The Langlands Program — Overview and Conjectures

One-line statement: The Langlands Program is a vast web of conjectures positing a deep correspondence between automorphic representations of reductive groups and Galois representations (or more precisely, parameters into a dual group), unifying number theory, representation theory, and harmonic analysis. Key result in words: Roughly, the "spectral" data of automorphic forms on a reductive group

G

over a global field should match the "Galois-theoretic" data of

L

-parameters into the Langlands dual group

G

, with

L

-functions and functoriality serving as the dictionary. Where it applies: Number theory (reciprocity,

L

-functions), automorphic representation theory, arithmetic geometry (Shimura varieties), the geometric Langlands correspondence, and mathematical physics (S-duality, conformal field theory). Who/when: Robert P. Langlands, in a 1967 letter to André Weil and subsequent papers; developed by Harish-Chandra, Jacquet, Shahidi, Arthur, Lafforgue, Ngô, Scholze, and many others, 1967–present.

The Langlands Program is a network of conjectures and theorems asserting a correspondence between two seemingly unrelated kinds of objects: automorphic representations — infinite-dimensional representations arising from harmonic analysis on adelic quotients $G (F) \ G (A_{F})$ — and Galois representations (or $L$ -parameters) valued in the dual group $G$ . The bridge between them is built from $L$ -functions, which can be attached to both sides and which the conjectures demand agree. It matters because it simultaneously generalizes class field theory (the abelian case), explains the appearance of modular forms in the proof of Fermat's Last Theorem, and predicts the analytic continuation and functional equations of an enormous class of $L$ -functions. A surprising fact: the structure on the Galois side is governed not by $G$ but by its Langlands dual $G$ , a different group whose root datum is dual to that of $G$ — so the dual of $SO_{2 n + 1}$ is $Sp_{2 n}$ , an exchange invisible until one views both through this lens.

The Core Insight

Reciprocity in its broadest sense: arithmetic information encoded in how primes split (Galois theory, motives) is the same data as the spectrum of natural differential and Hecke operators on spaces of automorphic forms. The two languages — algebraic (Galois) and analytic (automorphic) — are dual descriptions of a single underlying structure, mediated by

L

-functions. Functoriality, the master conjecture, says that any homomorphism of dual groups

H \to G

should induce a transfer of automorphic representations from

H

G

— and almost every classical reciprocity law is a special case.

Formal Statement

We work over a global field $F$ (a number field or function field of a curve over a finite field), with ring of adeles $A_{F}$ , and a connected reductive group $G$ over $F$ .

Automorphic Representation

An automorphic representation of $G$ is an irreducible constituent of the regular representation of $G (A_{F})$ on the space of automorphic forms — smooth functions

φ : G (F) \ G (A_{F}) ⟶ C

that are

K

-finite (for a maximal compact

K

Z

-finite (annihilated by a finite-codimension ideal of the center of the universal enveloping algebra), and of moderate growth. The cuspidal subspace

L_{cusp}^{2}

consists of those

φ

with vanishing constant terms:

\int_{N (F) \ N (A_{F})} φ (n g) d n = 0 for all unipotent radicals N of proper parabolics .

⟳[1] is the basic object — functions on the adelic quotient; [2] imposes the cuspidality condition by killing constant terms along every proper parabolic

P = M N

, isolating the discrete, "primitive" part of the spectrum.

By admissibility, an automorphic representation $π$ factors as a restricted tensor product over all places $v$ of $F$ :

π ≅ v ⨂^{'} π_{v},

where

π_{v}

is an irreducible admissible representation of

G (F_{v})

, and for almost all (finite, unramified)

v

π_{v}

is unramified — i.e. has a nonzero

G (O_{v})

-fixed vector.

⟳This is the tensor product theorem of Flath; it follows from the factorization of the global Hecke algebra into local components and the existence of spherical vectors away from a finite set of ramified places.

The Langlands Dual Group

Let $G$ have root datum $(X^{*}, Δ, X_{*}, Δ^{\lor})$ — characters, roots, cocharacters, coroots. The Langlands dual group (or $L$ -group's connected component) $G$ is the complex reductive group with the dual root datum $(X_{*}, Δ^{\lor}, X^{*}, Δ)$ — roots and coroots, characters and cocharacters interchanged. The full $L$ -group incorporates the Galois action on the based root datum:

^{L} G = G ⋊ Gal (\overline{F} / F) .

