Rings of Stable Range One

Q: Can cancellation hold without stable range one?

Yes. (20.11) is sufficient, not necessary. For example over Z, which has stable range 2, finitely generated modules cancel by the structure theorem. The theorem is valuable because it needs no hypothesis at all on B and C.

Q: How does (20.12) differ from (20.11)?

(20.11) imposes a hypothesis on End(A_R), which is often hard to check. (20.12) replaces it with checkable hypotheses on the ground ring and on A: the proof simply verifies that End(A_R) is module-finite over k and therefore semilocal by (20.6).

Q: What is the connection with stably free modules?

A stably free module satisfies P ⊕ R^m ≅ R^n. Cancelling the free summand — which stable range one permits — gives P ≅ R^n-m, so every stably free module is free. This is (20.13)(3) and it is the cleanest visible consequence of the hypothesis.

Executive Summary

Stable range is Bass' measure of how far a ring is from behaving like a field for elementary-matrix purposes. The extreme case, left stable range one, says that any relation $E a + E b = E$ can be collapsed to a single unit: some correction $a + e b$ is invertible.

Two facts make the condition central. First, Bass' Theorem $(20.9)$ says every semilocal ring has left stable range one — the class is therefore very large. Second, the condition is exactly what is needed to cancel a module from a direct sum: if $End (A_{R})$ has left stable range one then $A \oplus B ≅ A \oplus C$ implies $B ≅ C$ .

$a + e b \in U (E)$ The conclusion

$(20.10)$ Definition

$(20.11)$ Cancellation

$b = 0$ Recovers Dedekind finiteness

Overview

A pair $(a, b)$ with $E a + E b = E$ is a unimodular row of length two. Elementary row operations let us replace $a$ by $a + e b$ without changing the ideal generated. Stable range one asks that some such operation lands on a unit — that unimodular rows of length two can be shortened to length one.

E a + E b = E ⟹ \exists e \in E : a + e b \in U (E)

(20.10)

Left stable range one. Setting $b = 0$ gives: $E a = E$ implies $a \in U (E)$ , i.e. Dedekind finiteness.

Bass introduced the general condition — stable range at most $n$ — to obtain stability theorems in algebraic K-theory: once the stable range is bounded, ${GL}_{n}$ and $K_{1}$ stop changing as $n$ grows. Stable range one is the strongest such bound and gives the cleanest consequences: $K_{1} (E) = U (E)^{ab}$ , cancellation of modules, and invariant basis number.

The condition is strictly weaker than semilocality. A commutative von Neumann regular ring such as $\prod_{i = 1}^{\infty} 𝔽_{2}$ has stable range one but is neither artinian nor semilocal; Lam records this asymmetry without constructing an example.

Learning Objectives

State $(20.10)$ and verify that $b = 0$ recovers Dedekind finiteness.
Restate Bass' Theorem $(20.9)$ as "semilocal implies left stable range one".
Prove the Cancellation Theorem $(20.11)$ in full.
Prove Corollary $(20.12)$ and locate where the noetherian hypothesis is used.
Exhibit a ring of stable range one that is not semilocal.
Exhibit a ring whose stable range exceeds one and see cancellation fail.

Definitions

Definition(20.10)Left stable range one

A ring $E$ has left stable range one if for all $a, b \in E$ with $E a + E b = E$ there exists $e \in E$ such that $a + e b \in U (E)$ .

The special case $b = 0$ says: if $E a = E$ , i.e. $a$ has a left inverse, then $a$ is a unit. So left stable range one implies Dedekind finiteness.

DefinitionStable range at most

n

(Bass)

$E$ has left stable range at most $n$ if whenever $a_{1}, \dots, a_{n + 1} \in E$ satisfy $\sum_{i = 1}^{n + 1} E a_{i} = E$ , there exist $b_{1}, \dots, b_{n} \in E$ with

\sum_{i = 1}^{n} E (a_{i} + b_{i} a_{n + 1}) = E .

