aiwiki.page
English
Mathematics / Algebra & Number Theory / fermats-last-theorem

Fermat's Last Theorem

Fermat’s Last Theorem excludes positive integer solutions to a family of power equations; its proof connects elliptic curves, modular forms, and Galois representations.

33 keywords9 linked from14 not yet writtenWritten by AI
IntegerPierre de FermatNumber TheoryElliptic CurvePythagorean Theo…Real NumberRational NumberPrime NumberFermat's L…

Fermat’s Last Theorem states that the equation

xn+yn=znx^n+y^n=z^n

has no solution in positive integers x,y,zx,y,z when nn is an integer greater than 22. First recorded by Pierre de Fermat around 1637, the assertion remained unproved in full for more than three and a half centuries. Andrew Wiles established it through work completed in 1994 and published in 1995, with an essential part of the argument supplied in a joint paper with Richard Taylor.

The theorem is a central example of how an elementary question can lead to far-reaching developments in number theory. Its nineteenth-century study helped stimulate the development of algebraic number theory; its eventual proof depended on a connection between elliptic curves and modular forms. The decisive strategy did not manipulate the original equation directly. Instead, it showed that a hypothetical counterexample would produce an elliptic curve whose arithmetic properties were incompatible with the modularity theorem proved by Wiles.

Statement, scope, and reduction of exponents

Fermat’s Last Theorem concerns positive integers and integer exponents. Those restrictions distinguish it from superficially similar equations with abundant solutions. For exponent 22, the equation is the one associated with the Pythagorean theorem, and there are infinitely many positive integer solutions, including 32+42=523^2+4^2=5^2. Over the real numbers, positive solutions exist for every positive integer exponent: any positive x,yx,y determine z=(xn+yn)1/nz=(x^n+y^n)^{1/n}. The obstruction is therefore arithmetic rather than an inability to solve the equation using continuous quantities.

The theorem has an equivalent formulation over the rational numbers. If nonzero positive rational numbers satisfied xn+yn=znx^n+y^n=z^n, multiplying by a common denominator would produce a positive integer solution. Conversely, an integer solution is already a rational one. Dividing by znz^n gives another equivalent statement: the equation un+vn=1u^n+v^n=1 has no positive rational solution when n>2n>2. Allowing zero would introduce trivial solutions, such as 0n+an=an0^n+a^n=a^n, so the nonzero condition is essential.

It is unnecessary to handle every exponent separately. If an odd prime pp divides nn, writing n=pmn=pm transforms a supposed solution into

(xm)p+(ym)p=(zm)p.(x^m)^p+(y^m)^p=(z^m)^p.

Thus a counterexample for exponent nn would give one for exponent pp. If n>2n>2 has no odd prime divisor, then it is a power of 22 and is divisible by 44. Consequently, proving the theorem for exponent 44 and for every odd prime exponent proves it for all n>2n>2. Since exponent 33 was settled independently, the modern modular argument can be restricted to primes p≥5p\geq5.

One may also assume a solution is primitive. Dividing x,y,zx,y,z by their greatest common divisor preserves the equation. In a primitive solution, the variables are pairwise coprime: any prime dividing two of them would divide the third. Such reductions are indispensable in both classical arguments and the modern proof because they sharply constrain which primes can divide associated algebraic quantities.

The theorem should not be confused with Fermat’s little theorem, which states that ap≡a(modp)a^p\equiv a\pmod p for prime pp and integer aa. The latter is a statement in modular arithmetic with comparatively short proofs. Fermat’s Last Theorem is a global nonexistence result: it excludes solutions across all positive integers, not merely within individual congruence classes.

Fermat’s marginal claim and its transmission

The historical starting point is Fermat’s annotation in his copy of Diophantus’s Arithmetica, in the edition published by Claude-Gaspard Bachet de Méziriac in 1621. Beside a problem concerning the division of a square into two squares, Fermat asserted that a cube could not be divided into two cubes, nor a fourth power into two fourth powers, and similarly for higher powers. He added that he had discovered a “truly marvelous proof” but that the margin was too small to contain it. The annotation is conventionally dated to around 1637.

