J-Semisimplicity of Complex and Real Group Algebras

Q: Why does the proof need complex analysis at all?

To justify one evaluation. The series _n tr_g(a^n)z^n is only known to represent the resolvent near the origin, but the conclusion is needed at z=1. Since the resolvent is defined and differentiable on the whole plane, the composed scalar function is entire, and an entire function is represented by its Taylor series at the origin everywhere. Without this the argument stalls at |z| < |a|^-1.

Q: Does Rickart's theorem say CG is semisimple?

No, and for infinite G it cannot: no group ring of an infinite group is semisimple. It says the Jacobson radical vanishes, which is the same as semisimplicity only in the presence of the descending chain condition. For G infinite, CG is a J-semisimple non-artinian ring.

Q: What is the role of the involution, exactly?

It converts an arbitrary nonzero element into one whose trace is a sum of squared moduli, hence strictly positive. Positivity is what allows a normalisation with tr(a) = 1 and what makes the doubling inequality run. Any coefficient ring with an involution satisfying the same positivity condition supports the algebraic half of the argument, which is the content of (6.11).

Q: Why is QG harder than CG?

The later proofs replace analysis by transcendence: one picks a transcendence basis of k over Q and uses the fact that a transcendental scalar extension forces the contracted radical to be nil. A field algebraic over Q offers no transcendence basis to work with, so the mechanism disappears. That the smallest fields are the hardest is an inversion of the usual pattern.

Q: Is the characteristic-p statement simply the same proof with p in place of 2?

No. Positivity of a trace form is unavailable in characteristic p — the example (1+g)^2 = 0 in F_2C_2 kills it immediately. The substitute is an orbit count: for a p'-group, the p-tuples of group elements multiplying to 1 contribute in orbits of size p except for one singleton, giving tr(^p) = tr()^p. The hypothesis that G has no element of order p is what makes the singleton unique.

Q: How does the real case follow from the complex case?

CG is generated as a left RG-module by the two centralising elements 1 and i. The ascent result for such extensions gives rad(RG) ⊆ rad(CG), and the right-hand side is zero. The finiteness of the generating set is essential — the corresponding ascent fails for infinite extensions.

Executive Summary

For an infinite group no coefficient ring makes $k G$ semisimple, so the substitute question is J-semisimplicity: is $rad (k G) = 0$ ? Rickart settled the first serious case in 1950: for every group $G$ , the complex group algebra $ℂ G$ has zero Jacobson radical, and the real one follows.

The proof splits into two independent halves that contradict each other. The algebraic half manufactures, from any nonzero radical element, a self-adjoint element $a$ whose traces along the powers $a^{2^{m}}$ are real and at least $1$ . The analytic half shows every radical element has $tr (a^{n}) \to 0$ . Neither half is hard; the pairing is what does the work.

1950Rickart

$\sum_{g} | a_{g} |^{2}$

tr (a^{*} a)

$\geq 1$ vs $\to 0$ The contradiction

Open

ℚ G

for general

G

Overview

Maschke's theorem and its infinite-group complement leave a clean gap. Semisimplicity of $k G$ is settled — it happens exactly for finite $G$ with $| G |$ invertible — and for all other groups the natural weakening is $rad (k G) = 0$ . Since the radical is defined without chain conditions, this question makes sense for every group.

The historical order is worth respecting even though later results are stronger. Rickart's theorem is the first, is proved by analysis rather than algebra, and is what convinced the field that the general problem was worth attacking.

rad (ℂ G) = 0 and rad (ℝ G) = 0 for every group G .

(6.4)

No hypothesis on $G$ whatsoever: not finiteness, not countability, not torsion-freeness.

The same positivity, stated abstractly as a condition on an involution, yields a purely algebraic theorem: over a formally real commutative ring, or over an algebraically closed field of characteristic zero, $k G$ has no nonzero nil one-sided ideals. That result is what Amitsur's later extension to almost all fields of characteristic zero is built on.

Learning Objectives

State $(6.4)$ and explain why the real case follows from the complex one.
Prove the identity $tr (α^{*} α) = \sum_{g} | a_{g} |^{2}$ and the doubling inequality it gives.
Construct the element $a$ of $(6.5)$ from an arbitrary nonzero radical element.
Verify that $| \cdot |$ is a submultiplicative norm and that the resolvent is entire.
Prove $(6.6)$ : ${tr}_{g} (a^{n}) \to 0$ for every $a \in rad (ℂ G)$ and every $g \in G$ .
State $(6.11)$ with its involution hypothesis and identify the two standard cases where it applies.

