Nil and Nilpotent Ideals and the Radical

Q: Is the sum of all nilpotent ideals of a ring nilpotent?

It is always nil — any element lies in a finite sum, which is nilpotent by (4.10) — but it need not be nilpotent. That failure is exactly why Wedderburn's definition of the radical does not survive outside rings with chain conditions, and why the lower nilradical is defined by a transfinite iteration rather than a single sum.

Q: Does (4.11) have a converse?

Not in general: rad k[[x]] = (x) is not nil. It has a converse under hypotheses. For left artinian rings the radical is nilpotent (4.12); for algebraic algebras over a field, and for k-algebras of dimension less than |k| by Amitsur's theorem, the radical is the largest nil ideal.

Q: Why is the Koethe conjecture hard if (4.10) works so easily?

Because the absorb step in (4.10) needs a uniform exponent. For nil ideals each element carries its own exponent and no pigeonhole argument can bound the product length. Krempa showed the conjecture is equivalent to M_2(N) being nil for every nil ring N, which shows how far the difficulty is from a bookkeeping issue.

Q: Can a ring with identity be nil?

No: 1 is not nilpotent. Nil rings are studied without identity, which is why the Koethe literature works in the category of rngs and why Lam's exercises develop the radical for rings possibly lacking an identity.

Q: What is the relationship between nil ideals and nilpotent elements?

Weaker than intuition suggests. A nil ideal consists of nilpotent elements by definition, but in a noncommutative ring the nilpotent elements need not form an ideal or even an additive subgroup — M_2(k) makes this vivid — so there is no passage from a supply of nilpotent elements to a nil ideal.

Q: Does nilpotency of an ideal pass to matrix rings?

Yes. If A^n = 0 then M_m(A)^n = M_m(A^n) = 0, so nilpotency and its index are Morita-stable. The corresponding question for nil ideals is one of the standard equivalents of the Koethe conjecture.

Executive Summary

Two finiteness conditions on an ideal look alike and behave very differently. $𝔄$ is nilpotent if $𝔄^{n} = 0$ for some fixed $n$ — one bound serving all products of $n$ elements. $𝔄$ is nil if each individual element has some power equal to zero, with no bound across the ideal.

Lam's $(4.11)$ is the bridge to the radical: any nil left or right ideal is contained in $rad R$ , proved in two lines with a geometric series. The reverse containment fails badly — $rad k [[x]] = (x)$ contains no nonzero nilpotent element — and repairing it requires a chain condition, which is the subject of The Radical of a Left Artinian Ring Is Nilpotent.

(4.9)Definitions of nil and nilpotent

(4.10)Finite sums of nilpotent left ideals

(4.11)Nil implies inside the radical

OpenKoethe conjecture, since 1930

Overview

Wedderburn's original radical was the largest nilpotent ideal. That definition works for finite-dimensional algebras and for artinian rings and collapses outside them, because the sum of all nilpotent ideals of a general ring need not be nilpotent. The material on this page is the anatomy of that collapse.

Three facts are established. Nilpotency is stable under finite sums of left ideals $(4.10)$ . Nilness is enough to force membership of the Jacobson radical $(4.11)$ . And nilness is strictly weaker than nilpotency, with an explicit commutative witness.

𝔄^{n} = 0 ⟹ 𝔄 nil ⟹ 𝔄 \subseteq rad R,

(4.11)

Both implications are strict; neither reverses without extra hypotheses.

Learning Objectives

State $(4.9)$ precisely, including what $𝔄^{n} = 0$ means as a condition on products.
Prove $(4.10)$ by the pigeonhole-and-absorb argument and see where left ideal is used.
Prove $(4.11)$ using the finite geometric series inverse of $1 - x y$ .
Exhibit a nil ideal that is not nilpotent and verify both halves of the claim.
Show that a nilpotent element need not generate a nil left ideal.
Order the lower nilradical, Levitzki radical, upper nilradical and Jacobson radical by containment.

