Bass’s Theorem: Flat Modules Are Projective over Perfect Rings

Q: Why does the proof tensor with R/J rather than an arbitrary module?

Because N ⊗_R R/J ≅ N/NJ converts the flatness statement into a statement about radicals, which is where the smallness of the kernel can be used. Any other test module would leave the two hypotheses — flatness and smallness — with no common language.

Q: Is there a flat module that is projective over one ring and not over another?

Yes, in the natural sense: Q is flat and not projective over Z and over Z_(p), but it is projective — indeed free of rank one — over itself. Projectivity depends on the base ring, and the theorem says the base ring is exactly what decides whether flatness suffices.

Q: Why is Lam's third condition numbered (5)?

To align with (23.20), whose four conditions are numbered (1)–(4). Adding the flat criterion as a fifth condition emphasises that (24.25) extends the earlier theorem rather than replacing it, and that the present proof also fills the implication left open there.

Executive Summary

Projective modules are always flat; the converse fails over most rings. Bass identified exactly when it holds: **every flat right $R$ -module is projective if and only if $R$ is right perfect**, equivalently if and only if $R$ satisfies the descending chain condition on principal left ideals.

The cycle of implications is short. Right perfectness gives every module a projective cover; tensoring the cover's kernel sequence with $R / rad R$ and using flatness forces the kernel to vanish. Conversely, the flat module attached to a sequence $a_{1}, a_{2}, \dots$ is projective only when the chain $R a_{1} \supseteq R a_{2} a_{1} \supseteq \dots$ stabilises, and that chain condition returns right perfectness.

flat $\Rightarrow$ projCharacterising property

DCC, leftChain condition, opposite side

(24.25)Lam's numbering

1960Bass

Overview

Write $J = rad R$ . Two prior results converge here: the construction of projective covers over right perfect rings, and the flat module $F / K$ built from an arbitrary sequence of ring elements. The first drives the implication *perfect $\Rightarrow$ flat is projective*; the second drives its converse.

R right perfect ⟺ DCC on principal left ideals ⟺ every flat M_{R} is projective

(24.25)

Compare $(23.20)$ ; the proof given here is independent of it and supplies the implication left open there.

The side switch is the memorable feature. Perfectness is a right-handed condition — right T-nilpotence of $J$ , projective covers for right modules — yet the chain condition it is equivalent to concerns left ideals. The reason is visible in the construction: the relations $e_{i} = e_{i + 1} a_{i + 1}$ in a right module accumulate coefficients on the left.

A useful by-product: since every direct limit of projective modules is flat, over a right perfect ring direct limits of projectives are again projective — a closure property that fails badly over $ℤ$ , where $ℚ$ is a direct limit of copies of $ℤ$ .

Learning Objectives

State $(24.25)$ and lay out the implication cycle used to prove it.
Run the argument $K / K J ↪ P / P J$ and see why both injectivity and vanishing are available.
Show that any descending chain of principal left ideals has the form $R a_{1} \supseteq R a_{2} a_{1} \supseteq \dots$ .
Apply $(24.24)$ to convert flat implies projective into the chain condition.
Prove directly that DCC on principal left ideals makes $rad R$ right T-nilpotent.
Identify the flat non-projective module over $ℤ_{(p)}$ produced by the construction.

Definitions

Right perfect: $R / J$ semisimple and $J$ right T-nilpotent: every sequence $a_{1}, a_{2}, \dots \in J$ has $a_{n} \dots a_{1} = 0$ for some $n$ .
DCC on principal left ideals: Every chain $R b_{1} \supseteq R b_{2} \supseteq \dots$ stabilises. Equivalently, $R R$ has DCC on cyclic submodules.
$N \otimes_{R} R / J$: Naturally isomorphic to $N / N J$ for a right module $N$ ; this identification is what turns the flatness hypothesis into a statement about radicals.
Direct limit: A colimit over a directed system. Direct limits of flat modules are flat; direct limits of projectives are flat but generally not projective.
Semiprimary: $R / J$ semisimple with $J$ nilpotent. Every semiprimary ring is right and left perfect; Bass's perfect rings are the homological generalisation.