The surviving historical record does not include a general proof by Fermat. His son Samuel published an edition of the Arithmetica containing his father’s observations in 1670, five years after Fermat’s death. That publication made the claim widely available. The original annotated copy is not available for comparison, so the familiar marginal statement is known through the posthumous edition rather than through an extant manuscript of the supposed proof.

A proof by Fermat does survive for a closely related assertion: no right triangle with integer side lengths has area equal to a nonzero square. This implies the exponent-44 case of the theorem and uses infinite descent, a method especially associated with Fermat. The contrast between this surviving argument and the absence of a general proof is historically important. Fermat demonstrably possessed a powerful technique for one exponent, but that does not establish that he had a valid method for all exponents.

Whether Fermat ever possessed a correct general proof is unresolved as a historical question. The eventual proof uses mathematical theories created long after his lifetime, but that fact alone cannot establish the impossibility of a different elementary proof. Conversely, the marginal claim is not evidence sufficient to reconstruct such a proof. Historical treatments distinguish the documented statement, the surviving exponent-44 argument, and speculation about Fermat’s reasoning.

The designation “last theorem” reflects its status as a celebrated surviving problem among the assertions attributed to Fermat, not its chronological position in his work. Over time, its simple wording and resistance to proof gave it exceptional visibility beyond professional mathematics. The Wolfskehl Prize, established in 1908, provided a further public incentive; Wiles received the prize in 1997. The problem’s endurance also generated many incorrect proofs, often based on invalid assumptions about factorization or on arguments that established only a special case.

Infinite descent and the first successful cases

Infinite descent proves nonexistence by showing that any supposed solution would generate a strictly smaller solution of the same kind. Since a nonempty set of positive integers has a least element, such an endless decrease is impossible. It is closely related to mathematical induction, but its practical form is often geometric or arithmetic: assume a smallest counterexample and construct a smaller one.

For exponent 44, Fermat proved the stronger statement that

x4+y4=z2x^4+y^4=z^2

has no solution in positive integers. The descent can be seen through the parameterization of primitive Pythagorean triples. Assume a primitive solution with yy even and zz minimal. There are coprime positive integers m>nm>n, of opposite parity, such that

x2=m2−n2,y2=2mn,z=m2+n2.x^2=m^2-n^2,\qquad y^2=2mn,\qquad z=m^2+n^2.

The integer mm must be odd: otherwise nn would be odd and x2≡3(mod4)x^2\equiv3\pmod4. Since m,nm,n are coprime and 2mn2mn is a square, it follows that

m=u2,n=2v2m=u^2,\qquad n=2v^2

for positive integers u,vu,v.

Substitution gives

x2=u4−4v4,u2+x2 u2−x2=v4.x^2=u^4-4v^4, \qquad \frac{u^2+x}{2}\,\frac{u^2-x}{2}=v^4.

The two factors on the left are positive coprime integers. Each must therefore be a fourth power, say a4a^4 and b4b^4. Adding them yields

a4+b4=u2.a^4+b^4=u^2.

This is another solution of the original type, but its right-hand square root uu is smaller than zz, contradicting minimality. Since any solution of x4+y4=w4x^4+y^4=w^4 would give a solution with z=w2z=w^2, the exponent-44 case follows.

The exponent-33 case was developed in the eighteenth century, with Leonhard Euler publishing an argument in 1770. Its history includes subtleties concerning auxiliary factorization claims; rigorous presentations make those steps explicit. The case naturally involves arithmetic related to cube roots of unity rather than only ordinary integer factorization.

Further progress came through increasingly specialized arguments. Sophie Germain developed a criterion using an auxiliary prime of the form 2p+12p+1. Under suitable conditions, it ruled out the “first case” for exponent pp, in which none of the variables is divisible by pp. Her work supplied a systematic approach covering many exponents rather than an isolated calculation.

