Triangular Rings: Ideals, Modules and Chain Conditions

Q: Is a triangular ring ever semisimple?

Only when M = 0, in which case it is just R × S. For M ≠ 0 the copy of M is a nonzero square-zero ideal, hence lies in the radical, so rad A ≠ 0 and A cannot be semisimple. This is why triangular rings are the standard first examples of non-semisimple artinian rings.

Q: How do I see the simple modules?

Since M ⊆ rad A, every simple left A-module is a simple module over A/rad A ≅ (R/rad R) × (S/rad S). So the simple left A-modules are exactly the simple left R-modules and the simple left S-modules, pulled back along the two quotient maps. For R and S fields there are precisely two, both one-dimensional.

Executive Summary

Given rings $R$ and $S$ and an $(R, S)$ -bimodule $M$ , formal matrix multiplication makes the set of arrays $(\begin{matrix} r & m \\ 0 & s \end{matrix})$ into a ring $A$ . The construction costs nothing and delivers a great deal: the entire one-sided ideal theory of $A$ is readable off the module structures of $M$ .

That readability is the point. A left ideal of $A$ involves $M$ only through its left $R$ -module structure; a right ideal involves $M$ only through its right $S$ -module structure. Choosing a bimodule whose two structures behave differently — $ℝ$ as an $(ℝ, ℚ)$ -bimodule is the canonical choice — produces a ring whose left and right behaviour diverge as sharply as one likes. Nearly every one-sided counterexample in elementary ring theory is manufactured this way.

$R \oplus M \oplus S$ Additive decomposition

$M^{2} = 0$ The bimodule as an ideal

3Data items:

R

S

M

(1.14)–(1.24)Lam's numbering

Overview

Let $R$ and $S$ be rings with identity and let $M$ be an $(R, S)$ -bimodule: a left $R$ -module and a right $S$ -module such that $(r m) s = r (m s)$ for all $r \in R$ , $m \in M$ , $s \in S$ . The triangular ring is

A = (\begin{matrix} R & M \\ 0 & S \end{matrix}) = {(\begin{matrix} r & m \\ 0 & s \end{matrix}) : r \in R, m \in M, s \in S},

(1.14)

(\begin{matrix} r & m \\ 0 & s \end{matrix}) (\begin{matrix} r' & m' \\ 0 & s' \end{matrix}) = (\begin{matrix} r r' & r m' + m s' \\ 0 & s s' \end{matrix}) .

(1.15)

Addition is entrywise. The bimodule axiom $(r m) s = r (m s)$ is not needed to define this product, but it is needed for associativity.

Familiar rings are special cases. With $R = S = M = k$ a field, $A$ is the ring of upper triangular $2 \times 2$ matrices. With $R = S = ℤ$ and $M = ℤ / 2 ℤ$ one gets a small ring exhibiting a left zero-divisor that is not a right zero-divisor. With $R = ℝ$ , $S = ℚ$ , $M = ℝ$ one gets a ring that is left artinian but not right noetherian.

Learning Objectives

Verify that $(1.15)$ is associative and identify where the bimodule axiom is used.
Read off from the multiplication chart which of $R$ , $M$ , $S$ is a left ideal, a right ideal, or an ideal of $A$ .
State and prove the classification $(1.17)$ of left, right and two-sided ideals.
Compute $rad A$ and the simple left $A$ -modules.
Prove the chain condition theorem $(1.22)$ and deduce $(1.23)$ and $(1.24)$ .
Exhibit a concrete left noetherian ring that is not right noetherian.

Definitions

Construction(1.14)The triangular ring

Given rings $R$ , $S$ and an $(R, S)$ -bimodule $M$ , the abelian group $A = R \oplus M \oplus S$ with the multiplication $(1.15)$ is an associative ring with identity $(\begin{matrix} 1 & 0 \\ 0 & 1 \end{matrix})$ . It is customary to identify $R$ , $M$ and $S$ with the corresponding subgroups of $A$ and to write elements as $r + m + s$ .

