I. Introduction
The four prior papers in this suite have treated the Riemann hypothesis in various frames: historically, as a conjecture with a long pedigree; field-theoretically, as a statement about how zeta and L-functions relate to algebraic, geometric, and arithmetic objects; strategically, as a target of proof methods that have so far stalled; and prospectively, as the anchor of forward-looking conjectures whose investigation can advance the conditional theory even in the absence of a proof. What none of these papers has done is place RH within what is, on present evidence, the most ambitious and most likely productive organizing framework of modern number theory: the Langlands program.
The Langlands program is sometimes described as a “grand unified theory” of mathematics. The description is journalistic but not entirely inaccurate. Beginning with a 1967 letter from Robert Langlands to André Weil, the program has grown over six decades into a vast network of conjectures, partial results, and proof techniques connecting number theory, representation theory, harmonic analysis, automorphic forms, algebraic geometry, and mathematical physics. At its center is a single organizing principle: that the L-functions arising in number theory — Dirichlet L-functions, Hecke L-functions, Artin L-functions, modular form L-functions, and many others — should all be understood as L-functions attached to automorphic representations of reductive algebraic groups, and that the analytic, arithmetic, and Galois-theoretic properties of these L-functions are governed by a coherent set of correspondences and functorialities.
Within this framework, the Riemann zeta function is not a special object. It is the simplest possible case: the L-function of the trivial automorphic representation of GL(1) over Q. The Riemann hypothesis for ζ is the simplest case of the Grand Riemann Hypothesis, which asserts that all L-functions in the Selberg class — conjecturally, all automorphic L-functions — have their nontrivial zeros on the critical line. RH is, in this view, not the central problem of analytic number theory but rather one specimen of a much larger problem, and progress on RH is best understood as emerging from progress on the broader framework rather than from purely analytic techniques applied to ζ in isolation.
This paper takes up the Langlands framework as the proper setting for RH. The aim is structural: to make precise the sense in which RH is a specimen of a Langlands-organized family, to identify what kinds of progress on the Langlands program would bear on RH, and to be clear about what the Langlands framework supplies and what it does not. The paper does not claim that the Langlands program will, eventually, yield a proof of RH. It claims something weaker but still substantive: that the Langlands framework is the natural setting in which to understand RH, and that the directions of likely progress are best identified within that framework.
The structure of the paper proceeds outward from the origins of Langlands’s vision through the precise statement of the framework, the conjectures of functoriality and reciprocity, the trace formula and the role of endoscopy, the Galois side via Fontaine–Mazur, the geometric Langlands analog, and finally the specific implications for RH. Throughout, the emphasis is on how RH fits into the larger picture rather than on the details of the Langlands program itself, which is far too vast to be treated comprehensively in a single paper.
II. The Origins of Langlands’s Framework
Langlands’s 1967 Letter
In January 1967, Robert Langlands, then a young assistant professor at Princeton, wrote a seventeen-page letter to André Weil at the Institute for Advanced Study. The letter, handwritten and tentative in its formulations, sketched a series of conjectures connecting automorphic forms on reductive groups to representations of Galois groups, with predictions for the analytic continuation and functional equations of associated L-functions. Langlands prefaced the letter with the now-famous remark, “If you are willing to read it as pure speculation I would appreciate that; if not — I am sure you have a waste basket handy.”
The letter was, in retrospect, one of the seminal documents of twentieth-century mathematics. Weil’s response — which preserved rather than discarded the speculation — was that the conjectures, if even partially correct, would constitute a unification of number theory at a depth not previously imagined. The conjectures became the seed of what is now called the Langlands program, and the program has been the central organizing principle of an enormous fraction of subsequent number theory.
The letter contained several distinct conjectural components. There was a conjecture about automorphic L-functions: that they should admit analytic continuation and functional equations of a uniform shape. There was a conjecture about functoriality: that homomorphisms between certain “dual groups” should produce transfers between automorphic representations of different reductive groups, with corresponding identities of L-functions. There was a conjecture about reciprocity: that representations of Galois groups should correspond to automorphic representations, with matching L-functions. And there was a unifying vision: that all of these pieces should fit together into a single coherent theory.
The Motivating Examples
Langlands’s conjectures did not arise from nothing. They arose from a careful examination of what was already known and from the recognition that the known cases shared a structural pattern that suggested a much more general framework.
Class field theory, developed in the early twentieth century by Hilbert, Takagi, Artin, and others, gives a complete description of abelian extensions of number fields. The Artin reciprocity law, in particular, establishes a canonical correspondence between abelian Galois representations and Hecke characters (via the idele class group). The L-functions of Galois representations on the abelian side coincide with Hecke L-functions on the automorphic side. Class field theory is, in Langlands’s framework, the abelian case — the case where the Galois representations and automorphic representations are both one-dimensional.
Eichler–Shimura theory, developed in the 1950s and 1960s, gives a partial description of the non-abelian case for GL(2). To each holomorphic newform of weight 2 and level N, Eichler and Shimura attached a two-dimensional l-adic Galois representation, with L-functions matching. This case suggested that the abelian theory of class field theory might extend to a non-abelian theory in higher rank.
Jacquet and Langlands’s 1970 monograph on automorphic forms on GL(2) supplied the local representation theory and the global trace formula for the GL(2) case. The work made precise what an “automorphic representation” of GL(2) is, what its local components look like, and how the trace formula could be used to study the spectral decomposition of automorphic forms.
These three pieces — class field theory, Eichler–Shimura, Jacquet–Langlands — together suggested the shape of a much more general theory. Langlands’s conjecture was that the pattern they exhibited extends to all reductive groups, with corresponding correspondences and functorialities at every rank.