⟳The duality of root data swaps

Δ \leftrightarrow Δ^{\lor}

; the semidirect product encodes how the absolute Galois group acts on the Dynkin diagram (the "outer form" data), trivially when

G

is split.

Examples: $GL_{n} = GL_{n} (C)$ , $SL_{n} = PGL_{n} (C)$ , $SO_{2 n + 1} = Sp_{2 n} (C)$ , $Sp_{2 n} = SO_{2 n + 1} (C)$ , $SO_{2 n} = SO_{2 n} (C)$ .

L-Parameter and the Langlands Group

A (global) $L$ -parameter is a continuous homomorphism, commuting with projection to $Gal (\overline{F} / F)$ ,

ϕ : L_{F} ⟶^{L} G,

where

L_{F}

is the conjectural Langlands group — a global object whose finite-dimensional representations should classify automorphic representations of

GL_{n}

, generalizing the Weil group. Locally, for

v

archimedean or non-archimedean, the parameter is

ϕ_{v} : W_{F_{v}} \times SL_{2} (C) ⟶^{L} G,

where

W_{F_{v}}

is the Weil group and the

SL_{2}

-factor (the Arthur or Weil–Deligne

SL_{2}

) records unipotent/monodromy data.

⟳[5] is conjectural because

L_{F}

has no construction; [6] is well-defined locally — the Weil–Deligne group

W_{F_{v}} \times SL_{2}

replaces the elusive global group and captures ramification through the

SL_{2}

"Steinberg" direction.

The Langlands Correspondence (Conjectural Form)

There should be a canonical bijection (more precisely a finite-to-one map with $L$ -packets as fibers) between

{automorphic representations π of G (A_{F})} ⟷ {admissible L -parameters ϕ : L_{F} \to^{L} G},

compatible at every place with the local Langlands correspondence, and such that $L$ -functions and $ε$ -factors match:

L (s, π, r) = L (s, r \circ ϕ), ε (s, π, r) = ε (s, r \circ ϕ),

for every finite-dimensional representation

r

^{L} G

⟳[7] is the existence of the dictionary; [8] is its defining compatibility — the automorphic

L

-function (defined via the Langlands–Shahidi method or integral representations) must equal the Artin/motivic

L

-function of the composed parameter, place by place.

Functoriality (The Master Conjecture)

Given an $L$ -homomorphism of dual groups

ρ :^{L} H ⟶^{L} G

between

L

-groups of two reductive groups

H, G

over

F

, there exists a transfer of automorphic representations

π_{H} ⇝ π_{G} = ρ_{*} (π_{H})

such that, at every unramified place

v

, the Satake/Hecke parameters (semisimple conjugacy classes

t_{π_{H, v}} \in H

) map correctly:

t_{π_{G, v}} = ρ (t_{π_{H, v}}) in G (C) .

⟳[9] is just a group homomorphism on the dual side; [10] is the prescription that determines the transferred representation at almost all places via its unramified parameters — the Satake isomorphism turns the spherical Hecke algebra into

C [G]^{G}

, so

π_{v}

is encoded by a conjugacy class which

ρ

then transports.

Why It Holds / Derivation

The abelian case — class field theory — is fully proven and provides the template.

Class Field Theory as Langlands for GL₁

Step 1. Take

G = GL_{1} = G_{m}

. Then

G = C^{\times} = GL_{1} (C)

, and

^{L} G = GL_{1} (C) \times Gal

. Step 2. Automorphic representations of

GL_{1} (A_{F}) = A_{F}^{\times}

trivial on

F^{\times}

are exactly Hecke characters (Größencharaktere):

χ : F^{\times} \ A_{F}^{\times} ⟶ C^{\times} .

Step 3.

L

-parameters

ϕ : L_{F}^{ab} \to C^{\times}

are exactly continuous characters of the abelianized Weil group

W_{F}^{ab}

. Step 4. The Artin reciprocity isomorphism of class field theory,

Art_{F} : F^{\times} \ A_{F}^{\times} \sim W_{F}^{ab},

gives the bijection

χ \leftrightarrow χ \circ Art_{F}^{- 1}

. This is equation [7] for

G = GL_{1}

, and it is a theorem. Step 5. Matching of

L

-functions [8]: the Hecke

L

-function

L (s, χ)

equals the Artin

L

-function

L (s, χ \circ Art_{F}^{- 1})

, including the analytic continuation and functional equation — Tate's thesis (1950) supplies the local–global proof via adelic Fourier analysis.