Definitions

Definition—Trace, involution and norm on

ℂ G

For $α = \sum_{g \in G} a_{g} g \in ℂ G$ define

{tr}_{g} (α) = a_{g}, tr (α) = {tr}_{1} (α) = a_{1}, α^{*} = \sum_{g} \bar{a_{g}} g^{- 1}, | α | = \sum_{g} | a_{g} | .

(6.4a)

The map $α \mapsto α^{*}$ is an involution: it is additive, satisfies $α^{* *} = α$ , and reverses products. The function $| \cdot |$ is a norm with $| α β | \leq | α | | β |$ and $| {tr}_{g} (α) | \leq | α |$ , so $ℂ G$ is a normed algebra and each ${tr}_{g}$ is continuous and $ℂ$ -linear.

Self-adjoint: $α^{*} = α$ . Powers of a self-adjoint element are self-adjoint, since $(α^{n})^{*} = (α^{*})^{n}$ .
Formally real: A commutative ring $k$ with $\sum_{i} a_{i}^{2} = 0 \Rightarrow$ all $a_{i} = 0$ . Examples: $ℚ$ , $ℝ$ , any real-closed field, and $ℝ [t]$ .
Real-closed field: A formally real field admitting no formally real proper algebraic extension. By Artin–Schreier, an algebraically closed field of characteristic zero is $k_{0} (i)$ for a real-closed $k_{0}$ .
Resolvent: $ϕ (z) = (1 - z a)^{- 1}$ , defined for all $z \in ℂ$ when $a \in rad (ℂ G)$ , since $z a$ then lies in the radical and $1 - z a$ is a unit.
$ℓ^{1} (G)$: The completion of $ℂ G$ in the norm $| \cdot |$ , a Banach algebra under convolution; the analyst's version of the group algebra.

Nothing here requires $G$ to be countable or finitely generated. Every element of $ℂ G$ has finite support, so all sums written above are finite.

Core Concepts

The trace form is positive definite

Expanding $α^{*} α = \sum_{g, h} \bar{a_{g}} a_{h} g^{- 1} h$ and reading off the coefficient of the identity — which requires $h = g$ — gives

tr (α^{*} α) = \sum_{g \in G} \bar{a_{g}} a_{g} = \sum_{g \in G} | a_{g} |^{2} \geq | a_{1} |^{2} = | tr α |^{2} .

(6.5a)

Positive definiteness plus the crude bound that discarding all terms but one can only decrease the sum.

Two consequences follow at once. First, $tr (α^{*} α) = 0$ forces $α = 0$ : the trace form makes $ℂ G$ a pre-Hilbert space, and its completion is $ℓ^{2} (G)$ with $G$ as orthonormal basis. Second, for self-adjoint $α$ we get $tr (α^{2}) \geq | tr α |^{2}$ , and iterating along squares gives a sequence that cannot decay.

Why the radical is analytically small

If $a \in rad (ℂ G)$ then $z a$ is in the radical for every scalar $z$ , so $1 - z a$ is a unit and the resolvent $ϕ (z) = (1 - z a)^{- 1}$ is defined on all of $ℂ$ . The resolvent identity

ϕ (y) - ϕ (z) = ϕ (y) [(1 - z a) - (1 - y a)] ϕ (z) = (y - z) a ϕ (y) ϕ (z)

(6.7)

Valid because $ϕ (y)$ and $ϕ (z)$ commute — both are power series in $a$ in the formal sense.

shows $ϕ$ is locally bounded, continuous, and differentiable with $ϕ' (z) = a ϕ (z)^{2}$ . Composing with the continuous linear functional ${tr}_{g}$ produces an entire scalar function whose Taylor coefficients at $0$ are the numbers ${tr}_{g} (a^{n})$ .

The Banach algebra picture behind it

Rickart's original argument treats $G$ as a discrete locally compact group and works in the convolution Banach algebra $ℓ^{1} (G)$ . Two general facts are being specialised: the radical of a complex Banach algebra consists of topologically nil elements, and in a $C^{*}$ -algebra the only topologically nil one-sided ideal is zero. The second is where the $*$ -structure and positivity enter, and it also shows every $C^{*}$ -algebra is J-semisimple.

Key Results

Theorem(6.4)Rickart's theorem

For every group $G$ — finite or infinite, of arbitrary cardinality — the complex group algebra $ℂ G$ is J-semisimple, that is $rad (ℂ G) = 0$ . Consequently $rad (ℝ G) = 0$ as well.

Lemma(6.5)The algebraic part

Let $G$ be any group. If $rad (ℂ G) \neq 0$ , then there exists $a \in rad (ℂ G)$ with $a^{*} = a$ and $tr (a) = 1$ such that, for every $m \geq 1$ , the number $tr (a^{2^{m}})$ is real and $tr (a^{2^{m}}) \geq 1$ .

