Module Theory — Direct Sums and Free Modules

Direct sums assemble modules into larger modules with canonical inclusions and projections; free modules are direct sums of copies of the ring, possessing a basis and a universal mapping property that makes them the prototype for all modules. Key result: every module is a quotient of a free module, and over a field every module is free (a vector space). Applies throughout commutative algebra, homological algebra, and representation theory. Foundations laid by Dedekind, Noether, and the structural school of the 1920s–1930s.

A module over a ring $R$ generalizes a vector space by allowing scalars from an arbitrary ring rather than a field. The direct sum $⨁_{i \in I} M_{i}$ is the submodule of the direct product consisting of tuples with finitely many nonzero entries, equipped with universal properties dual to those of the product. A free module is one isomorphic to $⨁_{i \in I} R$ , equivalently one admitting a basis: a generating set that is also linearly independent. The surprising fact, invisible in linear algebra, is that over a general ring bases can have different cardinalities (failure of the invariant basis number property) and submodules of free modules need not be free — phenomena that are impossible over fields and that drive much of the subtlety of module theory.

The Core Insight

A free module is the most "unconstrained" module: it is built by stipulating a basis and imposing no relations whatsoever among the basis elements. This is captured precisely by the universal property — a homomorphism out of a free module is uniquely and freely determined by where it sends the basis, with no compatibility conditions to check. Every module arises as a free module modulo relations, so free modules and their kernels (relation modules) encode all of module theory.

Formal Statement

Module

Let $R$ be a ring with $1$ . A left $R$ -module is an abelian group $(M, +)$ together with a scalar multiplication $R \times M \to M$ , $(r, m) \mapsto r m$ , satisfying for all $r, s \in R$ , $m, n \in M$ :

r (m + n) = r m + r n, (r + s) m = r m + s m, (r s) m = r (s m), 1 \cdot m = m .

⟳these are the vector-space axioms with the field replaced by

R

; distributivity on both sides plus associativity of the action plus unitality.

Direct Product and Direct Sum

Let ${M_{i}}_{i \in I}$ be a family of left $R$ -modules. The direct product is

i \in I \prod M_{i} = {(m_{i})_{i \in I} : m_{i} \in M_{i}},

⟳componentwise operations

(m_{i}) + (n_{i}) = (m_{i} + n_{i})

and

r (m_{i}) = (r m_{i})

make this an

R

-module.

The direct sum (external) is the submodule

i \in I ⨁ M_{i} = {(m_{i})_{i \in I} : m_{i} = 0 for all but finitely many i} .

⟳the finite-support condition is closed under addition and scalar multiplication, hence a submodule of the product; for finite

I

the two coincide.

Internal Direct Sum

A module $M$ is the internal direct sum of submodules ${N_{i}}_{i \in I}$ , written $M = ⨁_{i \in I} N_{i}$ , if every $m \in M$ has a unique expression $m = \sum_{i} n_{i}$ with $n_{i} \in N_{i}$ and all but finitely many $n_{i} = 0$ . Equivalently:

M = i \in I \sum N_{i} and N_{j} \cap i \neq = j \sum N_{i} = 0 \forall j .

⟳the sum condition gives existence of the representation; the intersection condition forces uniqueness, since a nonzero overlap would produce two distinct decompositions of the same element.

Free Module and Basis

A subset $B = {b_{i}}_{i \in I} \subseteq M$ is a basis if it is:

generating: every $m \in M$ equals $\sum_{i} r_{i} b_{i}$ for some $r_{i} \in R$ (finite support), and
linearly independent: $\sum_{i} r_{i} b_{i} = 0 ⟹ r_{i} = 0$ for all $i$ .

A module is free if it admits a basis. Equivalently $M$ is free iff

M ≅ i \in I ⨁ R = R^{(I)},

⟳send

b_{i}

to the standard generator

e_{i}

(the tuple with

1

in slot

i

); the basis property makes this a bijective homomorphism, i.e. an isomorphism.