The Shift from Arithmetic to Automorphic
The deepest conceptual shift in Langlands’s framework is the replacement of “L-functions of arithmetic origin” by “L-functions of automorphic origin.” Before Langlands, L-functions were classified by what they came from: Dirichlet L-functions came from characters, Hecke L-functions from ideal class groups, Artin L-functions from Galois representations, modular L-functions from modular forms. Each had its own theory, with overlapping but distinct techniques.
Langlands’s reframing puts all of these on a single footing: each L-function should be understood as the L-function of an automorphic representation of some reductive algebraic group. The variety of L-functions is then explained by the variety of reductive groups and the variety of representations of their dual groups. The structural unity of the L-function landscape is, on this view, a reflection of the structural unity of the automorphic landscape.
This reframing has practical consequences. Methods developed for one type of L-function become applicable, in principle, to all L-functions in the framework, mediated by functoriality. Properties conjectured for one type — analytic continuation, functional equation, location of zeros — become conjecturally universal. The Selberg class, treated in Paper 2, can be reinterpreted: the conjecture is that the Selberg class equals the class of all automorphic L-functions, and the structural axioms of the Selberg class are reflections of the automorphic origin.
III. Automorphic Representations as the Right Objects
Reductive Groups and Adelic Points
The objects on which the Langlands program operates are reductive algebraic groups. A reductive group G over Q is, roughly, a matrix group defined by polynomial equations whose connected component is a product of a torus and a semisimple part. The basic examples are GL(n), SL(n), Sp(2n), SO(n), and various inner forms and twisted versions of these.
For a reductive group G over Q, one considers the adelic group G(A_Q), where A_Q is the ring of adeles of Q. The adeles are the restricted product of all completions of Q (the real numbers and the p-adic numbers for each prime p), with the restriction that almost all components must lie in the maximal compact subring. The adelic group G(A_Q) is then the restricted product of the local groups G(Q_v), where Q_v ranges over the completions.
The adelic framework supplies the right setting for putting all places — Archimedean and non-Archimedean — on the same footing. Local–global principles, harmonic analysis, and trace formula methods all benefit from this uniform treatment.
Automorphic Forms and Representations
An automorphic form on G is a function on G(Q)\G(A_Q) — that is, a function on the adelic group invariant under the rational points G(Q) — satisfying certain analytic conditions (smoothness, finiteness under center action, moderate growth). The space of automorphic forms admits a natural decomposition under the action of G(A_Q) by right translation, and the irreducible components of this decomposition are called automorphic representations.
The space L²(G(Q)\G(A_Q)), suitably defined, decomposes into a discrete part (which contains the cuspidal automorphic representations and the residual spectrum) and a continuous part. The discrete part is the analog of the discrete spectrum of the Laplacian on a compact manifold; the continuous part is the analog of the continuous spectrum on a non-compact manifold.
A cuspidal automorphic representation is one whose realization in L² is by genuinely L² functions (rather than by limits of Eisenstein series). The cuspidal representations are the “most generic” automorphic representations, and they are the central objects of the Langlands program. The L-functions attached to cuspidal representations form, conjecturally, the cuspidal members of the Selberg class.
Local–Global Decomposition
A central structural feature of automorphic representations is that they decompose as restricted tensor products over places:
π = ⊗’_v π_v,
where π_v is an irreducible admissible representation of the local group G(Q_v). For almost all places v, π_v is unramified, meaning it has a vector fixed by a maximal compact subgroup of G(Q_v); the rest are ramified.
The local components π_v are studied through local representation theory, which has been worked out in considerable detail. For non-Archimedean v, the unramified representations correspond to semisimple conjugacy classes in the dual group (treated below), and the ramified representations have a more complicated parameterization by Langlands parameters or, equivalently, by Weil–Deligne representations of the local Galois group.
The local-global principle is that automorphic representations are determined by their local components, subject to a global compatibility condition. This makes the study of automorphic representations decomposable into local and global pieces, with substantial machinery developed for each.
Why This Framework Supersedes Earlier Formulations
The shift to automorphic representations as the basic objects has several structural advantages over earlier formulations.
First, it provides a single language. Dirichlet characters, Hecke characters, modular forms, Maass forms, Hilbert modular forms, and many other objects all become special cases of automorphic representations on appropriate groups (GL(1), GL(2) over various base fields).
Second, it accommodates the full range of L-functions. The L-functions arising in the framework include not only the standard L-functions but also Rankin–Selberg products, symmetric powers, exterior squares, and many others, all defined by means of representations of the dual group.
Third, it provides a setting in which local methods (representation theory of p-adic groups, Whittaker models, intertwining operators) and global methods (trace formula, theta correspondence, period integrals) can be applied in concert.
Fourth, it supplies a natural framework for the Langlands functoriality conjectures, which are most naturally formulated in terms of homomorphisms between dual groups acting on automorphic representations.
The Langlands framework thus does not merely reorganize known number theory; it reveals a structural unity that was implicit in the prior fragmented theories and provides tools for proving things that were not accessible in those theories.
IV. L-Functions Attached to Automorphic Representations
The Construction
To each automorphic representation π of a reductive group G over Q, and to each finite-dimensional representation r of the Langlands dual group ^L G, the Langlands construction produces an L-function L(s, π, r). The construction proceeds locally: for each place v, one defines a local L-factor L(s, π_v, r) using the local Langlands correspondence (which expresses π_v in terms of a local Langlands parameter, on which r can act). The global L-function is then the product of local factors:
L(s, π, r) = ∏_v L(s, π_v, r).
For almost all v, the local factor takes the explicit form
L(s, π_v, r) = det(1 − r(t_{π_v}) q_v^{−s})^{−1},
where t_{π_v} is the Satake parameter (a semisimple conjugacy class in ^L G) and q_v is the residue field cardinality at v.