Proof

Pick $0 \neq β \in rad (ℂ G)$ and set $γ = β^{*} β$ . Since the radical is a two-sided ideal, $γ \in rad (ℂ G)$ , and $γ^{*} = β^{*} β^{* *} = γ$ . By $(6.5 a)$ , $tr (γ) = \sum_{g} | b_{g} |^{2}$ is a positive real number, so we may define

a = \frac{γ}{tr (γ)} \in rad (ℂ G), a^{*} = a, tr (a) = 1 .

(6.5b)

Self-adjointness is preserved because the scalar divisor is real. Now induct on $m$ . Each $a^{2^{m - 1}}$ is self-adjoint, so writing $b = a^{2^{m - 1}}$ we get $tr (a^{2^{m}}) = tr (b^{2}) = tr (b^{*} b) = \sum_{g} | b_{g} |^{2}$ , which is real and non-negative, and by $(6.5 a)$ it is at least $| tr (b) |^{2}$ .

The base case is $tr (a) = 1$ . If $tr (a^{2^{m - 1}}) \geq 1$ then $tr (a^{2^{m}}) \geq | tr (a^{2^{m - 1}}) |^{2} \geq 1$ . This closes the induction.

Lemma(6.6)The analytic part

Let $G$ be any group and $a \in rad (ℂ G)$ . Then for every $g \in G$ ,

\lim_{n \to \infty} {tr}_{g} (a^{n}) = 0 in ℂ .

(6.6a)

In particular $tr (a^{n}) \to 0$ , taking $g = 1$ .

Proof

The resolvent is defined everywhere. For $z \in ℂ$ the element $z a$ lies in the radical, so $1 - z a \in U (ℂ G)$ and $ϕ (z) = (1 - z a)^{- 1}$ makes sense for all $z$ .

Local boundedness and differentiability. From $(6.7)$ and submultiplicativity, $| ϕ (y) | \leq | ϕ (z) | + | y - z | | a | | ϕ (y) | | ϕ (z) |$ , hence $| ϕ (y) | (1 - | y - z | | a | | ϕ (z) |) \leq | ϕ (z) |$ . For $y$ close enough to $z$ the bracket exceeds $\frac{12}{}$ , giving $| ϕ (y) | \leq 2 | ϕ (z) |$ . Substituting back into $(6.7)$ yields $| ϕ (y) - ϕ (z) | \leq 2 | a | | y - z | | ϕ (z) |^{2}$ , so $ϕ$ is continuous, and dividing by $y - z$ and letting $y \to z$ gives $ϕ' (z) = a ϕ (z)^{2}$ for every $z \in ℂ$ .

A geometric series near the origin. Suppose $| z | | a | < 1$ . For any $N$ ,

ϕ (z) - \sum_{n = 0}^{N} z^{n} a^{n} = ϕ (z) [1 - (1 - z a) \sum_{n = 0}^{N} z^{n} a^{n}] = ϕ (z) (z a)^{N + 1},

(6.8)

whose norm is at most $| ϕ (z) | (| z | | a |)^{N + 1} \to 0$ . Hence $ϕ (z) = \sum_{n \geq 0} z^{n} a^{n}$ in the norm of $ℂ G$ for $| z | < | a |^{- 1}$ .

Conclusion. Fix $g$ and put $f = {tr}_{g} \circ ϕ : ℂ \to ℂ$ . Since ${tr}_{g}$ is $ℂ$ -linear and continuous and $ϕ$ is differentiable everywhere, $f$ is entire, and by the previous step

f (z) = \sum_{n = 0}^{\infty} {tr}_{g} (a^{n}) z^{n} for | z | < | a |^{- 1} .

(6.9)

That is the Taylor expansion of $f$ at the origin. An entire function's Taylor series at $0$ has infinite radius of convergence and represents the function everywhere, so the series converges at $z = 1$ . The terms of a convergent series tend to zero, giving ${tr}_{g} (a^{n}) \to 0$ .

Proofof Rickart's theorem

(6.4)

Suppose $rad (ℂ G) \neq 0$ . Lemma $(6.5)$ produces $a$ in the radical with $tr (a^{2^{m}}) \geq 1$ for all $m \geq 1$ . Lemma $(6.6)$ applied to the same $a$ gives $tr (a^{n}) \to 0$ , hence $tr (a^{2^{m}}) \to 0$ along the subsequence $n = 2^{m}$ . These are incompatible, so $rad (ℂ G) = 0$ .

