Loading field map
How can one prove that a specific number—say π or e—is not the root of any polynomial equation with integer coefficients? The difficulty is that the boundary between algebraic and transcendental numbers is invisible to elementary arithmetic. A number may satisfy no simple algebraic relation, yet still be algebraic; conversely, a number that looks complicated may turn out to be algebraic. Proving transcendence therefore requires methods that reach beyond the algebraic toolkit itself. The history of transcendental number theory is a history of inventing fundamentally different proof architectures—each designed to overcome the limitations of its predecessors—and of discovering that several of those architectures remain indispensable today.
The first successful strategy came from Joseph Liouville in 1844. Liouville observed that algebraic numbers cannot be approximated too well by rational numbers: if α is algebraic of degree d, then |α – p/q| > C/q^d for some constant C depending on α. By constructing numbers that violate this bound—for example, the sum ∑ 10^{-k!}—Liouville produced the first explicit transcendental numbers. The method was ingenious but narrow: it could only reach numbers deliberately engineered to be well-approximable, not the naturally occurring constants that mathematicians most wanted to classify.
Charles Hermite broke through this limitation in 1873 with a radically different approach for e. Instead of approximation, Hermite built an auxiliary function—a carefully chosen linear combination of exponentials and polynomials—and used its analytic properties to force a contradiction if e were algebraic. The technique was intricate, but it proved that e is transcendental. Ferdinand Lindemann extended Hermite's method in 1882 to handle π, showing that if π were algebraic then e^{iπ} = –1 would contradict Hermite's result. Lindemann's proof also settled the ancient problem of squaring the circle: because π is transcendental, the circle cannot be squared with straightedge and compass. Yet the Hermite–Lindemann approach remained tied to the exponential function; it gave no general criterion for deciding transcendence of numbers of the form a^b.
The exponential gap was closed in 1934, independently by Alexander Gelfond and Theodor Schneider. They proved that if a is algebraic (a ≠ 0,1) and b is algebraic and irrational, then a^b is transcendental. The result settled Hilbert's seventh problem and provided the first systematic transcendence criterion for a wide class of numbers. The proof architecture again relied on an auxiliary function, but now the function was built from interpolation: one constructed an entire function that vanished at many integer combinations of logarithms and then used growth estimates to force a contradiction. The Gelfond–Schneider method was powerful, but it had two limitations. First, it was ineffective: it proved existence of a lower bound for linear forms in logarithms without giving a way to compute that bound. Second, it handled only a single exponential relation at a time. Extending the method to several logarithms—to prove, for example, that e and π are algebraically independent—required a fundamentally new infrastructure.
While Gelfond and Schneider worked on a^b, Carl Ludwig Siegel had already launched a different program in 1929. Siegel observed that many transcendental numbers arise as values of solutions to linear differential equations. He introduced the classes of E-functions (entire functions with arithmetic growth conditions, like the exponential function) and G-functions (functions with finite radius of convergence, like hypergeometric series). The Siegel–Shidlovsky theory, completed by Andrey Shidlovsky in the 1950s, provides a general theorem: if a set of E-functions satisfies a linear differential system and the functions are algebraically independent over ℂ(z), then their values at any nonzero algebraic point are algebraically independent. This framework subsumes Hermite's and Lindemann's results as special cases and reaches far beyond them—it can handle values of Bessel functions, confluent hypergeometric functions, and many other special functions. Unlike the Gelfond–Schneider method, Siegel–Shidlovsky is built on differential equations rather than interpolation, and it produces algebraic independence results, not just transcendence of single numbers. The theory remains active today, with extensions to p-adic settings and to G-functions, and it coexists with other frameworks because it addresses a class of problems—values of differential-equation solutions—that the logarithm-based methods do not naturally cover.