Analytic Properties
The L-functions L(s, π, r) are conjectured to satisfy a uniform set of analytic properties: they admit meromorphic continuation to the entire complex plane, they satisfy a functional equation of a prescribed form, and they have Euler products as defined above. For “most” choices of π and r — specifically, when r is irreducible nontrivial — the L-function is conjectured to be entire.
The conjecture has been proved in various cases. For r = standard representation and π cuspidal on GL(n), the analytic properties were established by Godement–Jacquet (for general n) and earlier by Hecke and others (for small n). For Rankin–Selberg products L(s, π × π’), Jacquet, Piatetski-Shapiro, and Shalika established the analytic properties through the Rankin–Selberg integral method. For symmetric and exterior squares, the analytic properties are known through work of Shimura, Bump, Friedberg, Ginzburg, and others. For higher symmetric powers, the analytic properties remain open in many cases.
The Standard L-Function
The simplest case of the construction is r = standard representation of ^L G. For G = GL(n), the standard L-function L(s, π) is what one would call simply “the L-function of π.” The standard L-function is conjecturally in the Selberg class, satisfies the functional equation Λ(s, π) = ε(π) Λ(1 − s, π̃), where π̃ is the contragredient representation, and admits analytic continuation to an entire function (when π is cuspidal and not the trivial representation of GL(1)).
Examples
The variety of L-functions in the Langlands framework can be illustrated by examples.
For G = GL(1) over Q and π a Hecke character (which is, in the GL(1) case, just an idele class character), L(s, π) is the corresponding Hecke L-function. When π is the trivial character, L(s, π) = ζ(s). When π is a Dirichlet character viewed as an idele class character, L(s, π) is the Dirichlet L-function L(s, χ).
For G = GL(2) over Q and π corresponding to a holomorphic newform f of weight k and level N, L(s, π) is the L-function L(s, f) attached to f, with appropriate normalization. The analytic properties of L(s, f) — analytic continuation, functional equation, Euler product — are direct consequences of the automorphy of π.
For G = GL(n) and π a self-dual cuspidal representation, the symmetric square L(s, π, sym²) and the exterior square L(s, π, Λ²) are L-functions of the corresponding tensor product representations of the dual group. These play important roles in the theory.
For G a more general reductive group (Sp(2n), SO(n), etc.), the standard L-functions are attached to representations of the corresponding dual groups, and their analytic properties have been studied through the Langlands–Shahidi method, the Rankin–Selberg method, and the doubling method.
The unifying point is that all of these L-functions arise from the same construction: an automorphic representation, a representation of the dual group, and a product of local factors. The Selberg class, conjecturally, equals the set of L-functions arising in this way.
V. The Langlands Functoriality Conjectures
The Principle of Functoriality
The most far-reaching of Langlands’s conjectures is the principle of functoriality. In its general form, functoriality asserts the following: given two reductive groups G and H over Q, and a homomorphism of L-groups (the L-group is a more refined version of the dual group that incorporates the Galois action)
φ: ^L H → ^L G,
there should exist a “transfer” or “lifting” of automorphic representations from H to G. That is, to each automorphic representation π of H(A_Q), there should correspond an automorphic representation Π of G(A_Q) such that for every finite-dimensional representation r of ^L G,
L(s, Π, r) = L(s, π, r ∘ φ).
The transfer should be compatible with local-global decomposition: at each place v, the local representation Π_v should be obtained from π_v by a local lifting determined by φ.
The functoriality principle is enormous in scope. It says that the entire landscape of automorphic representations on different reductive groups is interconnected by a web of liftings, with each homomorphism of L-groups producing a corresponding lifting on automorphic representations.
Specific Cases of Conjectured Functoriality
Several specific cases of functoriality have been formulated and studied in detail.
Base change: For a finite extension E/F of number fields, base change is the lifting from automorphic representations of G(A_F) to automorphic representations of G(A_E). The L-group homomorphism is the natural inclusion. Cyclic base change for GL(2) was established by Langlands himself; cyclic base change for GL(n) was extended by Arthur and Clozel in 1989. Non-cyclic base change is open in general.
Automorphic induction: For a finite extension E/F, automorphic induction is the lifting from automorphic representations of GL(m) over E to automorphic representations of GL(m[E:F]) over F. The conjectured correspondence has been established in various cases.
Symmetric power liftings: For an automorphic representation π of GL(2), the n-th symmetric power lifting Sym^n π should be an automorphic representation of GL(n + 1). The symmetric power lifting is conjectured to exist for every n; it has been established for n = 2 (Gelbart–Jacquet), n = 3 (Kim–Shahidi), n = 4 (Kim), and n = 5 to 8 in various cases. The full symmetric power conjecture was established by Newton and Thorne in 2020 for cuspidal modular forms on GL(2), a major recent result.
Rankin–Selberg products: For automorphic representations π_1 of GL(m) and π_2 of GL(n), the Rankin–Selberg product π_1 ⊠ π_2 should be an automorphic representation of GL(mn). The L-function L(s, π_1 × π_2) is known to have the expected analytic properties (Jacquet–Piatetski-Shapiro–Shalika), but the full functoriality (existence of the automorphic representation π_1 ⊠ π_2) is open.
Endoscopic transfer: For G a reductive group with non-trivial endoscopy, the endoscopic transfer relates automorphic representations of G to automorphic representations of smaller groups (the endoscopic groups). The endoscopic classification, completed by Arthur in the 2010s for classical groups, is a substantial achievement of the Langlands program.
Modularity of Elliptic Curves as Functoriality
The modularity theorem (Wiles, Taylor–Wiles, Breuil–Conrad–Diamond–Taylor) — that every elliptic curve over Q is modular — can be reinterpreted as a functoriality statement. The Galois representation attached to an elliptic curve corresponds to a two-dimensional Galois representation; modularity asserts that this Galois representation comes from an automorphic representation of GL(2) over Q. In Langlands’s framework, this is the Galois-to-automorphic direction of reciprocity (treated below) for two-dimensional Galois representations of a specific kind.