Definitions

Definition(4.9)Nil and nilpotent

Let $𝔄$ be a left, right or two-sided ideal of $R$ . Then $𝔄$ is nil if every element of $𝔄$ is nilpotent, and nilpotent if $𝔄^{n} = 0$ for some $n \geq 1$ , where $𝔄^{n}$ denotes the additive subgroup generated by all products $a_{1} a_{2} \dots a_{n}$ with $a_{i} \in 𝔄$ .

So nilpotency says $a_{1} a_{2} \dots a_{n} = 0$ for every choice of $n$ elements, which is much stronger than requiring each single $a$ to satisfy $a^{m} = 0$ for some $m$ depending on $a$ .

Nilpotency index: The least $n$ with $𝔄^{n} = 0$ . Finite for a nilpotent ideal, undefined for a nil ideal that is not nilpotent.
Locally nilpotent: Every finitely generated subring is nilpotent. Strictly between nilpotent and nil; the largest locally nilpotent ideal is the Levitzki radical.
${Nil}_{*} R$: The lower nilradical or Baer radical: the intersection of all prime ideals of $R$ .
${Nil}^{*} R$: The upper nilradical: the sum of all nil two-sided ideals, itself nil, hence the largest nil two-sided ideal.
$Nil R$: For a commutative ring, the set of nilpotent elements, which is then an ideal. In a noncommutative ring this set need not be closed under addition.

Nil and nilpotent are conditions on ideals, not on rings; a ring with identity is never nil, since 1 is not nilpotent.

Core Concepts

Why nilpotency behaves and nilness does not

Nilpotency is a statement about a single integer, so it survives operations that are uniform in that integer: finite sums, passage to the generated two-sided ideal, and images under ring surjections. Nilness is a statement quantified separately over each element, and quantifiers that do not commute with sums are exactly what break under addition.

Concretely: if $𝔄$ is a nilpotent left ideal with $𝔄^{n} = 0$ , then the two-sided ideal it generates, $𝔄 R$ , satisfies $(𝔄 R)^{n} = 𝔄 (R 𝔄)^{n - 1} R \subseteq 𝔄^{n} R = 0$ , using $R 𝔄 \subseteq 𝔄$ . The same computation with nil in place of nilpotent is meaningless, and its conclusion is the Koethe conjecture.

$rad R$ contains every nil one-sided ideal

${Nil}^{*} R$ largest nil two-sided ideal

$Levitzki (R)$ largest locally nilpotent ideal

${Nil}_{*} R$ lower nilradical, intersection of the primes

Sum of all nilpotent idealsnil, but not necessarily nilpotent

Every containment shown is strict for suitable rings, and each of the inner three radicals gets its own page in this collection. The outermost containment, into $rad R$ , is $(4.11)$ .

One-sided nilpotency is not one-sided

$(4.11)$ is stated for a nil left ideal or a nil right ideal, and both give the same conclusion because $rad R$ is left-right symmetric. The proof, however, changes side: for a left ideal one inverts $1 - x y$ , for a right ideal one inverts $1 - y x$ .

Key Results

Lemma(4.10)Finite sums of nilpotent left ideals

Let $𝔄_{1}, \dots, 𝔄_{m}$ be finitely many nilpotent left ideals of $R$ . Then $𝔄_{1} + \dots + 𝔄_{m}$ is a nilpotent left ideal.

Proof

By induction it suffices to treat $m = 2$ . Let $𝔄, 𝔅$ be left ideals with $𝔄^{n} = 𝔅^{n} = 0$ , and set $ℭ = 𝔄 + 𝔅$ . We claim $ℭ^{2 n} = 0$ .

A generator of $ℭ^{2 n}$ expands into a sum of products $c_{1} c_{2} \dots c_{2 n}$ in which each $c_{i}$ lies in $𝔄$ or in $𝔅$ . By pigeonhole at least $n$ of the $2 n$ positions belong to the same one of the two ideals; say positions $i_{1} < i_{2} < \dots < i_{n}$ all carry elements of $𝔄$ .

