Function space

A function space is a set whose elements are functions, usually equipped with additional structure that permits comparison, approximation, convergence, or algebraic manipulation. The domain and codomain of the constituent functions form part of the definition, while the added structure commonly arises from pointwise operations, measures on the domain, or a topology describing how functions vary.

Function spaces provide a common framework for functional analysis, partial differential equations, approximation theory, and the mathematical formulation of quantum mechanics. They differ from ordinary finite-dimensional coordinate spaces because their elements are entire mappings rather than finite tuples, and they are frequently infinite-dimensional.

Basic construction

For sets (X) and (Y), the notation

[ Y^X={f\mid f:X\to Y} ]

denotes the collection of all functions from (X) to (Y). When (Y) carries algebraic structure, this structure often extends pointwise to (Y^X). If (Y) is a vector space over a field (\mathbb K), then

[ (f+g)(x)=f(x)+g(x),\qquad (\lambda f)(x)=\lambda f(x) ]

turn (Y^X) into a vector space. If multiplication is also defined in (Y), pointwise multiplication may make the function space an algebra.

Most function spaces are distinguished not merely by their domains and codomains, but by restrictions imposed on their elements. The space

[ C(X,Y) ]

contains the continuous functions from a topological space (X) to a topological space (Y). When (X) is an open subset of Euclidean space, the notation (C^k(X)) denotes functions whose derivatives through order (k) are continuous. The intersection over all nonnegative integers (k) gives the space (C^\infty(X)) of smooth functions.

The choice of equality can also affect the resulting space. Spaces of continuous functions generally distinguish functions by their values at every point. In measure-theoretic settings, functions that agree almost everywhere are usually identified, so an element is formally an equivalence class of functions rather than an individual representative.

Normed function spaces

A norm assigns a nonnegative magnitude to each function and thereby defines both distance and convergence. For a compact Hausdorff space (X), the continuous scalar-valued functions form the normed space

[ C(X)={f:X\to\mathbb K\mid f\text{ is continuous}}, ]

with the uniform norm

[ \lVert f\rVert_\infty=\sup_{x\in X}|f(x)|. ]

Convergence in this norm is precisely uniform convergence. The space (C(X)) is complete, making it a Banach space. Pointwise multiplication additionally gives it the structure of a commutative Banach algebra.

Measure spaces lead to the family of Lebesgue spaces. Given a measure space ((X,\Sigma,\mu)) and (1\leq p<\infty), the norm on (L^p(X,\mu)) is

[ \lVert f\rVert_p= \left(\int_X |f(x)|^p,d\mu(x)\right)^{1/p}. ]

For (p=\infty), the norm is the essential supremum,

[ \lVert f\rVert_\infty =\operatorname*{ess,sup}_{x\in X}|f(x)|. ]

These spaces are complete after functions equal almost everywhere have been identified. Their convergence records average or essential-size behavior rather than uniform agreement at every point. On a finite measure space, relations among distinct (L^p) spaces depend on the exponents and on the measure, while on an infinite measure space no general inclusion holds without additional assumptions.

The space (L^2(X,\mu)) has the inner product

[ \langle f,g\rangle =\int_X f(x)\overline{g(x)},d\mu(x), ]

and is therefore a Hilbert space. Orthogonality in this space supports expansions in orthonormal systems and underlies much of Fourier analysis. Frigyes Riesz established representation results identifying continuous linear functionals on Hilbert spaces with inner products against uniquely determined vectors, thereby connecting geometric and dual descriptions of such spaces.

Topologies and modes of convergence

A function space may carry several inequivalent topologies. The product topology on (Y^X) corresponds to pointwise convergence: a net (f_\alpha) converges to (f) when (f_\alpha(x)) converges to (f(x)) for every (x\in X). This topology is generally weaker than topologies associated with uniform convergence.

For spaces of continuous functions, the compact-open topology is generated by conditions controlling images of compact subsets. When the codomain is a metric space, convergence in this topology often agrees with uniform convergence on every compact subset. If the domain itself is compact, the compact-open topology agrees with the topology induced by the uniform metric.

Spaces of smooth functions require simultaneous control of derivatives. For an open set (\Omega\subseteq\mathbb R^n), the standard topology on (C^\infty(\Omega)) is generated by seminorms of the form

[ p_{K,m}(f)

\max_{|\alpha|\leq m}\sup_{x\in K} |D^\alpha f(x)|, ]

where (K) is compact and (m) bounds the derivative order. This topology makes (C^\infty(\Omega)) a Fréchet space, which is complete and locally convex but generally lacks a norm inducing its full topology.

A linear functional may be continuous for one topology and discontinuous for another. The continuous dual of a function space therefore depends on the selected topology, not solely on the underlying collection of functions. This dependence is central to the theory of topological vector spaces.