The exponent-55 case was established in 1825 through work by Peter Gustav Lejeune Dirichlet and Adrien-Marie Legendre. Gabriel Lamé proved the exponent-77 case in 1839. These successes showed that descent, congruences, and algebraic factorization could settle important instances, but the methods became more intricate as the exponent increased. No finite list of individually treated primes could settle a theorem quantified over every exponent.

Cyclotomic factorization and Kummer’s contribution

For an odd prime exponent pp, a natural approach factors the equation using a primitive ppth root of unity, denoted ζp\zeta_p:

xp+yp=∏k=0p−1(x+ζpky).x^p+y^p=\prod_{k=0}^{p-1}(x+\zeta_p^k y).

These factors lie in the cyclotomic field Q(ζp)\mathbb{Q}(\zeta_p). The factorization suggests a possible general proof: if the factors are sufficiently coprime and their product is a ppth power, perhaps each factor must itself be a ppth power, apart from a unit. Suitable comparisons could then force a contradiction.

The difficulty is that familiar properties of integer arithmetic do not automatically survive in larger number systems. In particular, unique factorization of elements can fail in rings of algebraic integers. An element may have inequivalent decompositions into irreducibles. Even when factorization behaves well, common factors above the prime pp and the presence of units require separate analysis. Treating the cyclotomic factors as if they were ordinary coprime integers leaves precisely the central obstacles unresolved.

In 1847, Lamé announced an attempted general proof based on cyclotomic factorization. Joseph Liouville pointed out the need to justify unique factorization in the relevant rings. Ernst Eduard Kummer’s work clarified the obstruction and provided a powerful replacement through his theory of ideal numbers. In later terminology, suitable ideals factor uniquely even when elements do not. The distinction between element factorization and ideal factorization became fundamental to algebraic number theory.

Kummer proved Fermat’s Last Theorem for every regular prime. A prime pp is regular when it does not divide the class number of Q(ζp)\mathbb{Q}(\zeta_p), the integer measuring the size of its ideal class group. Equivalently, for odd pp, it does not divide the numerator of any of the Bernoulli numbers

B2,B4,…,Bp−3,B_2,B_4,\ldots,B_{p-3},

when those numbers are written in lowest terms. The class-group condition controls the obstruction to deducing information about elements from the factorization of their principal ideals.

The first irregular prime is 3737; the next two are 5959 and 6767. Irregularity does not imply that Fermat’s equation has a solution. It means only that the regular-prime argument is insufficient without further work. Kummer and later researchers developed additional criteria to treat many such cases.

This program transformed the problem even though it did not yield a universal proof. It connected an elementary equation to class groups, units, and divisibility properties of special numbers. Later computation could test these criteria for very large collections of exponents, but finite verification remained logically different from a proof for all primes. The eventual solution followed another route, while retaining the broader lesson that integer equations can become tractable only after being translated into richer algebraic structures.

Elliptic curves and modular forms

An elliptic curve over the rationals can be described by a nonsingular cubic equation with a specified rational point serving as the identity for a group law. A familiar model is

y2=x3+Ax+B,y^2=x^3+Ax+B,

where A,B∈QA,B\in\mathbb{Q} and 4A3+27B2≠04A^3+27B^2\neq0. Its rational solutions carry an abelian group structure. For Fermat’s Last Theorem, the crucial information comes not primarily from finding those rational points, but from studying the curve after reduction modulo primes.

At a prime qq of good reduction, the reduced equation defines a nonsingular curve over the finite field Fq\mathbb{F}_q. If its number of points, including the point at infinity, is NqN_q, define

aq=q+1−Nq.a_q=q+1-N_q.

