian. What is unavailable is any form of the descending chain condition and hence any Wedderburn-style decomposition.

Executive Summary

Maschke's theorem settles semisimplicity for finite groups. This page settles the remaining case in one line of principle: for an infinite group $G$ and any nonzero coefficient ring $k$ , the group ring $k G$ is never semisimple.

The proof does not touch the coefficient ring at all. If $k G$ were semisimple, the augmentation ideal $Δ (G)$ would be complemented, and the complementary idempotent would have to be fixed by left multiplication by every element of $G$ . Its support would then be all of $G$ , which is infinite — but group ring elements have finite support by construction.

NeverSemisimple,

G

infinite

Finite supportThe whole obstruction

$rad = 0$ The surviving question

$k \neq 0$ Only hypothesis on

k

Overview

Semisimplicity is a very strong condition: it is equivalent to being left artinian with zero Jacobson radical. For group rings of infinite groups the artinian half is hopeless, and this proposition makes that concrete without any chain-condition machinery.

Combining with Maschke's theorem gives a complete answer to the semisimplicity question for group rings.

k G semisimple iff G is finite, k is semisimple, and | G | \cdot 1_{k} \in U (k) .

(6.3a)

The finite case is $(6.1)$ ; the infinite case is $(6.3)$ , which rules the possibility out unconditionally.

Once semisimplicity is unavailable, the natural weakening is J-semisimplicity: $rad (k G) = 0$ . That question is genuinely hard and is the subject of J-Semisimplicity of Complex and Real Group Algebras and Group Rings in Prime Characteristic. This page is the reason those pages exist.

Learning Objectives

State $(6.3)$ with its exact hypotheses: $k \neq 0$ arbitrary, $G$ infinite.
Run the finite-support argument and identify precisely where infiniteness of $G$ is used.
Strengthen the conclusion to: $Δ (G)$ is never a direct summand of $k G$ .
Deduce that the trivial module is never projective over $k G$ for infinite $G$ .
State the combined classification of semisimple group rings.
Verify everything by hand for $k [x, x^{- 1}]$ , the group ring of the infinite cyclic group.

Core Concepts

Finite support is a hypothesis, not an accident

An element of $k G$ is a function $G \to k$ with finite support. This is what distinguishes $k G$ from the full product $k^{G}$ , and it is the only property of the construction the proof uses. Every completion of $k G$ that allows infinite support — the $ℓ^{1}$ -algebra, the reduced group $C^{*}$ -algebra, the group von Neumann algebra — escapes this obstruction, and in those settings averaging-type idempotents genuinely exist.

$G$ -fixed elements of a group ring

Say $α \in k G$ is **left $G$ -fixed** if $σ α = α$ for every $σ \in G$ . Comparing coefficients, this says the coefficient function of $α$ is constant on $G$ — because left translation acts transitively on $G$ .

σ α = α \forall σ \in G ⟺ α \in k \hat{G} (G finite), ⟺ α = 0 (G infinite) .

(6.3b)

The dichotomy on which everything here rests. For finite $G$ the fixed elements are the multiples of the group sum; for infinite $G$ there is no group sum to be a multiple of.

Why $Δ (G) f = 0$ appears

A semisimple ring has every left ideal complemented, so $k G = Δ (G) \oplus 𝔅$ for some left ideal $𝔅$ . Complementary summands of a ring come from a pair of orthogonal idempotents summing to $1$ , and orthogonality gives $Δ (G) f = 0$ where $f$ generates $𝔅$ . Since every $σ - 1$ lies in $Δ (G)$ , this reads $σ f = f$ .

Key Results

Proposition(6.3)Infinite groups give non-semisimple group rings

Let $k \neq 0$ be any ring with identity and let $G$ be an infinite group. Then $R = k G$ is not semisimple.

Proof

Let $ε : k G \to k$ be the augmentation and $Δ (G) = \ker ε$ . Since $k \neq 0$ and $ε$ is surjective, $Δ (G)$ is a proper left ideal of $R$ .

Suppose $R$ were semisimple. Then every left ideal is a direct summand, so $R = Δ (G) \oplus 𝔅$ for some left ideal $𝔅$ . Write $1 = e + f$ with $e \in Δ (G)$ and $f \in 𝔅$ . Standard Peirce bookkeeping makes $e$ and $f$ orthogonal idempotents with $Δ (G) = R e$ and $𝔅 = R f$ ; in particular

Δ (G) f = R e f = 0 .