Group the product as $(c_{1} \dots c_{i_{1}}) (c_{i_{1} + 1} \dots c_{i_{2}}) \dots (c_{i_{n - 1} + 1} \dots c_{i_{n}}) (c_{i_{n} + 1} \dots c_{2 n})$ . Each of the first $n$ blocks ends in an element of $𝔄$ preceded by elements of $R$ , hence lies in $R 𝔄 \subseteq 𝔄$ because $𝔄$ is a left ideal. So the product lies in $𝔄^{n} R = 0$ . The case with $n$ positions in $𝔅$ is identical.

Lemma(4.11)Nil one-sided ideals live in the radical

Let $𝔄$ be a nil left ideal or a nil right ideal of $R$ . Then $𝔄 \subseteq rad R$ .

Proof

Suppose first that $𝔄$ is a left ideal and take $y \in 𝔄$ , $x \in R$ . Then $x y \in 𝔄$ , so $(x y)^{n} = 0$ for some $n$ , and

(1 - x y) (\sum_{i = 0}^{n - 1} (x y)^{i}) = 1 - (x y)^{n} = 1 = (\sum_{i = 0}^{n - 1} (x y)^{i}) (1 - x y) .

So $1 - x y$ is a unit, in particular left-invertible, for every $x \in R$ ; the element characterisation of the radical gives $y \in rad R$ .

If instead $𝔄$ is a right ideal, then for $y \in 𝔄$ and $x \in R$ the element $y x$ is nilpotent and the same series inverts $1 - y x$ . Applying the right-handed form of the characterisation — legitimate because $rad R$ coincides with the intersection of the maximal right ideals — again gives $y \in rad R$ .

Corollary—The upper nilradical sits inside the Jacobson radical

${Nil}^{*} R \subseteq rad R$ , and more generally every nil subideal on either side is contained in $rad R$ . In particular a semiprimitive ring has no nonzero nil one-sided ideals.

Remark—The containment is strict and cannot be reversed

$rad k [[x]] = (x)$ , and $k [[x]]$ is a domain, so its radical contains no nonzero nilpotent element at all. Hence $rad R$ is in general neither nil nor nilpotent, and $(4.11)$ is a one-way street. The converse becomes true under a chain condition, where by $(4.12)$ the radical is even nilpotent.

Counterexample—A nilpotent element need not generate a nil left ideal

In $R = M_{2} (k)$ the matrix unit $e_{12}$ satisfies $e_{12}^{2} = 0$ , but $R e_{12}$ consists of all matrices with zero first column and therefore contains the idempotent $e_{22}$ . So $R e_{12}$ is not nil — consistent with $rad M_{2} (k) = 0$ , which forbids any nonzero nil one-sided ideal.

Proof Techniques and Method

The reusable moves in these three proofs.

Move 1

Pigeonhole then absorb

To bound a product from a sum of ideals, count which summand supplies at least half the factors, then use the one-sided ideal property to absorb the intervening factors into that summand. This is the whole of $(4.10)$ .

Move 2

The finite geometric series

If $z^{n} = 0$ then $1 - z$ is a unit with inverse $1 + z + \dots + z^{n - 1}$ . Every proof that nilpotence implies membership of the radical is this identity plus the element characterisation.

Move 3

Test one-sided claims on matrix units

$M_{2} (k)$ is the standard laboratory: it has abundant nilpotent elements and zero radical, so any claim of the form nilpotent elements generate nil ideals dies there immediately.

Move 1 is worth isolating because it is the only place in $(4.10)$ where left ideal rather than subgroup is used, and it is where the analogous statement for nil ideals would have to be repaired.

Worked Example

A nil ideal that is not nilpotent

Let $k$ be a field and consider the commutative ring

R = k [x_{1}, x_{2}, x_{3}, \dots] / (x_{1}^{2}, x_{2}^{2}, x_{3}^{2}, \dots), 𝔄 = (x_{1}, x_{2}, x_{3}, \dots) .

(E.1)

Infinitely many square-zero variables; $𝔄$ is the augmentation ideal.