Modules tested for flatness and projectivity are right modules; the chain condition concerns left ideals. Keeping the two sides straight is the single most error-prone aspect of this theorem.

Core Concepts

Why a projective cover converts flatness into projectivity

Let $M$ be flat and let $θ : P \to M$ be a projective cover with kernel $K$ . Two independent facts collide. Flatness of $M$ , via the one-presentation criterion $(24.23)$ applied to $ε : 0 \to K \to P \to M \to 0$ with $P$ flat, says $K \otimes_{R} M' \to P \otimes_{R} M'$ is injective for every left module $M'$ . Smallness of $K$ says $K \subseteq rad P = P J$ .

K / K J ↪ P / P J is injective, and is the zero map since K \subseteq P J .

(24.25a)

Take $M' = R / J$ and use $N \otimes_{R} R / J ≅ N / N J$ .

An injective zero map has zero source, so $K = K J$ ; right T-nilpotence then forces $K = 0$ and $θ$ is an isomorphism.

Why the chain condition appears on the left

Any descending chain of principal left ideals can be normalised. If $R b_{i + 1} \subseteq R b_{i}$ then $b_{i + 1} = a_{i + 1} b_{i}$ for some $a_{i + 1} \in R$ , so after renaming the chain reads

R b_{1} \supseteq R b_{2} \supseteq \dots becomes R a_{1} \supseteq R a_{2} a_{1} \supseteq R a_{3} a_{2} a_{1} \supseteq \dots

(24.25b)

This is exactly the data feeding the construction $(24.24)$ : a sequence $a_{1}, a_{2}, \dots$ of ring elements. The construction produces a right module whose projectivity is equivalent to stationarity of a chain of left ideals, and that asymmetry is inherited by the theorem.

Independence from Theorem P

The cycle $(1) \Rightarrow (5) \Rightarrow (2) \Rightarrow (1)$ uses only projective covers, the flat construction, and the unit property of the radical. It never invokes $(23.20)$ , so it independently establishes the implication *right perfect $\Rightarrow$ DCC on principal left ideals*, which was left unproved there.

Key Results

Theorem(24.25)Bass: flat implies projective

For any ring $R$ with identity the following are equivalent:

$R$ is right perfect;
$R$ satisfies the descending chain condition on principal left ideals;
every flat right $R$ -module is projective.

The numbering follows Lam, who labels the third condition (5) to align with the four conditions of $(23.20)$ .

Proof

We prove $(1) \Rightarrow (3) \Rightarrow (2) \Rightarrow (1)$ .

** $(1) \Rightarrow (3)$ .** Let $M$ be a flat right $R$ -module. Since $R$ is right perfect, $(24.12)$ provides a projective cover $θ : P \to M$ ; write $ε : 0 \to K \to P \to M \to 0$ with $K = \ker θ \subseteq_{s} P$ . As $P$ is projective it is flat, so $(24.23)$ applies: because $M$ is flat, $ε \otimes_{R} M'$ is exact for every left $R$ -module $M'$ .

Take $M' = R / J$ with $J = rad R$ . Using $N \otimes_{R} R / J ≅ N / N J$ , exactness says the induced map $K / K J \to P / P J$ is injective. On the other hand $K \subseteq_{s} P$ gives $K \subseteq rad P$ by $(24.4) (1)$ , and $rad P = P J$ by $(24.7)$ ; hence the image of $K$ in $P / P J$ is zero and the map $K / K J \to P / P J$ is the zero map.

A map that is both injective and zero has zero domain, so $K / K J = 0$ , i.e. $K = K J$ . Since $J$ is right T-nilpotent, the criterion $(23.16)$ forces $K = 0$ . Therefore $θ : P \to M$ is an isomorphism and $M$ is projective.

** $(3) \Rightarrow (2)$ .** Let $R b_{1} \supseteq R b_{2} \supseteq \dots$ be a descending chain of principal left ideals. Writing $b_{i + 1} = a_{i + 1} b_{i}$ and $b_{1} = a_{1}$ , the chain becomes $R a_{1} \supseteq R a_{2} a_{1} \supseteq \dots$ . Attach to the sequence $a_{1}, a_{2}, \dots$ the module $M = F / K$ of $(24.24)$ , which is flat. By (3) it is projective, and then $(24.24)$ says the chain $R a_{1} \supseteq R a_{2} a_{1} \supseteq \dots$ is eventually stationary. Hence so is the original chain.