$e_{11}, e_{22}$: The idempotents $(\begin{matrix} 1 & 0 \\ 0 & 0 \end{matrix})$ and $(\begin{matrix} 0 & 0 \\ 0 & 1 \end{matrix})$ . They are orthogonal, sum to $1$ , and $e_{11} A e_{11} ≅ R$ , $e_{22} A e_{22} ≅ S$ , $e_{11} A e_{22} ≅ M$ , $e_{22} A e_{11} = 0$ .
Formal triangular matrix ring: A common alternative name, emphasising that the entries live in three different sets and that the product is formal rather than an actual matrix product.
$M$ as an ideal: The subgroup $(\begin{matrix} 0 & M \\ 0 & 0 \end{matrix})$ is a two-sided ideal of $A$ with square zero.
Trivial extension: The degenerate case $R = S$ with $M$ an $R$ -bimodule, giving the ring $R ⋉ M$ ; the triangular ring is the analogous construction with two different corner rings.

Everything below is stated for the upper triangular convention. The lower triangular ring is the opposite ring of the upper triangular one built from the same data with sides exchanged.

Core Concepts

The multiplication chart

Identifying $A = R \oplus M \oplus S$ , the entire multiplicative structure is captured by a nine-cell table recording which summand receives each product.

Products of the three summands (row times column)
	$R$	$M$	$S$
$R$	$R$	$M$	$0$
$M$	$0$	$0$	$M$
$S$	$0$	$0$	$S$

Reading the table by rows and columns gives four immediate consequences: $R$ is a left ideal (the $R$ -column is $R$ , $0$ , $0$ ), $S$ is a right ideal (the $S$ -row is $0$ , $0$ , $S$ ), $M$ is a two-sided ideal with $M^{2} = 0$ , and $R \oplus S$ is a subring. Moreover $R \oplus M$ and $M \oplus S$ are ideals, with

A / (R \oplus M) ≅ S, A / (M \oplus S) ≅ R .

(1.16)

So $R$ and $S$ are quotient rings of $A$ , which is how they inherit chain conditions from $A$ .

Why the two sides separate

A left ideal $I$ is stable under multiplication by $e_{11}$ and $e_{22}$ on the left, and those two idempotents split $A$ into $(R \oplus M) \oplus S$ . So $I = (I \cap (R \oplus M)) \oplus (I \cap S)$ , and the first summand is a left $R$ -module. The right $S$ -module structure of $M$ never appears. Dually for right ideals. The construction has therefore decoupled the two sides by design.

The radical and the simple modules

Because $M$ is a square-zero ideal it is contained in every prime ideal and in $rad A$ . Killing it leaves $R \times S$ , so every simple left $A$ -module is either a simple left $R$ -module (with $M$ and $S$ acting as zero) or a simple left $S$ -module pulled back along $A \to S$ . This is the fastest route to $rad A$ , computed below.

Key Results

Proposition(1.17)Complete description of the ideals

Let $A = (\begin{matrix} R & M \\ 0 & S \end{matrix})$ with $M$ an $(R, S)$ -bimodule.

The left ideals of $A$ are exactly the sets $I_{1} \oplus I_{2}$ where $I_{2}$ is a left ideal of $S$ and $I_{1}$ is a left $R$ -submodule of $R \oplus M$ with $M I_{2} \subseteq I_{1}$ .
The right ideals of $A$ are exactly the sets $J_{1} \oplus J_{2}$ where $J_{1}$ is a right ideal of $R$ and $J_{2}$ is a right $S$ -submodule of $M \oplus S$ with $J_{1} M \subseteq J_{2}$ .
The two-sided ideals of $A$ are exactly the sets $K_{1} \oplus K_{0} \oplus K_{2}$ where $K_{1}$ is an ideal of $R$ , $K_{2}$ is an ideal of $S$ , and $K_{0}$ is an $(R, S)$ -sub-bimodule of $M$ with $K_{1} M + M K_{2} \subseteq K_{0}$ .

Proof

Sufficiency in all three cases is a direct reading of the multiplication chart. For (1): $R I_{1} \subseteq I_{1}$ since $I_{1}$ is a left $R$ -submodule; $M I_{1} = S I_{1} = 0$ since $M \cdot (R \oplus M) = S \cdot (R \oplus M) = 0$ ; $R I_{2} \subseteq R S = 0$ ; $M I_{2} \subseteq I_{1}$ by hypothesis; $S I_{2} \subseteq I_{2}$ . So $A (I_{1} \oplus I_{2}) \subseteq I_{1} \oplus I_{2}$ .