For the real case, $ℂ G = ℝ G \cdot 1 + ℝ G \cdot i$ is generated as a left $ℝ G$ -module by the two elements $1$ and $i$ , both of which centralise $ℝ G$ . The ascent result for ring extensions then gives $rad (ℝ G) \subseteq rad (ℂ G) = 0$ .

Remark(6.10)Why the shortcut is illegal

The value $f (1) = {tr}_{g} ((1 - a)^{- 1})$ would follow immediately from $(1 - a)^{- 1} = \sum_{n} a^{n}$ if that series converged in $ℂ G$ — which needs $| a | < 1$ , a condition no one has established and which is false in general. The entire-function theorem is exactly the device that transfers a statement true near the origin to the point $z = 1$ without any control on $| a |$ .

Proposition(6.11)Involutions kill nil one-sided ideals

Let $k$ be a ring with an involution $*$ satisfying the positivity condition

\sum_{i} a_{i}^{*} a_{i} = 0 ⟹ a_{i} = 0 for all i (a_{i} \in k) .

(6.11a)

Then for every group $G$ , the group ring $R = k G$ has no nonzero nil left ideals. This applies in particular when (a) $k$ is a commutative formally real ring with $*$ the identity, and (b) $k$ is an algebraically closed field of characteristic zero.

Proof

Extend $*$ to $R$ by $(\sum a_{g} g)^{*} = \sum a_{g}^{*} g^{- 1}$ ; this is an involution on $R$ . Computing the identity coefficient of $α^{*} α$ as in $(6.5 a)$ gives $tr (α^{*} α) = \sum_{g} a_{g}^{*} a_{g}$ , so hypothesis $(6.11 a)$ says precisely

tr (α^{*} α) = 0 ⟹ α = 0 (α \in R) .

(6.11b)

Suppose $𝔅 \neq 0$ is a nil left ideal of $R$ and pick $0 \neq β \in 𝔅$ . Then $γ = β^{*} β \in 𝔅$ because $𝔅$ is a left ideal, $γ^{*} = γ$ , and $γ \neq 0$ by $(6.11 b)$ . Since $γ$ is nil, there is $n \geq 1$ with $γ^{n} \neq 0$ and $γ^{n + 1} = 0$ .

Set $α = γ^{n}$ . Then $α^{*} α = γ^{n} γ^{n} = γ^{2 n} = 0$ , because $2 n \geq n + 1$ for $n \geq 1$ . Applying $(6.11 b)$ gives $α = γ^{n} = 0$ , a contradiction. Hence no nonzero nil left ideal exists.

Case (a). With $*$ the identity on a commutative formally real $k$ , condition $(6.11 a)$ is the definition of formal reality.

Case (b). Let $k$ be algebraically closed of characteristic zero. By Artin–Schreier, $k = k_{0} (i)$ with $k_{0}$ real-closed and $i^{2} = - 1$ . Define $(a + b i)^{*} = a - b i$ for $a, b \in k_{0}$ . Then $\sum_{j} (a_{j} + b_{j} i)^{*} (a_{j} + b_{j} i) = \sum_{j} (a_{j}^{2} + b_{j}^{2})$ , and formal reality of $k_{0}$ forces every $a_{j}$ and $b_{j}$ to vanish.

Corollary—Characteristic-zero Maschke, again

Let $k$ be a field of characteristic $0$ and $G$ a finite group. Then $k G$ is semisimple. Indeed $k G$ is artinian, so $rad (k G)$ is nilpotent; extending scalars to the algebraic closure $\bar{k}$ makes $(rad k G) \cdot \bar{k}$ a nilpotent, hence nil, left ideal of $\bar{k} G$ . By $(6.11)$ (b) it is zero, so $rad (k G) = 0$ and $k G$ is semisimple.

Proof Techniques and Method

The reusable moves, and which of them survive outside characteristic zero.

Move 1

Normalise a radical element

From any $β \neq 0$ in a two-sided ideal, build $β^{*} β$ : it stays in the ideal, is self-adjoint, and has strictly positive trace. Dividing by that trace gives a canonical representative. The move needs only that the ideal be two-sided and the trace be positive definite.

Move 2

Square along $2^{m}$ , not $n$

The inequality $tr (b^{2}) \geq | tr b |^{2}$ chains only through squares, so the subsequence of powers $a^{2^{m}}$ is the one that can be controlled. Contradicting a limit along a subsequence is enough.

Move 3

Make a Banach-valued function scalar

Compose a vector-valued analytic function with a continuous linear functional and apply ordinary complex analysis. This is how the resolvent argument avoids developing vector-valued function theory from scratch.