** $(2) \Rightarrow (1)$ , T-nilpotence.** Let $a_{1}, a_{2}, \dots \in J$ . The chain $R a_{1} \supseteq R a_{2} a_{1} \supseteq \dots$ is stationary by (2), so for some $n$ there is $r \in R$ with $a_{n} \dots a_{1} = r a_{n + 1} a_{n} \dots a_{1}$ , that is

(1 - r a_{n + 1}) a_{n} \dots a_{1} = 0 .

(P.1)

Since $a_{n + 1} \in J$ and $J$ is an ideal, $r a_{n + 1} \in J$ , so $1 - r a_{n + 1} \in U (R)$ by the characterisation $(4.1)$ of the Jacobson radical. Multiplying by its inverse gives $a_{n} \dots a_{1} = 0$ , which is right T-nilpotence of $J$ .

** $(2) \Rightarrow (1)$ , semisimplicity of $R / J$ .** This is the remaining half and is the content of the implications $(2) \Rightarrow (3) \Rightarrow (4) \Rightarrow (1)$ of $(23.20)$ : DCC on principal left ideals gives DCC on cyclic submodules of every left module, hence simple submodules and the absence of infinite orthogonal families of idempotents, from which $R / J$ is semisimple. With both halves, $R$ is right perfect.

Corollary—Direct limits of projectives

If $R$ is right perfect, then every direct limit of projective right $R$ -modules is projective. Indeed direct limits of flat modules are flat, projective modules are flat, and $(24.25)$ returns projectivity. Over $ℤ$ the conclusion fails: $ℚ = varinjlim (ℤ \to ℤ \to \dots)$ is a direct limit of free modules and is not projective.

Corollary—Detecting failure of perfectness

If $R$ admits a sequence $a_{1}, a_{2}, \dots \in rad R$ with $a_{n} \dots a_{1} \neq 0$ for every $n$ , then $R$ is not right perfect and the module $F / K$ of $(24.24)$ is a flat, non-projective right $R$ -module. This is a recipe, not merely an existence statement.

Remark—What the theorem does not say

Flat right modules being projective is a condition on $R$ , not a symmetric one: it characterises right perfectness. The mirror statement — flat left modules are projective — characterises left perfectness, and the two are not equivalent.

Proof Techniques and Method

How these proofs work, and which move to reuse.

Cover the moduleRight perfectness supplies a projective cover for the arbitrary flat module

M

; this is the only place the hypothesis is used in

(1) \Rightarrow (3)

Tensor with $R / J$ Flatness makes the kernel sequence stay exact after tensoring; choosing

R / J

as the test module converts the statement into one about

K / K J

and

P / P J

Collide injectivity with smallnessSmallness of

K

puts it inside

P J

, making the injective map zero. The only module with an injective zero map out of it is

0

Kill the kernel with T-nilpotence

K = K J

plus right T-nilpotence gives

K = 0

(23.16)

; without T-nilpotence this step fails and the argument stops.

Reverse via a constructed moduleFor the converse, do not search for a flat non-projective module — build one from the offending chain using

(24.24)

The last step is the reusable idea: to prove that a homological property forces a chain condition, manufacture a module whose homological behaviour encodes the chain. Bass's module does this by presenting a direct limit $varinjlim (R \overset{a_{1}}{⟶} R \overset{a_{2}}{⟶} \dots)$ as an explicit quotient of a free module.

Worked Example

Failure over $ℤ_{(p)}$ , with the module made explicit

Let $R = ℤ_{(p)}$ , a local ring with $J = p R$ . The chain of principal ideals

R p ⊋ R p^{2} ⊋ R p^{3} ⊋ \dots

(E.1)

Strictly descending, since $p^{k} \notin R p^{k + 1}$ — otherwise $p$ would be a unit.

never stabilises, so $R$ fails condition (2) and is not right perfect. Applying $(24.24)$ with $a_{i} = p$ for all $i$ gives $F = ⨁_{i \geq 0} e_{i} R$ and $K = ⟨ e_{i} - p e_{i + 1} ⟩$ , and the quotient identifies $e_{i}$ with $p e_{i + 1}$ :

M = F / K ≅ varinjlim (R \overset{p}{⟶} R \overset{p}{⟶} R \to \dots) = R [\frac{1}{p}] = ℚ .

