The generalized Riemann hypothesis is a conjecture in number theory about the locations of zeros of certain complex analytic functions that encode arithmetic information. In its most common formulation, it asserts that every nontrivial zero of every Dirichlet L-function has real part . These functions distinguish integers by their residue classes, so the conjecture gives precise predictions about the distribution of prime numbers in arithmetic progressions. It includes the ordinary Riemann hypothesis as the special case associated with the Riemann zeta function.
The terminology is not completely uniform. Some authors use “generalized Riemann hypothesis” for analogous assertions about Dedekind zeta functions, Hecke L-functions, or still broader classes of L-functions. The designation “extended Riemann hypothesis,” abbreviated ERH, also has overlapping meanings. Any precise use therefore requires specifying the family of functions involved and the normalization of its critical line.
The conjecture remains unproved. Its importance extends beyond the location of zeros: it would yield strong, effective estimates for primes in prescribed congruence classes, rule out exceptional real zeros of Dirichlet L-functions, and justify deterministic bounds used in computational number theory. It also belongs to a larger framework connecting arithmetic, spectral theory, and the geometry of varieties over finite fields.
Dirichlet characters and the precise statement
A Dirichlet character modulo a positive integer is a periodic function that is completely multiplicative, vanishes on integers not relatively prime to , and restricts to a homomorphism from the group into the nonzero complex numbers. On this group its values are roots of unity. The principal character takes the value on integers relatively prime to and elsewhere.
For a complex number with , the corresponding Dirichlet L-function is defined by
Absolute convergence and unique factorization give its Euler product:
The product runs over all primes. It makes the arithmetic meaning of the function explicit: changing changes how different residue classes contribute.
Every Dirichlet L-function has analytic continuation beyond this initial half-plane. A nonprincipal character gives an entire function, whereas a principal character gives a function with a simple pole at . For the principal character modulo ,
Thus the zeta function occurs directly within the family.
The cleanest statement uses primitive characters. A character modulo is primitive if it is not induced from a character of a proper divisor of ; its conductor is then . Every character is induced from a unique primitive character of some conductor dividing its modulus. For a primitive character, the nontrivial zeros are those in the critical strip . The generalized Riemann hypothesis states that
for every primitive Dirichlet character , including the character of conductor .
Restricting the statement to primitive characters avoids an important ambiguity. An imprimitive L-function differs from its primitive counterpart by finitely many Euler factors. Those extra factors can introduce zeros on , and these are not the zeros to which GRH refers. Even for primitive functions, the qualification “nontrivial” is essential because other zeros arise from the gamma factors in the functional equation.
Functional equations and the critical line
The special role of comes from a symmetry of completed L-functions. Let be primitive of conductor , and define by
Using the gamma function, form
For a primitive nonprincipal character this completed function is entire and satisfies
The constant , called the root number, is determined by a Gauss sum. The completed zeta function has a corresponding functional equation, with poles handled separately.
The transformation exchanges the two sides of the critical strip and fixes its central vertical line. Complex conjugation supplies a second relation:
Together these identities imply that a nontrivial zero of is accompanied by a zero of the same L-function. For a real character there is also symmetry about the real axis. For a nonreal character, conjugation by itself generally transfers a zero to the L-function of the conjugate character.
These symmetries do not prove GRH. An off-line zero can occur together with its reflected partner without violating any functional equation. The conjecture asserts that every nontrivial zero lies on the symmetry line, not merely that the set of zeros is symmetric around it. This distinction is one reason that knowing an analytic continuation and a functional equation is insufficient to establish a Riemann hypothesis.
The gamma factor also explains the trivial zeros. Its poles occur when is a nonpositive even integer. For a primitive nonprincipal character, zeros of cancel these poles: even characters have trivial zeros at , and odd characters at . The zeta function has its familiar trivial zeros at negative even integers; its behavior at differs because the completed zeta function is meromorphic.
For a fixed primitive character, the number of nontrivial zeros with , counted with multiplicity, satisfies
as grows. This counts zeros throughout the critical strip. GRH makes the much stronger assertion that all of them have precisely the same real part.