Move 1 is what generalises: it is entirely algebraic and is the content of $(6.11)$ . Moves 2 and 3 are tied to $ℂ$ . The characteristic- $p$ theory replaces them by a combinatorial count on $p$ -tuples with product $1$ , which shows $tr (β^{p}) = tr (β)^{p}$ for $p'$ -groups — a Frobenius substitute for the doubling inequality.

Worked Example

Watching $(6.5)$ run on a two-element group

Take $G = C_{2} = {1, g}$ and $β = 1 + g \in ℂ G$ . Then $β^{*} = 1 + g^{- 1} = 1 + g$ , and

γ = β^{*} β = (1 + g)^{2} = 2 + 2 g, tr (γ) = 2 = | 1 |^{2} + | 1 |^{2},

(E.1)

confirming $tr (β^{*} β) = \sum_{g} | b_{g} |^{2}$ in the smallest possible case.

The normalisation gives $a = γ / 2 = 1 + g$ , which is self-adjoint with $tr (a) = 1$ . Since $a^{2} = 2 + 2 g = 2 a$ , induction gives $a^{n} = 2^{n - 1} a$ for $n \geq 1$ , so

tr (a^{2^{m}}) = 2^{2^{m} - 1} \geq 1 for every m \geq 1,

(E.2)

exactly the lower bound $(6.5)$ predicts — and here it diverges rather than merely staying above $1$ .

Exactly where characteristic $2$ breaks the argument

Repeat the computation over $k = 𝔽_{2}$ with the identity involution. Now $α = 1 + g$ satisfies $α^{*} = α$ and

α^{*} α = (1 + g)^{2} = 1 + g^{2} = 0, yet α \neq 0 .

(E.3)

So the positivity hypothesis $(6.11 a)$ fails for $(𝔽_{2}, id)$ , and indeed the conclusion fails: $𝔽_{2} C_{2}$ has the nonzero nil ideal $(1 + g)$ , which is its whole Jacobson radical. This one line explains why the entire characteristic-zero theory has to be rebuilt in characteristic $p$ , and why the rebuilt version excludes elements of order $p$ .

The same element in three coefficient rings
$k$	$tr (α^{*} α)$ for $α = 1 + g$	$rad (k C_{2})$
$ℂ$	$2$	$0$
$ℚ$	$2$	$0$
$𝔽_{2}$	$0$	$(1 + g)$ , of dimension $1$

Comparison and Classification

State of the J-semisimplicity problem by coefficient field
Field $k$	Groups covered	Answer	Due to
$ℂ$ and $ℝ$	all $G$	J-semisimple	Rickart $(6.4)$
Uncountable, characteristic $0$	all $G$	J-semisimple	Amitsur and Herstein, independently
Characteristic $0$ , not algebraic over $ℚ$	all $G$	J-semisimple	Amitsur $(6.12)$
Algebraic over $ℚ$ , including $ℚ$ itself	all $G$	open in general	—
Characteristic $p$ , not algebraic over $𝔽_{p}$	$p'$ -groups	J-semisimple	Passman $(6.15)$
Characteristic $p$	$G$ with a finite normal subgroup of order divisible by $p$	not J-semisimple	square-zero ideal argument

Which ingredient each step of the proof uses
	Involution on $k$	Positivity of the trace	Complex analysis	$rad$ is two-sided
$(6.5)$ algebraic lemma	yes	yes	no	yes
$(6.6)$ analytic lemma	no	no	yes	yes
$(6.4)$ Rickart	yes	yes	yes	yes
$(6.11)$ no nil left ideals	yes	yes	no	no

Which ingredient each step of the proof uses

$(6.11)$ needs no analysis and no two-sidedness — it applies to any nonzero nil left ideal — which is why it, rather than Rickart's theorem, is the result that generalises to arbitrary characteristic-zero coefficient fields.

Relationship Map

The logical shape of the proof is a collision between two independent estimates on the same element.

rad (ℂ G) \neq 0

\exists a = a^{*}

tr (a) = 1

tr (a^{2^{m}}) \geq 1

contradicts

tr (a^{n}) \to 0

Rickart, 1950

ℂ G

and

ℝ G

are J-semisimple, by Banach algebra methods.

Amitsur and HersteinAny uncountable field of characteristic zero works, by a cardinality argument on the inverses of

a - r

rather than by analysis.

Amitsur, 1959Any field of characteristic zero that is not algebraic over

ℚ

: take a transcendence basis, use

(6.11)

(a) over

ℚ

to kill the nil contraction, then ascend along a separable algebraic extension.

Passman, 1962The characteristic-

p

analogue for

p'

-groups, with a combinatorial substitute for the positivity lemma.