Implications for the Selberg Class
If functoriality holds in full generality, the Selberg class equals the class of automorphic L-functions. The implication is structural: the Selberg axioms (Dirichlet series, Euler product, functional equation, analytic continuation, Ramanujan bound) characterize precisely those L-functions that come from automorphic representations.
The conjecture that the Selberg class equals the automorphic class has substantial consequences. The Selberg orthogonality conjectures, which predict that distinct primitive L-functions in the Selberg class have orthogonal Dirichlet coefficients, follow from functoriality (combined with Rankin–Selberg results). The Grand Riemann Hypothesis for the Selberg class becomes the Grand Riemann Hypothesis for automorphic L-functions.
The conjecture also implies Artin’s holomorphy conjecture: every Artin L-function for a non-trivial irreducible representation is entire. The reason is that, under functoriality, every Artin representation corresponds to an automorphic representation, and the L-function on the automorphic side is known to be entire.
VI. The Implications of Functoriality for L-Function Theory
What Functoriality Buys Concretely
Beyond the structural unification, functoriality has concrete consequences for L-function theory. Some examples:
Artin holomorphy: Established for all one-dimensional Artin representations (via Hecke L-functions and the abelian case of class field theory). Established for two-dimensional representations of solvable image (Langlands–Tunnell). Established for symmetric square Galois representations of certain modular forms (via Gelbart–Jacquet). Open in general.
Sato–Tate conjecture: For an elliptic curve E without complex multiplication, the angles θ_p defined by a_p(E) = 2√p cos(θ_p) should be equidistributed in [0, π] with respect to the Sato–Tate measure (2/π) sin²(θ) dθ. This conjecture was proved for elliptic curves over totally real fields by Taylor and collaborators in the late 2000s, conditional on automorphy of all symmetric powers Sym^n of the relevant Galois representations. The unconditional case followed once the relevant symmetric powers were established to be automorphic.
Effective Chebotarev: Under GRH for Artin L-functions, effective forms of the Chebotarev density theorem can be proved with strong error terms. Under unconditional Artin holomorphy alone (without RH), the effective forms are weaker but still substantial.
Class number bounds: Under GRH for various L-functions, sharper bounds on class numbers of number fields can be obtained. Under functoriality alone (without RH), partial results follow.
The Broader Pattern
The pattern is that functoriality results, even without RH, supply substantial conditional and sometimes unconditional progress on classical questions. RH on top of functoriality supplies further sharpening, but functoriality alone is already substantial. This suggests a research strategy: pursue functoriality results as the primary target, with RH-conditional sharpening as a secondary target.
The strategy has been pursued, with substantial success, over the past several decades. The proofs of modularity of elliptic curves (1995–2001), of Sato–Tate (2008–2010), of higher symmetric power liftings for GL(2) (Newton–Thorne 2020), and of various endoscopic classifications (Arthur 2010s) are all functoriality results in this framework. Each has substantial consequences for L-function theory, even though none of them proves any case of RH.
The Sato–Tate Conjecture as a Case Study
The Sato–Tate conjecture is a useful case study because it illustrates the relationship between functoriality, L-function analytic properties, and explicit equidistribution results.
The original conjecture, stated by Mikio Sato and John Tate in the 1960s, predicts that the Frobenius angles of an elliptic curve without complex multiplication are equidistributed in a specific way. The conjecture has direct arithmetic content — it predicts the average behavior of point counts of E modulo p as p varies — but its proof requires substantial machinery.
The Tate-Serre approach to the conjecture uses L-functions: Sato–Tate is equivalent to certain L-functions (the symmetric power L-functions L(s, Sym^n E)) having no pole at s = 1 for n ≥ 1. Proving non-vanishing at s = 1 requires knowing the L-function is well-behaved analytically, which in turn requires the Galois representation to be automorphic.
The Taylor–Harris–Shepherd-Barron–Clozel proof of Sato–Tate for elliptic curves over totally real fields proceeds by establishing the requisite automorphy. The proof uses Galois deformation theory, modularity lifting theorems extending Wiles’s methods, and the trace formula. The proof is, in this sense, a functoriality result that yields Sato–Tate as a consequence.
Sato–Tate is not an RH-style result. It does not involve zeros on the critical line; it involves non-vanishing at the edge of the critical strip. But the structural pattern — automorphy yielding analytic properties yielding arithmetic consequences — is the pattern by which functoriality results bear on the broader L-function landscape.
VII. The Trace Formula and Endoscopy
The Arthur–Selberg Trace Formula
The central computational tool of the Langlands program is the trace formula. Originally formulated by Selberg in the 1950s for SL(2) and developed extensively by James Arthur over the following decades for general reductive groups, the trace formula is an identity of the form
(Geometric side) = (Spectral side),
where the geometric side is a sum over conjugacy classes of G(Q) and the spectral side is a sum over automorphic representations of G(A_Q). Each side is, in a precise sense, a distribution on the appropriate space, and the equality of distributions is the trace formula.
The trace formula’s significance is that it relates two very different kinds of data. The geometric side is, in principle, computable from the arithmetic of G — it involves orbital integrals at conjugacy classes, which are local geometric integrals. The spectral side is what one wants to compute — it involves traces of operators on automorphic representations, which encode the spectrum of automorphic forms.
The trace formula has been used to prove a wide range of results: cyclic base change (Arthur–Clozel, using the trace formula), Jacquet–Langlands correspondence, the endoscopic classification, and many functoriality results. It is, on present evidence, the principal tool by which functoriality results are obtained.