** $𝔄$ is nil.** An element $f \in 𝔄$ is a polynomial with zero constant term involving finitely many variables, say $x_{1}, \dots, x_{m}$ . Every monomial of degree $m + 1$ in $x_{1}, \dots, x_{m}$ must repeat a variable and hence vanishes, so $f^{m + 1} = 0$ .

** $𝔄$ is not nilpotent.** For every $n$ , the squarefree monomial $x_{1} x_{2} \dots x_{n}$ is a nonzero element of $𝔄^{n}$ , so no power of $𝔄$ vanishes.

Identification of the radical. $R / 𝔄 ≅ k$ is a field, so $𝔄$ is a maximal ideal; and $𝔄 \subseteq rad R$ by $(4.11)$ . Hence $rad R = 𝔄$ , and this is a ring whose Jacobson radical is nil but not nilpotent.

Sum of nilpotent ideals, checked

In $T_{3} (k)$ let $𝔄 = k e_{12} + k e_{13}$ and $𝔅 = k e_{13} + k e_{23}$ . Both are left ideals of $T_{3} (k)$ — a check on the three-by-three products — with $𝔄^{2} = 𝔅^{2} = 0$ . Their sum is the strictly upper triangular ideal $J$ , and $(4.10)$ predicts nilpotency with index at most $4$ ; in fact $J^{3} = 0$ , so the bound $2 n$ from the proof is not sharp.

Comparison and Classification

Which finiteness conditions hold for standard radicals and ideals
	Nilpotent	Locally nilpotent	Nil	Inside $rad R$
Strictly upper triangular ideal of $T_{n} (k)$	yes	yes	yes	yes
$𝔄$ of $(E . 1)$	no	yes	yes	yes
$(x) \subseteq k [[x]]$	no	no	no	yes
${Nil}_{*} R$ , general $R$	no	yes	yes	yes
${Nil}^{*} R$ , general $R$	no	no	yes	yes
$rad R$ , general $R$	no	no	no	yes
$rad R$ , $R$ left artinian	yes	yes	yes	yes
$R e_{12} \subseteq M_{2} (k)$	no	no	no	no

Which finiteness conditions hold for standard radicals and ideals

Nil against nilpotent, side by side
Property	Nilpotent ideals	Nil ideals
Closed under finite sums	yes, by $(4.10)$	yes for two-sided ideals; open for one-sided (Koethe)
Generates a two-sided ideal of the same type	yes: $(𝔄 R)^{n} = 0$	open in general
Closed under arbitrary sums	no — the sum can be non-nilpotent	yes for two-sided ideals
Contained in the Jacobson radical	yes, via nil and $(4.11)$	yes, by $(4.11)$
Largest one exists	only under a chain condition	yes among two-sided: ${Nil}^{*} R$
Survives to matrix rings	yes: $M_{n} (𝔄)$ is nilpotent	open — equivalent to Koethe

Relationship Map

𝔄

nilpotent

𝔄

locally nilpotent

𝔄

nil

𝔄 \subseteq rad R

None of the three arrows reverses. The first fails for $(E . 1)$ , the second for suitable Golod–Shafarevich style constructions of nil algebras that are not locally nilpotent, and the third for $k [[x]]$ .

You have a one-sided ideal $𝔄$ and want to know it is in $rad R$ .

It is nilpotentDone: nilpotent implies nil implies inside the radical, by

(4.11)

It is nilDone, by

(4.11)

directly. No chain condition and no two-sidedness is needed.

Neither, but every $1 - x y$ is invertibleUse the element characterisation of the radical instead; nilness was only ever a convenient sufficient condition.

You want the converseYou need

R

left artinian, and then

(4.12)

gives that

rad R

is nilpotent, hence nil.

Applications and Industry Use

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

Symbolic computation

Nilpotency as a termination certificate

Algorithms that filter a module by powers of an ideal terminate exactly when that ideal is nilpotent. A nil-but-not-nilpotent ideal gives a non-terminating filtration, which is why implementations require finite dimension over a field.