Necessity for (1). Let $I$ be a left ideal and let $a = (\begin{matrix} r & m \\ 0 & s \end{matrix}) \in I$ . Then $e_{11} a = (\begin{matrix} r & m \\ 0 & 0 \end{matrix})$ and $e_{22} a = (\begin{matrix} 0 & 0 \\ 0 & s \end{matrix})$ both lie in $I$ , and they sum to $a$ . Hence $I = I_{1} \oplus I_{2}$ with $I_{1} = I \cap (R \oplus M)$ and $I_{2} = I \cap S$ .

Now $I_{1}$ is stable under left multiplication by $R$ , so it is a left $R$ -submodule of $R \oplus M$ ; and $I_{2}$ is stable under left multiplication by $S$ , so it is a left ideal of $S$ . Finally

M I_{2} = M (I \cap S) \subseteq I \cap M \subseteq I \cap (R \oplus M) = I_{1},

using $M \cdot S \subseteq M$ . Part (2) is the same argument on the other side.

Necessity for (3). Let $K$ be an ideal and $a \in K$ as above. Then $e_{11} a e_{11} = (\begin{matrix} r & 0 \\ 0 & 0 \end{matrix})$ and $e_{22} a e_{22} = (\begin{matrix} 0 & 0 \\ 0 & s \end{matrix})$ lie in $K$ , hence so does $a$ minus their sum, namely $(\begin{matrix} 0 & m \\ 0 & 0 \end{matrix})$ . Therefore $K = K_{1} \oplus K_{0} \oplus K_{2}$ with $K_{1} = K \cap R$ , $K_{0} = K \cap M$ , $K_{2} = K \cap S$ . Since $K$ is two-sided, $K_{1}$ is an ideal of $R$ , $K_{2}$ an ideal of $S$ , and $K_{0}$ an $(R, S)$ -sub-bimodule; and $K_{1} M + M K_{2} \subseteq K \cap M = K_{0}$ . $□$

Proposition—Radical of a triangular ring

With $A$ as above, $rad A = (\begin{matrix} rad R & M \\ 0 & rad S \end{matrix})$ , and $A / rad A ≅ (R / rad R) \times (S / rad S)$ .

Proof

Write $K$ for the displayed set; by $(1.17) (3)$ it is an ideal, since $rad R$ and $rad S$ are ideals and $(rad R) M + M (rad S) \subseteq M$ . Clearly $A / K ≅ (R / rad R) \times (S / rad S)$ , whose radical is zero, so $rad A \subseteq K$ .

Conversely let $a = (\begin{matrix} r & m \\ 0 & s \end{matrix}) \in K$ and let $b = (\begin{matrix} r' & m' \\ 0 & s' \end{matrix}) \in A$ be arbitrary. Then $1 - b a$ has diagonal entries $1 - r' r$ and $1 - s' s$ , which are units of $R$ and of $S$ respectively because $r \in rad R$ and $s \in rad S$ . A triangular array with invertible diagonal entries is invertible: its inverse is $(\begin{matrix} u & - u v w \\ 0 & w \end{matrix})$ where $u, w$ are the inverses of the diagonal entries and $v$ is the off-diagonal entry. Hence $1 - b a \in U (A)$ for every $b$ , so $a \in rad A$ . $□$

Theorem(1.22)Chain conditions for triangular rings

Let $A = (\begin{matrix} R & M \\ 0 & S \end{matrix})$ with $M$ an $(R, S)$ -bimodule. Then $A$ is left noetherian if and only if $R$ and $S$ are left noetherian and $M$ is noetherian as a left $R$ -module. Dually, $A$ is right noetherian if and only if $R$ and $S$ are right noetherian and $M$ is noetherian as a right $S$ -module. Both statements remain true with noetherian replaced throughout by artinian.

Proof

We prove the left noetherian case; the other three are the same argument with the obvious substitutions.

Necessity. By $(1.16)$ , $R$ and $S$ are quotient rings of $A$ , and a quotient of a left noetherian ring is left noetherian. If $M_{1} \subseteq M_{2} \subseteq \dots$ is an ascending chain of left $R$ -submodules of $M$ , then $(\begin{matrix} 0 & M_{i} \\ 0 & 0 \end{matrix})$ is an ascending chain of left ideals of $A$ by $(1.17) (1)$ — take $I_{2} = 0$ . It stabilises, hence so does the original chain.

Sufficiency. Assume $R$ , $S$ left noetherian and $M$ noetherian as a left $R$ -module. Then $R \oplus M$ is a noetherian left $R$ -module, being a direct sum of two noetherian ones. Let $I^{(1)} \subseteq I^{(2)} \subseteq \dots$ be an ascending chain of left ideals of $A$ . By $(1.17) (1)$ , $I^{(n)} = (I^{(n)} \cap (R \oplus M)) \oplus (I^{(n)} \cap S)$ .