Historical development and broader formulations
Dirichlet introduced the L-functions bearing his name in his 1837 proof that every reduced arithmetic progression contains infinitely many primes. The essential analytic input was nonvanishing at , not a hypothesis about all zeros. In 1859, Bernhard Riemann studied the zeta function through analytic continuation and its functional equation, and proposed the critical-line assertion that became the ordinary Riemann hypothesis. Extending this assertion to Dirichlet L-functions connected Riemann’s viewpoint with the finer arithmetic distinctions in Dirichlet’s theorem.
The development of algebraic number theory supplied further natural families. For a number field , its Dedekind zeta function is
where the sum ranges over nonzero integral ideals and is the ideal norm. Its Euler product runs over prime ideals. A Riemann hypothesis for predicts that its nontrivial zeros have real part , just as for .
Cyclotomic fields directly connect the two formulations. If , where is a primitive -th root of unity, then factors into Dirichlet L-functions associated with characters of its Galois group. In a precise factorization, each character is represented by its primitive Dirichlet character and the conductor-dependent Euler factors are treated correctly. Consequently, the Riemann hypotheses for all Dedekind zeta functions imply GRH for Dirichlet L-functions. The converse for arbitrary number fields is not known.
Hecke L-functions replace Dirichlet characters by suitable characters associated with ideals or ideles of number fields. They include the Dirichlet family but encode more general abelian arithmetic. Still broader formulations concern Artin L-functions, constructed from representations of Galois groups, and automorphic L-functions, arising from automorphic representations. Their expected critical lines belong to a unified analytic picture, but the hypotheses for these families are not interchangeable.
Normalization matters when moving between these settings. A unitary normalization usually puts the critical line at ; an arithmetic normalization may place it elsewhere. For instance, the usual L-function of an elliptic curve over has a functional equation exchanging and , so its central line is . A shift of variable moves that line to . Likewise, a conjecture about the locations of Artin L-function zeros does not by itself establish the separate Artin holomorphy conjecture.
Prime numbers in arithmetic progressions
The central arithmetic application concerns primes satisfying
This condition is expressed through modular arithmetic. Dirichlet characters provide a finite Fourier analysis on the multiplicative group of reduced residue classes. Their orthogonality relation is
when is relatively prime to ; the identity also gives zero when is not relatively prime to . Here is Euler’s totient function.
A particularly useful counting function is
where the von Mangoldt function is defined by for prime powers and otherwise. The weighted function
then gives
Thus information about the zeros of the L-functions becomes information about primes in individual progressions.
The analytic bridge is the logarithmic derivative:
Contour integration turns this identity into explicit formulas. In these formulas, a nontrivial zero contributes a term of the form . Its size is governed by . GRH places every such contribution at the square-root scale , before the effects of summing over zeros and logarithmic factors are accounted for.
Under Dirichlet GRH, a standard uniform estimate is
The notation is big-O notation, describing a bound up to a constant factor. The main term expresses equal asymptotic shares among the reduced residue classes; the error measures the possible imbalance.
For fixed , partial summation yields
where counts primes in the progression and
This strengthens the prime number theorem for arithmetic progressions by giving a near-square-root error term. It does not imply exact equality between residue classes at finite , nor does the displayed bound provide a useful relative error for every possible modulus size.
Exceptional zeros and unconditional results
The main obstruction to unconditional GRH-quality estimates is the possibility of zeros very close to . Classical zero-free-region theorems show that Dirichlet L-functions have no zeros in a region of the approximate shape
for a suitable positive constant , with a possible exception involving a real zero of a real character. For a primitive nonprincipal character of this kind, the character is quadratic.
Such a hypothetical exceptional zero is commonly called a Siegel zero, or a Landau–Siegel zero. If it is denoted by , with close to , its contribution to a prime-counting formula is proportional to . This can be much larger than the square-root fluctuations predicted by GRH and can create a substantial temporary bias between classes distinguished by the associated quadratic character. GRH excludes exceptional zeros outright.
Siegel’s theorem gives an important but ineffective limitation. For every , there is a positive constant such that an exceptional zero of a primitive real character of conductor satisfies
The theorem does not provide an effectively computable value of the constant in its usual form. “Ineffective” refers to this missing numerical control, not to any uncertainty about the theorem’s validity. It prevents some asymptotic statements derived from the theorem from immediately becoming explicit finite bounds.
