An elementary function is a function obtainable from a specified collection of basic functions through finitely many arithmetic operations and compositions. The collection includes polynomials, rational functions, powers, the exponential and logarithm, and trigonometric functions together with their appropriate inverses. In the broader definition used in differential algebra, algebraic functions are also admitted. Familiar examples include , , , and .
Elementary functions supply much of the working vocabulary of calculus, mathematical modeling, and scientific computation. Their importance comes partly from their explicit construction and partly from their differentiation rules: wherever the relevant derivatives exist, differentiating an elementary expression produces another elementary expression. Integration behaves differently. Simple elementary functions, including and , have no elementary antiderivative.
The word elementary identifies a class of representations, not a level of difficulty. An elementary formula can have intricate behavior, require substantial numerical work, or conceal delicate domain restrictions. Conversely, a function outside this class may have a straightforward definition, an efficient numerical algorithm, and a central role in applications.
Definition and conventions
The central requirement in the definition is finite construction. Starting with the independent variable and permitted constants, one forms new expressions by addition, subtraction, multiplication, division where defined, and composition. Applying an exponential, logarithm, or permitted algebraic operation produces further expressions. Any expression obtained after finitely many such steps represents an elementary function on a suitable domain. For example,
is elementary wherever the chosen real or complex interpretations make its components meaningful and its denominator nonzero.
This definition is relative to a choice of constants. In ordinary real-variable calculus, arbitrary real constants are generally permitted; in complex analysis, complex constants are permitted. In symbolic integration, the constant field is specified more carefully because a computational procedure must represent its elements and decide certain equalities involving them. Whether a coefficient has an elementary description as a number is a separate question from whether an expression involving that coefficient is elementary as a function of .
There are two conventions concerning algebraic functions. Many introductory treatments describe elementary expressions using arithmetic, roots, exponentials, logarithms, and trigonometric functions. Such a definition directly includes functions expressible by radicals. In differential algebra, elementary extensions generally allow arbitrary algebraic extensions, including branches of algebraic equations that cannot be solved by radicals. These conventions coincide for most standard calculus examples but are not identical. A local branch of the inverse of , for example, is algebraic; it does not generally admit a radical formula. The broader convention includes it as elementary.
A finite expression must also be distinguished from an infinite process. An infinite series, an integral definition, or the solution of a differential equation does not automatically define an elementary function. Some such definitions happen to yield elementary results:
Others do not. The distinction depends on whether the resulting function admits a finite representation of the permitted kind, not on the notation in which it was first introduced.
Finally, an elementary expression is not identical to the function it represents. Different expressions can agree on one domain and disagree elsewhere because of branches or omitted points. The expressions and , for instance, agree for , but the first displayed expression is undefined at . Elementary-function classification therefore concerns expressions together with their interpretations, rather than formulas stripped of domain information.
Principal families and their relationships
The most basic elementary functions are polynomials and rational functions. A polynomial has the form
with finitely many coefficients and nonnegative integer exponents. A rational function is a quotient , with not identically zero. These functions require only arithmetic operations. Their zeros, poles, factorization, and behavior at infinity connect elementary-function theory directly with algebra.
An algebraic function satisfies a polynomial equation involving its argument and value:
Examples include , , and local solutions of equations such as . A single polynomial equation can describe several branches rather than one globally single-valued function. Radical expressions are algebraic, but not every algebraic function is expressible through radicals. The distinction is explained by Galois theory, which studies the algebraic symmetries governing solvability.
The exponential function is characterized by its derivative and normalization, or equivalently by its convergent power series. The natural logarithm is its real inverse on positive arguments. Exponentials with other positive bases are obtained from
Likewise, logarithms to another positive base satisfy . Variable powers such as , interpreted for , are elementary because . This does not imply that exponentiation has an unambiguous global meaning over all real or complex arguments.
The trigonometric functions and their inverses constitute another familiar family. Over the complex numbers, sine and cosine can be expressed using exponentials:
Tangent, cotangent, secant, and cosecant are quotients or reciprocals of these functions. Hyperbolic functions are similarly given by
Inverse trigonometric and inverse hyperbolic functions can locally be written with logarithms and square roots.