The chain of contractions $I^{(n)} \cap S$ consists of left ideals of $S$ , so it stabilises. The chain of contractions $I^{(n)} \cap (R \oplus M)$ consists of left $R$ -submodules of the noetherian module $R \oplus M$ , so it stabilises too. Beyond the larger of the two stabilisation indices, the direct sum decomposition shows $I^{(n)}$ is constant. Hence $A$ is left noetherian. $□$

Corollary(1.23)Left noetherian without right noetherian

Let $S$ be a commutative noetherian domain that is not a field, and let $R$ be its field of fractions, regarded as an $(R, S)$ -bimodule via $M = R$ . Then $A = (\begin{matrix} R & R \\ 0 & S \end{matrix})$ is left noetherian but not right noetherian, and is neither left nor right artinian.

Proof

$R$ is a field and $S$ is noetherian, and $M = R$ is one-dimensional over $R$ , hence a noetherian left $R$ -module; $(1.22)$ gives left noetherian.

For the right-hand failure it suffices, by $(1.22)$ , to show $R$ is not noetherian as a right $S$ -module. If it were, it would in particular be finitely generated over $S$ , so all its elements would share a common denominator $s \in S$ , say $R = s^{- 1} S$ after clearing. But then $1 / s^{2} = s' / s$ for some $s' \in S$ , giving $1 = s s'$ , so $s \in U (S)$ and $R = S$ — contradicting that $S$ is not a field.

For artinian: $S$ is a noetherian domain that is not a field, so it has a nonzero nonunit $s$ , and $(s) ⊋ (s^{2}) ⊋ (s^{3}) ⊋ \dots$ is a strictly descending chain (strict because $S$ is a domain and $s$ is not a unit). So $S$ is not artinian, and $(1.22)$ rules out $A$ being artinian on either side. $□$

Corollary(1.24)Left artinian without right noetherian

Let $S \subseteq R$ be fields with $\dim_{S} R = \infty$ , and set $M = R$ as an $(R, S)$ -bimodule. Then $A = (\begin{matrix} R & R \\ 0 & S \end{matrix})$ is left noetherian and left artinian, but neither right noetherian nor right artinian. As a left module over itself $A$ has a composition series of length $3$ .

Proof

$R$ and $S$ are fields, hence left noetherian and left artinian, and $M = R$ is a one-dimensional left $R$ -vector space, hence a simple and therefore noetherian and artinian left $R$ -module. Apply $(1.22)$ . On the right, $M = R$ is an $S$ -vector space of infinite dimension, so it satisfies neither chain condition; apply $(1.22)$ again.

For the composition series, take

A ⊋ (\begin{matrix} 0 & R \\ 0 & S \end{matrix}) ⊋ (\begin{matrix} 0 & R \\ 0 & 0 \end{matrix}) ⊋ 0 .

(1.24a)

The successive quotients are $R$ , $S$ and $R$ respectively as left $A$ -modules, with $A$ acting through $A \to R$ on the first and third and through $A \to S$ on the second. Each is a one-dimensional vector space over a field, hence a simple $A$ -module, so the chain is a composition series of length $3$ . $□$

Remark—Left Goldie, not right Goldie

In the ring of $(1.24)$ , choose $S$ -subspaces $V_{1}, V_{2}, \dots$ of $R$ whose sum inside $R$ is direct — possible exactly because $\dim_{S} R = \infty$ . Then $(\begin{matrix} 0 & V_{i} \\ 0 & 0 \end{matrix})$ are right ideals of $A$ whose sum is direct, so $A$ contains an infinite direct sum of nonzero right ideals. On the left no such thing exists, since $A$ is left noetherian. $A$ is thus left Goldie and not right Goldie.

Proof Techniques and Method

How these proofs work, and which move to reuse.

Move 1

Split with orthogonal idempotents

Multiplying a one-sided ideal by $e_{11}$ and $e_{22}$ decomposes it. This is the Peirce decomposition in its simplest form and is the only trick needed for $(1.17)$ .

Move 2

Contract the chain to each corner

A chain of left ideals of $A$ produces two chains — one in $S$ , one in the left $R$ -module $R \oplus M$ . Stabilising both stabilises the original, because the decomposition is functorial in the ideal.