Kurt Mahler introduced yet another proof architecture in the 1930s, based on functional equations rather than differential equations or interpolation. Mahler's method applies to functions f(z) that satisfy a functional equation of the form f(z^m) = R(z, f(z)) for an integer m ≥ 2 and a rational function R. The prototypical example is the Fredholm series ∑ z^{2^k}, which satisfies f(z^2) = f(z) – z. By evaluating such functions at algebraic points, Mahler proved transcendence and algebraic independence results for numbers arising from automatic sequences, digital expansions, and certain infinite products. The method is narrower than Siegel–Shidlovsky in the functions it can handle, but it covers territory that differential-equation methods cannot reach—for instance, the generating functions of automatic sequences have no natural differential equation. Mahler's method has seen a revival since the 1990s, driven by connections to automata theory and to the arithmetic of dynamical systems. It remains a living tradition, complementary to both Siegel–Shidlovsky and Baker's theory.
Alan Baker transformed the field in the 1960s by extending the Gelfond–Schneider method from a single logarithm to several logarithms and, crucially, by making the bounds effective. Baker proved that if α₁,…,αn are nonzero algebraic numbers and β₁,…,βn are algebraic numbers with 1, β₁,…,βn linearly independent over ℚ, then the absolute value of β₁ log α₁ + … + βn log αn is bounded below by an explicit positive constant depending on the heights of the αi and β_i. The proof combined the auxiliary-function technique of Gelfond–Schneider with new estimates from the theory of linear forms, and it produced a computable constant—something earlier methods could not do. This effectiveness transformed the relationship between transcendental number theory and Diophantine geometry. Baker's bounds could be plugged directly into the theory of Thue equations, elliptic curves, and exponential Diophantine equations, yielding explicit upper bounds on the number of solutions and, in many cases, actually determining all solutions. Baker's work earned him the Fields Medal in 1970 and effectively replaced the Gelfond–Schneider method as the primary tool for logarithm-based transcendence. Yet it did not replace Siegel–Shidlovsky or Mahler's method, because those frameworks address different classes of functions and produce algebraic independence results that Baker's theory does not directly give.
Stephen Schanuel formulated a conjecture in the 1960s that, if true, would unify virtually all of the above results. The conjecture states: if x₁,…,xn are complex numbers linearly independent over ℚ, then the transcendence degree of the field ℚ(x₁,…,xn, e^{x₁},…,e^{xn}) is at least n. This single statement would imply the Hermite–Lindemann theorem (take n=1, x₁=1), the Gelfond–Schneider theorem (take n=2 with x₁=1, x₂=b log a), and Baker's theorem (take xi = βi log αi with appropriate linear independence). It would also settle many open algebraic independence problems—for example, whether e and π are algebraically independent (take x₁=1, x₂=iπ) and whether e^{e} is transcendental. Schanuel's conjecture is not a method but a structural prediction about the exponential function. It remains unproven, though partial results exist: Boris Zilber's work on pseudo-exponentiation shows that a model-theoretic analogue of the conjecture is consistent, and there are known lower bounds on the transcendence degree in special cases. If proved, Schanuel's conjecture would not render the existing frameworks obsolete; rather, it would explain why they work and provide a unified foundation for future transcendence results.
Today, three major frameworks coexist, each with its own domain of strength. Baker's theory of linear forms in logarithms is the most widely applied tool in Diophantine geometry: it gives effective bounds for exponential Diophantine equations, for the solution of Thue equations, and for problems in arithmetic geometry such as bounding the number of integral points on elliptic curves. Siegel–Shidlovsky theory remains the central framework for algebraic independence of values of special functions defined by differential equations, and it has been extended to p-adic and global-function-field settings. Mahler's method, revived and refined, handles transcendence problems arising from functional equations, automatic sequences, and dynamical systems. The three frameworks agree on the fundamental role of auxiliary functions and height estimates, but they disagree on which source of auxiliary functions—differential equations, functional equations, or logarithm linear forms—is most natural for a given problem. Researchers often combine them: for instance, one might use Baker's bounds to control the arithmetic part of a problem and Siegel–Shidlovsky to handle the analytic part. Schanuel's conjecture hovers over the field as a unifying hypothesis, guiding expectations even though it remains out of reach. The leading open problems—such as proving the algebraic independence of e and π, or determining the transcendence of values of the gamma function at rational points—continue to drive the development of all three frameworks, and the field's vitality comes from their productive coexistence rather than from any single method's dominance.