Deformation theory

Square-zero extensions

Obstruction calculus is built on nilpotent ideals with $𝔄^{2} = 0$ , where the geometric series inverse of $1 - z$ truncates after one term and lifting is controlled by a single cohomology class.

Coding theory

Chain rings and their filtrations

A finite chain ring has nilpotent maximal ideal $(π)$ , and codes over it are analysed through the finite tower $(π) \supset (π^{2}) \supset \dots$ . Nilpotency, not merely nilness, is what makes the tower finite.

Representation theory

Loewy layers

For a finite-dimensional algebra the powers of the radical give the Loewy filtration of every module, and the number of layers is the nilpotency index of the radical.

Failure Modes and Common Mistakes

Do not use $𝔄^{n} = 0$ to mean $a^{n} = 0$ for all $a \in 𝔄$ ; the former is strictly stronger and the notation invites the confusion.
Do not assume the sum of all nilpotent ideals is nilpotent; it is always nil, and nilpotent only under a chain condition.
Do not quote a nil-implies-nilpotent result without its hypothesis: the true statements are for left artinian rings $(4.13)$ , and for finitely generated ideals of noetherian rings.

Historical Notes and Lessons Learned

1908–27The nilpotent radicalWedderburn defines the radical of a finite-dimensional algebra as its largest nilpotent ideal; Artin extends the theory to rings with the descending chain condition, where such a largest nilpotent ideal still exists.
1930Koethe's questionKoethe asks whether a ring with a nonzero nil one-sided ideal must have a nonzero nil two-sided ideal. The question remains open.
1939–45Levitzki and JacobsonLevitzki studies locally nilpotent ideals, producing the radical that bears his name; Jacobson's 1945 definition replaces nilpotence by quasi-regularity and works for all rings.
1943–56Baer and AmitsurBaer introduces the lower nilradical as the intersection of the prime ideals; Amitsur analyses the radical of polynomial rings and shows the radical of an algebra of small dimension over a large field is nil.
1972Krempa's reformulationsKrempa shows the Koethe conjecture is equivalent to the statement that M_2(N) is nil for every nil ring N, and to a statement about polynomial rings over nil rings.
2000SmoktunowiczSmoktunowicz constructs a nil ring whose polynomial ring is not nil, settling Amitsur's conjecture negatively and sharpening the landscape around Koethe's question without resolving it.

The methodological lesson is Jacobson's: an invariant defined by a uniform bound is fragile, one defined by the action on modules is not. The nil ideals remain the place where that fragility is still visible as an open problem.

Quick Reference

Nilpotent

𝔄^{n} = 0

: every product of

n

elements vanishes

Nilevery

a \in 𝔄

has

a^{m (a)} = 0

(4.10)a finite sum of nilpotent left ideals is nilpotent

(4.11)nil left or right ideal

\Rightarrow

inside

rad R

Inverse used

(1 - z)^{- 1} = 1 + z + \dots + z^{n - 1}

when

z^{n} = 0

Largest nil ideal

{Nil}^{*} R

, the upper nilradical

Radical not nil

rad k [[x]] = (x)

Open problemKoethe: is every nil one-sided ideal inside a nil two-sided ideal?

Standard witnesses
Phenomenon	Witness	Check
nilpotent ideal	strictly upper triangular in $T_{n} (k)$	$J^{n} = 0$ , $J^{n - 1} \neq 0$
nil, not nilpotent	$(x_{1}, x_{2}, \dots)$ in $(E . 1)$	$x_{1} \dots x_{n} \neq 0$ for all $n$
radical not nil	$(x) \subseteq k [[x]]$	the ring is a domain
nilpotent element, non-nil ideal	$e_{12} \in M_{2} (k)$	$R e_{12} ∋ e_{22} = e_{22}^{2}$
nilpotents not closed under addition	$e_{12} + e_{21} \in M_{2} (k)$	its square is the identity

Frequently Asked Questions

Is the sum of all nilpotent ideals of a ring nilpotent?

