It seems you’re interested in the rather convoluted history of how humanity decided to wrestle with the concept of a function when reality refused to conform to its neat definitions. Fine. Let’s delve into these “generalized functions,” as if adding a new adjective somehow makes the universe less messy.

Objects extending the notion of functions

In the grand, often frustrating, arena of mathematics , generalized functions emerge as necessary contrivances, extending the rather quaint notion of ordinary functions defined over real or complex numbers. It’s a testament to the persistent inadequacy of our initial conceptual tools, forcing us to invent more robust frameworks to describe a world that clearly wasn’t designed for our convenience. There isn’t just one monolithic theory, of course; that would be far too simple. Instead, we have a collection of approaches, with the theory of distributions standing as a particularly prominent example, a testament to the idea that if you can’t solve a problem directly, you just redefine the problem.

These generalized functions prove themselves especially useful when dealing with mathematical entities that stubbornly refuse to behave nicely. Consider the unruly nature of discontinuous functions , which classical calculus often struggles to handle with grace. Generalized functions allow us to treat these jagged, broken things with a semblance of the elegance usually reserved for smooth functions , making them amenable to the sophisticated tools of analysis. Beyond mere mathematical convenience, they are indispensable for describing discrete, often singular, physical phenomena. Imagine trying to model a point charge , a theoretical construct of infinite density at a single, infinitesimal point, using a conventional function – it’s like trying to catch smoke with a sieve. Generalized functions provide the mathematical scaffolding for such concepts, proving their worth extensively, particularly within the demanding fields of physics and engineering . The primary motivations for these developments weren’t born of idle curiosity, but rather from the relentless technical requirements imposed by advanced theories, notably in the realm of partial differential equations and the intricate structures of group representations . These fields simply demanded more from their mathematical tools, and generalized functions, with their expanded expressive power, reluctantly obliged.

A common, almost predictable, feature uniting several of these disparate approaches is their shift in focus. Rather than obsessing over the point-wise values of functions – a futile exercise for anything truly generalized – they build upon the operator aspects of everyday, numerical functions. This essentially means moving from asking “what is its value here?” to “what does it do to other things?” It’s a subtle but profound reorientation, acknowledging that the interaction, the effect, is often more fundamental than the isolated existence. The early history of this conceptual evolution is inextricably connected with some rather ambitious, and at times mathematically dubious, ideas on operational calculus . Fast forward, and some contemporary developments find themselves closely related to Mikio Sato ’s rather abstract, yet elegant, framework of algebraic analysis , demonstrating how even the most pragmatic needs can eventually lead to highly abstract theoretical constructions.

Some early history

The seeds of generalized function theory were scattered throughout the mathematics of the nineteenth century, though they remained largely unrecognized as part of a coherent whole at the time. Like disparate whispers of a future revelation, aspects of these theories appeared in various guises. For instance, the very definition of the Green’s function , a fundamental tool for solving inhomogeneous differential equations, inherently hinted at singular behavior that defied standard functional descriptions. Similarly, the Laplace transform , which transmutes differential equations into algebraic ones, often required handling inputs and outputs that weren’t strictly “functions” in the classical sense. Then there was Riemann ’s groundbreaking work on the theory of trigonometric series . These series were not necessarily the well-behaved Fourier series of an integrable function but could represent functions with discontinuities or even more pathological behavior, forcing mathematicians to confront the limits of their definitions. At the time, these were merely disconnected aspects of mathematical analysis , intriguing anomalies rather than pieces of a grander, unifying puzzle. The intellectual landscape simply wasn’t ready to connect these dots, preferring the comfort of well-defined, continuous functions.

The intense, often impatient, use of the Laplace transform within engineering disciplines inadvertently spurred the development of these ideas. Engineers, driven by practical necessity rather than mathematical rigor, embraced what they termed operational calculus – a collection of heuristic symbolic methods. These techniques, while undeniably effective in solving complex problems, often relied on justifications that employed mathematically suspect divergent series . Unsurprisingly, from the fastidious perspective of pure mathematics , these methods were, to put it mildly, questionable. Yet, their very existence and utility foreshadowed the later applications of generalized function methods, demonstrating that sometimes, usefulness precedes formal justification. A profoundly influential book that championed this operational calculus, effectively bridging the gap between mathematical theory and practical application, was Oliver Heaviside ’s Electromagnetic Theory of 1899. Heaviside, a self-taught genius, pushed the boundaries of what was considered mathematically acceptable, much to the chagrin of his more orthodox contemporaries, but his methods undeniably worked.