Universal Property of Free Modules

Let $F$ be free on basis $B = {b_{i}}_{i \in I}$ with inclusion $ι : B ↪ F$ . For any $R$ -module $N$ and any set map $f : B \to N$ , there exists a unique $R$ -homomorphism $\overset{ˉ}{f} : F \to N$ with $\overset{ˉ}{f} \circ ι = f$ :

\overset{ˉ}{f} (i \sum r_{i} b_{i}) = i \sum r_{i} f (b_{i}) .

⟳linear independence guarantees the formula is well-defined (each element has a unique coordinate expression); the generating property guarantees

\overset{ˉ}{f}

is defined everywhere; the formula is forced by

R

-linearity, giving uniqueness.

Every Module is a Quotient of a Free Module

For any $R$ -module $M$ there is a free module $F$ and a surjection $π : F ↠ M$ . Hence

M ≅ F / K, K = ker π,

⟳take

F = R^{(M)}

free on a generating set

S

M

(e.g.

S = M

); the universal property extends the inclusion

S ↪ M

π : F \to M

, which is surjective because

S

generates.

K

is the "module of relations."

Why It Holds / Derivation

Universal Property ⟹ Uniqueness of Free Objects

We show free modules are determined up to unique isomorphism by their universal property.

Step 1. Suppose

F_{1}

(basis

B_{1}

, map

ι_{1}

) and

F_{2}

(basis

B_{2}

, map

ι_{2}

) are both free on a set

S

via bijections

β_{1} : S \to B_{1}

β_{2} : S \to B_{2}

. Step 2. By the universal property of

F_{1}

applied to the map

ι_{2} \circ β_{2} \circ β_{1}^{- 1} : B_{1} \to F_{2}

, get a unique

ϕ : F_{1} \to F_{2}

. Symmetrically get

ψ : F_{2} \to F_{1}

. Step 3. The composite

ψ \circ ϕ : F_{1} \to F_{1}

fixes

B_{1}

pointwise. But

id_{F_{1}}

also fixes

B_{1}

. By the uniqueness clause of the universal property,

ψ \circ ϕ = id_{F_{1}}

. Step 4. Symmetrically

ϕ \circ ψ = id_{F_{2}}

. Hence

ϕ

is an isomorphism, and it is the unique one respecting the bases.

■

Direct Sum Universal Property

The direct sum is the coproduct in $R -Mod$ . Given inclusions $λ_{j} : M_{j} \to ⨁_{i} M_{i}$ and any family of homomorphisms $g_{j} : M_{j} \to N$ , there exists a unique $g : ⨁_{i} M_{i} \to N$ with $g \circ λ_{j} = g_{j}$ :

g ((m_{i})_{i}) = i \sum g_{i} (m_{i}) .

⟳the finite-support condition makes the sum finite hence well-defined; linearity is checked componentwise; uniqueness follows since the

λ_{j} (M_{j})

generate the direct sum.

Contrast. The direct product is the product (universal for maps into it via projections). For finite index sets they agree; for infinite sets the coproduct (sum) is strictly smaller.

■

Emmy Noether | 1882–1935

Though "module" predates her, Emmy Noether transformed it from a computational device of Dedekind and Kronecker into the central organizing concept of abstract algebra. In her 1921 Idealtheorie in Ringbereichen and her Göttingen lectures, she recast ideals, groups, and linear systems uniformly as modules over rings, emphasizing homomorphisms and quotients over explicit elements. Her insistence that "we should study structures via their maps" is precisely the spirit of the universal property of free modules. The modern habit of writing $M ≅ F / K$ — presenting a module by generators and relations — is her conceptual legacy.

Building Intuition

Think of a free module $R^{(I)}$ as a bag of independent dials, one per index $i$ , each turnable by any scalar in $R$ , with no dial constraining any other. A homomorphism out of it is just a choice of "where each dial points" in the target — total freedom, hence the name.

