Normed vector space

In mathematics, a normed vector space or normed space is a vector space over the real or complex numbers, on which a norm is defined.[1] A norm is the formalization and the generalization to real vector spaces of the intuitive notion of "length" in the real (physical) world. A norm is a real-valued function defined on the vector space that is commonly denoted and has the following properties:[2]

  1. It is nonnegative, meaning that for every vector
  2. It is positive on nonzero vectors, that is,
  3. For every vector and every scalar
  4. The triangle inequality holds; that is, for every vectors and
Hierarchy of mathematical spaces. Normed vector spaces are a superset of inner product spaces and a subset of metric spaces, which in turn is a subset of topological spaces.

A norm induces a distance, called its (norm) induced metric, by the formula

which makes any normed vector space into a metric space and a topological vector space. If this metric space is complete then the normed space is a Banach space. Every normed vector space can be "uniquely extended" to a Banach space, which makes normed spaces intimately related to Banach spaces. Every Banach space is a normed space but converse is not true. For example, the set of the finite sequences of real numbers can be normed with the Euclidean norm, but it is not complete for this norm.

An inner product space is a normed vector space whose norm is the square root of the inner product of a vector and itself. The Euclidean norm of a Euclidean vector space is a special case that allows defining Euclidean distance by the formula

The study of normed spaces and Banach spaces is a fundamental part of functional analysis, which is a major subfield of mathematics.


A normed vector space is a vector space equipped with a norm. A seminormed vector space is a vector space equipped with a seminorm.

A useful variation of the triangle inequality is

for any vectors   and  

This also shows that a vector norm is a (uniformly) continuous function.

Property 3 depends on a choice of norm   on the field of scalars. When the scalar field is   (or more generally a subset of  ), this is usually taken to be the ordinary absolute value, but other choices are possible. For example, for a vector space over   one could take   to be the  -adic absolute value.

Topological structureEdit

If   is a normed vector space, the norm   induces a metric (a notion of distance) and therefore a topology on   This metric is defined in the natural way: the distance between two vectors   and   is given by   This topology is precisely the weakest topology which makes   continuous and which is compatible with the linear structure of   in the following sense:

  1. The vector addition   is jointly continuous with respect to this topology. This follows directly from the triangle inequality.
  2. The scalar multiplication   where   is the underlying scalar field of   is jointly continuous. This follows from the triangle inequality and homogeneity of the norm.

Similarly, for any seminormed vector space we can define the distance between two vectors   and   as   This turns the seminormed space into a pseudometric space (notice this is weaker than a metric) and allows the definition of notions such as continuity and convergence. To put it more abstractly every seminormed vector space is a topological vector space and thus carries a topological structure which is induced by the semi-norm.

Of special interest are complete normed spaces, which are known as Banach spaces. Every normed vector space   sits as a dense subspace inside some Banach space; this Banach space is essentially uniquely defined by   and is called the completion of  

Two norms on the same vector space are called equivalent if they define the same topology. On a finite-dimensional vector space, all norms are equivalent but this is not true for infinite dimensional vector spaces.

All norms on a finite-dimensional vector space are equivalent from a topological viewpoint as they induce the same topology (although the resulting metric spaces need not be the same).[3] And since any Euclidean space is complete, we can thus conclude that all finite-dimensional normed vector spaces are Banach spaces. A normed vector space   is locally compact if and only if the unit ball   is compact, which is the case if and only if   is finite-dimensional; this is a consequence of Riesz's lemma. (In fact, a more general result is true: a topological vector space is locally compact if and only if it is finite-dimensional. The point here is that we don't assume the topology comes from a norm.)

The topology of a seminormed vector space has many nice properties. Given a neighbourhood system   around 0 we can construct all other neighbourhood systems as


Moreover, there exists a neighbourhood basis for the origin consisting of absorbing and convex sets. As this property is very useful in functional analysis, generalizations of normed vector spaces with this property are studied under the name locally convex spaces.

A norm (or seminorm)   on a topological vector space   is continuous if and only if the topology   that   induces on   is coarser than   (meaning,  ), which happens if and only if there exists some open ball   in   (such as maybe   for example) that is open in   (said different, such that  ).