The stable range $sr (E)$ is the least such $n$ . The case $n = 1$ is $(20.10)$ , since $E (a_{1} + b_{1} a_{2}) = E$ together with Dedekind finiteness makes $a_{1} + b_{1} a_{2}$ a unit.

$End (A_{R})$: The ring of R-linear endomorphisms of the right R-module A, with multiplication given by composition.
Split epimorphism: A surjection h with h s = identity for some s; its kernel is then a direct summand.
Stably free module: A module P with P direct sum R^m isomorphic to R^n for some m and n.
Unimodular pair: A pair (a,b) generating E as a left ideal.

Vaserstein proved that the left and right stable ranges of a ring coincide, so the adjective left is dispensable; it is kept here to match the phrasing of $(20.10)$ .

Core Concepts

Bass' Theorem restated

In $(20.9)$ take $𝔅 = R b$ , a principal left ideal. The hypothesis becomes $R a + R b = R$ and the conclusion produces a unit in $a + R b$ , that is, an element $a + e b \in U (R)$ . Conversely stable range one for principal $𝔅$ implies the general statement, because $R a + 𝔅 = R$ already forces $R a + R b = R$ for some single $b \in 𝔅$ : write $1 = r a + b$ .

R

semilocalBass

(20.9)

left stable range

1

cancellation

(20.11)

Why endomorphism rings are the right place to put the hypothesis

Cancelling $A$ from $A \oplus B ≅ A \oplus C$ is a question about maps into and out of $A$ , and all of those are recorded in $E = End (A_{R})$ . The isomorphism supplies a split epimorphism $A \oplus B \to A$ ; its components give two elements of $E$ that generate $E$ ; stable range one corrects them to a single unit; and the unit reorganises the direct sum decomposition so that $B$ and $C$ appear as complements of the same submodule.

Sources of stable range one besides semilocality

Class 1

Semilocal rings

By Bass' Theorem $(20.9)$ . Includes local rings, artinian rings, finite rings, finite-dimensional algebras and module-finite algebras over commutative semilocal rings.

Class 2

Unit-regular rings

Every von Neumann regular ring in which each element satisfies $a = a u a$ with $u$ a unit has stable range one. This includes all commutative von Neumann regular rings and all finite von Neumann algebras' underlying regular rings.

Class 3

Rings of continuous functions and $π$ -regular rings

Strongly $π$ -regular rings have stable range one. These classes overlap with, but are not contained in, the semilocal ones.

Key Results

Theorem(20.9)Bass' Theorem, stable range form

Every semilocal ring has left stable range one. Explicitly: if $R$ is semilocal and $R a + R b = R$ , then $a + e b \in U (R)$ for some $e \in R$ .

The converse fails: there are rings of left stable range one that are not semilocal.

Theorem(20.11)Cancellation Theorem

Let $R$ be a ring and let $A, B, C$ be right $R$ -modules. Suppose the ring $E = End (A_{R})$ has left stable range one — for instance, suppose $E$ is semilocal. Then

A \oplus B ≅ A \oplus C ⟹ B ≅ C .

No finiteness hypothesis is imposed on $B$ or $C$ .

Proof

Setting up. Compose an isomorphism $A \oplus B \to A \oplus C$ with the projection onto $A$ . This gives a split epimorphism $h_{0} = (f, g) : A \oplus B \to A$ , where $f \in End (A_{R})$ and $g \in Hom (B, A)$ , whose kernel is carried isomorphically onto $C$ . Let $s = (\binom{f'}{g'}) : A \to A \oplus B$ be a splitting, so $f' \in E$ , $g' \in Hom (A, B)$ and

(f, g) \circ s = f f' + g g' = 1_{A} .

Applying stable range one. Both $f'$ and $g g'$ lie in $E$ — note $g g'$ is the composite $A \to B \to A$ , an endomorphism of $A$ , even though $g$ and $g'$ individually are not. The displayed identity shows $1_{A} \in E f' + E (g g')$ , hence $E f' + E (g g') = E$ . By hypothesis there is $e \in E$ with

u : = f' + e (g g') \in U (E) .

