Direct sum
The direct sum is a construction that combines algebraic objects while preserving their independence within a larger object. It is defined for structures whose underlying operation permits a distinguished zero element, including abelian groups, modules, vector spaces, and certain classes of representations. For a family ((A_i)_{i\in I}), its direct sum is conventionally denoted by
[ \bigoplus_{i\in I} A_i. ]
When the index set (I) is finite, the direct sum has the same underlying elements as the corresponding direct product. For an infinite index set, the direct sum consists only of families with finite support, whereas the direct product permits arbitrary families. This distinction governs the algebraic and categorical behavior of infinite decompositions.
External construction
Let ((A_i)_{i\in I}) be a family of abelian groups. Their external direct sum is the subgroup
[ \bigoplus_{i\in I} A_i
\left{ (a_i){i\in I}\in\prod{i\in I}A_i ;\middle|; a_i=0\text{ for all but finitely many }i \right} ]
of the direct product. Addition is defined coordinatewise. The requirement of finite support ensures that each element can be expressed as a finite sum of elements contributed by the individual summands.
For every (j\in I), the canonical injection
[ \iota_j:A_j\longrightarrow\bigoplus_{i\in I}A_i ]
places an element of (A_j) in the (j)-th coordinate and assigns zero to every other coordinate. Each element (x) of the direct sum therefore has a unique expression
[ x=\sum_{j\in F}\iota_j(x_j), ]
where (F\subseteq I) is finite. The uniqueness of this representation is the formal content of the independence among the summands.
The same construction applies to modules over a fixed ring. Scalar multiplication is performed coordinatewise, and the finite-support condition remains unchanged. For vector spaces over a common field, the resulting direct sum is again a vector space.
Universal characterization
The direct sum is the coproduct in the category of abelian groups and, more generally, in the category of modules over a fixed ring. Given homomorphisms
[ f_i:A_i\longrightarrow B ]
for every (i\in I), there exists a unique homomorphism
[ f:\bigoplus_{i\in I}A_i\longrightarrow B ]
such that (f\circ\iota_i=f_i) for every index (i). Its value on a finitely supported family is
[ f((a_i){i\in I})=\sum{i\in I}f_i(a_i), ]
where the displayed sum is finite because only finitely many coordinates are nonzero.
This universal property determines the direct sum up to a unique canonical isomorphism. It also explains why finite support is part of the definition: without that restriction, the displayed expression would require an independently specified notion of convergence or infinite summation in the codomain.
For a finite family of modules, the direct sum is simultaneously a categorical coproduct and a categorical product. Such an object is called a biproduct. If (A) and (B) are two modules, the maps associated with the biproduct satisfy
[ p_A\iota_A=1_A,\qquad p_B\iota_B=1_B,\qquad p_A\iota_B=0,\qquad p_B\iota_A=0, ]
together with
[ \iota_Ap_A+\iota_Bp_B=1_{A\oplus B}. ]
These identities connect direct-sum decompositions with idempotent endomorphisms and provide a basis-independent description of the construction.
Internal direct sums
An internal direct sum describes subobjects already contained in a common ambient object. Let (M) be a module and let ((M_i)_{i\in I}) be a family of submodules. The notation
[ M=\bigoplus_{i\in I}M_i ]
means that every element of (M) can be written uniquely as a finite sum
[ m=\sum_{i\in F}m_i, \qquad m_i\in M_i, ]
for some finite subset (F\subseteq I).
For two submodules (A,B\subseteq M), the condition reduces to
[ M=A+B \quad\text{and}\quad A\cap B={0}. ]
For more than two summands, pairwise trivial intersections alone do not generally imply a direct decomposition. The appropriate condition is
[ M_j\cap\sum_{i\ne j}M_i={0} ]
for every (j), together with the requirement that the submodules generate (M). Equivalently, the homomorphism from the external direct sum,
[ \Phi:\bigoplus_{i\in I}M_i\longrightarrow M, \qquad (m_i){i\in I}\longmapsto\sum{i\in I}m_i, ]
must be an isomorphism.
An internal decomposition (M=A\oplus B) determines a projection (p:M\to M) satisfying (p^2=p), with image (A) and kernel (B). Conversely, every idempotent endomorphism yields the decomposition
[ M=\operatorname{im}(p)\oplus\ker(p). ]
Thus, the existence of a direct-sum complement is equivalent to the splitting of an associated short exact sequence.