Thus the usual list of elementary families is redundant when complex constants and branches are available: exponentials, logarithms, and algebraic functions suffice. The longer list remains useful in real-variable work because it preserves geometric meaning, real domains, and convenient identities. An oscillation is more naturally described by than by a quotient of complex exponentials, even though the two representations are mathematically equivalent.
Real domains, branches, and piecewise definitions
An elementary formula defines a real-valued function only where every operation in the formula is valid. The domain is therefore part of its mathematical meaning. For real expressions, denominators must be nonzero, even roots require nonnegative radicands, and logarithms require positive arguments. Conditions introduced at different stages of a composition must hold simultaneously.
For example,
has real domain . The logarithm requires , while the square root in the denominator requires , not merely . Determining this domain is an intersection problem, not a property that can be read from the outermost operation alone. A different example, , is defined on two disjoint intervals, and .
Algebraic and inverse functions require branch choices. The relation determines two real values when ; the symbol conventionally means the nonnegative one. Similarly, sine is not one-to-one on the entire real line. Its usual inverse, , is obtained by restricting sine to . Consequently,
whereas holds only on the selected interval. Outside it, the composition folds values back into the principal range.
Elementary does not mean differentiable at every point of the real domain. The absolute value function is expressible as
but is not differentiable at zero. The function is continuous on the real line but has no finite derivative at zero. Such examples occur where algebraic branches meet or become singular. On intervals where all component operations are analytic and all relevant denominators remain nonzero, their composition is analytic.
Piecewise definitions require a separate distinction. A function with elementary formulas on each piece is often called piecewise elementary, but arbitrary case distinctions are not among the operations in the basic definition. Some finite piecewise constructions nevertheless reduce to elementary formulas. For instance,
By contrast, floor and ceiling functions are not standard elementary functions. Their infinitely repeated jumps also distinguish them from the analytic local behavior central to elementary-function theory. Thus “given by a familiar formula” and “elementary” are not interchangeable descriptions.
Complex interpretation and analytic continuation
Working over the complex numbers unifies elementary families but makes branch structure unavoidable. The complex exponential is single-valued and entire, with period :
It is therefore not globally one-to-one. Any logarithm satisfying must accommodate the values
There is no continuous single-valued logarithm on the whole punctured complex plane.
A branch of the logarithm is a single-valued analytic choice on a suitable region. A standard principal branch takes the argument in and is analytic on the plane with the nonpositive real axis removed. Other cuts are possible. More generally, on a simply connected open region that excludes zero, an analytic logarithm exists. The obstruction on the punctured plane is topological: going once around zero changes a continued logarithm by .
This affects familiar identities. The equation
need not hold for independently chosen principal values, although the two sides differ by an integer multiple of when defined. Likewise, complex powers are typically interpreted locally by
so their values depend on the logarithm branch. When is an integer, this ambiguity disappears; for rational noninteger exponents, finitely many values occur; for general exponents, the branch behavior can be more extensive.
Algebraic functions have their own branching. Analytically continuing a square root around zero changes its sign. The language of analytic continuation and Riemann surfaces treats such behavior systematically: a multivalued expression on the plane can become a single-valued analytic function on a suitable surface. Elementary-function theory in complex analysis is consequently often formulated locally, using analytic germs or chosen branches rather than a single global formula on all of .
Branch points, poles, and essential singularities must also be distinguished. A rational function can have poles; a square root can have branch points; and has an essential singularity at zero despite being elementary. Thus elementary functions are not limited to mild singular behavior. Conversely, a function can be entire without being elementary. The error function, defined by an integral of , is a standard example. Analyticity describes local regularity, whereas elementary representability describes construction from a specified set of operations.
Differentiation, local behavior, and limits
The elementary class is closed under differentiation in its appropriate local sense. Starting from the usual derivatives of powers, exponentials, logarithms, and trigonometric functions, the product rule, quotient rule, and chain rule produce elementary expressions for derivatives. For example, if ,
The formula illustrates why logarithmic representation is useful even when the original expression uses only a power notation.
Algebraic functions obey the same principle through implicit differentiation. If satisfies and , then
The right-hand side is algebraic in and . At a point where , this formula no longer establishes the existence of a finite derivative; the branch may require separate analysis. This qualification explains how derivative closure is compatible with examples such as at zero.
On a region where an elementary function is analytic, it has a local Taylor series. Standard expansions include
The expansion is an analytical description of the function, not the finite construction that makes it elementary. A convergent power series may represent either an elementary or a nonelementary function.