The Stabilization Problem
A central technical issue in applying the trace formula is the stabilization of orbital integrals. The geometric side of the trace formula is a sum over conjugacy classes, and individual conjugacy classes contribute orbital integrals that depend on the choice of representative. To extract invariant information, one needs to organize the conjugacy classes into stable conjugacy classes (orbits under a larger equivalence relation) and produce stable orbital integrals.
The stabilization problem is the problem of expressing the geometric side of the trace formula in terms of stable orbital integrals on G and on its endoscopic groups (smaller groups that capture the difference between conjugacy and stable conjugacy). The stabilization, when achieved, decomposes the trace formula into a “stable” part for G itself and “endoscopic” contributions from smaller groups, with the endoscopic contributions providing the structure for endoscopic transfer.
The Fundamental Lemma
The technical centerpiece of the stabilization is the fundamental lemma. Conjectured by Langlands and Shelstad in the 1980s, the fundamental lemma is an identity between certain orbital integrals on G and corresponding orbital integrals on its endoscopic groups. The identity is purely local in nature — it concerns integrals at a single place — but its truth is essential for the global stabilization to work.
The fundamental lemma was proved by Ngô Bao Châu in 2008–2010, in a proof of remarkable depth and originality. Ngô’s proof uses the geometry of the Hitchin fibration on the moduli space of Higgs bundles, with a perverse sheaf decomposition that translates the orbital integral identities into geometric statements about the Hitchin fibration. The proof was awarded the Fields Medal in 2010.
The proof of the fundamental lemma cleared a substantial technical obstacle that had blocked progress on the trace formula for two decades. With the fundamental lemma established, the stabilization of the trace formula could proceed, and many functoriality results that had been pending the fundamental lemma became accessible.
What the Fundamental Lemma Buys
Several major functoriality results followed, more or less directly, from the proof of the fundamental lemma.
Endoscopic classification of representations of classical groups: Arthur’s monograph “The Endoscopic Classification of Representations” (2013) gave a complete description of the discrete spectrum of automorphic forms on classical groups (orthogonal, symplectic, and unitary groups) in terms of automorphic forms on GL(n). The classification depends essentially on the fundamental lemma.
Sato–Tate for higher genus: The methods used to prove Sato–Tate for elliptic curves, when combined with the fundamental lemma, extend to give Sato–Tate-type results for Hilbert modular forms and certain higher-rank cases.
Symmetric power liftings: The Newton–Thorne 2020 result establishing all symmetric power liftings for cuspidal modular forms on GL(2) uses methods that depend on the trace formula machinery the fundamental lemma supports.
The fundamental lemma is a useful case study because it illustrates how progress on a single deep technical problem in the Langlands program can have cascading consequences across many functoriality results. It also illustrates the depth of the methods involved: a proof requiring perverse sheaves on Hitchin fibrations is not the kind of proof that was anticipated in 1967, and its discovery represents a substantial expansion of the methods available to the program.
VIII. Galois Representations and the Fontaine–Mazur Conjecture
The Galois Side
The Langlands program has two sides. The automorphic side, treated above, involves automorphic representations of reductive groups. The Galois side involves continuous representations of the absolute Galois group Gal(Q̄/Q).
For each prime l, one considers continuous l-adic representations
ρ: Gal(Q̄/Q) → GL_n(Q̄_l),
where Q̄_l is an algebraic closure of the l-adic numbers. Such representations arise from many sources: l-adic cohomology of varieties over Q, modular forms via Eichler–Shimura and Deligne, Artin representations (which are representations with finite image, equivalent to representations factoring through a finite Galois group), and many others.
Each l-adic Galois representation has a corresponding L-function L(s, ρ), defined as an Euler product over primes:
L(s, ρ) = ∏_p det(1 − ρ(Frob_p) p^{−s} | V^{I_p})^{−1},
where Frob_p is the Frobenius at p, I_p is the inertia subgroup, and V^{I_p} is the inertia-fixed subspace. The L-function L(s, ρ) is conjectured to admit analytic continuation, satisfy a functional equation, and lie in the Selberg class.
Reciprocity: Galois ↔ Automorphic
The reciprocity conjecture of the Langlands program asserts that there is a correspondence between certain Galois representations and certain automorphic representations, with matching L-functions. Specifically, for an n-dimensional l-adic Galois representation ρ that is “geometric” in the Fontaine–Mazur sense (a precise technical condition), there should exist an automorphic representation π of GL(n) over Q such that L(s, ρ) = L(s, π) (suitably normalized).
The reciprocity conjecture has been established in various cases.
One-dimensional: All one-dimensional Galois representations correspond to Hecke characters by class field theory. This is the abelian reciprocity case.
Two-dimensional, with finite image (odd Artin representations): Established by Khare and Wintenberger (2009) for odd Artin representations, conditional on Serre’s modularity conjecture, which they also proved.
Two-dimensional, from elliptic curves: This is the modularity theorem for elliptic curves over Q, proved by Wiles, Taylor–Wiles, and Breuil–Conrad–Diamond–Taylor between 1995 and 2001.
Two-dimensional, from modular forms: The Eichler–Shimura–Deligne construction goes from modular forms to Galois representations; the converse, from suitable two-dimensional Galois representations to modular forms, is the modularity conjecture.
Higher-dimensional, special cases: Various cases of higher-dimensional reciprocity have been established, including some cases of automorphic Galois representations attached to Hilbert modular forms, certain self-dual representations, and others. The general case remains open.
The Fontaine–Mazur Conjecture
The Fontaine–Mazur conjecture, formulated by Jean-Marc Fontaine and Barry Mazur in 1995, is one of the most precise statements of the reciprocity conjecture. It asserts that an l-adic Galois representation ρ comes from an automorphic representation if and only if ρ is “geometric” in a specific technical sense:
(i) ρ is unramified outside a finite set of primes. (ii) ρ is de Rham at l (a condition from p-adic Hodge theory on the local representation at l).