Still openThe case where

k

is algebraic over its prime field — above all

ℚ G

itself. Lam records that it is precisely the prime fields that resist.

The step from $(6.11)$ to Amitsur's theorem is worth isolating. Writing $F = ℚ ({x_{i}})$ for a transcendence basis of $K / ℚ$ , the contraction $rad (F G) \cap ℚ G$ is a nil ideal of $ℚ G$ by the transcendental scalar-extension results; $(6.11)$ (a) with $k = ℚ$ forces it to vanish, hence $rad (F G) = 0$ ; and $K / F$ is separable algebraic in characteristic zero, so J-semisimplicity ascends to $K G$ .

Applications and Industry Use

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

Operator algebras

Semiprimitivity of $C^{*}$ -algebras

The argument of $(6.5)$ , read in $ℓ^{1} (G)$ and its $C^{*}$ -completion, shows that a $C^{*}$ -algebra has no nonzero topologically nil one-sided ideal, hence is J-semisimple. The group algebra result is the discrete special case of a general principle about $*$ -algebras with positive-definite trace.

Representation theory

Faithful families of representations

$rad (ℂ G) = 0$ says the simple $ℂ G$ -modules separate elements: no nonzero element of $ℂ G$ acts as zero on every irreducible representation. For infinite $G$ this is the closest available substitute for the Wedderburn decomposition.

Harmonic analysis

Convolution algebras on discrete groups

The norm $| α | = \sum | a_{g} |$ is the $ℓ^{1}$ -norm and the multiplication is convolution. Facts proved here about $ℂ G$ are the algebraic core of results about $ℓ^{1} (G)$ , and the resolvent function is the standard tool of spectral theory in that setting.

Ring theory

A benchmark for radical computations

Rickart's theorem supplies an infinite family of J-semisimple, non-artinian, generally noncommutative rings, against which conjectures about radicals and nil ideals can be tested.

The honest summary is that this result is internal to algebra and analysis. Its practical significance is that it makes character-theoretic and operator-theoretic methods for infinite discrete groups viable at all, since a nonzero radical would mean irreducible representations lose information.

Computational Notes

Computational notes cover algorithms, cost and library behaviour rather than manufacturing process.

There is no algorithm that decides J-semisimplicity of $k G$ from a presentation of $G$ : for finitely presented groups even the word problem is undecidable, so membership in $rad (k G)$ cannot be decided in general.
Radical questions do localise. For any $α \in k G$ there is a finitely generated subgroup $G_{0}$ with $α \in k G_{0}$ , and $k G_{0} \cap rad (k G) \subseteq rad (k G_{0})$ ; so if $k H$ is J-semisimple for every finitely generated $H \leq G$ then $k G$ is J-semisimple. This reduction is how the locally finite cases are handled.
For $G$ finite the computation is entirely effective: $\dim_{k} k G = | G |$ , and the radical of a finite-dimensional algebra is computed in $O (| G |^{3})$ field operations in characteristic zero, or by the Friedl–Rónyai method in characteristic $p$ .
The traces $tr (a^{n})$ appearing in $(6.6)$ are computable numerically for any explicit $a$ of finite support, but observing decay is not a proof of membership in the radical — the implication runs the other way.

Failure Modes and Common Mistakes

$(6.11)$ concerns nil left ideals, not the radical. Absence of nil ideals does not by itself give $rad (k G) = 0$ — the radical need not be nil. The bridge to J-semisimplicity is supplied by scalar-extension results that force the relevant contraction to be nil.
The footnote in Lam is important: the conclusion of $(6.11)$ holds for every field of characteristic zero, not only for algebraically closed ones, but the stronger statement is not proved there. Do not cite the general form as though it were established by this argument.
In case (b) of $(6.11)$ , the involution is not the identity — it is complex conjugation relative to a real-closed subfield. Applying the identity involution to an algebraically closed field fails immediately, since $i^{2} + 1^{2} = 0$ .
Passman's characteristic- $p$ theorem needs $G$ to be a $p'$ -group, but that hypothesis is sufficient rather than necessary: the infinite dihedral group has elements of order $2$ and yet $k G$ is J-semisimple in characteristic $2$ , as Wallace showed.