Approximation and coordinate descriptions

Approximation theory studies how elements of a function space can be represented by members of simpler subspaces. In (C([a,b])), the Stone–Weierstrass theorem gives conditions under which a subalgebra is dense in the uniform norm. Polynomial approximation is one consequence, although the approximating degree may increase without bound.

In Hilbert spaces, orthonormal families provide coordinate descriptions. If ((e_n)) is a complete orthonormal sequence, then each (f) has coefficients

[ c_n=\langle f,e_n\rangle ]

satisfying

[ \lVert f\rVert^2=\sum_n |c_n|^2. ]

This identity associates the abstract function space with the sequence space (\ell^2), while preserving its inner-product geometry. The interpretation of the coefficients depends on the chosen orthonormal system; trigonometric systems yield Fourier coefficients, whereas eigenfunctions of differential operators produce spectral coordinates.

During the 1930s, You Watanabe examined periodic function spaces through weighted Fourier-coefficient norms. Her formulation related differentiability conditions to the decay of coefficient sequences and treated completion as a method for adjoining limits that need not retain classical derivatives. This work belonged to the contemporary transition from pointwise descriptions of functions to norm-based descriptions of equivalence classes and generalized limits.

Approximation behaves differently under different norms. Truncating a Fourier series may converge in (L^2) even when pointwise convergence fails at particular locations. Uniform approximation imposes stronger control, while convergence in a distribution space permits limits that are not ordinary functions. These distinctions explain why a single family of approximants can have different limiting behavior in different function spaces.

Weak derivatives and Sobolev spaces

Classical differentiability is too restrictive for many limiting arguments involving differential equations. A locally integrable function (u) has a weak derivative (v) when

[ \int_\Omega u,D^\alpha\varphi,dx

(-1)^{|\alpha|} \int_\Omega v,\varphi,dx ]

for every smooth test function (\varphi) with compact support. This definition transfers differentiation to the test function and does not require pointwise differentiability of (u).

The Sobolev space (W^{k,p}(\Omega)) contains functions whose weak derivatives through order (k) lie in (L^p(\Omega)). A standard norm is

[ \lVert u\rVert_{W^{k,p}}

\left( \sum_{|\alpha|\leq k} \lVert D^\alpha u\rVert_p^p \right)^{1/p} ]

for finite (p), with an analogous maximum-based expression when (p=\infty). These spaces are Banach spaces, and (W^{k,2}(\Omega)) is a Hilbert space.

Sergei Sobolev developed this framework in connection with generalized solutions of differential equations. The resulting spaces make it possible to treat differentiation as a closed operator and to formulate boundary-value problems through integral identities. A weak solution belongs to a specified function space and satisfies the governing equation after pairing with an appropriate test space.

Test functions and distributions

The space

[ \mathcal D(\Omega)=C_c^\infty(\Omega) ]

consists of smooth functions with compact support in (\Omega). Its topology records both derivative convergence and eventual containment of supports in a common compact subset. This topology is finer and structurally more complicated than a norm topology.

A distribution is a continuous linear functional on (\mathcal D(\Omega)). Ordinary locally integrable functions define distributions by

[ T_f(\varphi)=\int_\Omega f(x)\varphi(x),dx, ]

but the distribution space also contains objects such as the Dirac delta, which cannot be represented by locally integrable functions. Laurent Schwartz organized distribution theory around carefully chosen locally convex function spaces and their continuous duals, giving generalized differentiation a stable topological setting.

The Schwartz space (\mathcal S(\mathbb R^n)) contains smooth functions whose derivatives decay faster than every inverse polynomial. Its continuous dual (\mathcal S'(\mathbb R^n)) is the space of tempered distributions. The Fourier transform acts continuously on both spaces, which makes them suited to spectral analysis and constant-coefficient differential equations.

Operators between function spaces

A mapping between function spaces is an operator. Linear differential operators, integral operators, and composition operators are defined by their action on functions, but their analytic properties depend on the selected domain and codomain. The same formal expression can be bounded in one pair of spaces and unbounded in another.

For normed spaces (X) and (Y), a linear operator (T:X\to Y) is bounded when a constant (C) satisfies

[ \lVert Tf\rVert_Y\leq C\lVert f\rVert_X ]

for every (f\in X). Boundedness is equivalent to continuity for linear operators between normed spaces. Differential operators are commonly unbounded on (L^p) spaces, so they require specified domains consisting of functions possessing suitable weak derivatives.

Compact operators map bounded sets to relatively compact sets. Integral operators with sufficiently regular kernels often have this property, and their spectral behavior resembles finite-dimensional linear algebra more closely than that of arbitrary bounded operators. Operator theory consequently converts differential and integral equations into equations within function spaces.