Elementary functions also provide standard scales for comparing growth. For fixed positive and ,
and for fixed real ,
Repeated exponentials create still faster scales. Such comparisons are important in analysis, asymptotic estimation, and algorithmic complexity. They depend on limits and growth rates rather than on elementary status alone.
There is no general closure under limits of sequences of elementary functions. Polynomial partial sums of a convergent power series are elementary, but their limit need not be. For example, the series
converges for every real and equals , which is not elementary. Even uniform convergence on compact intervals does not change that conclusion. Finite symbolic construction and convergence in a function space are fundamentally different kinds of closure.
Integration and the boundary of elementary expressions
Unlike differentiation, integration does not preserve elementary functions. Every continuous real function has local antiderivatives by the fundamental theorem of calculus, but those antiderivatives need not lie in the elementary class. Thus existence of an antiderivative is an analytical question, while elementary expressibility is an additional algebraic constraint.
Rational functions form a particularly favorable case. Over the complex numbers, partial-fraction decomposition reduces their integration to polynomial terms, powers of linear factors, and logarithms. Over the reals, irreducible quadratic factors introduce inverse tangents. Consequently every rational function has an elementary antiderivative, although factoring the denominator and expressing its roots can introduce algebraic constants. A representative identity is
Under the broad algebraic convention, no radical solvability assumption is needed for the roots arising in a general denominator.
Many expressions involving exponentials or trigonometric functions are also elementary integrable. For example,
on . The first works because the factor matches the derivative of the exponent up to a constant; the second follows from integration by parts. These mechanisms explain particular successes but do not supply a universal method for finding elementary primitives.
Classical counterexamples include
None has an elementary antiderivative on an interval where the corresponding integrand is defined. Their primitives lead to the error function, logarithmic integral, and sine integral. The first is essential in the normal distribution: Gaussian probabilities involve an integral whose exact functional dependence cannot be written using elementary functions alone. Algebraic integrands also need not have elementary primitives; many lead instead to elliptic integrals.
A definite integral must be distinguished from an indefinite one. Although has no elementary primitive,
A special definite value can simplify through symmetry or a multidimensional argument even when the general antiderivative is nonelementary. Conversely, the appearance of an unfamiliar constant in a definite integral is not by itself proof that its integrand lacks an elementary primitive. Classification of functions, exact evaluation of individual integrals, and classification of constants are related but separate problems.
Differential algebra and symbolic decision methods
Differential algebra makes elementary integration precise by combining a field with a derivation. A differential field has an operation satisfying
The basic example is the field of rational functions , with . Its constants are the elements for which . An elementary extension is built by successively adjoining algebraic elements, logarithms, or exponentials, subject to the corresponding differentiation relations.
For example, adjoining gives , while adjoining gives . These relations retain the information needed for differentiation without requiring a globally single-valued complex formula. Algebraic extensions are handled by differentiating their defining polynomial equations. This formalism separates the algebraic structure of expressions from questions about particular real intervals or branch cuts.
Liouville’s theorem on elementary integration provides the central structural restriction. In a standard formulation, under appropriate characteristic-zero and constant-field assumptions, if has an elementary primitive in an extension introducing no new constants, then
where the are constants and the elements can be taken in a suitable finite algebraic extension of . Integrating yields an algebraic-field term together with a finite sum of constant multiples of logarithms.
This theorem is not merely a recipe for guessing substitutions. It limits the possible form of every elementary answer. For example, the analysis of a prospective primitive of leads, in the relevant exponential-extension setting, to a rational-function equation of the form
A finite pole of a rational solution would produce an uncancelled higher-order pole through differentiation, so such a solution would have to be polynomial. Polynomial degree comparison then rules it out: for nonzero polynomial , the term has greater degree than and cannot yield the constant . Differential-algebraic reduction supplies the justification for using this obstruction to prove nonelementarity.
The Risch algorithm, developed by Robert H. Risch in work published in 1969 and 1970, turns this structural approach into decision procedures for elementary integration under specified effective assumptions. It reduces integration to problems involving polynomial arithmetic, differential equations, algebraic extensions, and constants. The theoretical achievement does not mean that every computer algebra system implements every case. Algebraic-function integration and recognition of constants remain substantial implementation issues, and an unevaluated output from a particular program is not itself a proof of nonelementarity.