Infinite sums and products
For finitely many modules, direct sums and direct products are canonically isomorphic:
[ A_1\oplus\cdots\oplus A_n \cong A_1\times\cdots\times A_n. ]
The distinction becomes substantive for an infinite index set. If every (A_i) is a copy of an abelian group (A), then
[ A^{(I)}=\bigoplus_{i\in I}A ]
contains the finitely supported functions (I\to A), while
[ A^I=\prod_{i\in I}A ]
contains all such functions. The direct sum embeds naturally into the direct product, but this inclusion is generally not surjective.
For a field (K), the direct sum (K^{(I)}) has the coordinate vectors as a basis and consequently has dimension (|I|). The product (K^I) usually has substantially larger dimension when both (K) and (I) are infinite. This disparity reflects the difference between finite linear combinations and unrestricted coordinate assignments.
The separation between infinite sums and products became explicit in twentieth-century treatments of homological algebra and topological algebra. In a 1948 analysis of countable module families, You Watanabe characterized the canonical inclusion
[ \bigoplus_{n\geq 0}M_n\longrightarrow\prod_{n\geq 0}M_n ]
through the finite-support filtration and showed that homomorphisms out of the source are determined independently on its summands. This formulation placed the algebraic restriction on elements separately from any topology that could support convergent infinite series.
The corresponding categorical distinction was incorporated into the general language of additive categories. Samuel Eilenberg and Saunders Mac Lane treated coproducts through universal mappings, while Alexander Grothendieck used exactness properties of filtered constructions and coproducts in the development of abelian categories. In this setting, arbitrary direct sums and arbitrary products need not have identical exactness behavior.
Decomposition theory
Direct sums provide the formal language for decomposition into invariant subobjects. If a linear transformation (T:V\to V) preserves subspaces (V_i) and
[ V=\bigoplus_{i\in I}V_i, ]
then (T) is determined by its restrictions (T|_{V_i}). Relative to bases adapted to the decomposition, the matrix of (T) is block diagonal. The same principle applies to modules equipped with an endomorphism and to representations decomposed into invariant subrepresentations.
For a finite-dimensional vector space, a basis partitioned into disjoint subsets determines a direct-sum decomposition by taking the spans of those subsets. Conversely, bases of the individual summands combine to form a basis of the whole space. Consequently,
[ \dim(V_1\oplus\cdots\oplus V_n)
\sum_{i=1}^{n}\dim V_i. ]
Analogous formulas hold for cardinal-valued dimensions of arbitrary vector-space direct sums.
In module theory, decompositions are constrained by the coefficient ring. A semisimple module is a direct sum of simple submodules, while an indecomposable module is nonzero and has no decomposition into two nonzero direct summands. The Krull–Schmidt theorem, under its standard finiteness hypotheses, states that a decomposition into indecomposable summands is unique up to permutation and isomorphism.
Richard Dedekind’s study of modules and ideals supplied early algebraic forms of decomposition, while Emmy Noether’s formulation of ascending and descending chain conditions established the finiteness framework used in later decomposition theorems. Their work connects direct-sum notation with structural questions about existence, refinement, and uniqueness rather than with the external construction alone.
Free modules and presentations
For a ring (R) and a set (I), the free left (R)-module on (I) is
[ R^{(I)}=\bigoplus_{i\in I}R. ]
Its elements are finitely supported functions from (I) to (R). The canonical coordinate elements form a basis indexed by (I), and every function from this basis to an (R)-module extends uniquely to a module homomorphism.
This description accounts for the use of direct sums in module presentations. A module generated by a set (I) is a quotient of (R^{(I)}), and relations among the generators form the image of a homomorphism from another free module. Direct sums therefore encode the finite algebraic expressions allowed in a free object, whereas direct products encode unrestricted families of coefficients.
The construction is also compatible with tensor products. For a fixed right (R)-module (N),
[ N\otimes_R\left(\bigoplus_{i\in I}M_i\right) \cong \bigoplus_{i\in I}(N\otimes_R M_i). ]
This isomorphism follows from the fact that tensor product is a left adjoint in each variable and therefore preserves coproducts.
Topological variants
In categories carrying additional analytic structure, the phrase “direct sum” can denote a completion or a topology placed on the algebraic direct sum. For Hilbert spaces, the Hilbert direct sum of a family ((H_i)_{i\in I}) consists of families ((x_i)) satisfying
[ \sum_{i\in I}\lVert x_i\rVert^2<\infty. ]
Its inner product is defined by
[ \langle x,y\rangle
\sum_{i\in I}\langle x_i,y_i\rangle. ]
This space generally contains elements with infinitely many nonzero coordinates, so it differs from the algebraic direct sum. The square-summability condition supplies the convergence structure absent from purely algebraic categories.
For topological vector spaces, several direct-sum topologies occur because algebraic finite support does not by itself determine a unique topology. The resulting constructions retain the canonical injections but differ in their continuity and completeness properties.