(E.2)

The class of $e_{i}$ corresponds to $p^{- i}$ .

So the construction reproduces the classical example: $ℚ$ is flat over $ℤ_{(p)}$ — it is a localisation — and is not projective, because projective modules over a local ring are free and $ℚ$ is not free over $ℤ_{(p)}$ . Everything is consistent with $(24.25)$ .

Success over an artinian ring: $R = k [x] / (x^{3})$

Here $R$ is local with $J = (x) / (x^{3})$ and $J^{3} = 0$ , hence nilpotent and in particular right T-nilpotent, and $R / J ≅ k$ is semisimple. So $R$ is right perfect and every flat $R$ -module is projective — and, being projective over a local ring, free.

Check the chain condition directly: the principal left ideals of $R$ are $R$ , $(x)$ , $(x^{2})$ and $0$ , only four of them, so DCC is trivially satisfied. Running the construction $(24.24)$ with $a_{i} = x$ gives $M = varinjlim (R \overset{x}{⟶} R \to \dots) = 0$ , because $x^{3} = 0$ makes every element eventually die — the construction produces nothing new, exactly as it must over a perfect ring.

Reading the theorem as a test

$ℤ$ : the chain $2 ℤ ⊋ 4 ℤ ⊋ \dots$ fails DCC, and indeed $ℚ$ and $ℤ [1 / 2]$ are flat non-projective.
$k [[x]]$ : the chain $(x) ⊋ (x^{2}) ⊋ \dots$ fails DCC; $k ((x))$ is flat and not projective.
$T_{2} (k)$ : left artinian, so DCC holds on all left ideals in particular, and flat equals projective.
Any semiprimary ring: $J$ nilpotent, so DCC on principal left ideals holds and flat equals projective.

Comparison and Classification

Bass's conditions across standard rings
	Right perfect	DCC on principal left ideals	Flat $\Rightarrow$ projective	Every module has a projective cover
$k$ a field	yes	yes	yes	yes
$k [x] / (x^{n})$	yes	yes	yes	yes
$T_{2} (k)$	yes	yes	yes	yes
$ℤ_{(p)}$	no	no	no	no
$k [[x]]$	no	no	no	no
$ℤ$	no	no	no	no

Bass's conditions across standard rings

The columns move together by $(24.25)$ and $(24.18)$ ; the table is really a check that the theorem's four conditions never disagree. The first three rows are semiprimary, the last three are semiperfect or worse with a non-T-nilpotent radical.

Which theorem supplies which implication
Implication	Tool	Where the hypothesis is spent
Right perfect $\Rightarrow$ flat is projective	(24.12), (24.23), (23.16)	Existence of a cover, and $K = K J \Rightarrow K = 0$
Flat is projective $\Rightarrow$ DCC	(24.24)	Projectivity of the constructed module
DCC $\Rightarrow$ T-nilpotent radical	(4.1)	$1 - r a_{n + 1}$ is a unit
DCC $\Rightarrow$ $R / J$ semisimple	(23.20)	Simple submodules and finite orthogonal families

Relationship Map

R

right perfectevery flat

M_{R}

is projectiveDCC on principal left ideals

R

right perfect

The cycle is the proof. Adding the results of the previous pages gives a longer list of equivalents for right perfectness: every right module has a projective cover $(24.18)$ ; every left module has DCC on cyclic submodules; $R$ has no infinite orthogonal family of idempotents and every nonzero left module has a simple submodule $(23.20)$ .