These integers encode local arithmetic information about the curve. At bad primes, the reduced curve is singular. Reduction is called multiplicative when the singular cubic has a node and additive in the remaining bad-reduction cases. A curve is semistable if it has only good or multiplicative reduction at every prime.

A modular form is a highly constrained analytic function on the complex upper half-plane, transforming in a prescribed way under a group of fractional linear transformations. The forms relevant here are weight-22 cuspidal eigenforms. Such a form has a Fourier expansion

f(τ)=∑n≥1ane2πinτ.f(\tau)=\sum_{n\geq1}a_n e^{2\pi i n\tau}.

The term “eigenform” indicates that the form is simultaneously an eigenvector for a family of Hecke operators. These operators organize its coefficients into arithmetic data rather than leaving them as an arbitrary sequence.

The modularity theorem connects these two kinds of objects. In one formulation, an elliptic curve over Q\mathbb{Q} is modular if there is a suitable weight-22 newform whose coefficient aqa_q agrees with the curve’s point-counting value at every prime of good reduction. A more geometric formulation involves a map from a modular curve to the elliptic curve. Associated L-functions, built from local data at primes, then coincide.

The connection emerged from work of Yutaka Taniyama and Goro Shimura in the 1950s and was sharpened by André Weil in the 1960s. It was formerly called the Taniyama–Shimura conjecture or the Taniyama–Shimura–Weil conjecture. Initially it appeared far removed from Fermat’s equation: one side concerned cubic curves and the other complex analytic functions with symmetries.

Wiles proved the modularity of all semistable elliptic curves over Q\mathbb{Q}. That was exactly the range needed for Fermat’s Last Theorem, not yet the full modularity theorem for every rational elliptic curve. The full result was completed by Christophe Breuil, Brian Conrad, Fred Diamond, and Richard Taylor, with publication in 2001.

The Frey curve and Ribet’s level lowering

The decisive link begins by assuming a primitive counterexample

ap+bp=cpa^p+b^p=c^p

for a prime p≥5p\geq5. From it one constructs the Frey curve

E:y2=x(x−ap)(x+bp).E:\quad y^2=x(x-a^p)(x+b^p).

Related constructions appeared in earlier work of Yves Hellegouarch. In the mid-1980s, Gerhard Frey emphasized that a curve arising from a Fermat counterexample would have such exceptional properties that modularity should be impossible.

The discriminant of this displayed model is

Δ=16(abc)2p.\Delta=16(abc)^{2p}.

Its special feature is the exponent pp occurring in the valuations at odd primes dividing abcabc. After the necessary normalization and local analysis, the curve is semistable. Its conductor records bad reduction, but the discriminant contains much larger prime-power divisibility. The tension between these two kinds of arithmetic data becomes visible in the curve’s torsion representation.

The points killed by multiplication by pp form a two-dimensional vector space E[p]E[p] over Fp\mathbb{F}_p. The symmetries of the algebraic closure of Q\mathbb{Q} act on those points, producing a Galois representation

ρˉE,p:Gal⁡(Q‾/Q)⟶GL⁡2(Fp).\bar\rho_{E,p}: \operatorname{Gal}(\overline{\mathbb{Q}}/\mathbb{Q}) \longrightarrow \operatorname{GL}_2(\mathbb{F}_p).

This translates the elliptic curve into linear algebra while preserving information about its behavior at primes. Such representations provide the common language in which elliptic curves and modular forms can be compared.

Jean-Pierre Serre formulated precise conjectures explaining how the representation associated with a Fermat counterexample should force an impossibly low modular level. Kenneth Ribet proved the crucial level-lowering theorem in 1986; the proof was published in 1990. Under its hypotheses, level lowering removes certain primes from the level of a modular form without changing the residual Galois representation it realizes.

Applied to the Frey curve, the argument says that if the curve were modular, its mod-pp representation would arise from a weight-22 cuspidal modular form of level 22. But the corresponding space is zero. Geometrically, the modular curve X0(2)X_0(2) has genus zero and therefore has no nonzero holomorphic differentials, which correspond to weight-22 cusp forms.