A pivotal moment arrived with the introduction of the Lebesgue integral . For the first time, a truly central notion of generalized function began to crystallize within the heart of mathematics . Lebesgue’s theory offered a radical perspective: an integrable function, under his framework, is considered equivalent to any other function that is the same almost everywhere . This implies a profound shift in focus: the value of a function at each individual point is, in a profound sense, no longer its most important or defining feature. Instead, its behavior over sets of non-zero measure takes precedence. This conceptual leap paved the way for functional analysis , where a far clearer and more powerful formulation of the essential characteristic of an integrable function emerged. This essential feature is precisely the way it defines a linear functional on other functions, effectively mapping a function to a scalar value through integration. This abstract but powerful perspective not only clarified existing concepts but also allowed for a rigorous definition of the weak derivative , enabling differentiation even for functions that lacked classical differentiability, pushing the boundaries of what could be subjected to calculus.

The late 1920s and 1930s witnessed further decisive strides in this evolving understanding. It was during this period that Paul Dirac , with characteristic boldness and a flair for scientific formalism , defined the now-iconic Dirac delta function . This revolutionary concept allowed physicists to treat mathematical measures , such as charge density , as if they were genuine, albeit highly singular, functions. It was a pragmatic invention that worked wonders in quantum mechanics, even if mathematicians initially recoiled from its apparent lack of rigor. Simultaneously, and perhaps more rigorously, Sergei Sobolev , while immersed in the complex world of partial differential equation theory , developed the first truly rigorous theory of generalized functions. His objective was to provide a solid foundation for defining weak solutions of partial differential equations – solutions that might be generalized functions and therefore might not conform to the classical definition of an ordinary function. This was a critical step in expanding the solvability of many important equations. Others like Salomon Bochner and Kurt Friedrichs were proposing related, convergent theories around the same time, indicating a collective intellectual push towards these concepts. Sobolev’s pioneering work, a cornerstone in the field, was subsequently extended and refined by the brilliant Laurent Schwartz , laying the groundwork for what would become the dominant theory. 1 2

Schwartz distributions

The most definitive and widely adopted development in the realm of generalized functions arrived with the theory of distributions , meticulously crafted by Laurent Schwartz . Schwartz’s genius lay in systematically working out the principle of duality for topological vector spaces , transforming the somewhat ad-hoc notions into a coherent, rigorous mathematical framework. Instead of trying to define the value of a singular function at a point, Schwartz defined it by its action on a class of “test functions,” essentially characterizing it by its overall “behavior” rather than its problematic individual points. It was an elegant solution to a very stubborn problem.

This powerful theory, while dominant, wasn’t without its contenders, particularly in the more pragmatic domain of applied mathematics . Its main rival often presented itself in the form of mollifier theory. This alternative approach, often associated with the ‘James Lighthill’ explanation, eschews the abstract elegance of duality for a more intuitive, though perhaps less profound, method: using sequences of smooth approximations. Essentially, instead of directly defining the generalized function, one approximates it with a sequence of increasingly “spiky” smooth functions, and then takes a limit. It’s a bit like trying to define a knife edge by describing a sequence of progressively sharper butter knives. 3

While Schwartz’s theory was remarkably successful and remains widely used today, a significant drawback soon became apparent: distributions, unlike most classical function spaces , cannot generally be multiplied. This isn’t a minor inconvenience; it means they do not form an algebra in the conventional sense, severely limiting their utility in many practical applications, especially those involving non-linear operations. For instance, the seemingly straightforward operation of squaring the Dirac delta function becomes mathematically meaningless within this framework. This wasn’t an oversight or a temporary bug; work by Schwartz himself, around 1954, demonstrated this to be an intrinsic, fundamental difficulty, a stubborn refusal of the mathematical landscape to conform to simple expectations.