A second split epimorphism. Define $h = (1_{A}, e g) : A \oplus B \to A$ , $(x, y) \mapsto x + e (g (y))$ . Then $h \circ s = f' + e g g' = u$ is an automorphism of $A$ , so $h$ is a split epimorphism with splitting $s u^{- 1}$ , and

A \oplus B = im (s u^{- 1}) \oplus \ker h = im (s) \oplus \ker h,

the last equality because $u^{- 1}$ is an automorphism of $A$ and hence does not change the image. From the original splitting we also have $A \oplus B = im (s) \oplus \ker h_{0}$ .

Comparing complements. Two complements of the same submodule are isomorphic to the same quotient, so

\ker h ≅ (A \oplus B) / im (s) ≅ \ker h_{0} ≅ C .

**Identifying $\ker h$ .** By definition $\ker h = {(x, y) \in A \oplus B : x = - e (g (y))}$ , and $y \mapsto (- e (g (y)), y)$ is an isomorphism $B \to \ker h$ with inverse the projection to the second coordinate. Hence $B ≅ \ker h ≅ C$ . $□$

Corollary(20.12)Cancellation over module-finite algebras

Let $k$ be a commutative noetherian semilocal ring and let $R$ be a $k$ -algebra which is **finitely generated as a $k$ -module**. Let $A$ be a finitely generated right $R$ -module and let $B, C$ be arbitrary right $R$ -modules. Then $A \oplus B ≅ A \oplus C$ implies $B ≅ C$ .

Proof

It suffices to prove that $E = End (A_{R})$ is semilocal, for then $E$ has left stable range one by $(20.9)$ and $(20.11)$ applies.

Since $A$ is finitely generated over $R$ and $R$ is finitely generated over $k$ , the module $A$ is finitely generated over $k$ ; say $A$ is a quotient of $k^{m}$ . Restricting along the surjection $k^{m} ↠ A$ embeds ${End}_{k} (A)$ into ${Hom}_{k} (k^{m}, A) ≅ A^{m}$ , a finitely generated $k$ -module. As $k$ is noetherian, submodules of finitely generated modules are finitely generated, so ${End}_{k} (A)$ is a finitely generated $k$ -module.

Now $E = End (A_{R})$ is a $k$ -submodule of ${End}_{k} (A)$ , hence also finitely generated over $k$ by noetherianity. Thus $E$ is a $k$ -algebra which is module-finite over the commutative semilocal ring $k$ , and $(20.6)$ makes $E$ semilocal. $□$

CorollaryModules of finite length cancel

Let $R$ be any ring and let $A$ be a right $R$ -module possessing a composition series. Then $A \oplus B ≅ A \oplus C$ implies $B ≅ C$ for arbitrary right $R$ -modules $B, C$ .

Sketch. By Fitting's Lemma a module of finite length decomposes into finitely many indecomposables each with local endomorphism ring, and the endomorphism ring of such a finite direct sum is semiperfect, hence semilocal. Apply $(20.11)$ .

RemarkStable range one is strictly weaker than semilocal

$E = \prod_{i = 1}^{\infty} 𝔽_{2}$ is commutative von Neumann regular, hence unit-regular, hence of stable range one. But $rad E = 0$ and $E$ is not artinian, so $E$ is not semilocal. The implication in $(20.9)$ therefore does not reverse.

Proof Techniques and Method

How this proof works, and which move to reuse.

1. Turn the isomorphism into a split epimorphism

A \oplus B ≅ A \oplus C

plus projection gives

h_{0} : A \oplus B \to A

split, with

\ker h_{0} ≅ C

. All information about

C

is now stored as a kernel.

2. Extract an equation in $End (A)$ The splitting yields

f f' + g g' = 1_{A}

, so the pair

(f', g g')

is unimodular in

E

. This is the only place the module data is converted into ring data.

3. Correct to a unitStable range one replaces the pair by a single unit

u = f' + e (g g')

. The correction

e

is what makes a second, simpler epimorphism

h = (1, e g)

split.

