In mathematics, **generalized functions** are objects extending the notion of functions. There is more than one recognized theory, for example the theory of distributions. Generalized functions are especially useful in making discontinuous functions more like smooth functions, and describing discrete physical phenomena such as point charges. They are applied extensively, especially in physics and engineering.

A common feature of some of the approaches is that they build on operator aspects of everyday, numerical functions. The early history is connected with some ideas on operational calculus, and more contemporary developments in certain directions are closely related to ideas of Mikio Sato, on what he calls algebraic analysis. Important influences on the subject have been the technical requirements of theories of partial differential equations, and group representation theory.

In the mathematics of the nineteenth century, aspects of generalized function theory appeared, for example in the definition of the Green's function, in the Laplace transform, and in Riemann's theory of trigonometric series, which were not necessarily the Fourier series of an integrable function. These were disconnected aspects of mathematical analysis at the time.

The intensive use of the Laplace transform in engineering led to the heuristic use of symbolic methods, called operational calculus. Since justifications were given that used divergent series, these methods had a bad reputation from the point of view of pure mathematics. They are typical of later application of generalized function methods. An influential book on operational calculus was Oliver Heaviside's *Electromagnetic Theory* of 1899.

When the Lebesgue integral was introduced, there was for the first time a notion of generalized function central to mathematics. An integrable function, in Lebesgue's theory, is equivalent to any other which is the same almost everywhere. That means its value at a given point is (in a sense) not its most important feature. In functional analysis a clear formulation is given of the *essential* feature of an integrable function, namely the way it defines a linear functional on other functions. This allows a definition of weak derivative.

During the late 1920s and 1930s further steps were taken, basic to future work. The Dirac delta function was boldly defined by Paul Dirac (an aspect of his scientific formalism); this was to treat measures, thought of as densities (such as charge density) like genuine functions. Sergei Sobolev, working in partial differential equation theory, defined the first adequate theory of generalized functions, from the mathematical point of view, in order to work with weak solutions of partial differential equations.^{[1]} Others proposing related theories at the time were Salomon Bochner and Kurt Friedrichs. Sobolev's work was further developed in an extended form by Laurent Schwartz.^{[2]}

The realization of such a concept that was to become accepted as definitive, for many purposes, was the theory of distributions, developed by Laurent Schwartz. It can be called a principled theory, based on duality theory for topological vector spaces. Its main rival, in applied mathematics, is to use sequences of smooth approximations (the 'James Lighthill' explanation), which is more *ad hoc*. This now enters the theory as mollifier theory.^{[3]}

This theory was very successful and is still widely used, but suffers from the main drawback that it allows only linear operations. In other words, distributions cannot be multiplied (except for very special cases): unlike most classical function spaces, they are not an algebra. For example, it is not meaningful to square the Dirac delta function. Work of Schwartz from around 1954 showed that was an intrinsic difficulty.

Some solutions to the multiplication problem have been proposed. One is based on a very simple and intuitive definition a generalized function given by Yu. V. Egorov^{[4]} (see also his article in Demidov's book in the book list below) that allows arbitrary operations on, and between, generalized functions.

Another solution of the multiplication problem is dictated by the path integral formulation of quantum mechanics.
Since this is required to be equivalent to the Schrödinger theory of quantum mechanics which is invariant under coordinate transformations, this property must be shared by path integrals. This fixes all products of generalized functions
as shown by H. Kleinert and A. Chervyakov.^{[5]} The result is equivalent to what can be derived from
dimensional regularization.^{[6]}

Several constructions of algebras of generalized functions have been proposed, among others those by Yu. M. Shirokov
^{[7]} and those by E. Rosinger, Y. Egorov, and R. Robinson.^{[citation needed]}
In the first case, the multiplication is determined with some regularization of generalized function. In the second case, the algebra is constructed as *multiplication of distributions*. Both cases are discussed below.

The algebra of generalized functions can be built-up with an appropriate procedure of projection of a function to its smooth and its singular parts. The product of generalized functions and appears as

(1) |

Such a rule applies to both the space of main functions and the space of operators which act on the space of the main functions.
The associativity of multiplication is achieved; and the function signum is defined in such a way, that its square is unity everywhere (including the origin of coordinates). Note that the product of singular parts does not appear in the right-hand side of (**1**); in particular, . Such a formalism includes the conventional theory of generalized functions (without their product) as a special case. However, the resulting algebra is non-commutative: generalized functions signum and delta anticommute.^{[7]} Few applications of the algebra were suggested.^{[8]}^{[9]}

The problem of *multiplication of distributions*, a limitation of the Schwartz distribution theory, becomes serious for non-linear problems.

Various approaches are used today. The simplest one is based on the definition of generalized function given by Yu. V. Egorov.^{[4]} Another approach to construct associative differential algebras is based on J.-F. Colombeau's construction: see Colombeau algebra. These are factor spaces

of "moderate" modulo "negligible" nets of functions, where "moderateness" and "negligibility" refers to growth with respect to the index of the family.

A simple example is obtained by using the polynomial scale on **N**,
. Then for any semi normed algebra (E,P), the factor space will be

In particular, for (*E*, *P*)=(**C**,|.|) one gets (Colombeau's) generalized complex numbers (which can be "infinitely large" and "infinitesimally small" and still allow for rigorous arithmetics, very similar to nonstandard numbers). For (*E*, *P*) = (*C ^{∞}*(

This algebra "contains" all distributions *T* of * D' * via the injection

*j*(*T*) = (φ_{n}∗*T*)_{n}+*N*,

where ∗ is the convolution operation, and

- φ
_{n}(*x*) =*n*φ(*nx*).

This injection is *non-canonical *in the sense that it depends on the choice of the mollifier φ, which should be *C ^{∞}*, of integral one and have all its derivatives at 0 vanishing. To obtain a canonical injection, the indexing set can be modified to be

If (*E*,*P*) is a (pre-)sheaf of semi normed algebras on some topological space *X*, then *G _{s}*(

- For the subsheaf {0}, one gets the usual support (complement of the largest open subset where the function is zero).
- For the subsheaf
*E*(embedded using the canonical (constant) injection), one gets what is called the singular support, i.e., roughly speaking, the closure of the set where the generalized function is not a smooth function (for*E*=*C*^{∞}).

The Fourier transformation being (well-)defined for compactly supported generalized functions (component-wise), one can apply the same construction as for distributions, and define Lars Hörmander's *wave front set* also for generalized functions.

This has an especially important application in the analysis of propagation of singularities.

These include: the *convolution quotient* theory of Jan Mikusinski, based on the field of fractions of convolution algebras that are integral domains; and the theories of hyperfunctions, based (in their initial conception) on boundary values of analytic functions, and now making use of sheaf theory.

Bruhat introduced a class of test functions, the Schwartz–Bruhat functions as they are now known, on a class of locally compact groups that goes beyond the manifolds that are the typical function domains. The applications are mostly in number theory, particularly to adelic algebraic groups. André Weil rewrote Tate's thesis in this language, characterizing the zeta distribution on the idele group; and has also applied it to the explicit formula of an L-function.

A further way in which the theory has been extended is as **generalized sections** of a smooth vector bundle. This is on the Schwartz pattern, constructing objects dual to the test objects, smooth sections of a bundle that have compact support. The most developed theory is that of De Rham currents, dual to differential forms. These are homological in nature, in the way that differential forms give rise to De Rham cohomology. They can be used to formulate a very general Stokes' theorem.