Move 3

Choose $M$ to be lopsided

The construction is a counterexample generator. Pick a bimodule finitely generated on one side and not the other, and the theorem $(1.22)$ delivers the asymmetry automatically.

Move 1 generalises: whenever $1 = e_{1} + \dots + e_{n}$ with orthogonal idempotents, $A$ decomposes as $⨁_{i, j} e_{i} A e_{j}$ , and one-sided ideals decompose along the row or column direction. The triangular ring is the case $n = 2$ with $e_{2} A e_{1} = 0$ .

It is worth noticing what the proof of $(1.22)$ does not use: nothing about $M$ beyond its two module structures, and nothing about $R$ and $S$ beyond their chain conditions. The theorem is therefore exactly as strong as the input data allows.

Worked Example

A left zero-divisor that is not a right zero-divisor

Take $R = S = ℤ$ and $M = ℤ / 2 ℤ$ , with both actions the obvious ones. So $A = (\begin{matrix} ℤ & ℤ / 2 ℤ \\ 0 & ℤ \end{matrix})$ , and the product rule $(1.15)$ reads $(x, y, z) (x', y', z') = (x x', x y' + y z', z z')$ with $y$ reduced modulo $2$ . Put

a = (\begin{matrix} 2 & 0 \\ 0 & 1 \end{matrix}), b = (\begin{matrix} 0 & 1 \\ 0 & 0 \end{matrix}) .

(E.1)

Then $a b = (\begin{matrix} 0 & 2 \cdot 1 \\ 0 & 0 \end{matrix}) = 0$ because $2 \equiv 0$ in $ℤ / 2 ℤ$ , so $a$ is a left zero-divisor. But if $c = (\begin{matrix} x & y \\ 0 & z \end{matrix})$ satisfies $c a = 0$ , then $c a = (\begin{matrix} 2 x & y \\ 0 & z \end{matrix}) = 0$ forces $x = z = 0$ in $ℤ$ and $y = 0$ in $ℤ / 2 ℤ$ , so $c = 0$ : $a$ is not a right zero-divisor. Incidentally $b^{2} = 0$ , so $b$ is a zero-divisor on both sides.

Left noetherian, not right noetherian

Instantiate $(1.23)$ with $S = ℤ$ and $R = ℚ$ :

A = (\begin{matrix} ℚ & ℚ \\ 0 & ℤ \end{matrix}) .

(E.2)

Left: $ℚ$ and $ℤ$ are noetherian rings and $M = ℚ$ is a one-dimensional $ℚ$ -vector space, so $A$ is left noetherian. Right: $ℚ$ is not a finitely generated abelian group, so it is not a noetherian right $ℤ$ -module and $A$ is not right noetherian. Artinian: $ℤ$ has the strictly descending chain $2 ℤ ⊋ 4 ℤ ⊋ 8 ℤ ⊋ \dots$ , so $A$ is artinian on neither side.

Its radical is $rad A = (\begin{matrix} 0 & ℚ \\ 0 & 0 \end{matrix})$ , since $rad ℚ = 0$ and $rad ℤ = 0$ . So $(rad A)^{2} = 0$ and $A / rad A ≅ ℚ \times ℤ$ , semiprimitive but not semisimple — as it must be, since $A$ is not artinian.

Left artinian, neither right noetherian nor right artinian

Instantiate $(1.24)$ with $S = ℚ \subseteq R = ℝ$ , so $\dim_{ℚ} ℝ = \infty$ :

A = (\begin{matrix} ℝ & ℝ \\ 0 & ℚ \end{matrix}), A ⊋ (\begin{matrix} 0 & ℝ \\ 0 & ℚ \end{matrix}) ⊋ (\begin{matrix} 0 & ℝ \\ 0 & 0 \end{matrix}) ⊋ 0,

(E.3)

A composition series of length $3$ with factors $ℝ$ , $ℚ$ , $ℝ$ — so $A$ is left artinian and left noetherian.

On the right, $ℝ$ is an infinite-dimensional $ℚ$ -vector space, so it satisfies neither chain condition and $A$ is neither right noetherian nor right artinian. Note the consequence for the Hopkins–Levitzki theorem: left artinian implies left noetherian, and this example shows it says nothing at all about the right side.

Process and Workflow

Name the property you want to breakNoetherian, artinian, Goldie, self-injective — anything whose definition refers to one-sided ideals.