It is always nil — any element lies in a finite sum, which is nilpotent by $(4.10)$ — but it need not be nilpotent. That failure is exactly why Wedderburn's definition of the radical does not survive outside rings with chain conditions, and why the lower nilradical is defined by a transfinite iteration rather than a single sum.

Does

(4.11)

have a converse?

Not in general: $rad k [[x]] = (x)$ is not nil. It has a converse under hypotheses. For left artinian rings the radical is nilpotent $(4.12)$ ; for algebraic algebras over a field, and for $k$ -algebras of dimension less than $| k |$ by Amitsur's theorem, the radical is the largest nil ideal.

Why is the Koethe conjecture hard if

(4.10)

works so easily?

Because the absorb step in $(4.10)$ needs a uniform exponent. For nil ideals each element carries its own exponent and no pigeonhole argument can bound the product length. Krempa showed the conjecture is equivalent to $M_{2} (N)$ being nil for every nil ring $N$ , which shows how far the difficulty is from a bookkeeping issue.

Can a ring with identity be nil?

No: $1$ is not nilpotent. Nil rings are studied without identity, which is why the Koethe literature works in the category of rngs and why Lam's exercises develop the radical for rings possibly lacking an identity.

What is the relationship between nil ideals and nilpotent elements?

Weaker than intuition suggests. A nil ideal consists of nilpotent elements by definition, but in a noncommutative ring the nilpotent elements need not form an ideal or even an additive subgroup — $M_{2} (k)$ makes this vivid — so there is no passage from a supply of nilpotent elements to a nil ideal.

Does nilpotency of an ideal pass to matrix rings?

Yes. If $𝔄^{n} = 0$ then $M_{m} (𝔄)^{n} = M_{m} (𝔄^{n}) = 0$ , so nilpotency and its index are Morita-stable. The corresponding question for nil ideals is one of the standard equivalents of the Koethe conjecture.

References

T. Y. Lam, A First Course in Noncommutative Rings, Graduate Texts in Mathematics 131, Springer-Verlag, 1991, §4, (4.9)–(4.11) and §10 (pp. 56–58).
N. Jacobson, Structure of Rings, American Mathematical Society Colloquium Publications 37, revised edition, 1964, Chapter I.
I. N. Herstein, Noncommutative Rings, Carus Mathematical Monographs 15, Mathematical Association of America, 1968, Chapter 1.
L. H. Rowen, Ring Theory, Volume I, Academic Press, 1988, Chapter 2.
A. Smoktunowicz, “Polynomial rings over nil rings need not be nil”, Journal of Algebra 233 (2000), 427–436.

AI Suggested Questions

Prove that the two-sided ideal generated by a nilpotent left ideal is nilpotent, with an explicit index.
Show that the sum of two nil two-sided ideals is nil, and explain why the argument fails for one-sided ideals.
Construct a nil algebra that is not locally nilpotent, and identify which radical separates the two conditions.
State three equivalent forms of the Koethe conjecture and prove one equivalence.
For which classes of rings is the Jacobson radical known to equal the upper nilradical?
How does Smoktunowicz's example of a nil ring with non-nil polynomial ring bear on the Koethe conjecture?

01Executive Summary

02Overview

03Learning Objectives

04Definitions

05Core Concepts

Why nilpotency behaves and nilness does not

One-sided nilpotency is not one-sided

06Key Results

07Proof Techniques and Method

Pigeonhole then absorb

The finite geometric series

Test one-sided claims on matrix units

08Worked Example

A nil ideal that is not nilpotent

Sum of nilpotent ideals, checked

09Comparison and Classification

10Relationship Map

11Applications and Industry Use

Nilpotency as a termination certificate

Square-zero extensions

Chain rings and their filtrations

Loewy layers

12Failure Modes and Common Mistakes

13Historical Notes and Lessons Learned

14Quick Reference

15Frequently Asked Questions

16Related KEVOS Topics

17References

18AI Suggested Questions

Continue learning