Historical Notes and Lessons Learned

1945The radical is defined for arbitrary ringsJacobson's radical makes a semisimplicity-like notion available without chain conditions, so the question of $rad (k G) = 0$ for infinite $G$ becomes meaningful.
1950RickartUsing Banach algebra methods on $ℓ^{1} (G)$ , Rickart proves $ℂ G$ and $ℝ G$ are J-semisimple for every group $G$ . The proof is analytic and gives the problem its momentum.
1950sAmitsur and HersteinIndependently, they extend the conclusion to arbitrary uncountable fields of characteristic zero, replacing analysis by a cardinality argument on the family of inverses.
1959AmitsurFor any field of characteristic zero that is not an algebraic extension of $ℚ$ , every group ring is J-semisimple. The proof runs through the absence of nil left ideals over $ℚ$ and scalar extension along a transcendence basis.
1962Passman, and ConnellThe characteristic- $p$ analogue: over a nonalgebraic extension of $𝔽_{p}$ , group rings of $p'$ -groups are J-semisimple. The positivity lemma is replaced by an orbit count on $p$ -tuples.
SinceThe prime-field obstructionEvery known class of groups gives an affirmative answer, but the cases $ℚ G$ and $𝔽_{p} G$ remain unresolved in general — an inversion of the usual situation, where the smallest fields are the easiest.

The lesson is about the division of labour between analysis and algebra. Rickart's analytic half was later replaced entirely — by cardinality arguments, then by transcendence arguments — while the algebraic half, positivity of $tr (α^{*} α)$ , survives untouched in $(6.11)$ and is still the engine of the characteristic-zero theory.

Quick Reference

Rickart $(6.4)$

rad (ℂ G) = rad (ℝ G) = 0

for every

G

Trace identity

tr (α^{*} α) = \sum_{g} | a_{g} |^{2}

Doubling

α^{*} = α \Rightarrow tr (α^{2}) \geq | tr α |^{2}

Algebraic lemma $(6.5)$ radical

\neq 0

gives

a = a^{*}

tr (a^{2^{m}}) \geq 1

Analytic lemma $(6.6)$

a \in rad \Rightarrow {tr}_{g} (a^{n}) \to 0

Resolvent

ϕ (z) = (1 - z a)^{- 1}

, entire,

ϕ' (z) = a ϕ (z)^{2}

Norm

| α | = \sum_{g} | a_{g} |

, submultiplicative

Involution test $(6.11)$

\sum a_{i}^{*} a_{i} = 0 \Rightarrow

all

a_{i} = 0

, then no nil left ideals in

k G

Statement map for §6, pages 86–92
Reference	Statement	Method
$(6.4)$	$ℂ G$ and $ℝ G$ are J-semisimple	combination of $(6.5)$ and $(6.6)$
$(6.5)$	a normalised self-adjoint radical element with traces at least $1$	positivity of the trace form
$(6.6)$	${tr}_{g} (a^{n}) \to 0$ for radical $a$	entire resolvent, Taylor expansion at $0$ , evaluate at $1$
$(6.7)$ – $(6.9)$	resolvent identity, local bound, geometric series	elementary estimates in the norm $\| \cdot \|$
$(6.10)$	why the Neumann series shortcut is invalid	commentary
$(6.11)$	no nonzero nil left ideals under a positive involution	purely algebraic; covers formally real and characteristic-zero algebraically closed $k$

Frequently Asked Questions

Why does the proof need complex analysis at all?

To justify one evaluation. The series $\sum_{n} {tr}_{g} (a^{n}) z^{n}$ is only known to represent the resolvent near the origin, but the conclusion is needed at $z = 1$ . Since the resolvent is defined and differentiable on the whole plane, the composed scalar function is entire, and an entire function is represented by its Taylor series at the origin everywhere. Without this the argument stalls at $| z | < | a |^{- 1}$ .

Does Rickart's theorem say

ℂ G

is semisimple?

No, and for infinite $G$ it cannot: no group ring of an infinite group is semisimple. It says the Jacobson radical vanishes, which is the same as semisimplicity only in the presence of the descending chain condition. For $G$ infinite, $ℂ G$ is a J-semisimple non-artinian ring.

What is the role of the involution, exactly?

It converts an arbitrary nonzero element into one whose trace is a sum of squared moduli, hence strictly positive. Positivity is what allows a normalisation with $tr (a) = 1$ and what makes the doubling inequality run. Any coefficient ring with an involution satisfying the same positivity condition supports the algebraic half of the argument, which is the content of $(6.11)$ .

Why is

ℚ G

harder than

ℂ G

The later proofs replace analysis by transcendence: one picks a transcendence basis of $k$ over $ℚ$ and uses the fact that a transcendental scalar extension forces the contracted radical to be nil. A field algebraic over $ℚ$ offers no transcendence basis to work with, so the mechanism disappears. That the smallest fields are the hardest is an inversion of the usual pattern.