Translate it into a condition on $M$ By

(1.17)

, a left-sided property of

A

becomes a property of

M

as a left

R

-module, and a right-sided one a property as a right

S

-module.

Find a lopsided bimoduleField extensions are the reliable source:

M = R

as an

(R, S)

-bimodule for

S \subseteq R

fields with

\dim_{S} R = \infty

. Fraction fields over a domain are the other.

Read off the conclusionApply

(1.22)

or the relevant analogue; no further computation is needed.

Which chain conditions does $A = (\begin{matrix} R & M \\ 0 & S \end{matrix})$ satisfy?

$R, S$ left noetherian and $_{R} M$ noetherian

A

is left noetherian. Nothing follows about the right side.

$R, S$ left artinian and $_{R} M$ artinian

A

is left artinian, hence also left noetherian by Hopkins–Levitzki. Still nothing follows on the right.

Both sides hold for $M$

A

is two-sided noetherian or artinian. This needs

M

to be finitely generated over both

R

and

S

— for instance

M

finite-dimensional over a common field.

$S$ not artinian

A

is artinian on neither side, whatever

M

and

R

do, since

S

is a quotient of

A

Comparison and Classification

Chain conditions for four instantiations
	left noeth.	right noeth.	left art.	right art.
$(\begin{matrix} k & k \\ 0 & k \end{matrix})$ , $k$ a field	yes	yes	yes	yes
$(\begin{matrix} ℚ & ℚ \\ 0 & ℤ \end{matrix})$	yes	no	no	no
$(\begin{matrix} ℝ & ℝ \\ 0 & ℚ \end{matrix})$	yes	no	yes	no
$(\begin{matrix} ℤ & ℤ / 2 ℤ \\ 0 & ℤ \end{matrix})$	yes	yes	no	no

Chain conditions for four instantiations

What each ingredient controls
Ingredient	Controls on the left	Controls on the right
$R$	left ideals of $A$ inside $R \oplus M$ ; a quotient ring of $A$	right ideals of $A$ inside $R$
$S$	left ideals of $A$ inside $S$	right ideals of $A$ inside $M \oplus S$ ; a quotient ring of $A$
$M$ as a left $R$ -module	everything one-sided on the left	nothing
$M$ as a right $S$ -module	nothing	everything one-sided on the right
$M$ as a bimodule	the two-sided ideals, via $K_{1} M + M K_{2} \subseteq K_{0}$	the same

Relationship Map

The triangular ring sits between the direct product and the full matrix ring.

R \times S

(\begin{matrix} R & M \\ 0 & S \end{matrix})

(\begin{matrix} R & M \\ N & S \end{matrix})

M_{2} (R)

Setting $M = 0$ gives the direct product; filling in the lower-left corner with a bimodule $N$ and a pairing gives a Morita context ring; taking $R = S$ with $M = N = R$ and the obvious pairings gives $M_{2} (R)$ . The triangular case is the one where the ideal theory stays completely transparent.

$A$ the triangular ring

$M \oplus S$ an ideal, with

A / (M \oplus S) ≅ R

$M$ an ideal, square zero, contained in

rad A

$0$ the bottom of the filtration

The two-step filtration $A \supseteq M \supseteq 0$ with $M^{2} = 0$ makes $A$ a nilpotent extension of $R \times S$ . Every structural question about $A$ therefore splits into a semisimple-quotient question, answered by Wedderburn–Artin, and a lifting question across a square-zero ideal.

Applications and Industry Use

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

Internal to algebra

Counterexample engineering

The honest primary use. Left–right asymmetry of noetherian, artinian, Goldie, coherent, self-injective and perfect conditions is nearly always demonstrated with a triangular ring, because $(1.17)$ turns the question into a module-theoretic one.

Representation theory

Quiver algebras and path algebras

The path algebra of the quiver with two vertices and one arrow is exactly $(\begin{matrix} k & k \\ 0 & k \end{matrix})$ . Triangular rings are the smallest nontrivial examples of finite-dimensional algebras of finite representation type, and the base case for tilting theory.

Homological algebra

Rings of finite global dimension

Triangular rings realise prescribed global dimensions cheaply: if $R$ and $S$ are semisimple and $M \neq 0$ , then $A$ is hereditary with global dimension $1$ . This makes them standard test objects for homological conjectures.

