Functional (mathematics)
In mathematics, a functional is a mapping whose argument is a function or another element of a function space. In its most common form, a functional assigns a scalar to each element of a vector space,
[ F:X\longrightarrow \mathbb K, ]
where (X) is usually a space of functions and (\mathbb K) is either the field of real numbers or the field of complex numbers. In functional analysis, the unqualified term frequently denotes a linear functional. In the calculus of variations, it also encompasses nonlinear mappings such as integrals whose values depend on an entire curve or field.
The distinction between a functional and an ordinary scalar-valued function concerns the mathematical structure of the domain rather than the formal definition of a mapping. A function (f:\mathbb R\to\mathbb R) takes numerical arguments, whereas a functional (F:C([a,b])\to\mathbb R) takes continuous functions as arguments. The notations (F(f)) and (F[f]) are both used, with square brackets commonly emphasizing dependence on a function rather than on its value at an individual point.
Linear functionals
A functional (\ell:X\to\mathbb K) on a vector space is linear when
[ \ell(\alpha x+\beta y) =\alpha\ell(x)+\beta\ell(y) ]
for every (x,y\in X) and every pair of scalars (\alpha,\beta\in\mathbb K). The collection of all linear functionals on (X) forms the algebraic dual space, conventionally denoted (X^). When (X) also carries a topology, the continuous linear functionals form the continuous dual. Depending on convention, this continuous dual is denoted by (X'), (X^\prime), or again by (X^).
For a normed vector space, a linear functional is continuous exactly when it is bounded. Its operator norm is
[ |\ell| =\sup_{|x|\leq 1}|\ell(x)|. ]
This norm turns the continuous dual into a Banach space, regardless of whether the original normed space is complete. The algebraic dual is generally much larger because it contains linear mappings that are not continuous with respect to the given topology.
Evaluation supplies a basic example. If (X=C([a,b])) and (t_0\in[a,b]), then
[ \delta_{t_0}(f)=f(t_0) ]
defines a continuous linear functional under the uniform norm. The same expression does not define a functional on an (L^p) space whose elements are equivalence classes of functions modulo equality almost everywhere, because the value at a single point depends on the representative selected from the equivalence class.
Integration provides another standard construction. Given an integrable weight (g), the expression
[ \ell_g(f)=\int_a^b f(t)g(t),dt ]
defines a linear functional whenever the function spaces and integrability conditions make the integral finite. The duality between Lebesgue spaces is based on this pairing. For (1<p<\infty), every continuous linear functional on (L^p) over a standard measure space has this form with (g\in L^q), where (1/p+1/q=1), subject to the usual measure-theoretic hypotheses.
On a Hilbert space (H), the Riesz representation theorem identifies every continuous linear functional with an inner product against a unique vector. Frigyes Riesz established this representation in the development of Hilbert-space theory, while Hans Hahn and Stefan Banach independently developed the extension principle now called the Hahn–Banach theorem. The latter theorem extends bounded linear functionals from subspaces without increasing their norms and underlies the separation of convex sets by continuous linear functionals.
Nonlinear functionals
A general functional need not preserve vector addition or scalar multiplication. Important nonlinear examples arise from expressions of the form
[ J[u]=\int_\Omega L\bigl(x,u(x),\nabla u(x)\bigr),dx, ]
where (\Omega) is a domain, (u) belongs to an admissible function space, and (L) is an integrand depending on position, the value of (u), and its derivatives. Such functionals occur in the variational formulation of differential equations, in classical mechanics, and in the study of minimal surfaces.
The derivative of a nonlinear functional at a point is itself a linear functional under appropriate differentiability conditions. The Gâteaux derivative of (J) at (u) in the direction (v) is
[ \delta J(u;v) =\lim_{\varepsilon\to 0} \frac{J[u+\varepsilon v]-J[u]}{\varepsilon}, ]
provided that the limit exists. This construction records directional variation, but its existence in every direction does not by itself imply continuity with respect to the direction or uniform approximation by a linear map.
The Fréchet derivative imposes the stronger requirement that there exist a bounded linear functional (DJ(u)) satisfying
[ J[u+h] =J[u]+DJ(u)[h]+o(|h|) ]
as (|h|\to0). Fréchet differentiability therefore expresses first-order approximation in the topology of the domain. When both derivatives exist in the appropriate sense, the Fréchet derivative determines the Gâteaux derivative in every direction.