The equivalent conditions for right perfectness
Condition	Type	Source
$R / J$ semisimple and $J$ right T-nilpotent	internal	Definition (23.18)
DCC on principal left ideals	chain condition	(23.20), (24.25)
Every left module has DCC on cyclic submodules	chain condition	(23.20)
No infinite orthogonal idempotents; every nonzero left module has a simple submodule	mixed	(23.20)
Every right module has a projective cover	homological	(24.18)
Every flat right module is projective	homological	(24.25)

Note the pattern of sides: the homological conditions are stated for right modules, the chain conditions for left ideals and left modules. This is not a misprint anywhere in the list; it is a genuine feature of the theory.

Applications and Industry Use

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

Homological dimension theory. Bass introduced perfect rings to extend the good behaviour of finitistic dimension from semiprimary rings; flat equals projective is what makes flat and projective dimensions agree over such rings.
Representation theory of finite-dimensional algebras. Every such algebra is semiprimary, hence perfect, so flat modules over group algebras and quiver algebras are automatically projective. Computational packages may therefore treat the two notions interchangeably in that setting.
Approximation theory and cotorsion pairs. The contrast between projective covers, which need perfectness, and flat covers, which exist over every ring, is the historical starting point of the theory of covers and envelopes relative to a class of modules.
Descent and localisation. In commutative algebra and algebraic geometry, flatness is the workhorse and projectivity the exception; Bass's theorem explains why the two notions can only be conflated over very special, essentially finite-dimensional-like rings.

Honest summary: the theorem is internal to algebra. Its practical effect is that whenever one works over a finite-dimensional algebra — the setting of nearly all computational representation theory — the distinction between flat and projective can be ignored, and that licence is exactly what Bass's theorem certifies.

Failure Modes and Common Mistakes

Do not conclude from $(24.25)$ that flat modules over a perfect ring are free; they are projective, and freeness requires the ring to be local or otherwise special.
Do not use $(24.24)$ as an equivalence: stationarity of the chain is derived as a necessary condition for projectivity of $F / K$ , not asserted as sufficient.
Do not assume the theorem is symmetric — a right perfect ring that is not left perfect has all flat right modules projective while some flat left module is not.
Do not overlook the role of $(4.1)$ in the last step: without $1 - r a_{n + 1}$ being a unit the chain argument gives nothing, which is why the elements must be taken inside the radical.

Quick Reference

TheoremRight perfect

iff

DCC on principal left ideals

iff

flat right modules are projective

Proof cycle

(1) \Rightarrow (3) \Rightarrow (2) \Rightarrow (1)

Key step

K / K J ↪ P / P J

is injective and zero, so

K = K J

Finish

K = K J

and

J

right T-nilpotent

\Rightarrow K = 0

Converse device

(24.24)

: the flat module attached to

a_{1}, a_{2}, \dots

Chain normalisation

R b_{1} \supseteq R b_{2} \supseteq \dots

becomes

R a_{1} \supseteq R a_{2} a_{1} \supseteq \dots

Unit trick

(1 - r a_{n + 1}) a_{n} \dots a_{1} = 0

with

r a_{n + 1} \in J

CorollaryOver a right perfect ring, direct limits of projectives are projective

Always trueFinitely presented flat

\Rightarrow

projective, over any ring

Diagnostic checklist
Question	Answer if $R$ is right perfect	Answer otherwise
Does every flat right module split off a free one?	Yes, it is projective	Not in general
Is there a flat non-projective right module?	No	Yes, built by (24.24)
Does $varinjlim$ of projectives stay projective?	Yes	Not in general
Do principal left ideals satisfy DCC?	Yes	No
Does every right module have a projective cover?	Yes	No

Frequently Asked Questions

Why does the proof tensor with

R / J

rather than an arbitrary module?

Because $N \otimes_{R} R / J ≅ N / N J$ converts the flatness statement into a statement about radicals, which is where the smallness of the kernel can be used. Any other test module would leave the two hypotheses — flatness and smallness — with no common language.

Where exactly does right T-nilpotence get used?

Twice. Once implicitly, to produce the projective cover via $(24.12)$ ; and once explicitly, to pass from $K = K J$ to $K = 0$ using the criterion $(23.16)$ . Semiperfectness alone gives neither step for an arbitrary flat module.

Is there a flat module that is projective over one ring and not over another?