Computing

Block-triangular linear algebra

Numerical and symbolic linear algebra routinely exploits block-triangular structure — the same algebra with $R$ , $S$ matrix rings over a field and $M$ a rectangular block. Invertibility of a block-triangular array reduces to invertibility of the diagonal blocks, exactly as in the radical computation above.

The honest summary: this is infrastructure, not an application area. Triangular rings exist so that general claims about rings can be tested against a construction whose ideal theory is fully known in advance.

Design Considerations

Design considerations here means the choices made when modelling a problem with these algebraic structures.

Which corner takes which ring? The upper-left corner $R$ acts on $M$ from the left and so governs the left ideal theory. If the property you care about is a left property, put the well-behaved ring in the upper-left corner and the pathology in $S$ .
**How big should $M$ be?** $M$ finitely generated on both sides gives a two-sided noetherian ring and no asymmetry. Asymmetry costs exactly one infinite-generation condition, which is why $ℝ$ over $ℚ$ appears in so many examples.
Upper or lower triangular? The two conventions give opposite rings, so every theorem transfers with left and right exchanged. Fix one convention per document; mixing them is the most common source of sign-of-side errors.
When to generalise to a Morita context. Filling the lower-left corner buys generality but loses $(1.17)$ : the ideal theory of a Morita context ring depends on the pairings $M \otimes_{S} N \to R$ and $N \otimes_{R} M \to S$ , and is no longer a simple direct sum decomposition.
Larger triangular arrays. The $n \times n$ version with rings $R_{1}, \dots, R_{n}$ on the diagonal and bimodules above it behaves identically, and is the right model for the path algebra of a linear quiver.

Failure Modes and Common Mistakes

$A$ is a domain only in degenerate cases: $M \neq 0$ always yields $M^{2} = 0$ , so $A$ has nonzero nilpotents whenever $M \neq 0$ .
$rad A$ contains all of $M$ regardless of $R$ and $S$ , so $A$ is never semiprimitive when $M \neq 0$ — and never semisimple.
The chain conditions in $(1.22)$ see $M$ only as a one-sided module — but the lattice of left ideals still feels the right $S$ -action, through the coupling condition $M I_{2} \subseteq I_{1}$ in $(1.17) (1)$ . Do not conflate the two statements.
The idempotents $e_{11}, e_{22}$ are not central. $A$ does not decompose as a direct product unless $M = 0$ .

Quick Reference

Construction

A = (\begin{matrix} R & M \\ 0 & S \end{matrix})

M

(R, S)

-bimodule

Additive form

A = R \oplus M \oplus S

Left ideals

I_{1} \oplus I_{2}

I_{2} ⊴_{ℓ} S

I_{1} \subseteq R \oplus M

a left

R

-submodule,

M I_{2} \subseteq I_{1}

Right ideals

J_{1} \oplus J_{2}

J_{1}

a right ideal of

R

J_{2} \subseteq M \oplus S

a right

S

-submodule,

J_{1} M \subseteq J_{2}

Ideals

K_{1} \oplus K_{0} \oplus K_{2}

with

K_{1} M + M K_{2} \subseteq K_{0}

Radical

rad A = (\begin{matrix} rad R & M \\ 0 & rad S \end{matrix})

Left chain conditions

R

S

left noetherian (artinian) and

_{R} M

noetherian (artinian)

Right chain conditions

R

S

right noetherian (artinian) and

M_{S}

noetherian (artinian)

Standard instantiations
$R$	$M$	$S$	What it demonstrates
$k$	$k$	$k$	Upper triangular $2 \times 2$ matrices; $rad$ is square-zero
$ℤ$	$ℤ / 2 ℤ$	$ℤ$	Left zero-divisor that is not a right zero-divisor
$ℚ$	$ℚ$	$ℤ$	Left noetherian, not right noetherian, not artinian
$ℝ$	$ℝ$	$ℚ$	Left artinian, neither right noetherian nor right artinian; left Goldie not right Goldie

Frequently Asked Questions

Why is the lower-left corner zero?

Because that is what makes the ideal theory transparent. Filling it with a second bimodule $N$ requires pairings $M \otimes_{S} N \to R$ and $N \otimes_{R} M \to S$ satisfying associativity constraints — this is a Morita context ring — and then one-sided ideals no longer split as a direct sum along the diagonal idempotents. The triangular case is the degenerate one in which both pairings are zero.

Is a triangular ring ever semisimple?