A non-free module is what you get when you glue dials together with relations: $Z /6 Z$ as a $Z$ -module is $Z$ (one dial) with the relation $6 \cdot 1 = 0$ welded in. No basis can survive, because the single generator $\overset{ˉ}{1}$ satisfies $6 \overset{ˉ}{1} = 0$ with $6 \neq = 0$ — it is torsion, the opposite of independence.

The direct sum is side-by-side assembly with no interaction: $M \oplus N$ lets you operate on the $M$ -part and $N$ -part completely independently. Over a field, every module decomposes this way down to one-dimensional pieces (a basis), which is why linear algebra is so clean. Over $Z$ or $k [x]$ , indecomposable torsion pieces like $Z / p^{k}$ refuse to break apart, and the structure theory becomes rich.

External direct sum vs. direct product: finite-support vs. full tuples

<svg viewBox="0 0 640 360" xmlns="http://www.w3.org/2000/svg">
  <rect width="640" height="360" fill="#0a0e17"/>
  <text x="320" y="28" fill="#e2e8f0" font-size="16" text-anchor="middle" font-family="serif">Product ∏Mᵢ ⊇ Direct Sum ⊕Mᵢ</text>

  <!-- big box: product -->
  <rect x="40" y="60" width="560" height="240" fill="none" stroke="#94a3b8" stroke-width="1.5"/>
  <text x="50" y="80" fill="#94a3b8" font-size="13" font-family="serif">∏ Mᵢ : all tuples (m₁,m₂,m₃,…)</text>

  <!-- inner box: direct sum -->
  <rect x="70" y="110" width="320" height="160" fill="none" stroke="#d4a853" stroke-width="2"/>
  <text x="80" y="130" fill="#d4a853" font-size="13" font-family="serif">⊕ Mᵢ : finite support</text>

  <!-- sample finite-support tuple inside -->
  <text x="90" y="170" fill="#e2e8f0" font-size="13" font-family="monospace">(a, 0, b, 0, 0, 0, …)</text>
  <text x="90" y="200" fill="#e2e8f0" font-size="13" font-family="monospace">(0, 0, 0, c, 0, 0, …)</text>
  <text x="90" y="230" fill="#e2e8f0" font-size="13" font-family="monospace">eᵢ = (0,…,1,…,0)  basis</text>

  <!-- sample infinite-support tuple outside the sum but in product -->
  <text x="410" y="180" fill="#e2e8f0" font-size="13" font-family="monospace">(1, 1, 1, 1, …)</text>
  <text x="410" y="210" fill="#94a3b8" font-size="12" font-family="serif">∈ ∏  but  ∉ ⊕</text>

  <!-- arrow -->
  <line x1="405" y1="180" x2="395" y2="185" stroke="#94a3b8" stroke-width="1.2" marker-end="url(#a)"/>
  <defs>
    <marker id="a" markerWidth="8" markerHeight="8" refX="6" refY="3" orient="auto">
      <path d="M0,0 L6,3 L0,6 Z" fill="#94a3b8"/>
    </marker>
  </defs>

  <text x="320" y="330" fill="#94a3b8" font-size="12" text-anchor="middle" font-family="serif">For finite I they coincide; for infinite I the sum is a proper submodule.</text>
</svg>

Worked Examples

Free Resolution of $\mathbb{Z}/6\mathbb{Z}$ over $\mathbb{Z}$

Given:

M = Z /6 Z

as a

Z

-module. Present it via free modules. Step 1. Choose generator

\overset{ˉ}{1}

. The free module

F_{0} = Z

surjects via

π (n) = \overset{n}{ˉ}

, so

M ≅ Z / ker π

. Step 2. Compute

ker π = {n : \overset{n}{ˉ} = 0} = 6 Z

. Step 3.

6 Z

is itself free of rank 1, generated by

6

. So set

F_{1} = Z

, with map

d : F_{1} \to F_{0}

given by multiplication by

6

. Step 4. Assemble the free resolution:

0 \to Z \cdot 6 Z π Z /6 Z \to 0.