Numerical evaluation and computational representation
An elementary expression is an exact symbolic description, not a guarantee of easy or stable numerical evaluation. Scientific libraries evaluate exponentials, logarithms, and trigonometric functions using combinations of range reduction, polynomial or rational approximation, table lookup, and reconstruction. Direct summation of a defining series is often inefficient or inaccurate when its argument is large or when intermediate terms nearly cancel.
For an exponential, a typical strategy writes
with confined to a small interval. An approximation for can then be reused over the full supported range. Trigonometric evaluation similarly reduces an argument modulo a fraction of a full period. Accurate reduction becomes demanding for very large arguments because the small remainder can depend on many digits of the constant . These are problems in numerical analysis, not questions about whether sine or the exponential is elementary.
Equivalent elementary formulas can have very different numerical stability. Near , direct computation of loses relative accuracy through subtraction of nearly equal numbers. Libraries therefore provide a dedicated expm1 function. Likewise, log1p evaluates accurately for small . For large positive , the apparently straightforward formula
can overflow before the logarithm is taken. A stable real identity is
The identity changes the evaluation path without changing the exact function.
Floating-point arithmetic also separates a mathematical operation from its machine implementation. The exact result usually has to be rounded, and intermediate overflow, underflow, or cancellation can affect the answer. Correctly rounded transcendental functions are difficult to implement because the exact value may lie extremely close to a rounding boundary. Being elementary gives no automatic guarantee of low cost or a fixed accuracy.
Symbolic computation faces different representation problems. Simplification must preserve domain and branch conditions: is valid for nonnegative real , while is the appropriate real identity without that restriction. The logarithm of a product cannot be expanded unconditionally over complex arguments. Meanwhile, automatic differentiation uses the finite computational structure of elementary expressions to propagate derivatives by local rules. Its effectiveness follows from composition and derivative closure, but numerical conditioning and nondifferentiable branch points still require separate treatment.
Historical development and the wider function landscape
Elementary functions emerged from several older traditions rather than from a single definition. Polynomial and rational expressions developed within algebra; trigonometric tables arose from geometry and astronomy; logarithms were introduced as computational instruments. John Napier published his logarithmic work in 1614, and Henry Briggs subsequently developed common logarithms. These early constructions were not presented as members of a formally closed class of functions. Their unification came through analytic notation, calculus, and the study of functional relationships.
The seventeenth-century development of calculus by Isaac Newton and Gottfried Wilhelm Leibniz made differentiation and integration central organizing operations. The contrast between their behavior gradually became apparent: differentiation followed systematic rules, while integration repeatedly produced expressions beyond the established repertoire. Leonhard Euler’s Introductio in analysin infinitorum of 1748 helped consolidate the analytic treatment of exponential, logarithmic, and trigonometric functions. Complex identities, especially those relating exponentials to trigonometry, showed that apparently separate families belonged to a unified framework.
The nineteenth century sharpened the limits of this framework. Joseph Liouville’s work in the 1830s established structural results about elementary antiderivatives. Investigations of elliptic integrals by Adrien-Marie Legendre, Niels Henrik Abel, and Carl Gustav Jacob Jacobi demonstrated that functions outside the elementary class had rich theories of their own. The modern special functions tradition grew from such problems: a nonelementary function could be studied through differential equations, recurrence relations, integral representations, and convergent or asymptotic expansions.
Several distinctions remain important in this larger landscape. An inverse of an elementary function need not be elementary. The Lambert function, defined by
inverts the elementary map but is not itself elementary. Likewise, solving an ordinary differential equation with elementary coefficients need not produce elementary solutions. Differential equations generate a much broader world that includes Bessel functions, Airy functions, and many other functions indispensable in mathematical physics.
Applications therefore use elementary functions as a basic vocabulary rather than a universal endpoint. Exponentials model constant proportional growth and decay; logarithms convert multiplicative relationships into additive ones; trigonometric functions describe periodic motion. Gaussian models introduce nonelementary cumulative probabilities, and geometric arc lengths can introduce elliptic integrals. The elementary class remains especially useful because its construction is explicit and its derivatives remain within the class, while the wider function landscape supplies the objects required when finite combinations of those basic operations no longer suffice.