The conjecture is, in this form, a precise criterion for which Galois representations are expected to be automorphic. The “if” direction (geometric implies automorphic) is the deep content of the conjecture; the “only if” direction (automorphic implies geometric) is well understood.
Recent progress on the Fontaine–Mazur conjecture has been substantial. Calegari and Geraghty have developed methods that establish the conjecture in many additional cases. The work of Allen, Calegari, Caraiani, Gee, Helm, Le Hung, Newton, Scholze, Taylor, Thorne, and others over the past decade has substantially extended the proven domain of the conjecture, including establishing automorphy lifting theorems in dimensions higher than 2.
Implications for L-Function Theory
If the Fontaine–Mazur conjecture is true, every “geometric” Galois representation has an automorphic L-function, with all the analytic properties that automorphic L-functions are conjectured to have. This implies, in particular:
Artin holomorphy: Every Artin representation is finite-image and hence geometric. Under Fontaine–Mazur, every Artin L-function is an automorphic L-function and hence entire (for non-trivial irreducible Artin representations).
Analytic continuation of cohomological L-functions: For varieties X over Q, the L-functions L(s, H^i(X)) are L-functions of Galois representations. Under Fontaine–Mazur, these are automorphic L-functions, with the expected analytic properties.
Riemann hypothesis for cohomological L-functions: The Grand Riemann Hypothesis applies to all automorphic L-functions, hence (under Fontaine–Mazur) to all L-functions of geometric Galois representations.
The implications connect the analytic theory (Riemann hypothesis, functional equations) to the geometric theory (cohomology of varieties) through the Galois-to-automorphic bridge supplied by Fontaine–Mazur.
IX. Geometric Langlands and Its Bearing on the Arithmetic Case
The Geometric Analog
The geometric Langlands program is a function-field, categorical analog of the arithmetic Langlands program. The motivating idea is that many of the constructions of the arithmetic Langlands program — automorphic forms, Galois representations, L-functions — have geometric counterparts that can be defined and studied using the methods of algebraic geometry.
In the geometric setting, one replaces:
- Number fields with function fields of curves over a base.
- Galois representations with local systems (or D-modules) on the curve.
- Automorphic forms with sheaves on the moduli space of bundles on the curve.
- The Langlands correspondence with a correspondence between local systems and sheaves on bundle moduli spaces.
The geometric Langlands program has produced substantial mathematics. The work of Drinfeld in the 1970s and 1980s, of Beilinson, Bernstein, Deligne, and Drinfeld in subsequent decades, and of many others, has established large parts of the geometric Langlands correspondence in various forms.
Lafforgue’s Theorem
The most striking arithmetic application of geometric methods came from Vladimir Drinfeld and Laurent Lafforgue. Drinfeld in the 1970s proved the Langlands correspondence for GL(2) over function fields of curves over finite fields. Lafforgue in 2002 extended this to GL(n) for arbitrary n, establishing the Langlands correspondence for GL(n) over function fields of curves over finite fields.
Lafforgue’s proof uses the moduli space of “shtukas,” a function-field analog of Drinfeld’s earlier moduli of elliptic modules. The proof is geometric and substantial; it was awarded the Fields Medal in 2002.
In 2018, Vincent Lafforgue (Laurent’s brother) extended the framework still further, establishing one direction of the Langlands correspondence for general reductive groups over function fields of curves over finite fields. Vincent Lafforgue’s work builds on his brother’s methods but introduces new techniques connecting moduli of shtukas to representations of reductive groups.
What the Function Field Cases Supply
The function field cases of the Langlands correspondence, where they have been proved, supply two things to the arithmetic case.
First, they supply evidence. The fact that the Langlands correspondence is true in the function field setting is strong evidence that the corresponding arithmetic conjectures are true. The function field setting is, structurally, easier than the arithmetic setting (for reasons discussed in Paper 2: the absence of the Archimedean place, the availability of geometric methods), but it is still substantial enough that proof there is a meaningful achievement.
Second, they supply templates for arithmetic methods. Methods developed in the function field setting often suggest, by analogy, what arithmetic methods might look like. The use of moduli spaces in Lafforgue’s work, for instance, has parallels in arithmetic Langlands through the use of Shimura varieties as moduli spaces of abelian varieties with extra structure.
The relationship between function field Langlands and arithmetic Langlands is, in this respect, parallel to the relationship between the function field Riemann hypothesis (proved by Weil and Deligne) and the arithmetic Riemann hypothesis (open). The function field case provides evidence, methods, and structural templates, but does not directly transfer to the arithmetic case. The arithmetic case, on present evidence, requires additional structures that have not yet been constructed in compatible form.
Geometric Langlands and Mathematical Physics
The geometric Langlands program has, in recent years, developed substantial connections to mathematical physics — particularly to gauge theory and string theory. The work of Anton Kapustin and Edward Witten in the late 2000s reformulated geometric Langlands as a statement about S-duality of certain four-dimensional gauge theories. The connection is deep: geometric Langlands becomes, in this framework, a manifestation of physical dualities that have independent motivation in string theory.
The arithmetic implications of these connections are still being worked out. The hope is that physical methods, which have produced sharp predictions in geometric settings, might eventually contribute to the arithmetic theory. Whether this hope is realized, and whether it bears on RH specifically, remains open. But the geometric–arithmetic interface is one of the most active frontiers of the broader Langlands program.
X. The Riemann Hypothesis as a Langlands-Theoretic Statement
ζ as the Simplest Case
In the Langlands framework, the Riemann zeta function ζ(s) is the L-function of the trivial automorphic representation of GL(1) over Q. More precisely, the trivial idele class character (which sends every idele to 1) corresponds to the automorphic representation π_0 of GL(1) over Q whose L-function is
L(s, π_0) = ζ(s) · (factors at finitely many places).