Normable spacesEdit

A topological vector space   is called normable if there exists a norm   on   such that the canonical metric   induces the topology   on   The following theorem is due to Kolmogorov:[4]

Kolmogorov's normability criterion: A Hausdorff topological vector space is normable if and only if there exists a convex, von Neumann bounded neighborhood of  

A product of a family of normable spaces is normable if and only if only finitely many of the spaces are non-trivial (that is,  ).[4] Furthermore, the quotient of a normable space   by a closed vector subspace   is normable, and if in addition  's topology is given by a norm   then the map   given by   is a well defined norm on   that induces the quotient topology on  [5]

If   is a Hausdorff locally convex topological vector space then the following are equivalent:

  1.   is normable.
  2.   has a bounded neighborhood of the origin.
  3. the strong dual space   of   is normable.[6]
  4. the strong dual space   of   is metrizable.[6]

Furthermore,   is finite dimensional if and only if   is normable (here   denotes   endowed with the weak-* topology).

The topology   of the Fréchet space   as defined in the article on spaces of test functions and distributions, is defined by a countable family of norms but it is not a normable space because there does not exist any norm   on   such that the topology that this norm induces is equal to  

Even if a metrizable topological vector space has a topology that is defined by a family of norms, then it may nevertheless still fail to be normable space (meaning that its topology can not be defined by any single norm). An example of such a space is the Fréchet space   whose definition can be found in the article on spaces of test functions and distributions, because its topology   is defined by a countable family of norms but it is not a normable space because there does not exist any norm   on   such that the topology this norm induces is equal to   In fact, the topology of a locally convex space   can be a defined by a family of norms on   if and only if there exists at least one continuous norm on  [7]

Linear maps and dual spacesEdit

The most important maps between two normed vector spaces are the continuous linear maps. Together with these maps, normed vector spaces form a category.

The norm is a continuous function on its vector space. All linear maps between finite dimensional vector spaces are also continuous.

An isometry between two normed vector spaces is a linear map   which preserves the norm (meaning   for all vectors  ). Isometries are always continuous and injective. A surjective isometry between the normed vector spaces   and   is called an isometric isomorphism, and   and   are called isometrically isomorphic. Isometrically isomorphic normed vector spaces are identical for all practical purposes.

When speaking of normed vector spaces, we augment the notion of dual space to take the norm into account. The dual   of a normed vector space   is the space of all continuous linear maps from   to the base field (the complexes or the reals) — such linear maps are called "functionals". The norm of a functional   is defined as the supremum of   where   ranges over all unit vectors (that is, vectors of norm  ) in   This turns   into a normed vector space. An important theorem about continuous linear functionals on normed vector spaces is the Hahn–Banach theorem.

Normed spaces as quotient spaces of seminormed spacesEdit

The definition of many normed spaces (in particular, Banach spaces) involves a seminorm defined on a vector space and then the normed space is defined as the quotient space by the subspace of elements of seminorm zero. For instance, with the   spaces, the function defined by

is a seminorm on the vector space of all functions on which the Lebesgue integral on the right hand side is defined and finite. However, the seminorm is equal to zero for any function supported on a set of Lebesgue measure zero. These functions form a subspace which we "quotient out", making them equivalent to the zero function.

Finite product spacesEdit

Given   seminormed spaces   with seminorms   denote the product space by

where vector addition defined as
and scalar multiplication defined as

Define a new function   by

which is a seminorm on   The function   is a norm if and only if all   are norms.

More generally, for each real   the map   defined by

is a semi norm. For each   this defines the same topological space.

A straightforward argument involving elementary linear algebra shows that the only finite-dimensional seminormed spaces are those arising as the product space of a normed space and a space with trivial seminorm. Consequently, many of the more interesting examples and applications of seminormed spaces occur for infinite-dimensional vector spaces.

See alsoEdit


  1. ^ Callier, Frank M. (1991). Linear System Theory. New York: Springer-Verlag. ISBN 0-387-97573-X.
  2. ^ Rudin 1991, pp. 3–4.
  3. ^ Kedlaya, Kiran S. (2010), p-adic differential equations, Cambridge Studies in Advanced Mathematics, vol. 125, Cambridge University Press, CiteSeerX, ISBN 978-0-521-76879-5, Theorem 1.3.6
  4. ^ a b Schaefer 1999, p. 41.
  5. ^ Schaefer 1999, p. 42.
  6. ^ a b Trèves 2006, pp. 136–149, 195–201, 240–252, 335–390, 420–433.
  7. ^ Jarchow 1981, p. 130.


External linksEdit