(6.3c)

Because $Δ (G)$ is proper, $f \neq 0$ : otherwise $1 = e \in Δ (G)$ and $Δ (G) = R$ .

Now every $σ - 1$ with $σ \in G$ lies in $Δ (G)$ , so $(σ - 1) f = 0$ , that is $σ f = f$ for all $σ \in G$ . Write $f = \sum_{g} a_{g} g$ and choose $τ$ with $a_{τ} \neq 0$ . Comparing the coefficient of $σ τ$ on both sides of $σ f = f$ gives $a_{σ τ} = a_{τ} \neq 0$ for every $σ \in G$ .

As $σ$ ranges over $G$ , the products $σ τ$ range over all of $G$ . So every group element occurs in $f$ with nonzero coefficient, and $supp (f) = G$ is infinite — contradicting the definition of $k G$ , whose elements have finite support. Hence $R$ is not semisimple.

Corollary—The augmentation ideal is never complemented

Let $k \neq 0$ and let $G$ be infinite. Then $Δ (G)$ is not a direct summand of $k G$ as a left $k G$ -module, and the short exact sequence

0 ⟶ Δ (G) ⟶ k G \overset{ε}{⟶} k ⟶ 0

(6.3d)

does not split. Equivalently, the trivial module $k$ is not projective over $k G$ .

Proof

The proof of $(6.3)$ used semisimplicity only to complement $Δ (G)$ ; the contradiction is reached from the existence of $f$ alone. So no complement exists. A surjection onto a projective module splits, so if $k$ were projective the sequence would split and produce that complement.

Corollary—Classification of semisimple group rings

Let $k \neq 0$ be a ring and $G$ a group. Then $k G$ is semisimple if and only if $G$ is finite, $k$ is semisimple and $| G | \cdot 1_{k}$ is a unit of $k$ . For $k$ a field this reads: $k G$ is semisimple if and only if $G$ is finite and $char k$ does not divide $| G |$ .

Proof

If $G$ is infinite, $(6.3)$ rules semisimplicity out and the right-hand condition also fails. If $G$ is finite, this is exactly Maschke's theorem $(6.1)$ .

Remark—How far the failure goes

Semisimple means left artinian together with zero radical. Connell proved the sharper statement that $k G$ is left artinian if and only if $k$ is left artinian and $G$ is finite, so for infinite $G$ it is the chain condition that fails, not necessarily the radical. Indeed $rad (k G)$ is often zero for infinite $G$ — that is the content of Rickart's and Amitsur's theorems.

Remark—Regularity survives where semisimplicity does not

Semisimple is equivalent to von Neumann regular plus left artinian, and only the second half is lost. For $k$ a field and $G$ locally finite with no elements of order $char k$ , the ring $k G$ is von Neumann regular even when $G$ is infinite. So infinite group rings can retain the element-wise regularity of semisimple rings while losing the finiteness.

Proof Techniques and Method

The reusable moves behind the argument.

Move 1

Turn a splitting into an idempotent

A direct-sum decomposition of a ring as a left module over itself is the same data as a pair of orthogonal idempotents summing to $1$ . Any argument that can constrain idempotents can constrain splittings.

Move 2

Test with $σ - 1$

The elements $σ - 1$ generate $Δ (G)$ as a left ideal, so a statement about $Δ (G)$ annihilating something becomes a $G$ -invariance statement, which is combinatorial rather than algebraic.

Move 3

Count the support

Invariance under a transitive action forces the support to be a full orbit. Comparing that with the finiteness built into the group ring closes the argument. This move recurs throughout infinite group ring theory.

Move 3 is the characteristic technique of the subject. It reappears in the proof that certain group rings have only trivial units, in Passman's work on the f.c. subgroup, and in the analysis of central idempotents; in each case one shows that an algebraic hypothesis makes a support $G$ -invariant and therefore too large.

Worked Example

The infinite cyclic group: everything checkable by hand

Let $k$ be a field and $G = ⟨ x ⟩$ infinite cyclic. Then $k G$ is the Laurent polynomial ring $k [x, x^{- 1}]$ , and the augmentation is evaluation at $x = 1$ , so $Δ (G) = (x - 1)$ .

No complement. $k [x, x^{- 1}]$ is an integral domain, so its only idempotents are $0$ and $1$ . A complement to $(x - 1)$ would be generated by an idempotent $f \notin {0, 1}$ : if $f = 1$ then $Δ (G) = 0$ , false; if $f = 0$ then $Δ (G) = k G$ , false. This is $(6.3)$ verified without any support counting.