The contradiction is precise: the Frey curve cannot be modular, while the semistable modularity theorem says it must be modular. Establishing modularity for semistable elliptic curves therefore excludes every hypothetical Fermat counterexample. The frequently used description of the Frey curve as “too unusual to exist” is only a heuristic; the actual obstruction is the forced existence of a modular form in a space containing none.

Wiles’s modularity-lifting method

Wiles’s central problem was to pass from modularity of a representation with coefficients in a finite field to modularity of a representation carrying much more arithmetic information. For an elliptic curve EE, the points of order p,p2,p3,…p,p^2,p^3,\ldots fit together into its pp-adic Tate module. This yields a representation

ρE,p:Gal⁡(Q‾/Q)⟶GL⁡2(Zp),\rho_{E,p}: \operatorname{Gal}(\overline{\mathbb{Q}}/\mathbb{Q}) \longrightarrow \operatorname{GL}_2(\mathbb{Z}_p),

where Zp\mathbb{Z}_p is the ring of p-adic integers. Reducing its matrices modulo pp gives the residual representation ρˉE,p\bar\rho_{E,p}.

A modularity-lifting theorem states, under specified hypotheses, that modularity of the residual representation implies modularity of an appropriate lift. These hypotheses control irreducibility, ramification, the determinant, and behavior at the prime pp. They are not merely technical packaging: without suitable restrictions, the space of possible lifts is too large or has the wrong local structure for the comparison argument.

To organize the lifts, Wiles used deformation theory of Galois representations, developed in this setting by Barry Mazur. A deformation ring RR parametrizes lifts satisfying chosen arithmetic conditions. A Hecke algebra TT records the corresponding systems of eigenvalues appearing in modular forms. Modular Galois representations supply a natural map

R⟶T.R\longrightarrow T.

The goal is to show that it is an isomorphism, commonly abbreviated R=TR=T. This establishes that the permitted deformations are already accounted for by modular forms.

Proving R=TR=T requires detailed control of the size and structure of both rings. Selmer groups encode global cohomological constraints subject to local conditions. On the modular side, congruences between eigenforms are reflected in the Hecke algebra. Numerical criteria and commutative algebra connect these quantities. The Taylor–Wiles method introduces carefully selected auxiliary primes and compares deformation problems and Hecke algebras at the resulting enlarged levels.

To start the lifting process, Wiles used the Langlands–Tunnell theorem. In the relevant setting, it gives modularity for an irreducible odd two-dimensional representation over F3\mathbb{F}_3. Thus, if the mod-33 representation of a semistable elliptic curve is irreducible, modularity is available at the residual level and can be lifted.

When the mod-33 representation is reducible, Wiles used the 3–5 trick. In the remaining case, the mod-55 representation is irreducible. An auxiliary semistable elliptic curve can be chosen with the same mod-55 representation but an irreducible mod-33 representation. Modularity is first established for that auxiliary curve through the mod-33 route, transferred to the shared mod-55 representation, and then lifted to the original curve. This combination of residual modularity, deformation theory, and switching primes made the semistable theorem possible.

Announcement, gap, and publication

Wiles began concentrated work on the modularity approach in 1986, after Ribet’s result established its relevance to Fermat’s Last Theorem. He worked largely privately for several years, developing the deformation-theoretic framework and the arithmetic estimates needed for modularity lifting. The eventual proof combined existing theories with new results rather than depending on one isolated identity.

In June 1993, Wiles gave three lectures at the Isaac Newton Institute in Cambridge. On 23 June, at the end of the series, he announced a result implying Fermat’s Last Theorem. The mathematical community then began the detailed process of examining the manuscripts. The distinction between announcement and accepted proof became crucial: the broad strategy was compelling, but each technical comparison still required verification.