Z /6 Z ≅ coker (\cdot 6 : Z \to Z), projective dimension 1

This module is not free (it is finite, but $Z^{n}$ is infinite for $n \geq 1$ and $0$ has no nonzero torsion).

Free vs. Non-Free Ideals over a Number Ring

Given:

R = Z [- 5]

, and the ideal

I = (2, 1 + - 5)

. Is

I

free as an

R

-module? Step 1.

I

requires 2 generators and cannot be generated by 1 (it is not principal — a classical fact since

Z [- 5]

has class number 2). Step 2. If

I

were free, being a rank-1 ideal it would have to be free of rank 1, i.e. principal

I = (a)

— contradiction. A free module of rank

\geq 2

inside

R

(rank 1 as

R

-module) is impossible since any two elements of

R

are

R

-linearly dependent: for

x, y \in R

y \cdot x - x \cdot y = 0

is a nontrivial relation. Step 3. Hence

I

is non-free but projective (it is a direct summand of

R^{2}

— invertible ideals are projective).

I = (2, 1 + - 5) is projective of rank 1 but NOT free over Z [- 5]

This is the prototypical example separating "projective" from "free."

Common Misconceptions

"Every linearly independent set extends to a basis" / "every generating set contains a basis." Both fail badly outside vector spaces. Over

Z

, the set

{2} \subseteq Z

is linearly independent but does not extend to a basis (

{2}

is not generating, and adding any

b

gives the relation

b \cdot 2 - 2 \cdot b = 0

). The generating set

{2, 3}

Z

contains no basis:

{2}

and

{3}

each fail to generate. Bases over general rings are rigid: free modules have them, but you cannot freely manufacture or trim them. "Submodules of free modules are free." False in general. The submodule

{(a, b) \in R^{2} : a + b \in I}

over a non-PID, or simply the ideal

I

in the previous example, shows this. The implication holds precisely when

R

is a PID (submodules of free modules are free — even over non-finitely-generated free modules, by a theorem requiring the axiom of choice / well-ordering), and more generally over principal ideal domains and, for the analogous "projective ⟹ free" question, over local rings (Kaplansky) and polynomial rings over fields (Quillen–Suslin).

Failure of Invariant Basis Number

There exist rings $R$ with $R^{m} ≅ R^{n}$ as $R$ -modules for $m \neq = n$ . Let $V$ be an infinite-dimensional $k$ -vector space and $R = End_{k} (V)$ . Then $V ≅ V \oplus V$ (since $\infty = \infty + \infty$ ), which translates to

R ≅ R^{2} as left R -modules,

⟳

R = Hom (V, V)

; decomposing

V ≅ V \oplus V

on the source gives

Hom (V, V) ≅ Hom (V, V)^{2}

, so a free module of rank 1 is isomorphic to one of rank 2.

Thus rank is not well-defined over such rings. Rings where it is well-defined are said to satisfy invariant basis number (IBN) — all commutative rings (≠0), all Noetherian rings, and all division rings do.

Connections

Free modules sit at the base of an importance hierarchy: free ⟹ projective ⟹ flat ⟹ torsion-free (the last over domains). See Projective Modules, where freeness is relaxed to "direct summand of a free module," and Flat Modules, characterized by exactness of $- \otimes M$ . The presentation $M ≅ F / K$ launches Homological Algebra via Free Resolutions and the derived functors $Tor$ and $Ext$ . Over a field, every module is free, recovering Vector Spaces and Linear Algebra; over a PID, the Structure Theorem for Finitely Generated Modules over a PID classifies all such modules as $R^{n} \oplus ⨁ R / (d_{i})$ , the cornerstone of Jordan and Rational Canonical Forms. The direct sum is the coproduct in $R -Mod$ , dual to the product, linking to Universal Properties and Adjoint Functors and the Tensor–Hom Adjunction.