Not artinian. The units of $k [x, x^{- 1}]$ are $c x^{n}$ with $c \in k^{\times}$ , so $x - 1$ is a non-unit and

(x - 1) ⊋ ((x - 1)^{2}) ⊋ ((x - 1)^{3}) ⊋ \dots

(E.1)

A strictly descending chain of ideals, since $k [x, x^{- 1}]$ is a unique factorisation domain and $x - 1$ is irreducible.

But J-semisimple. $k [x, x^{- 1}]$ is a principal ideal domain whose maximal ideals are generated by the irreducible polynomials of $k [x]$ other than $x$ . There are infinitely many such irreducibles by Euclid's argument, while a nonzero Laurent polynomial has only finitely many irreducible factors; hence no nonzero element lies in every maximal ideal and $rad (k G) = 0$ .

An infinite group where even the radical is nonzero

Let $k = 𝔽_{2}$ and let $G$ be an infinite direct sum of copies of $C_{2}$ , so $G$ is abelian with every nonidentity element of order $2$ . Then $k G$ is commutative, and for each $g \in G$ ,

(g - 1)^{2} = g^{2} - 2 g + 1 = g^{2} + 1 = 0 in 𝔽_{2} G .

(E.2)

In a commutative ring the ideal generated by nilpotent elements is nil, so $Δ (G)$ is a nil ideal, hence contained in $rad (k G)$ . Since $k G / Δ (G) ≅ 𝔽_{2}$ is a field, $Δ (G)$ is maximal and therefore $rad (k G) = Δ (G) \neq 0$ .

This ideal is nil but not nilpotent: any product of $n$ generators $g_{i} - 1$ with the $g_{i}$ independent is nonzero, so $Δ (G)^{n} \neq 0$ for every $n$ . Infinite groups can therefore fail J-semisimplicity outright, and the necessary condition visible here — a finite normal subgroup whose order is divisible by the characteristic — reappears in Group Rings in Prime Characteristic.

Process and Workflow

$G$ is infinite and I need structural information about $k G$ — what should I ask instead of semisimplicity?

Is $rad (k G) = 0$ ?The J-semisimplicity problem. Affirmative for

k = ℂ

ℝ

by Rickart, and for any field of characteristic zero that is not algebraic over

ℚ

by Amitsur.

Does $k G$ have nonzero nil left ideals?Often answerable when the coefficient ring carries a positive-definite involution — this is Lam's

(6.11)

, and it is the algebraic engine behind the characteristic-zero results.

Is $k G$ a domain, or reduced?Kaplansky's zero-divisor and related problems, open for torsion-free

G

in general and known for orderable groups.

Are all units trivial?The unit problem, which implies the reduced and J-semisimplicity answers under mild hypotheses.

Confirm $G$ is infiniteThen stop asking about semisimplicity, artinian-ness, or Wedderburn decompositions — all are unavailable.

Reduce to finitely generated subgroupsFor any

α \in k G

there is a finitely generated

G_{0} \leq G

with

α \in k G_{0}

, and

k G_{0} \cap rad (k G) \subseteq rad (k G_{0})

. Radical questions localise.

Exploit special structure of $G$ Orderable, locally finite, abelian, torsion-free — each class has its own theorem, and the answers so far are uniformly affirmative.

Transfer along field extensionsScalar extension results move J-semisimplicity between

k G

and

K G

; separability of

K / k

is the usual hypothesis.

Comparison and Classification

What survives and what fails, $k$ a field
Property of $k G$	$G$ finite, $char k ∤ \| G \|$	$G$ finite, $char k$ divides $\| G \|$	$G$ infinite
Semisimple	yes	no	never
Left artinian	yes	yes	never
$rad (k G) = 0$	yes	no	sometimes — the open problem
$Δ (G)$ complemented	yes	no	never
Trivial module projective	yes	no	never
Can be von Neumann regular	yes	no	yes, e.g. locally finite $p'$ -groups

Which hypothesis each conclusion consumes
	$k \neq 0$	$G$ infinite	Finite support	Semisimplicity of $k G$
$Δ (G)$ is a proper left ideal	yes	no	no	no
$σ f = f$ for all $σ$	no	no	no	partial
$supp (f) = G$	yes	no	no	no
Contradiction	yes	yes	yes	no

Which hypothesis each conclusion consumes

The last row is the point: semisimplicity is consumed only to produce $f$ , and once $f$ exists the contradiction comes from the group and the construction, never from $k$ .