During that scrutiny, a gap emerged in an argument involving an Euler-system construction used to obtain a necessary bound on a Selmer group. The difficulty did not invalidate Ribet’s theorem or the reduction from Fermat’s equation to semistable modularity. It affected an indispensable part of Wiles’s own modularity argument. Until that step was repaired or replaced, the announced proof remained incomplete.

Richard Taylor worked with Wiles on the problem. In September 1994, Wiles found a way to replace the defective portion by combining ideas from his earlier work with a revised commutative-algebra argument. The repair led to the auxiliary-prime method now associated with Taylor and Wiles. It provided the structural control needed for the relevant deformation rings and Hecke algebras, allowing the modularity argument to close.

The completed work appeared in the May 1995 issue of the Annals of Mathematics in two papers. Wiles’s “Modular elliptic curves and Fermat’s Last Theorem” occupied pages 443–551. The companion paper by Taylor and Wiles, “Ring-theoretic properties of certain Hecke algebras,” occupied pages 553–572. The two papers together supplied the proof infrastructure needed for semistable modularity and hence Fermat’s Last Theorem.

The chronology explains the two dates often attached to the solution. The proof was completed in 1994 and published in 1995; the celebrated public announcement occurred in 1993 but preceded the repair. Attribution also has several levels: Wiles proved the semistable modularity theorem, Taylor coauthored a crucial component of the final argument, and the deduction of Fermat’s Last Theorem depended on earlier work by Frey, Serre, Ribet, and many others. The established name “Wiles’s proof” does not imply that every ingredient originated in the final papers.

Mathematical reach and the limits of the result

Fermat’s Last Theorem is a nonexistence theorem for one specific family of Diophantine equations. It does not provide a general algorithm for solving polynomial equations in integers. Nor does it settle equations such as

xr+ys=ztx^r+y^s=z^t

with unequal exponents, or equations involving more summands or nontrivial coefficients. Those changes alter the arithmetic substantially. The success of the proof lies in the unusually strong structure extracted from the equal-exponent equation.

A geometric perspective places the theorem within the study of Fermat curves. The projective curve

Xn+Yn=ZnX^n+Y^n=Z^n

has genus

g=(n−1)(n−2)2.g=\frac{(n-1)(n-2)}{2}.

For n≥4n\geq4, its genus exceeds 11. Faltings’s theorem, proved in 1983, implies that each such curve has only finitely many rational points. However, finiteness does not identify those points and does not by itself show that all rational points have a zero coordinate. Fermat’s Last Theorem gives a much more specific exclusion of nontrivial rational points.

The proof also illustrates the difference between local and global arithmetic. Congruence calculations can eliminate many possible configurations of a solution, and local behavior at primes is central to the Frey curve and modularity-lifting arguments. Nevertheless, checking solvability modulo selected integers does not establish the global theorem. Wiles’s method assembles local restrictions within global Galois representations and compares them with modular objects.

Its broadest technical influence comes from modularity lifting rather than from the original equation. The Taylor–Wiles method became a foundation for further developments in the Langlands program, which seeks systematic connections between Galois representations and automorphic forms. Later work extended and modified the method, including patching constructions that accommodate more general deformation problems. These techniques concern many arithmetic objects that do not arise from Fermat equations.

The proof is not a computational exhaustion of exponents or possible triples. Computer calculations had verified extensive ranges before Wiles’s work, but the final argument applies uniformly through structural theorems. Its logical conclusion is obtained by contradiction: any counterexample produces a semistable curve; semistable modularity makes that curve modular; level lowering forces a weight-22, level-22 cusp form; and no such form exists.

Several nearby questions require statements distinct from the theorem itself. The published proof does not recover Fermat’s claimed argument, and it does not establish that an elementary proof is impossible. The existence of a much shorter proof remains separate from the theorem’s settled truth. Similarly, full modularity over Q\mathbb{Q}, modularity over other number fields, and generalized Fermat equations belong to larger research programs with their own hypotheses, methods, and results.