Algebras of generalized functions

Given the intrinsic difficulty of multiplying distributions, it was inevitable that various ingenious solutions would be proposed to overcome this limitation. After all, what’s a mathematical problem if not an invitation for more abstraction? One notable solution is based on a rather direct definition by Yu. V. Egorov. 4 This approach, detailed further in his article within Demidov’s collected works, bravely allows for arbitrary operations, including multiplication, on and between generalized functions, effectively trying to force them into an algebraic straitjacket.

Another intriguing solution that permits multiplication draws its inspiration from the esoteric world of quantum mechanics , specifically the path integral formulation . This formulation, a cornerstone of modern physics, is fundamentally required to be equivalent to the more traditional Schrödinger theory of quantum mechanics . Crucially, the Schrödinger theory possesses an inherent invariance under coordinate transformations. This invariance, a fundamental symmetry of nature, must therefore be shared by path integrals. This deep physical requirement, it turns out, is powerful enough to fix all products of generalized functions, providing a robust, physically motivated framework for their multiplication. This significant insight was demonstrated by H. Kleinert and A. Chervyakov. 5 The resulting mathematical framework, quite remarkably, proves to be equivalent to what can be derived using the technique of dimensional regularization , a method frequently employed to handle infinities in quantum field theories. 6

Beyond these, several other significant constructions of algebras of generalized functions have been put forth, reflecting the persistent need to address the multiplication problem. Among these, the proposals by Yu. M. Shirokov 7 and a collaborative effort by E. Rosinger, Y. Egorov, and R. Robinson stand out. citation needed In Shirokov’s approach, the definition of multiplication is intricately tied to a process of regularization applied to the generalized functions, essentially smoothing out their singularities before attempting to multiply them. In contrast, the second case constructs the algebra specifically as a framework for the multiplication of distributions themselves. Both of these cases, with their distinct methodologies, warrant further exploration and are indeed discussed in more detail below, because apparently, one solution is never enough.

Non-commutative algebra of generalized functions

One particularly interesting construction for an algebra of generalized functions involves a somewhat surgical procedure: the appropriate projection of a function, let’s call it $F = F(x)$, into two distinct components. You have its smooth part, denoted $F_{\rm {smooth}}$, which behaves predictably and politely, and its singular part, $F_{\rm {singular}}$, which embodies all the unruly, discontinuous, or infinitely sharp aspects. The product of two such generalized functions, $F$ and $G$, then emerges not as a simple point-wise multiplication, but through a specific, carefully defined combination of these parts:

$FG~=~F_{\rm {smooth}}G_{\rm {smooth}}+~F_{\rm {smooth}}G_{\rm {singular}}+F_{\rm {singular}}~G_{\rm {smooth}}.$ (1)

This rule, a rather explicit blueprint for multiplication, is not confined merely to the space of the main functions themselves; it extends its reach to the space of operators that act upon these functions, ensuring a consistent framework. Crucially, this construction successfully achieves the property of associativity for multiplication, a fundamental requirement for any respectable algebra. Furthermore, within this framework, the peculiar signum function (which is +1 for positive numbers, -1 for negative, and 0 at the origin) is defined in such a way that its square is unity everywhere, including the tricky origin of coordinates, effectively smoothing over a potential singularity.

It’s worth noting the absence of a term involving the product of the singular parts ($F_{\rm {singular}}~G_{\rm {singular}}$) on the right-hand side of equation (1). This omission leads to some rather counter-intuitive, yet mathematically consistent, results. For instance, in this algebra, the square of the Dirac delta function , $\delta(x)^2$, is precisely 0. This is a stark departure from classical intuition but a necessary consequence of the construction. Such a formalism is robust enough to include the conventional theory of generalized functions (the one without a defined product) as a special, more restricted case. However, there’s a significant trade-off for this newfound multiplicative capability: the resulting algebra is non-commutative. Specifically, the generalized functions signum and delta are found to anticommute, meaning their order of multiplication matters. 7 While the theoretical elegance of this algebra is clear, only a few applications have been suggested thus far, perhaps indicating the practical challenges of working with non-commutative structures. 8 9

Multiplication of distributions

The inherent problem of multiplication of distributions , a fundamental limitation that plagued Schwartz’s otherwise elegant distribution theory, becomes a particularly serious impediment when one ventures into the complex domain of non-linear problems. The universe, it seems, delights in making things complicated, and non-linearity often means that simple additive principles no longer hold, demanding a robust multiplicative structure.