Real-World Applications

Coding theory. Linear codes are submodules of $F_{q}^{n}$ (free modules over finite fields); generator and parity-check matrices encode bases and relation modules. Cyclic codes are ideals in $F_{q} [x] / (x^{n} - 1)$ .
Canonical forms in linear algebra. A vector space with a linear operator $T$ is a $k [x]$ -module; the structure theorem yields Jordan and rational canonical forms directly from invariant factors.
Algebraic number theory. The ring of integers $O_{K}$ of a number field is a free $Z$ -module of rank $[K : Q]$ ; its ideals are projective modules, and the class group measures failure of freeness — central to factorization.
Topology / persistent homology. Homology groups $H_{n} (X; R)$ are $R$ -modules; for $R = Z$ , the structure theorem splits them into free (Betti numbers) and torsion parts. Persistence modules over $k [t]$ underlie topological data analysis (barcodes).
Algebraic geometry. Sheaves of $O_{X}$ -modules locally look like free modules; vector bundles correspond to locally free (= projective) modules of finite rank (Serre–Swan).
Representation theory. Group representations are $k [G]$ -modules; free modules $k [G]^{n}$ are the regular representations, and induction $Ind_{H}^{G} = k [G] \otimes_{k [H]} -$ is built on free-module tensoring.
Control theory. Linear time-invariant systems over $k [x]$ (or rings of stable functions) are modules; controllability and observability translate to freeness/projectivity of associated modules (the behavioral approach of Willems and Oberst).

Historical Development

1871. Richard Dedekind introduces modules (German Modul) as additive subgroups of $C$ closed under multiplication by an order — the seed of module theory, in his supplements to Dirichlet's Vorlesungen über Zahlentheorie.
1890s. Dedekind and Heinrich Weber develop ideal theory; ideals as modules over rings of integers, including the recognition that ideals need not be principal (non-free).
1921. Emmy Noether, Idealtheorie in Ringbereichen, establishes the ascending chain condition and the abstract module-theoretic framework; the rise of the homomorphism-and-quotient viewpoint.
1927–1932. Noether's Göttingen school (van der Waerden's Moderne Algebra, 1930–31) codifies modules, direct sums, and free modules as standard tools; the universal property viewpoint matures.
1937. Hans Hasse, Wolfgang Krull, and others extend module structure theory over Dedekind domains; finitely generated modules over PIDs fully classified.
1945–1956. Eilenberg–Mac Lane and Cartan–Eilenberg (Homological Algebra, 1956) make free and projective resolutions central, defining $Ext$ and $Tor$ .
1955. Jean-Pierre Serre poses whether finitely generated projective modules over $k [x_{1}, \dots, x_{n}]$ are free (Serre's conjecture).
1958. Irving Kaplansky proves projective modules over local rings are free, and that every projective module is a direct sum of countably generated ones.
1976. Daniel Quillen and Andrei Suslin independently resolve Serre's conjecture: finitely generated projectives over a polynomial ring over a field are free (the Quillen–Suslin theorem).

Subtleties & Limits

Where freeness becomes hard. Several deep questions surround the boundary of freeness:

The rank/IBN landscape. Classifying which rings satisfy IBN, the stronger stably free ⟹ free property, and the Hermite and Bass cancellation properties is intricate; counterexamples (e.g. tangent bundle of $S^{2}$ giving a stably free non-free module over $R [x, y, z] / (x^{2} + y^{2} + z^{2} - 1)$ ) reveal genuine geometry.
Generalizations of Quillen–Suslin. The Bass–Quillen conjecture — that finitely generated projective modules over $R [x_{1}, \dots, x_{n}]$ are extended from $R$ for regular $R$ — remains open in general dimension.
Infinitely generated projectives. Bass's question on whether "big" projective modules over Noetherian rings are free or determined by simple invariants leads to subtle set-theoretic dependence; Whitehead's problem (is every Whitehead $Z$ -module free?) was shown by Shelah to be independent of ZFC.
Trace ideals and non-free generation. Characterizing modules with few generators, or the minimal number of generators of an ideal (Forster–Swan bound), remains an active arena between commutative algebra and algebraic geometry.