4. Compare complements

h

and

h_{0}

split along the same image

im (s)

, so their kernels are isomorphic. One kernel is visibly

B

, the other is

C

Worked Example

Checking $(20.10)$ by hand in $ℤ / 12$

$ℤ / 12$ is semilocal with $U = {1, 5, 7, 11}$ , so $(20.9)$ guarantees stable range one. Verifying instances shows what the correction $e$ does.

Unimodular pairs in $ℤ / 12$ and their corrections
$(a, b)$	$(a) + (b)$	Choice of $e$	$a + e b$	Unit?
$(2, 3)$	$(1)$	$e = 1$	$5$	yes
$(3, 8)$	$(1)$	$e = 1$	$11$	yes
$(4, 3)$	$(1)$	$e = 1$	$7$	yes
$(6, 4)$	$(2) \neq (1)$	—	—	hypothesis fails

A ring of stable range greater than one

$ℤ$ does not have stable range one. Take $a = 3$ , $b = 5$ : then $3 ℤ + 5 ℤ = ℤ$ , but $3 + 5 e \equiv 3 (\mod 5)$ for every $e$ , while the units $\pm 1$ are $\equiv 1, 4 (\mod 5)$ . So no correction works. By Bass' stable range theorem a commutative noetherian ring of Krull dimension $d$ has stable range at most $d + 1$ , giving $sr (ℤ) = 2$ .

Cancellation failing when stable range is larger

Let $S = ℝ [x, y, z] / (x^{2} + y^{2} + z^{2} - 1)$ , the coordinate ring of the real $2$ -sphere, and let $P$ be the module of algebraic tangent vector fields, i.e. the kernel of $S^{3} \to S$ , $(u, v, w) \mapsto x u + y v + z w$ . Then $P \oplus S ≅ S^{3} ≅ S^{2} \oplus S$ , yet $P not ≅ S^{2}$ : a free module of rank $2$ would supply two everywhere-independent tangent fields on $S^{2}$ , contradicting the hairy ball theorem. So

P \oplus S ≅ S^{2} \oplus S, P not ≅ S^{2} .

(20.12a)

Cancellation of the single free summand $S$ fails. Here $End (S_{S}) ≅ S$ has stable range greater than $1$ ; $S$ is noetherian of Krull dimension $2$ but not semilocal.

A positive instance of $(20.11)$

Let $R = ℤ / 4$ and $A = ℤ / 4 \oplus ℤ / 2$ as a right $R$ -module. $A$ is finite, so $E = End (A_{R})$ is a finite ring, hence artinian, hence semilocal. By $(20.11)$ , $A$ cancels from direct sums of arbitrary $ℤ / 4$ -modules — including infinitely generated ones, where no counting argument is available.

Process and Workflow

You want to cancel $A$ from $A \oplus B ≅ A \oplus C$ . What do you need?

$End (A_{R})$ is semilocalApply

(20.9)

then

(20.11)

. Sufficient conditions:

A

has finite length;

A

is finitely generated over a module-finite algebra over a commutative noetherian semilocal ring

(20.12)

$End (A_{R})$ has stable range one but is not semilocal

(20.11)

still applies verbatim — the theorem is stated for the stable range hypothesis, not for semilocality.

$A$ is finitely generated projective and $R$ has stable range oneUse

(20.13) (1)

: embed

A

in a free module and cancel free summands one at a time.

None of theseCancellation may genuinely fail — see the tangent module of the real

2

-sphere. Look instead for a Krull–Schmidt theorem, which requires indecomposables with local endomorphism rings.