■

Sketch: Why Functoriality Implies Reciprocity (Function-Field Case)

Step 1. Over a function field

F = F_{q} (X)

, V. Drinfeld (

n = 2

) and L. Lafforgue (general

n

) proved the global Langlands correspondence for

GL_{n}

using moduli of shtukas

Sht_{n}

— function-field analogues of elliptic curves with extra structure. Step 2. The

ℓ

-adic cohomology of

Sht_{n}

carries commuting actions of

GL_{n} (A_{F})

(Hecke) and

Gal (\overline{F} / F)^{n}

(Frobenius). Decomposing it realizes the correspondence [7] geometrically:

H_{c}^{*} (Sht_{n}, \overline{Q}_{ℓ}) \supset π ⨁ π \otimes σ (π)^{\otimes∙},

where

σ (π)

is the associated

n

-dimensional Galois representation. Step 3. Matching of Frobenius eigenvalues at unramified

v

with Satake parameters yields [10]; the Ramanujan–Petersson bound (purity, by Weil II) follows from the geometry.

■

Robert P. Langlands | 1936–

A Canadian-American mathematician, Langlands was a young faculty member at Princeton when, over the 1966–67 winter, he wrote a now-legendary 17-page handwritten letter to André Weil sketching the conjectures. He had developed the theory of Eisenstein series and the constant term, leading to the spectral decomposition of $L^{2} (G (F) \ G (A_{F}))$ — the analytic foundation. His subsequent works defined the $L$ -group, $L$ -functions for all reductive groups, and the principle of functoriality. With Hervé Jacquet he proved the correspondence for $GL_{2}$ (the "Jacquet–Langlands" theory). He received the Abel Prize in 2018 for "his visionary program connecting representation theory to number theory." His later "beyond endoscopy" proposal seeks to attack functoriality directly through the trace formula and analytic number theory.

Building Intuition

Think of three parallel worlds, each with its own "spectrum":

The Galois world — how primes factor in extensions, encoded by Frobenius conjugacy classes $Frob_{p}$ in a Galois group. This is arithmetic.
The automorphic world — eigenvalues of Hecke operators $T_{p}$ acting on modular/automorphic forms. This is analytic/harmonic.
The geometric/motivic world — the cohomology of algebraic varieties, where both of the above appear simultaneously.

Langlands says these are the same spectrum viewed three ways. Concretely, for an elliptic curve $E / Q$ , the number $a_{p} = p + 1 - # E (F_{p})$ (Galois/geometric data) equals the $p$ -th Hecke eigenvalue of a weight-2 modular form (automorphic data) — this is modularity, proven by Wiles–Taylor–Breuil–Conrad–Diamond. The Langlands Program asserts such a dictionary for every reductive group and every compatible system of representations.

The role of the dual group is the deepest surprise. Why should the automorphic side of $G$ be parametrized into $G$ , not $G$ ? Because the Satake isomorphism identifies the spherical Hecke algebra of $G (F_{v})$ with the representation ring of $G$ — Hecke eigenvalues are semisimple conjugacy classes in the dual group. The arithmetic naturally lives on the dual side.

The Langlands correspondence as a dictionary between two worlds, mediated by L-functions

Worked Examples

Modularity of an Elliptic Curve: $E: y^2 + y = x^3 - x^2$

Given. The elliptic curve

E

of conductor

N = 11

(Cremona label 11a). We test modularity via point counts mod small primes, computing

a_{p} = p + 1 - # E (F_{p})

. Step 1 — count points mod 2. Over

F_{2}

: solutions to

y^{2} + y = x^{3} - x^{2}

. Checking

(x, y)

x = 0 \Rightarrow y^{2} + y = 0 \Rightarrow y \in {0, 1}

;

x = 1 \Rightarrow y^{2} + y = 0 \Rightarrow y \in {0, 1}

. With the point at infinity,

# E (F_{2}) = 2 + 2 + 1 = 5

. So

a_{2} = 2 + 1 - 5 = - 2.

Step 2 — count points mod 3. Tabulating

x = 0, 1, 2

: one finds

# E (F_{3}) = 5

, giving

a_{3} = 3 + 1 - 5 = - 1.

Step 3 — count points mod 5. Direct enumeration yields

# E (F_{5}) = 5

, so

a_{5} = 5 + 1 - 5 = 1.