Consequently, a variety of sophisticated approaches are employed today to circumvent this issue. The simplest among these, if “simple” can ever truly describe such mathematical endeavors, is rooted in the definition of generalized functions provided by Yu. V. Egorov. 4 This method offers a more direct pathway to defining products, striving for an intuitive extension of classical multiplication.

Another powerful and widely recognized approach to constructing associative differential algebras that accommodate multiplication is based on the seminal work of J.-F. Colombeau. This framework, often referred to as Colombeau algebra , constructs these algebras as factor spaces , typically represented as $G=M/N$. Here, $M$ represents a space of “moderate” nets of functions, while $N$ denotes a space of “negligible” nets. The crucial distinction between “moderateness” and “negligibility” lies in how these functions grow (or shrink) with respect to an indexing parameter of the family. Effectively, you’re classifying functions based on their asymptotic behavior, allowing you to discard what’s “too small” to matter in the grand scheme of things, much like sweeping dust under the rug of infinity.

Example: Colombeau algebra

To offer a glimpse into the mechanics of Colombeau algebra , consider a relatively straightforward example: utilizing the polynomial scale on the set of natural numbers, $\mathbb{N}$. This scale is defined as $s={a_{m}:\mathbb {N} \to \mathbb {R} ,n\mapsto n^{m};~m\in \mathbb {Z} }$, essentially a collection of power functions. Given any semi-normed algebra $(E,P)$, the resulting factor space is constructed as:

$G_{s}(E,P)={\frac {{f\in E^{\mathbb {N} }\mid \forall p\in P,\exists m\in \mathbb {Z} :p(f_{n})=o(n^{m})}}{{f\in E^{\mathbb {N} }\mid \forall p\in P,\forall m\in \mathbb {Z} :p(f_{n})=o(n^{m})}}}.$

This rather intimidating expression essentially divides the “moderate” sequences of functions (those whose semi-norm doesn’t grow faster than some polynomial in $n$) by the “negligible” sequences (those whose semi-norm grows slower than any polynomial in $n$).

The power of this construction becomes evident in specific instances. For example, if we take $(E,P) = (\mathbb{C},|\cdot|)$, where $\mathbb{C}$ represents the complex numbers and $|\cdot|$ is the absolute value, we obtain Colombeau’s generalized complex numbers. These are fascinating entities that can be “infinitely large” or “infinitesimally small” and yet still allow for rigorous arithmetic operations, offering a robust framework remarkably similar to the concept of nonstandard numbers . Furthermore, if we consider $(E,P) = (C^\infty(\mathbb{R}),{p_k})$, where $C^\infty(\mathbb{R})$ denotes the space of infinitely differentiable functions on the real line and $p_k$ represents the supremum of all derivatives up to order $k$ on a ball of radius $k$, we then arrive at Colombeau’s simplified algebra . This provides a concrete algebraic structure for generalized functions, capable of handling the multiplication that eluded earlier theories.

Injection of Schwartz distributions

One of the strengths of the Colombeau algebra is its ability to “contain” all Schwartz distributions $T$ from $D’$. This is achieved through a process called injection, defined as $j(T) = (\varphi_n \ast T)_n + N$. Here, the asterisk $\ast$ denotes the convolution operation, a fundamental tool in analysis that essentially “blends” two functions. The $\varphi_n(x)$ functions are a sequence of approximations, specifically $\varphi_n(x) = n \varphi(nx)$, derived from a base function $\varphi$.

However, this particular injection is noteworthy for being non-canonical. This means its specific form depends on the arbitrary choice of the mollifier $\varphi$. For this $\varphi$ to be valid, it must be infinitely differentiable ($C^\infty$), have an integral of one (ensuring it doesn’t add or subtract “mass”), and crucially, all its derivatives must vanish at the origin. The dependence on this arbitrary choice introduces a degree of non-uniqueness. To achieve a truly canonical injection, one that is independent of such arbitrary choices, the indexing set can be modified. Instead of simply $\mathbb{N}$, it can be expanded to $\mathbb{N} \times D(\mathbb{R})$, where $D(\mathbb{R})$ represents the space of smooth functions with compact support. This modification is then coupled with a carefully chosen filter base on $D(\mathbb{R})$, specifically functions that possess vanishing moments up to a certain order $q$. This more intricate construction removes the mollifier dependence, providing a more universally applicable embedding of Schwartz distributions into the Colombeau framework.