After appropriate normalization, L(s, π_0) is essentially ζ(s), up to a finite product of local factors that does not affect the location of zeros.
This identification places ζ as the simplest possible automorphic L-function: GL(1) is the smallest reductive group, and the trivial character is the simplest representation. Every other automorphic L-function is more complicated than ζ.
The Reframing
The reframing supplied by the Langlands framework is that RH for ζ is the simplest case of the Grand Riemann Hypothesis for the Selberg class (or, equivalently, for automorphic L-functions). The hypothesis is not about ζ specifically; it is about a general feature of L-functions arising from automorphic origin, with ζ as the simplest specimen.
This reframing has several consequences for how one thinks about RH.
First, it suggests that proof methods specific to ζ are unlikely to be the ultimate path to a proof. If RH holds for the same structural reasons that GRH holds for the entire automorphic class, then a method that works only for ζ is, in some sense, missing the structural reason. The function field case, where RH for varieties was proved by methods that generalize from curves to higher-dimensional varieties, illustrates this: the proof methods are uniform across the family, not specific to a single specimen.
Second, the reframing suggests that progress on RH is best pursued through progress on the Langlands program. Functoriality results, even when they do not directly involve RH, contribute to the structural understanding that may eventually support a proof. The cumulative effect of many functoriality results — modularity of elliptic curves, Sato–Tate, symmetric power liftings, the endoscopic classification — is to bring the automorphic landscape into clearer view, and clearer view of the landscape is a prerequisite for proving structural theorems about the landscape.
Third, the reframing suggests that any proof of RH will be embedded in a substantially advanced state of the Langlands program. A proof of RH that bypassed the Langlands program would be remarkable and structurally surprising; the Langlands framework so thoroughly organizes the L-function landscape that a proof outside the framework would have to explain why the framework is not, in fact, the right setting.
The Structural Argument
The structural argument for the Langlands framework as the proper setting for RH has several components.
First, the uniformity argument. The Selberg class is conjectured to satisfy the Grand Riemann Hypothesis uniformly: every L-function in the class has its zeros on the critical line, with no L-function being exceptional. This uniformity suggests that the reason for the location of zeros is structural, applying to all L-functions of automorphic origin in the same way. A proof method that explained ζ but not, say, L(s, χ) for Dirichlet characters would have to explain why ζ is special, and there is no apparent structural reason for ζ to be special.
Second, the function field argument. In the function field setting, the corresponding Riemann hypothesis is proved uniformly across the relevant class (smooth projective varieties over finite fields, in Deligne’s generalization). The proof is by methods that apply uniformly — étale cohomology, monodromy, weight filtrations — rather than by methods specific to particular varieties. The structural template suggests that the arithmetic case, when it is eventually proved, will also be by uniform methods.
Third, the Langlands–Shahidi argument. The Langlands–Shahidi method, developed by Shahidi from Langlands’s earlier work, supplies a method for proving the analytic continuation and functional equations of certain automorphic L-functions through the analysis of Eisenstein series. The method is uniform across the relevant class. While the method does not prove RH, it illustrates how uniform methods produce uniform results across the automorphic landscape, and it suggests that a uniform method for RH should exist.
The structural argument is not a proof. It is a framework within which proof methods are likely to be sought. The framework is, on present evidence, the most likely setting for eventual progress, but it is not the only possible setting, and surprises are possible.
XI. What Proof of Various Functorialities Would Buy
Cumulative Progress
Each functoriality result establishes a piece of the broader Langlands picture. Cumulatively, these results bring the picture into clearer focus and constrain the space of possible behaviors of L-functions. The cumulative effect, on present evidence, is what is most likely to produce eventual progress on RH.
Symmetric power functoriality for GL(2) → GL(n) for all n: Established by Newton and Thorne in 2020 for cuspidal modular forms on GL(2). The result has substantial consequences: it implies the Sato–Tate conjecture in a much wider range of settings than was previously accessible, it implies non-vanishing of symmetric power L-functions at the edge of the critical strip, and it provides tools for analyzing higher-rank L-functions through their relationship to GL(2) data.
Full Rankin–Selberg functoriality: Would imply that the product of two cuspidal automorphic representations is automorphic, with the corresponding L-function being the product of local Rankin–Selberg factors. This would close many open cases of the Selberg orthogonality conjectures and supply tools for moments of L-functions in much greater generality than is currently available.
Full automorphy of geometric Galois representations (Fontaine–Mazur): Would imply Artin’s holomorphy for all Artin representations, would imply analytic continuation of all cohomological L-functions, and would establish the full reciprocity correspondence between Galois and automorphic sides.
Full functoriality across all reductive groups: Would establish the entire Langlands picture, with all automorphic L-functions on all reductive groups governed by a single coherent theory.
The Path to RH
A natural question is whether full Langlands functoriality, if proved, would imply RH. The answer is: not directly, but it would substantially constrain the problem.
If the Selberg class equals the automorphic class (as functoriality would imply), then RH for ζ is one specimen of a uniform conjecture across the automorphic class. A proof of RH for any non-trivial automorphic L-function (say, a Dirichlet L-function or a modular L-function) by methods that exploit the automorphic structure would, by uniformity, suggest that the same methods should work for ζ. The methods might not directly transfer, but they would constrain what a proof could look like.
The function field analog illustrates this pattern. The function field RH is proved uniformly across the relevant class, and the proof methods exploit the geometric structure. Once the methods were available for one variety (curves, in Weil’s proof), they extended to higher-dimensional varieties (in Deligne’s proof). The arithmetic case, if it follows this pattern, will also have proof methods that apply uniformly across the automorphic class.