Step 4 — compare with the weight-2 newform $f \in S_{2} (Γ_{0} (11))$ . This form has

q

-expansion

f = q n \geq 1 \prod (1 - q^{n})^{2} (1 - q^{11 n})^{2} = q - 2 q^{2} - q^{3} + 2 q^{4} + q^{5} + 2 q^{6} - 2 q^{7} - \dots

Its Fourier coefficients are

a_{2} (f) = - 2

a_{3} (f) = - 1

a_{5} (f) = 1

. Answer.

a_{p} (E) = a_{p} (f) for p = 2, 3, 5 (and all p \neq = 11)

This is the

GL_{2} / Q

Langlands correspondence: the 2-dimensional

ℓ

-adic Galois representation on

T_{ℓ} E

matches the automorphic representation generated by

f

Satake Parameters and Unramified Local Langlands for $\mathrm{GL}_2(\mathbb{Q}_p)$

Given. An unramified principal series representation

π_{p} = Ind_{B}^{GL_{2}} (χ_{1}, χ_{2})

GL_{2} (Q_{p})

, with unramified characters

χ_{i}

Q_{p}^{\times}

determined by

α_{i} = χ_{i} (p)

. Take the situation of the prime

p = 5

for the curve

E

above, where

a_{5} = 1

. Step 1 — Satake parameter. The Satake isomorphism assigns to

π_{p}

the semisimple conjugacy class in

GL_{2} = GL_{2} (C)

t_{π_{p}} = (α_{1} 0 0 α_{2}) .

Step 2 — match to Hecke eigenvalues. For the spherical (unramified) vector, the Hecke operator

T_{p}

acts with eigenvalue

a_{p} = p^{1/2} (α_{1} + α_{2})

(in suitable normalization), and

⟨ T_{p, p} ⟩ = p α_{1} α_{2}

. For the modular form normalization the cleaner statement is the local $L$ -factor:

L (s, π_{p}) = [(1 - α_{1} p^{- s}) (1 - α_{2} p^{- s})]^{- 1} = (1 - a_{p} p^{- s} + p^{1 - 2 s})^{- 1} .

Step 3 — insert numbers for $p = 5$ . With

a_{5} = 1

L (s, π_{5}) = (1 - 5^{- s} + 5^{1 - 2 s})^{- 1} .

The roots satisfy

α_{1} + α_{2} = 1

and

α_{1} α_{2} = 5

(after un-normalizing by

p^{(s)}

), so

{α_{1}, α_{2}}

are roots of

T^{2} - T + 5 = 0

, giving

α = \frac{1 \pm - 19}{2}

, with

∣ α_{i} ∣ = 5 = p^{1/2}

— the Ramanujan–Petersson bound verified. Answer.

t_{π_{5}} \sim diag (\frac{1 + - 19}{2}, \frac{1 - - 19}{2}) \in GL_{2} (C), ∣ α_{i} ∣ = 5

These eigenvalues are exactly the Frobenius eigenvalues of

Frob_{5}

acting on

T_{ℓ} E

— equation [10] in action.

Common Misconceptions

"The Langlands correspondence is a bijection." Only for $GL_{n}$ is it (essentially) a bijection. In general it is finite-to-one: a single $L$ -parameter $ϕ$ corresponds to an $L$ -packet of several automorphic representations, organized by the component group $S_{ϕ} = π_{0} (Cent (ϕ, G))$ . The internal structure of packets is governed by the Arthur conjectures and characters of $S_{ϕ}$ .
"The dual group is $G$ over $C$ ." No — it is the group with dual root datum. For $G = Sp_{2 n}$ , $G = SO_{2 n + 1} (C)$ , a genuinely different group. Forgetting this is the single most common error.
"Automorphic = cuspidal." Cuspidal representations are the discrete, "atomic" pieces, but the full spectrum includes the continuous spectrum (Eisenstein series) and residual discrete spectrum. The Langlands spectral decomposition organizes all of these.
" $L$ -functions automatically have analytic continuation." This is part of what must be proven; functoriality and the Langlands–Shahidi method are precisely the tools that establish continuation and functional equations for broad families.
"The geometric Langlands program is just the function-field case." It is a categorical, sheaf-theoretic reformulation (an equivalence of derived categories) that is conceptually distinct, though deeply analogous, and now has its own near-complete proof (Gaitsgory–Raskin et al.).