Sheaf structure

In the ever-expanding universe of mathematical abstraction, if $(E, P)$ constitutes a (pre-)sheaf of semi-normed algebras defined over some topological space $X$, then it logically follows that $G_s(E, P)$ — the Colombeau algebra constructed from it — will also inherit this fundamental property. This is not merely an elegant theoretical observation; it has profound practical implications. Specifically, it means that the crucial notion of restriction will be well-defined within this generalized function framework.

The ability to restrict a generalized function is vital, as it allows for the precise definition of the support of such a function with respect to a given subsheaf. This provides a nuanced way to understand where a generalized function is “active” or “non-zero.” To be more precise:

• For the trivial subsheaf {0}: When considering the simplest possible subsheaf, the one containing only the zero element, we recover the conventional, familiar definition of a function’s support. This is simply the complement of the largest open subset where the function is identically zero. It’s where the function actually exists in a non-trivial way.
• For the subsheaf $E$ (embedded using the canonical injection): When we embed the original algebra $E$ (for instance, $C^\infty$) into the generalized function algebra using the canonical (constant) injection, we obtain what is termed the singular support . Roughly speaking, this refers to the closure of the set of points where the generalized function fails to be a smooth function. For example, if $E = C^\infty$, the singular support pinpoints precisely where the generalized function exhibits its “non-smooth” or “singular” behavior, effectively highlighting the problematic areas that necessitated the generalized function theory in the first place. It’s where the function truly deviates from polite, classical behavior.

Microlocal analysis

With the Fourier transformation being (thankfully) well-defined for compactly supported generalized functions, operating on them component-wise, one gains access to a powerful analytical toolkit. This allows us to apply the very same sophisticated construction initially developed for distributions to these more general objects. Specifically, this means one can rigorously define Lars Hörmander ’s wave front set for generalized functions as well.

The wave front set is not some abstract mathematical curiosity; it’s a remarkably precise tool that provides information not just about where a generalized function is singular (its singular support), but also about the directions in which these singularities propagate. It’s like having a map that not only shows you where the cracks are but also how they’re spreading. This capability has an especially important and profound application in the detailed analysis of the propagation of singularities within solutions to partial differential equations. Understanding how these “rough edges” or abrupt changes move through a system is crucial in many areas of physics and engineering, making microlocal analysis an indispensable tool for taming the universe’s inherent messiness.

Other theories

Of course, the mathematical community, ever restless, has not confined itself to just a few approaches. The quest for the ultimate generalized function theory has led to several other notable frameworks. These include the rather elegant convolution quotient theory, pioneered by Jan Mikusinski . This theory is built upon the concept of the field of fractions derived from convolution algebras that exhibit the property of being integral domains . It’s a highly algebraic approach, turning complex analytical problems into more manageable algebraic ones.

Then there are the captivating theories of hyperfunctions . In their initial conception, these were based on the ingenious idea of representing generalized functions as boundary values of analytic functions , extending the domain of classical analysis into regions where functions might otherwise be ill-defined. More recently, these theories have evolved, now making extensive use of the sophisticated machinery of sheaf theory , demonstrating how different branches of mathematics often converge to provide deeper insights into similar problems. It’s a testament to the fact that there’s rarely just one way to skin a mathematical cat.

Topological groups

Venturing beyond the familiar terrain of Euclidean spaces and differentiable manifolds , Bruhat introduced a specialized class of test functions —the Schwartz–Bruhat functions . These functions are defined not just on typical function domains but on a broader class of locally compact groups . This expansion of the domain allows for the application of generalized function theory to structures far more intricate than simple geometric spaces.

The primary applications of this advanced framework are predominantly found in the often-abstract world of number theory , specifically in the study of adelic algebraic groups . These groups provide a powerful way to unify number-theoretic information across different valuations. A prime example of their utility is André Weil ’s seminal rewriting of Tate’s thesis using this very language. Weil’s work elegantly characterized the zeta distribution on the idele group , providing profound insights into number fields. He further applied these concepts to the explicit formula of an L-function , demonstrating how the seemingly disparate worlds of analysis and number theory can be intricately linked through generalized functions on topological groups. It’s a subtle but powerful unification, revealing hidden symmetries in the fabric of numbers.