Relationship Map

The proposition sits at a junction. Above it, Maschke's theorem handles the finite case; below it, the whole theory of infinite group rings takes over.

Is $G$ finite?
- Yes
  - $| G |$ invertible in $k$ : semisimple, Wedderburn applies
  - $char k$ divides $| G |$ : modular theory, $rad \neq 0$ but nilpotent
- No
  - never semisimple, never artinian
  - $rad (k G) = 0$ is the J-semisimplicity problem
  - units, zero divisors and idempotents become open problems

SemisimpleJ-semisimpleno nonzero nil left idealsreduced

Reading the chain from right to left shows the hierarchy of substitutes. For infinite groups the leftmost condition is unavailable, so the literature works at the second and third positions: Rickart's theorem gives J-semisimplicity of $ℂ G$ , and Lam's $(6.11)$ gives the absence of nil left ideals over formally real and characteristic-zero algebraically closed coefficients.

Applications and Industry Use

Applications here means where this structure is used — inside mathematics and in the engineering and computing disciplines that consume it.

This is a negative result, and its honest use is as a boundary marker: it tells three different communities that a construction they might reach for is unavailable.

Operator algebras

Why completions are needed

The failure is caused by finite support, so analysts replace $ℂ G$ by $ℓ^{1} (G)$ , the reduced group $C^{*}$ -algebra, or the group von Neumann algebra. Projections onto invariant subspaces then exist, and the Kadison–Kaplansky idempotent conjecture asks how much of the algebraic obstruction persists.

Homological algebra

Group cohomology is nontrivial exactly here

The trivial module fails to be projective precisely for infinite groups, so $H^{n} (G; M)$ can be nonzero in positive degrees and group cohomology becomes a subject. For finite groups with invertible order the cohomology vanishes above degree zero for the same reason Maschke's theorem holds.

Coding and computation

No idempotent generators

Codes given as ideals in a group ring are generated by idempotents only in the semisimple case. Working with infinite or modular group rings means working with ideals that are not generated by idempotents, which changes the decoding algorithms available.

Ring theory

A supply of non-artinian examples

Group rings of infinite groups are the standard source of rings with zero radical and no chain conditions, precisely the setting Jacobson's radical was invented to handle.

Failure Modes and Common Mistakes

Do not conclude that $k G$ has no nontrivial idempotents for infinite $G$ ; the proof rules out one specific idempotent, not all of them. Whether $k G$ has nontrivial idempotents for $G$ torsion-free is Kaplansky's idempotent problem and is open.
Do not assume $k G$ fails to be noetherian. For $G$ polycyclic-by-finite and $k$ noetherian, $k G$ is noetherian — it is the descending chain condition that is impossible.
The corollary that the trivial module is not projective is about the trivial module. Other $k G$ -modules can perfectly well be projective, starting with $k G$ itself.
For finite $G$ with $char k$ dividing $| G |$ the trivial module is also non-projective, but for a different reason — dimension divisibility in modular representation theory, not support counting.

Quick Reference

Statement $(6.3)$

k \neq 0

G

infinite

\Rightarrow

k G

not semisimple

Sharper form

Δ (G)

is never a direct summand of

k G

Module formthe trivial module

k

is never projective over

k G

Key identity

Δ (G) f = 0 \Rightarrow σ f = f

for all

σ \in G

Contradiction

σ f = f

and

f \neq 0

force

supp (f) = G

Classification

k G

semisimple

iff

G

finite,

k

semisimple,

| G | \cdot 1_{k} \in U (k)

What failsthe descending chain condition, not necessarily the radical

Successor questionis

rad (k G) = 0

? — the J-semisimplicity problem

Reference examples for infinite $G$
$k$ , $G$	$k G$	$rad (k G)$	Comment
$k$ a field, $G = ℤ$	$k [x, x^{- 1}]$	$0$	J-semisimple domain, not artinian
$ℂ$ , $G$ any infinite group	$ℂ G$	$0$	Rickart's theorem
$𝔽_{2}$ , $G$ infinite elementary abelian $2$ -group	commutative	$Δ (G) \neq 0$	nil but not nilpotent
$k$ a field, $G$ locally finite $p'$ -group	von Neumann regular	$0$	regular but never artinian

Frequently Asked Questions

Does the proof need

k

to be commutative or semisimple?