Yes, in the natural sense: $ℚ$ is flat and not projective over $ℤ$ and over $ℤ_{(p)}$ , but it is projective — indeed free of rank one — over itself. Projectivity depends on the base ring, and the theorem says the base ring is exactly what decides whether flatness suffices.

How does this theorem relate to the Govorov–Lazard theorem?

Lazard's theorem says every flat module is a direct limit of finitely generated free modules. Combined with $(24.25)$ , over a right perfect ring every such direct limit is projective, so perfect rings are precisely those where the class of projectives is closed under direct limits.

Why is Lam's third condition numbered (5)?

To align with $(23.20)$ , whose four conditions are numbered (1)–(4). Adding the flat criterion as a fifth condition emphasises that $(24.25)$ extends the earlier theorem rather than replacing it, and that the present proof also fills the implication left open there.

Does the theorem help decide whether a specific module is projective?

Yes, in one direction: over a right perfect ring, verifying flatness — often easy, via a direct limit presentation — establishes projectivity. Over other rings it warns that no such shortcut exists, and $(24.24)$ even manufactures the counterexample.

References

T. Y. Lam, A First Course in Noncommutative Rings, Graduate Texts in Mathematics 131, Springer-Verlag, 1991, §24, (24.25) (p. 369); see also §23, (23.16) and (23.20).
H. Bass, “Finitistic dimension and a homological generalization of semi-primary rings”, Transactions of the American Mathematical Society 95 (1960), 466–488.
F. W. Anderson and K. R. Fuller, Rings and Categories of Modules, 2nd edition, Graduate Texts in Mathematics 13, Springer-Verlag, 1992, §28.
T. Y. Lam, Lectures on Modules and Rings, Graduate Texts in Mathematics 189, Springer-Verlag, 1999, §4 (flat modules, Lazard's theorem).
D. Lazard, “Autour de la platitude”, Bulletin de la Société Mathématique de France 97 (1969), 81–128.

AI Suggested Questions

Write out the proof that a finitely presented flat module is projective over an arbitrary ring.
Verify directly that the module $F / K$ of $(24.24)$ is the direct limit of $R \to R \to \dots$ along multiplication by the $a_{i}$ .
Give a right perfect ring that is not left perfect and exhibit a flat left module that is not projective.
Deduce from Bass's theorem that flat dimension and projective dimension agree over a right perfect ring.
How does the equivalence interact with Morita equivalence — is flat implies projective a Morita invariant condition?
Compare $(24.25)$ with the theorem that all modules have flat covers, and explain why the two are not in tension.
Which of the equivalences in $(24.25)$ survive for rings that are not assumed to have an identity?

Flat Implies Projective

Executive Summary

Overview

Learning Objectives

Definitions

Core Concepts

Why a projective cover converts flatness into projectivity

Why the chain condition appears on the left

Independence from Theorem P

Key Results

Proof Techniques and Method

Worked Example

Failure over $ℤ_{(p)}$ , with the module made explicit

Success over an artinian ring: $R = k [x] / (x^{3})$

Reading the theorem as a test

Comparison and Classification

Relationship Map

Applications and Industry Use

Failure Modes and Common Mistakes

Quick Reference

Frequently Asked Questions

References

AI Suggested Questions

Continue learning

01Executive Summary

02Overview

03Learning Objectives

04Definitions

05Core Concepts

Why a projective cover converts flatness into projectivity

Why the chain condition appears on the left

Independence from Theorem P

06Key Results

07Proof Techniques and Method

08Worked Example

Failure over ℤ(p), with the module made explicit

Success over an artinian ring: R=k[x]/(x3)

Reading the theorem as a test

09Comparison and Classification

10Relationship Map

11Applications and Industry Use

12Failure Modes and Common Mistakes

13Quick Reference

14Frequently Asked Questions

15Related KEVOS Topics

16References

17AI Suggested Questions

Continue learning

Executive Summary

Overview

Learning Objectives

Definitions

Core Concepts

Key Results

Proof Techniques and Method

Worked Example

Failure over $ℤ_{(p)}$ , with the module made explicit

Success over an artinian ring: $R = k [x] / (x^{3})$

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