Only when $M = 0$ , in which case it is just $R \times S$ . For $M \neq 0$ the copy of $M$ is a nonzero square-zero ideal, hence lies in the radical, so $rad A \neq 0$ and $A$ cannot be semisimple. This is why triangular rings are the standard first examples of non-semisimple artinian rings.

How do I see the simple modules?

Since $M \subseteq rad A$ , every simple left $A$ -module is a simple module over $A / rad A ≅ (R / rad R) \times (S / rad S)$ . So the simple left $A$ -modules are exactly the simple left $R$ -modules and the simple left $S$ -modules, pulled back along the two quotient maps. For $R$ and $S$ fields there are precisely two, both one-dimensional.

Does

(1.22)

have an analogue for other finiteness conditions?

Yes, and the pattern is the same for any condition defined by one-sided ideals or by finitely generated one-sided modules: $A$ has the property on the left iff $R$ and $S$ do and $M$ has the corresponding property as a left $R$ -module. Global dimension is the notable exception — it satisfies inequalities rather than an equality, since $M$ contributes homologically.

Why does the fraction field example in

(1.23)

need

S

to be a domain?

Two reasons. The field of fractions requires it, and the strictly descending chain $(s) ⊋ (s^{2}) ⊋ \dots$ used to show $S$ is not artinian requires cancellation. Without the domain hypothesis a noetherian ring can certainly be artinian — any finite ring is — and the conclusion would fail.

Can a triangular ring be right noetherian but not left noetherian?

Yes, by symmetry: build the same examples with the roles of the two sides exchanged, or equivalently pass to the opposite ring. The opposite of $(\begin{matrix} R & M \\ 0 & S \end{matrix})$ is the triangular ring built from $S^{op}$ , $M$ and $R^{op}$ , and taking opposites converts every left statement into the corresponding right statement.

References

T. Y. Lam, A First Course in Noncommutative Rings, Graduate Texts in Mathematics 131, Springer-Verlag, 1991, §1, results (1.14), (1.17) and (1.22)–(1.24) (pp. 17–22).
T. Y. Lam, Lectures on Modules and Rings, Graduate Texts in Mathematics 189, Springer-Verlag, 1999, §1 and §2.
F. W. Anderson and K. R. Fuller, Rings and Categories of Modules, 2nd edition, Graduate Texts in Mathematics 13, Springer-Verlag, 1992, §1 and §11.
L. H. Rowen, Ring Theory, Volume I, Academic Press, 1988, Chapter 1.
M. Auslander, I. Reiten and S. O. Smalo, Representation Theory of Artin Algebras, Cambridge Studies in Advanced Mathematics 36, Cambridge University Press, 1995 (triangular matrix algebras and quivers).

AI Suggested Questions

Compute the global dimension of the triangular ring in terms of the global dimensions of $R$ , $S$ and the projective dimension of $M$ .
Describe the finitely generated indecomposable modules over the upper triangular $2 \times 2$ matrix ring over a field.
Generalise $(1.17)$ to an $n \times n$ triangular ring with diagonal rings $R_{1}, \dots, R_{n}$ .
When is a triangular ring left self-injective, and why does that condition fail so often?
Give a triangular ring that is left coherent but not right coherent.
How does the Morita context ring differ from the triangular ring in its one-sided ideal theory?
Show that the path algebra of a finite acyclic quiver is isomorphic to a generalised triangular ring.

01Executive Summary

02Overview

03Learning Objectives

04Definitions

05Core Concepts

The multiplication chart

Why the two sides separate

The radical and the simple modules

06Key Results

07Proof Techniques and Method

Split with orthogonal idempotents

Contract the chain to each corner

Choose M to be lopsided

08Worked Example

A left zero-divisor that is not a right zero-divisor

Left noetherian, not right noetherian

Left artinian, neither right noetherian nor right artinian

09Process and Workflow

10Comparison and Classification

11Relationship Map

12Applications and Industry Use

Counterexample engineering

Quiver algebras and path algebras

Rings of finite global dimension

Block-triangular linear algebra

13Design Considerations

14Failure Modes and Common Mistakes

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

Choose $M$ to be lopsided

Worked Example

Process and Workflow

Comparison and Classification

Relationship Map

Applications and Industry Use

Design Considerations

Failure Modes and Common Mistakes

Quick Reference

Frequently Asked Questions

Related KEVOS Topics

References

AI Suggested Questions