Failure of naive 1–1 matching: $L$-packets for $\mathrm{SL}_2$

For $G = GL_{2}$ the local correspondence is a clean bijection. But descend to $G = SL_{2} (Q_{p})$ , whose dual is $G = PGL_{2} (C)$ . A single discrete $L$ -parameter $ϕ : W_{Q_{p}} \to PGL_{2} (C)$ whose image is the normalizer of a torus (dihedral type) has centralizer component group $S_{ϕ} ≅ Z /2 Z$ . Correspondingly the $L$ -packet contains two inequivalent supercuspidal representations of $SL_{2} (Q_{p})$ , distinguished by a character of $S_{ϕ}$ . Thus the parameter does not determine the representation uniquely — a fact invisible at the level of $GL_{n}$ and a warning against expecting bijections in general.

Connections

The Langlands Program sits at the crossroads of much of modern mathematics:

Class Field Theory is the abelian ( $GL_{1}$ ) case, fully proven; the program is its non-abelian generalization.
Modular Forms and Automorphic Representations supply the classical avatar: holomorphic modular forms are automorphic representations of $GL_{2} / Q$ with specific archimedean type.
Galois Representations and Étale Cohomology furnish the arithmetic side; modularity lifting theorems (Taylor–Wiles patching) are the engine connecting them to automorphy.
The Arthur–Selberg Trace Formula is the principal analytic tool for proving functoriality and characterizing the discrete spectrum.
Shimura Varieties geometrize the correspondence over number fields, their cohomology realizing both Hecke and Galois actions simultaneously.
Geometric Langlands Correspondence reframes everything categorically: $D$ -modules on $Bun_{G}$ versus quasi-coherent sheaves on the moduli of $G$ -local systems.
The Fundamental Lemma (Ngô's theorem) is the keystone of endoscopy, comparing orbital integrals across groups.
S-duality in Gauge Theory in physics (Kapustin–Witten) realizes geometric Langlands as electric–magnetic duality of $N = 4$ super Yang–Mills.

Real-World Applications

Though abstract, the program has decisive concrete consequences:

Fermat's Last Theorem (Wiles, 1994). The proof rests on modularity of semistable elliptic curves over $Q$ — a case of the Langlands correspondence ( $GL_{2}$ reciprocity), bridged to Frey's curve via Ribet's level-lowering.

The Sato–Tate Conjecture (Taylor et al., 2008). The equidistribution of $a_{p} / (2 p)$ for non-CM elliptic curves follows from automorphy of all symmetric powers $Sym^{k}$ , a functoriality instance, giving analytic continuation of the relevant $L$ -functions.

The Ramanujan–Petersson Conjecture. Optimal bounds $∣ a_{p} ∣ \leq 2 p^{(k - 1) /2}$ for Hecke eigenvalues of cusp forms — proven by Deligne via the Weil conjectures — are the archimedean Ramanujan bound predicted by the correspondence (tempered packets).

Function-field arithmetic and coding/cryptography. Drinfeld's and Lafforgue's work on $GL_{n}$ over function fields underlies the theory of Drinfeld modules and shtukas, with downstream uses in explicit class field theory of function fields.

Zeros of $L$ -functions and analytic number theory. Functoriality (e.g., Rankin–Selberg, symmetric powers) provides the analytic input — meromorphic continuation, non-vanishing on $Re (s) = 1$ — needed for prime-counting and subconvexity results.

Mathematical physics: S-duality. Kapustin–Witten's gauge-theoretic interpretation made geometric Langlands a precise prediction of quantum field theory dualities, linking to topological field theory and the geometric Satake equivalence.

The Birch–Swinnerton-Dyer landscape. The very definition of $L (E, s)$ as a motivic $L$ -function, and its analytic continuation (a consequence of modularity), is prerequisite to even stating BSD at $s = 1$ .

Historical Development

1880s–1920s — Reciprocity foundations. Hilbert's reciprocity laws and the development of class field theory (Takagi, Artin) settle the abelian case, culminating in Artin reciprocity (1927), the prototype of all Langlands reciprocity.

1950 — Tate's thesis. John Tate recasts Hecke $L$ -functions adelically, proving analytic continuation and functional equation via local–global Fourier analysis — the $GL_{1}$ template.

1956–1965 — Harish-Chandra & Selberg. Harish-Chandra builds the harmonic analysis of real reductive groups (discrete series, Plancherel); Selberg introduces the trace formula. Together they create the analytic engine.

1967 — The Langlands letter. Robert Langlands writes to André Weil, conjecturing functoriality, defining the $L$ -group, and predicting the $L$ -function correspondence. Weil has it typed; it circulates as a foundational document.

1970 — Jacquet–Langlands. *Automorphic Forms on $\mathrm{G