Functions with nonelementary indefinite integrals Although the methods in standard calculus textbooks allow one to find the indefinite integrals (or antiderivatives) of many functions that arise in the subject, there are also many examples that cannot be handled using elementary techniques like change of variables, integration by parts, partial fraction expansions or trigonometric substitutions. One particularly noteworthy example is the function f (x) = exp ( ax 2) (where a is an arbitrary nonzero number), and the purpose of this document is to provide some further information and references concerning this fact. The meaning of mathematical impossibility Everyone knows of examples where things previously thought to be impossible were eventually done, and it is natural to challenge such an impossibility assertion; even though no one has found a way to write the indefinite integral of exp ( ax 2) in terms of the usual elementary functions in calculus, it may seem plausible that someone will eventually be able to do so. However, the question here is much narrower than asking whether someone can come up with a method for evaluating an integral whose integrand is equal to such a function; we want to do so using a function of a specific type. To illustrate this, think about the following statement: It is mathematically impossible to express the indefinite integral of the cosine function as a polynomial function. In the preceding statement, the problem is not that no one has been clever enough to find the right polynomial, but rather that no such polynomial can exist. For suppose that one could be found. Then its derivative would be the cosine function, but it would also be a polynomial. We know that the 4th, 8th, 12th, etc. derivatives of the cosine function are equal to the cosine function itself, and thus infinitely many higher order derivatives are nonzero. In contrast, only finitely many higher derivatives of a polynomial are zero. Thus when we say that something of this sort is mathematically impossible, it does not reflect a sense of defeatism, but rather a convincing demonstration that any potential solution would lead to conclusions that are logically inconsistent. The normal distribution properties of exp ( – ax 2) One reason for interest in the function described above is its relevance to probability theory and statistics. The function exp ( – ax 2) is positive valued for all real numbers x, and if a is positive it is fairly easy to prove that the improper integral of this function from 0 to ∞ converges by a comparison with exp ( – ax), whose improper integral over the same values is easily checked to converge; more precisely, we have the inequality exp ( – ax 2) < exp ( – ax) if x > 1, and the integral of the latter from 0 to ∞ equals 1/a.In fact, one can evaluate the improper integral of exp ( – ax 2) explicitly using methods from multivariable calculus: The appearance of π in this formula suggests an unexpected significance of π for probability theory that we shall not try to explain here. It follows immediately that if we modify the original function to obtain then the integral of the latter from – ∞ to + ∞ is equal to 1. If we graph this function we obtain the following bell shaped curve which is fundamentally important in probability and statistics. (source: http://www.tushar-mehta.com/excel/charts/normal_distribution/ ) The proof of mathematical impossibility The first rigorous mathematical proof that the integral of exp ( – ax 2) is not an elementary function was due to J. Liouville (1809 – 1882) and published during the late 1830s; biographical information on Liouville is available at the following online site:http://www-gap.dcs.st-and.ac.uk/~history/Mathematicians/Liouville.html A precise description of Liouville’s proof is beyond the scope of a first year calculus course, but a relatively elementary and accessible account appears in the following expository paper: M. Rosenlicht, Integration in finite terms. American Mathematical Monthly 79 (1972), 963 – 972. More general discussions (without proofs) of functions with nonelementary integrals appear in the following online sites: http://mathworld.wolfram.com/ElementaryFunction.html http://www.sosmath.com/calculus/integration/fant/fant.html http://en.wikipedia.org/wiki/Nonelementary_integral http://www.math.unt.edu/integration_bee/AwfulTruth.html The main results on functions with nonelementary integrals were obtained during the 19th century beginning with Liouville’s results. About 40 years ago such questions were reconsidered from an algorithmic perspective, and the question of whether a function has an elementary integral was completely solved. The following online sites discuss these results without going into extensive detail: http://mathworld.wolfram.com/RischAlgorithm.html http://en.wikipedia.org/wiki/Risch_algorithm http://www.freesoft.org/Classes/Risch2006/ Final remark. Although the integral of exp ( – ax 2) is not an elementary function, one can write down a Maclaurin series expansion for this integral fairly easily starting with the usual power series expansion for the exponential function. Specifically, aside from the constant of integration this power series expansion only involves odd powers of x, and the coefficient of x2n + 1 is given as follows: (– 1) n a n_n! (2n + 1) If  x  is reasonably small, this series converges quite rapidly; in fact, by the basic results on alternating series, the error in approximating the integral by the Maclaurin polynomial of degree 2n + 1 has an absolute value less than or equal to the absolute value of the term of degree 2n + 3