Comparison and Classification

Stable range of standard rings
Ring	Stable range	Comment
Any semilocal ring	$1$	Bass' Theorem $(20.9)$ .
Local ring, finite ring, finite-dimensional algebra	$1$	Special cases of semilocal.
Unit-regular ring	$1$	Includes commutative von Neumann regular rings.
$\prod_{i = 1}^{\infty} 𝔽_{2}$	$1$	Stable range one without being semilocal.
$ℤ$	$2$	Krull dimension $1$ ; the pair $(3, 5)$ obstructs stable range one.
$k [x_{1}, \dots, x_{d}]$ , $k$ a field	$\leq d + 1$	Bass' stable range theorem for noetherian rings of Krull dimension $d$ .
${End}_{k} (V)$ , $\dim_{k} V$ infinite	not $1$	Not even Dedekind-finite.

The pattern is that stable range one behaves like a low-dimensionality condition. Bass' stable range theorem bounds $sr (R)$ by $d + 1$ for a commutative noetherian ring of Krull dimension $d$ , so dimension zero already forces stable range one; conversely a positive-dimensional ring such as $ℤ$ can fail it, and the failure is detected by an explicit unimodular pair.

Relationship Map

Left stable range one

(20.10)

Semilocalby

(20.9)

Left artinianchain condition

Semisimple

rad R = 0

and artinian

Localdivision ring modulo the radical

Unit-regularvon Neumann regular with

a = a u a

u

a unit

The two inner families overlap only in small cases: a ring that is both semilocal and von Neumann regular is semisimple. Everything inside the outer band cancels modules and has invariant basis number.

Applications and Industry Use

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

Algebraic K-theory. Stable range bounds are the hypotheses of Bass' stability theorems: for $n > sr (R)$ the maps ${GL}_{n} (R) \to {GL}_{n + 1} (R)$ induce isomorphisms on $K_{1}$ . Stable range one collapses the whole tower to $U (R)$ .
Serre's problem and vector bundles. Over a ring of stable range one every stably free module is free, so no exotic projective modules exist. Over rings of larger stable range they do, and the tangent bundle of $S^{2}$ is the standard witness.
Operator algebras. Rings of stable range one appear as the algebraic invariant behind cancellation of projections in C*-algebras; the property is a key input in the Elliott classification programme.
Direct-sum decomposition theory. Facchini's theory of modules with semilocal endomorphism rings uses stable range one to control when Krull–Schmidt-type uniqueness holds and how badly it can fail.
Symbolic computation. Algorithms that complete a unimodular row to an invertible matrix terminate in one step over a ring of stable range one; over $ℤ [x_{1}, \dots, x_{d}]$ they require the full Quillen–Suslin machinery.

Failure Modes and Common Mistakes

Quick Reference

$(20.10)$

E a + E b = E \Rightarrow a + e b \in U (E)

for some

e \in E

$(20.9)$ Semilocal

\Rightarrow

left stable range one. Not reversible.

$(20.11)$

End (A_{R})

of stable range one

\Rightarrow

A

cancels from direct sums, with

B, C

arbitrary.

$(20.12)$

k

commutative noetherian semilocal,

R

module-finite over

k

A

f.g.

\Rightarrow

A

cancels.

Degenerate case

b = 0

gives Dedekind finiteness.

SymmetryLeft and right stable range agree (Vaserstein).

What each hypothesis buys
Hypothesis	Conclusion available
$R$ semilocal	Stable range one for $R$ ; cancellation of $R$ itself and of f.g. projectives.
$End (A_{R})$ semilocal	$A$ cancels from arbitrary direct sums.
$A$ of finite length	$End (A_{R})$ semiperfect; $A$ cancels.
$A$ f.g. over a module-finite algebra over commutative noetherian semilocal $k$	$(20.12)$ : $A$ cancels.
$R$ of stable range one, $A$ f.g. projective	$(20.13) (1)$ : $A$ cancels; stably free modules are free.

Frequently Asked Questions

Why is the condition called "stable range"?

Because it measures the point at which unimodular rows stabilise: for rows longer than the stable range, one can always shorten by elementary operations. Bass introduced the numerical invariant so that ${GL}_{n}$ and $K_{1}$ would stop changing once $n$ exceeds it.

Does stable range one have a left and a right version?

