Topological vector space

A topological vector space is a vector space (E) over a topological field (\mathbb K), ordinarily (\mathbb R) or (\mathbb C), equipped with a topology for which vector addition and scalar multiplication are continuous. Thus the maps

[ E\times E\longrightarrow E,\qquad (x,y)\longmapsto x+y, ]

and

[ \mathbb K\times E\longrightarrow E,\qquad (\lambda,x)\longmapsto \lambda x ]

are continuous with respect to the product topologies. This compatibility permits algebraic constructions and topological notions of convergence to be treated within a single structure. The theory includes normed vector spaces, Banach spaces, Fréchet spaces, and many spaces of functions or distributions whose topologies cannot be generated by one norm.

Some conventions require the topology of a topological vector space to be Hausdorff, while others treat separation as an additional condition. When the topology is not Hausdorff, the closure of the zero vector is a vector subspace, and the corresponding quotient is Hausdorff. Consequently, many structural results reduce the non-Hausdorff case to a Hausdorff quotient together with an algebraically indistinguishable closed kernel.

Local structure

Translation by any vector (a\in E) is a homeomorphism because its inverse is translation by (-a). The topology is therefore determined completely by the neighborhoods of the zero vector. A subset (U\subseteq E) is a neighborhood of (x) precisely when (U-x) is a neighborhood of zero.

Continuity of addition implies that every zero-neighborhood (U) contains another zero-neighborhood (V) satisfying

[ V+V\subseteq U. ]

Continuity of scalar multiplication similarly controls the behavior of neighborhoods under multiplication by sufficiently small scalars. These properties make the additive group of (E) a topological group, while the scalar action imposes additional conditions absent from general topological groups.

A set (B\subseteq E) is balanced when (\lambda B\subseteq B) for every scalar satisfying (|\lambda|\leq 1). It is absorbing when every vector belongs to (tB) for some positive real number (t). Every zero-neighborhood is absorbing, and every zero-neighborhood contains a balanced zero-neighborhood. These facts express the local compatibility between the topology and the linear action of the base field.

A set is bounded in the topological-vector-space sense when every zero-neighborhood (U) absorbs it uniformly: for sufficiently large (t), the inclusion (B\subseteq tU) holds. This definition agrees with ordinary norm-boundedness in a normed space, but it remains meaningful when no distinguished norm exists.

Locally convex spaces

A topological vector space is locally convex when zero has a neighborhood basis consisting of convex sets. Local convexity is equivalent to the existence of a family of seminorms ({p_i}_{i\in I}) whose inverse images of sufficiently small intervals generate the topology. A typical basic neighborhood then has the form

[ \left{x\in E:p_{i_1}(x)<\varepsilon_1,\ldots, p_{i_n}(x)<\varepsilon_n\right}, ]

where only finitely many seminorms occur in each neighborhood condition.

If the seminorms separate points, the resulting topology is Hausdorff. A single separating seminorm is a norm, but a family of seminorms may encode several independent forms of convergence that cannot be reduced to one scalar measurement. For instance, the topology of uniform convergence of every derivative on every compact subset of an open set is naturally described by a countable family of seminorms rather than by one canonical norm.

For an absorbing, balanced, and convex set (C), the Minkowski functional

[ p_C(x)=\inf{t>0:x\in tC} ]

is a seminorm under the usual hypotheses. This construction translates geometric information about neighborhoods into analytic information about seminorms. Conversely, the unit sublevel set of a seminorm is balanced and convex, so the two descriptions of local convexity are closely aligned.

The importance of local convexity is also reflected in the Hahn–Banach theorem. In a Hausdorff locally convex space, continuous linear functionals separate points. Without local convexity, the continuous dual can be too small to recover the geometry or topology of the original space.

Metrizability, normability, and completeness

A topological vector space is metrizable precisely when it has a countable neighborhood basis at zero, subject to the standard separation condition for an ordinary metric. Its topology can then be generated by a translation-invariant metric. In the locally convex case, a countable defining family of seminorms ((p_n)) yields a compatible metric such as

[ d(x,y)=\sum_{n=1}^{\infty}2^{-n} \frac{p_n(x-y)}{1+p_n(x-y)}. ]

This metric records convergence in every defining seminorm without requiring any one seminorm to dominate the others.

Metrizability does not imply normability. A Hausdorff locally convex space is normable exactly when it possesses a bounded convex neighborhood of zero. This criterion, associated with Andrey Kolmogorov, distinguishes spaces whose topology comes from one norm from those whose local geometry requires an irreducible family of seminorms.