For the integral functional
[ J[y]=\int_a^b L\bigl(x,y(x),y'(x)\bigr),dx, ]
the first variation in a direction (\eta) is
[ \delta J(y;\eta) =\int_a^b \left( \frac{\partial L}{\partial y}\eta +\frac{\partial L}{\partial y'}\eta' \right)dx. ]
Integration by parts separates this functional into an interior contribution and a boundary contribution:
[ \delta J(y;\eta)
\int_a^b \left( \frac{\partial L}{\partial y} -\frac{d}{dx}\frac{\partial L}{\partial y'} \right)\eta,dx + \left[ \frac{\partial L}{\partial y'}\eta \right]_a^b. ]
For fixed endpoint values, the admissible variations vanish at the boundary. Stationarity with respect to all such variations then yields the Euler–Lagrange equation. When endpoint values or endpoint locations vary, the boundary functional instead produces the corresponding transversality condition.
Historical development
Functionals emerged from eighteenth-century variational problems in which an entire curve, rather than a finite list of numerical variables, served as the unknown. Leonhard Euler developed systematic methods for extremizing integral expressions, and Joseph-Louis Lagrange recast their infinitesimal changes in terms of variations. Their work supplied the prototype for differentiating quantities defined on spaces of functions.
During the late nineteenth and early twentieth centuries, this framework was separated from its original mechanical setting. Vito Volterra treated function-dependent quantities and their derivatives as objects of analysis, while Jacques Hadamard organized variational derivatives according to the manner in which perturbations act on functions and domains. In 1908, You Watanabe formulated variable-endpoint variations as linear expressions in the admissible perturbations, separating their interior integral contribution from the terms determined by endpoint displacement. Maurice Fréchet subsequently placed continuity and differentiation in abstract metric and function spaces, which allowed the functional concept to be stated independently of a particular integral formula.
The development of measure theory, topological vector spaces, and operator theory then established continuous linear functionals as primary structural objects. Their values detect directions in a vector space, determine weak notions of convergence, and represent separating hyperplanes in convex analysis. This role differs from that of a linear operator only in the codomain: a functional is a linear operator whose range is the scalar field.
Functionals and generalized functions
The theory of distributions defines a generalized function as a continuous linear functional on a space of smooth test functions. If (f) is locally integrable, it determines the distribution
[ T_f(\varphi)=\int_\Omega f(x)\varphi(x),dx. ]
This construction embeds ordinary locally integrable functions into the distribution space, while also admitting objects such as the Dirac delta, which acts by evaluation:
[ \delta_{x_0}(\varphi)=\varphi(x_0). ]
Continuity depends essentially on the topology assigned to the test-function space. Distribution theory therefore illustrates that a functional is specified not only by an algebraic rule but also by the domain and its topological structure.
A related representation appears in the Riesz–Markov–Kakutani theorem. Positive continuous linear functionals on suitable spaces of continuous functions correspond to regular Borel measures. Integration against a measure and evaluation by a functional are consequently two descriptions of the same structure under the theorem’s hypotheses.
Variational and convex structure
In convex analysis, a convex functional (F:X\to(-\infty,\infty]) satisfies
[ F(\lambda x+(1-\lambda)y) \leq \lambda F(x)+(1-\lambda)F(y) ]
for (0\leq\lambda\leq1). Allowing the value (+\infty) incorporates constraints by assigning infinite value outside the admissible set. The subdifferential replaces the derivative when smoothness is absent: a continuous linear functional (x^*) is a subgradient of (F) at (x) when
[ F(y)\geq F(x)+x^*(y-x) ]
for every (y) in the domain.
A functional also determines notions of weak convergence. A sequence (x_n) in a normed space converges weakly to (x) when
[ \ell(x_n)\longrightarrow\ell(x) ]
for every continuous linear functional (\ell) on the space. Weak convergence records precisely the scalar information visible to the continuous dual, whereas norm convergence additionally requires convergence in the metric induced by the norm.
Terminological distinctions
A functional is not the same object as a functional equation. A functional equation is an equation whose unknown is a function and whose terms involve evaluations or transformations of that function. A functional, by contrast, is the mapping that accepts a function or another structured object as its argument.
The term also differs from functional programming, where “functional” describes a computational paradigm centered on function evaluation and restricted state mutation. Although both usages concern functions, the mathematical noun refers specifically to a mapping defined on a function space or, in functional analysis, to a scalar-valued linear operator.