The definition is stated on one side, but Vaserstein proved the two agree for every ring, so $sr (R) = sr (R^{op})$ . Lam states the left version to match the left ideal $R a$ appearing in Bass' Theorem.

Can cancellation hold without stable range one?

Yes. $(20.11)$ is sufficient, not necessary. For example over $ℤ$ , which has stable range $2$ , finitely generated modules cancel by the structure theorem. The theorem is valuable because it needs no hypothesis at all on $B$ and $C$ .

Why must

B

and

C

be allowed to be arbitrary?

Because that is exactly where naive arguments break. Counting invariants — length, rank, dimension — cancel finitely generated modules easily. The content of $(20.11)$ is that infinitely generated $B$ and $C$ are also cancelled, with no finiteness assumption.

How does

(20.12)

differ from

(20.11)

$(20.11)$ imposes a hypothesis on $End (A_{R})$ , which is often hard to check. $(20.12)$ replaces it with checkable hypotheses on the ground ring and on $A$ : the proof simply verifies that $End (A_{R})$ is module-finite over $k$ and therefore semilocal by $(20.6)$ .

What is the connection with stably free modules?

A stably free module satisfies $P \oplus R^{m} ≅ R^{n}$ . Cancelling the free summand — which stable range one permits — gives $P ≅ R^{n - m}$ , so every stably free module is free. This is $(20.13) (3)$ and it is the cleanest visible consequence of the hypothesis.

References

T. Y. Lam, A First Course in Noncommutative Rings, Graduate Texts in Mathematics 131, Springer-Verlag, 1991, §20, (20.10)–(20.12), pp. 314–316.
H. Bass, “K-theory and stable algebra”, Publications Mathématiques de l'IHÉS 22 (1964), 5–60.
L. N. Vaserstein, “Stable rank of rings and dimensionality of topological spaces”, Functional Analysis and its Applications 5 (1971), 102–110.
E. G. Evans, Jr., “Krull–Schmidt and cancellation over local rings”, Pacific Journal of Mathematics 46 (1973), 115–121.
K. R. Goodearl, Von Neumann Regular Rings, 2nd edition, Krieger, 1991, Chapter 4, for unit-regularity and stable range one.
A. Facchini, Module Theory: Endomorphism Rings and Direct Sum Decompositions in Some Classes of Modules, Progress in Mathematics 167, Birkhäuser, 1998.

AI Suggested Questions

Prove Vaserstein's theorem that the left and right stable ranges of a ring coincide.
Show that a unit-regular ring has stable range one.
Compute the stable range of $ℤ [x]$ and of $ℝ [x, y, z] / (x^{2} + y^{2} + z^{2} - 1)$ .
Give an example of a module $A$ whose endomorphism ring is semilocal but not semiperfect.
State Bass' stability theorem for $K_{1}$ precisely and identify where stable range one is used.
Explain why a ring that is both semilocal and von Neumann regular must be semisimple.
Construct a ring of stable range one that is not semilocal and not von Neumann regular.

01Executive Summary

02Overview

03Learning Objectives

04Definitions

05Core Concepts

Bass' Theorem restated

Why endomorphism rings are the right place to put the hypothesis

Sources of stable range one besides semilocality

Semilocal rings

Unit-regular rings

Rings of continuous functions and π-regular rings

06Key Results

07Proof Techniques and Method

08Worked Example

Checking (20.10) by hand in ℤ/12

A ring of stable range greater than one

Cancellation failing when stable range is larger

A positive instance of (20.11)

09Process and Workflow

10Comparison and Classification

11Relationship Map

12Applications and Industry Use

13Failure Modes and Common Mistakes

14Quick Reference

15Frequently Asked Questions

16Related KEVOS Topics

17References

18AI Suggested Questions

Continue learning

Executive Summary

Overview

Learning Objectives

Definitions

Core Concepts

Rings of continuous functions and $π$ -regular rings

Key Results

Proof Techniques and Method

Worked Example

Checking $(20.10)$ by hand in $ℤ / 12$

A positive instance of $(20.11)$

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