No. The only hypothesis is $k \neq 0$ , which is used once, to know that the augmentation ideal is a proper left ideal. Everything else happens in the group: the invariance $σ f = f$ and the finiteness of supports. This is unusually clean — most results about $k G$ depend heavily on $k$ .

k G

is never semisimple for infinite

G

, why does anyone study representations of infinite groups over

k G

Because interesting module categories do not need semisimplicity. The failure means modules are not determined by composition factors and extensions carry information, which is what group cohomology measures. It also means one asks weaker questions, chiefly whether the radical vanishes.

Where does the argument break if

G

is finite?

At the last step only. For finite $G$ the element $f$ with $σ f = f$ for all $σ$ does exist — it is a multiple of the group sum $\hat{G}$ — and it is an idempotent precisely when $| G |$ is invertible, giving $| G |^{- 1} \hat{G}$ . So the finite case is where the argument turns into Maschke's construction rather than a contradiction.

Can

k G

still be a nice ring when

G

is infinite?

Yes. For $G = ℤ$ and $k$ a field, $k G$ is a principal ideal domain, hence noetherian, hereditary and J-semisimple. For $G$ polycyclic-by-finite, $k G$ is noetherian. What is unavailable is any form of the descending chain condition and hence any Wedderburn-style decomposition.

Does the same argument apply to the

ℓ^{1}

-algebra or the group von Neumann algebra?

No, and that is the useful contrast. Those completions contain elements of infinite support, so the final contradiction disappears. Group von Neumann algebras of infinite groups are far from semisimple in the ring-theoretic sense, but they do contain many projections, and the question of which ones exist is the subject of the Kadison–Kaplansky and Atiyah-type conjectures.

What replaces the Wedderburn decomposition for infinite group rings?

Nothing as strong. The practical substitutes are: the structure of $k G$ modulo its radical when the radical is known; reduction to finitely generated subgroups, since radicals contract along the inclusion $k G_{0} \subseteq k G$ ; and specific structure theorems for orderable, abelian or locally finite groups.

References

T. Y. Lam, A First Course in Noncommutative Rings, Graduate Texts in Mathematics 131, Springer-Verlag, 1991, §6, (6.3) (pp. 85–86).
D. S. Passman, The Algebraic Structure of Group Rings, Wiley-Interscience, 1977, Chapters 2 and 4.
I. G. Connell, “On the group ring”, Canadian Journal of Mathematics 15 (1963), 650–685.
K. S. Brown, Cohomology of Groups, Graduate Texts in Mathematics 87, Springer-Verlag, 1982, Chapters I–III.
D. S. Passman, Infinite Crossed Products, Academic Press, 1989.

AI Suggested Questions

Prove Connell's theorem that $k G$ is left artinian if and only if $k$ is left artinian and $G$ is finite.
For which infinite groups $G$ and fields $k$ is $k G$ von Neumann regular?
State the Kadison–Kaplansky idempotent conjecture and its relation to the finite-support obstruction.
Show that $ℤ G$ is noetherian when $G$ is polycyclic-by-finite, and give an infinite $G$ for which it is not.
Compute the global dimension of $k [x, x^{- 1}]$ and relate it to the cohomological dimension of $ℤ$ .
Give an infinite group $G$ and a field $k$ of characteristic $p$ with $rad (k G) \neq 0$ but $G$ containing no finite normal $p$ -subgroup, or explain why none exists.
How does the non-projectivity of the trivial module relate to the existence of a finite free resolution of $ℤ$ over $ℤ G$ ?

01Executive Summary

02Overview

03Learning Objectives

04Core Concepts

Finite support is a hypothesis, not an accident

G-fixed elements of a group ring

Why Δ(G)f=0 appears

05Key Results

06Proof Techniques and Method

Turn a splitting into an idempotent

Test with σ−1

Count the support

07Worked Example

The infinite cyclic group: everything checkable by hand

An infinite group where even the radical is nonzero

08Process and Workflow

09Comparison and Classification

10Relationship Map

11Applications and Industry Use

Why completions are needed

Group cohomology is nontrivial exactly here

No idempotent generators

A supply of non-artinian examples

12Failure Modes and Common Mistakes

13Quick Reference

14Frequently Asked Questions

15Related KEVOS Topics

16References

17AI Suggested Questions

Continue learning

Executive Summary

Overview

Learning Objectives

Core Concepts

$G$ -fixed elements of a group ring

Why $Δ (G) f = 0$ appears

Key Results

Proof Techniques and Method

Test with $σ - 1$

Worked Example

Process and Workflow

Comparison and Classification

Relationship Map

Applications and Industry Use

Failure Modes and Common Mistakes

Quick Reference

Frequently Asked Questions

Related KEVOS Topics

References

AI Suggested Questions