A proof of RH, on this picture, is not separable from progress on the broader Langlands program. The proof, when it comes, is likely to be a uniform proof for the automorphic class, with RH for ζ as the simplest specimen. Such a proof requires the automorphic class to be in clear view, and bringing it into clear view is the work of the Langlands program.
XII. Limitations of the Langlands Framework
Open Cases of Langlands Itself
The Langlands program is itself open in most of its central conjectures. Functoriality is established only in special cases. Reciprocity (the Galois-to-automorphic correspondence) is established in restricted ranges. The Fontaine–Mazur conjecture is established for many but not all geometric Galois representations. The endoscopic classification is established for classical groups but not yet for exceptional groups in full generality.
This means that the Langlands framework, as a setting for RH, is itself a partially conjectural setting. If one assumes the full Langlands picture, RH becomes a specimen of GRH for the Selberg class. But the full Langlands picture is not established. Progress on RH within the framework requires, in some form, simultaneous progress on the framework itself.
The Missing Geometry
The Langlands framework organizes the L-function landscape, but it does not, in itself, supply the geometric or cohomological structure that the function field proof of RH used. The function field proof succeeded because Spec of a smooth projective variety over a finite field is a geometric object, with finite-dimensional cohomology and a Frobenius operator and a Hodge-theoretic positivity. The Langlands framework, even in its full form, does not supply analogs of these for Spec(Z).
The “missing geometry” problem is not solved by Langlands. It must be addressed by some additional construction — Arakelov theory, the F_1 program, the Connes program, or some new framework — that supplies the geometric ingredients needed for a Riemann hypothesis-style proof.
The relationship between Langlands progress and missing-geometry progress is, on present evidence, complementary. Langlands progress organizes the L-function side; missing-geometry progress supplies the structural ingredients on the arithmetic side. A proof of RH likely requires both kinds of progress to converge.
Whether Progress Will Converge
The convergence of Langlands progress and missing-geometry progress is conjectural. The two programs have, historically, developed largely independently. Langlands progress has come from representation theory, automorphic forms, and the trace formula; missing-geometry progress has come from Arakelov theory, noncommutative geometry, and the F_1 program. The methods are different; the practitioners are largely different communities.
There are some signs of convergence. The geometric Langlands program connects automorphic forms to algebraic geometry, and recent work in the program has involved methods (perverse sheaves, derived categories) that also appear in arithmetic geometry. The Connes program has connections to automorphic forms through the adèle class space. The F_1 program has connections to combinatorial structures that arise in representation theory.
Whether these signs of convergence eventually produce a unified framework that supports a proof of RH is open. The framework, when it comes, is likely to be substantial — incorporating Langlands functoriality, missing geometry, and possibly additional structures not yet identified. The framework’s emergence would be a major mathematical event, comparable in scope to the development of étale cohomology in the 1960s.
XIII. Conclusion
The Langlands framework places the Riemann hypothesis as the simplest specimen of a much larger conjectural family. The reframing has structural consequences for how one thinks about RH: as a uniform feature of automorphic L-functions rather than a special property of ζ; as a target whose pursuit is best embedded in the broader Langlands program rather than pursued independently; as a problem whose resolution is likely to come from the convergence of multiple programs rather than from any single line of attack.
The reframing does not provide a proof. It does provide a framework within which proof methods can be evaluated. A proposed proof of RH that does not engage with the Langlands framework — that proves RH for ζ specifically, by methods that do not generalize to other automorphic L-functions — would be structurally surprising. Such a proof would have to explain why ζ is special, and the Langlands framework supplies no apparent reason for ζ to be special.
The directions in which Langlands-program progress is most likely to bear on RH are several. Symmetric power functoriality, having recently been established for GL(2) by Newton and Thorne, has cascading consequences for L-function theory and constrains the L-function landscape further. Rankin–Selberg functoriality, if established in full, would close many open cases of the Selberg orthogonality conjectures and supply tools for moments and zero statistics. Reciprocity, if extended to higher-dimensional Galois representations, would establish analytic continuation and functional equations for a much wider class of L-functions. The trace formula, post-fundamental-lemma, supplies methods for proving these functoriality results, and continued progress on the trace formula is likely to yield further functoriality results.
The cumulative progress on these fronts brings the L-function landscape into clearer focus. Each functoriality result narrows the space of possible behaviors and constrains the form that a proof of RH can take. The eventual proof, when it comes, will likely be a proof of GRH for the entire automorphic class — uniform, structural, and embedded in a substantially advanced state of the Langlands program. The proof will not be discovered in isolation from the program; it will emerge from the program as a corollary of structural theorems about automorphic L-functions.
The historical record supports this picture. Major theorems in number theory have, for several decades, come predominantly from the Langlands program and its offshoots: the modularity theorem, Sato–Tate, the Sato–Tate generalizations, the Newton–Thorne symmetric power result, the endoscopic classification, the proof of the fundamental lemma. Each of these is a Langlands-program result. Each has consequences beyond its immediate statement. Each contributes to the gradual construction of a unified picture in which RH is a specimen.
A proof of RH, on this picture, is unlikely to be soon. The Langlands program is enormous, and many of its central conjectures are open. The “missing geometry” problem is unresolved. The convergence of programs is at best partial. But the trajectory is identifiable: progress on the Langlands program, on missing geometry, and on their convergence is the most likely path to a proof, and the work of the next several decades is likely to be along that path.
What can be said with confidence is that the Riemann hypothesis is no longer best understood as Riemann understood it: as an isolated remark in an 1859 memoir on prime numbers. It is, in current understanding, the simplest case of a vast conjectural framework whose investigation has organized a substantial portion of modern number theory. The framework is the Langlands program; the case is RH for ζ; the proof, when it comes, is likely to be a proof for the framework.
═══════════════════════════════════════════════