Generalized section

A further, and perhaps even more abstract, extension of this theory involves conceptualizing generalized functions as generalized sections of a smooth vector bundle . This approach retains the fundamental ‘Schwartz pattern’ – constructing mathematical objects that are dual to a carefully chosen set of ’test objects.’ In this case, the test objects are smooth sections of a bundle that possess compact support , ensuring they behave nicely and don’t stretch off to infinity.

The most developed and widely recognized theory within this framework is that of De Rham currents . These currents are precisely the dual objects to differential forms . What makes them particularly compelling is their inherent homological nature, mirroring the way differential forms give rise to De Rham cohomology , a powerful tool for studying the topological properties of manifolds. This deep connection to topology and geometry allows De Rham currents to be used for formulating a remarkably general Stokes’ theorem . This generalized theorem extends the classical relationship between integrals over a region and integrals over its boundary to a far wider array of objects and spaces, providing a unifying principle that spans diverse areas of mathematics and physics. It’s a testament to the power of abstraction, revealing the underlying simplicity in seemingly complex phenomena.

See also

For those who simply can’t get enough of mathematical extensions and generalizations, consider exploring:

• Beppo-Levi space
• Dirac delta function
• Generalized eigenfunction
• Distribution (mathematics)
• Hyperfunction
• Laplacian of the indicator
• Rigged Hilbert space
• Limit of a distribution
• Generalized space
• Ultradistribution

Books

If you insist on further immersion into these dense topics, a selection of the foundational and more comprehensive texts includes:

• Schwartz, L. (1950). Théorie des distributions. Vol. 1. Paris: Hermann. OCLC 889264730. Vol. 2. OCLC 889391733
• Beurling, A. (1961). On quasianalyticity and general distributions (multigraphed lectures). Summer Institute, Stanford University. OCLC 679033904.
• Gelʹfand, Izrailʹ Moiseevič ; Vilenkin, Naum Jakovlevič (1964). Generalized Functions. Vol. I–VI. Academic Press. OCLC 728079644.
• Hörmander, L. (2015) [1990]. The Analysis of Linear Partial Differential Operators (2nd ed.). Springer. ISBN 978-3-642-61497-2.
• H. Komatsu, Introduction to the theory of distributions, Second edition, Iwanami Shoten, Tokyo, 1983.
• Colombeau, J.-F. (2000) [1983]. New Generalized Functions and Multiplication of Distributions. Elsevier. ISBN 978-0-08-087195-0.
• Vladimirov, V.S.; Drozhzhinov, Yu. N.; Zav’yalov, B.I. (2012) [1988]. Tauberian theorems for generalized functions. Springer. ISBN 978-94-009-2831-2.
• Oberguggenberger, M. (1992). Multiplication of distributions and applications to partial differential equations. Longman. ISBN 978-0-582-08733-0. OCLC 682138968.
• Morimoto, M. (1993). An introduction to Sato’s hyperfunctions. American Mathematical Society. ISBN 978-0-8218-8767-7.
• Demidov, A.S. (2001). Generalized Functions in Mathematical Physics: Main Ideas and Concepts. Nova Science. ISBN 9781560729051.
• Grosser, M.; Kunzinger, M.; Oberguggenberger, Michael; Steinbauer, R. (2013) [2001]. Geometric theory of generalized functions with applications to general relativity. Springer. ISBN 978-94-015-9845-3.
• Estrada, R.; Kanwal, R. (2012). A distributional approach to asymptotics. Theory and applications (2nd ed.). Birkhäuser Boston. ISBN 978-0-8176-8130-2.
• Vladimirov, V.S. (2002). Methods of the theory of generalized functions. Taylor & Francis. ISBN 978-0-415-27356-5.
• Kleinert, H. (2009). Path Integrals in Quantum Mechanics, Statistics, Polymer Physics, and Financial Markets (5th ed.). World Scientific. ISBN 9789814273572. (online here Archived 2008-06-15 at the Wayback Machine ). See Chapter 11 for products of generalized functions.
• Pilipovi, S.; Stankovic, B.; Vindas, J. (2012). Asymptotic behavior of generalized functions. World Scientific. ISBN 9789814366847.