Is the characteristic-

p

statement simply the same proof with

p

in place of

2

No. Positivity of a trace form is unavailable in characteristic $p$ — the example $(1 + g)^{2} = 0$ in $𝔽_{2} C_{2}$ kills it immediately. The substitute is an orbit count: for a $p'$ -group, the $p$ -tuples of group elements multiplying to $1$ contribute in orbits of size $p$ except for one singleton, giving $tr (β^{p}) = tr (β)^{p}$ . The hypothesis that $G$ has no element of order $p$ is what makes the singleton unique.

How does the real case follow from the complex case?

$ℂ G$ is generated as a left $ℝ G$ -module by the two centralising elements $1$ and $i$ . The ascent result for such extensions gives $rad (ℝ G) \subseteq rad (ℂ G)$ , and the right-hand side is zero. The finiteness of the generating set is essential — the corresponding ascent fails for infinite extensions.

References

T. Y. Lam, A First Course in Noncommutative Rings, Graduate Texts in Mathematics 131, Springer-Verlag, 1991, §6, (6.4)–(6.11) (pp. 86–92).
C. E. Rickart, “The uniqueness of norm problem in Banach algebras”, Annals of Mathematics 51 (1950).
S. A. Amitsur, “On the semi-simplicity of group algebras”, Michigan Mathematical Journal 6 (1959).
D. S. Passman, The Algebraic Structure of Group Rings, Wiley-Interscience, 1977, Chapter 7 and pp. 269–271.
I. N. Herstein, Noncommutative Rings, Carus Mathematical Monographs 15, Mathematical Association of America, 1968, Chapter 1.
C. E. Rickart, General Theory of Banach Algebras, Van Nostrand, 1960, Chapter II.

AI Suggested Questions

Write out the proof that the radical of a complex Banach algebra consists of topologically nil elements.
Prove that every $C^{*}$ -algebra is J-semisimple using the argument of the algebraic lemma.
State Amitsur's theorem on group rings over nonalgebraic extensions of $ℚ$ and identify the scalar-extension results it consumes.
Give Passman's orbit-counting proof that $k G$ has no nonzero nil left ideals for $k$ reduced of characteristic $p$ and $G$ a $p'$ -group.
Which classes of groups are known to give J-semisimple $ℚ G$ , and what obstructs the general case?
Explain Wallace's proof that the infinite dihedral group has J-semisimple group algebra in characteristic 2 despite having elements of order 2.
Compare the trace form on $ℂ G$ with the canonical trace on the group von Neumann algebra and its role in $ℓ^{2}$ -Betti numbers.

J-Semisimplicity of Group Algebras

Executive Summary

Overview

Learning Objectives

Definitions

Core Concepts

The trace form is positive definite

Why the radical is analytically small

The Banach algebra picture behind it

Key Results

Proof Techniques and Method

Normalise a radical element

Square along $2^{m}$ , not $n$

Make a Banach-valued function scalar

Worked Example

Watching $(6.5)$ run on a two-element group

Exactly where characteristic $2$ breaks the argument

Comparison and Classification

Relationship Map

Applications and Industry Use

Semiprimitivity of $C^{*}$ -algebras

Faithful families of representations

Convolution algebras on discrete groups

A benchmark for radical computations

Computational Notes

Failure Modes and Common Mistakes

Historical Notes and Lessons Learned

Quick Reference

Frequently Asked Questions

References

AI Suggested Questions

Continue learning

01Executive Summary

02Overview

03Learning Objectives

04Definitions

05Core Concepts

The trace form is positive definite

Why the radical is analytically small

The Banach algebra picture behind it

06Key Results

07Proof Techniques and Method

Normalise a radical element

Square along 2m, not n

Make a Banach-valued function scalar

08Worked Example

Watching (6.5) run on a two-element group

Exactly where characteristic 2 breaks the argument

09Comparison and Classification

10Relationship Map

11Applications and Industry Use

Semiprimitivity of C∗-algebras

Faithful families of representations

Convolution algebras on discrete groups

A benchmark for radical computations

12Computational Notes

13Failure Modes and Common Mistakes

14Historical Notes and Lessons Learned

15Quick Reference

16Frequently Asked Questions

17Related KEVOS Topics

18References

19AI Suggested Questions

Continue learning

Executive Summary

Overview

Learning Objectives

Definitions

Core Concepts

Key Results

Proof Techniques and Method

Square along $2^{m}$ , not $n$

Worked Example

Watching $(6.5)$ run on a two-element group

Exactly where characteristic $2$ breaks the argument

Comparison and Classification

Relationship Map

Applications and Industry Use

Semiprimitivity of $C^{*}$ -algebras

Computational Notes

Failure Modes and Common Mistakes

Historical Notes and Lessons Learned

Quick Reference

Frequently Asked Questions

Related KEVOS Topics

References

AI Suggested Questions