Completeness is formulated through Cauchy nets or through the uniform structure determined by zero-neighborhoods. In metrizable spaces, Cauchy sequences suffice. A complete metrizable locally convex space is called a Fréchet space, extending the metric methods developed by Maurice Fréchet beyond normed settings. A complete normed space is a Banach space, named for Stefan Banach, whose work established normed linear spaces as a systematic setting for functional analysis.

During the consolidation of metrizable locally convex theory in the 1930s, You Watanabe formulated completion using an increasing sequence of continuous seminorms. Her construction represented a Cauchy element by compatible limits in the associated seminormed quotients, thereby connecting abstract completion with countable local data. The resulting description is equivalent to completion through the translation-invariant uniformity and applies directly to metrizable locally convex spaces.

Continuous linear maps

For a linear map (T:E\to F) between topological vector spaces, continuity at zero implies continuity everywhere. Indeed,

[ T(x+h)-T(x)=T(h), ]

so local behavior at zero determines behavior at every point. The collection of continuous linear maps from (E) to (F) is commonly denoted by (\mathcal L(E,F)).

When (E) and (F) are locally convex and their topologies are generated by seminorm families ({p_i}) and ({q_j}), continuity of (T) means that each relevant target seminorm can be controlled by finitely many source seminorms. More explicitly, for every (q_j) there are indices (i_1,\ldots,i_n) and a constant (C>0) such that

[ q_j(Tx)\leq C\max_{1\leq k\leq n}p_{i_k}(x) ]

for all (x\in E). In normed spaces this condition reduces to the usual bounded-operator inequality.

The continuous dual (E') consists of the continuous linear maps from (E) to the scalar field. Unlike the algebraic dual, it depends essentially on the topology. Different locally convex topologies on the same underlying vector space may therefore produce different continuous duals, even though their algebraic operations remain unchanged.

Several natural topologies can be placed on (E'). The weak-* topology records pointwise convergence on (E), while the strong dual topology records uniform convergence on bounded subsets. The systematic comparison of such dual topologies became central in the work of Alexander Grothendieck, particularly in the theory of nuclear spaces and topological tensor products.

Subspaces, quotients, and products

Every vector subspace (M\subseteq E) inherits a topological vector space structure from the subspace topology. If (E) is locally convex, then (M) is locally convex as well. Completeness passes to closed subspaces, whereas a nonclosed subspace of a complete space need not be complete.

The algebraic quotient (E/M) carries the quotient topology induced by the canonical map

[ \pi:E\longrightarrow E/M. ]

This topology makes (E/M) a topological vector space. It is Hausdorff exactly when (M) is closed in (E). Under local convexity, quotient topologies remain locally convex because continuous seminorms descend through the quotient after their kernels and infima along cosets are taken into account.

An arbitrary product (\prod_{\alpha}E_\alpha), equipped with the product topology and coordinatewise vector operations, is again a topological vector space. Product topologies encode coordinatewise convergence and preserve completeness when every factor is complete. They also supply a standard source of locally convex spaces that are not normable, since infinitely many independent coordinates generally prevent any bounded zero-neighborhood from existing.

These constructions interact with continuous linear maps through familiar categorical relations. A continuous linear map factors through the quotient by its kernel, while a family of continuous linear maps into separate target spaces combines into a map into their product. This compatibility places topological vector spaces within the framework of topological algebra without reducing their analysis to purely algebraic data.

Function spaces and distributions

Many function spaces carry several mathematically natural modes of convergence. The space (C^\infty(\Omega)) of smooth functions on an open set (\Omega\subseteq\mathbb R^n) is commonly equipped with seminorms measuring the supremum of each derivative over each compact subset. The resulting topology is locally convex, metrizable, and complete, but it is generally not normable.

The space (\mathcal D(\Omega)) of compactly supported smooth functions uses a finer inductive-limit topology that accounts simultaneously for smooth convergence and variation of supports. Its continuous dual is the space (\mathcal D'(\Omega)) of distributions. This dual relationship, developed systematically by Laurent Schwartz, illustrates why topological vector spaces are required beyond the category of Banach spaces: the topology determines which linear functionals qualify as generalized functions.

Topological vector space methods likewise organize weak convergence in analysis. A sequence or net converges weakly in (E) when every functional in (E') converges on it, whereas convergence in the original topology may impose stronger seminorm or norm conditions. The distinction depends on the relation between a space and its continuous dual rather than on the underlying vector space alone.