A complementary phenomenon, the Deuring–Heilbronn phenomenon, says that an exceptional real zero would repel other zeros away from . This makes exceptional zeros unusually influential: their existence would worsen one part of the picture while improving zero-free regions elsewhere. Arguments can exploit this repulsion to prove results without first resolving whether exceptional zeros exist.
Strong average results also bypass the need for individual GRH estimates. The Bombieri–Vinogradov theorem states, in a standard form, that for every there is a such that
when
There are equivalent formulations with additional maxima over the counting range. The theorem delivers strong average control over moduli up to nearly , sufficient for many sieve applications. Its averaged conclusion must not be confused with the uniform estimate for each individual modulus that follows from GRH.
Least primes, nonresidues, and algorithms
One concrete consequence of GRH is a strong bound for the least prime in a reduced arithmetic progression. Let be the smallest prime congruent to , with . Conditional on Dirichlet GRH, refinements of the explicit-formula method give
for , with an absolute implied constant. For prime moduli, this has essentially quadratic size in the modulus, apart from logarithmic factors.
An unconditional theorem of Linnik states that
for some absolute constant . The distinction is more than qualitative existence: Dirichlet’s theorem ensures that exists, while Linnik’s theorem and the conditional GRH bound control how far a search might have to proceed. The strongest proofs use more than a direct comparison of the simplest unsmoothed counting estimate with its main term; smoothing and careful treatment of zeros improve the resulting logarithmic factors.
A second example is the least quadratic nonresidue. For an odd prime , an integer relatively prime to is a quadratic nonresidue if has no solution. A quadratic Dirichlet character records residues by and nonresidues by . Under GRH, the least positive quadratic nonresidue is
This is an example of how analytic information can guarantee the existence of a small arithmetic witness.
The same principle supports conditional deterministic algorithms. In primality testing, a composite number can often be exposed by a small witness to the failure of a modular identity. Gary Miller’s 1976 primality test has deterministic polynomial running time under an appropriate extended Riemann hypothesis, because the hypothesis supplies bounds on how many small witnesses must be checked. Michael Rabin’s later randomized version leads to the widely used Miller–Rabin test; its probabilistic error guarantee is unconditional.
The 2002 Agrawal–Kayal–Saxena result established deterministic polynomial-time primality testing without GRH. Consequently, GRH is not required for the existence of such a primality test. It remains useful for explicit witness bounds and for other algorithms involving finite fields, algebraic number fields, and class groups.
A conditional running-time result must also be distinguished from conditional correctness. An algorithm may verify its final output unconditionally while using GRH only to justify a stopping threshold or a search bound. In other cases, failure to find an object within a GRH-based range supports a conclusion only under the hypothesis. The exact dependency is part of the mathematical specification of the algorithm.
Number fields and the Chebotarev density theorem
The number-field version of the conjecture concerns prime ideals and the way rational primes split in extensions. For a finite Galois extension , an unramified prime ideal of determines a Frobenius conjugacy class in the Galois group
This class encodes how splits in . The construction brings the analytic study of primes into direct contact with Galois theory.
The Chebotarev density theorem states that, for a conjugacy class , the unramified prime ideals whose Frobenius class is have density
In counting form, if counts such prime ideals of norm at most , the leading term is
This generalizes prime equidistribution in arithmetic progressions: for cyclotomic extensions of , Frobenius classes correspond to residue classes modulo the conductor.
Effective Chebotarev theorems quantify the error and bound the norm of the smallest prime ideal with a prescribed Frobenius class. Under GRH for the relevant Dedekind zeta functions, they give square-root-type counting errors, multiplied by logarithmic and field-dependent factors. Two central parameters are the degree and the absolute discriminant , which measures arithmetic complexity and ramification.
For a nontrivial finite Galois extension of number fields, a standard conditional least-prime consequence has the scale
for an unramified prime ideal with a specified conjugacy class. Precise formulations specify the permissible prime ideals and the hypothesis on the zeta function. Explicit bounds and variants with additional restrictions may include further degree terms. The significant feature is the quadratic dependence on the logarithm of the discriminant, rather than a bound comparable to a fixed power of the discriminant itself.
These statements require careful identification of the assumed GRH. Dirichlet GRH suffices for the classical arithmetic-progression setting, but it is not a substitute for the hypotheses needed for arbitrary nonabelian extensions. Conversely, effective Chebotarev results can often be proved under GRH for a Dedekind zeta function without assuming Artin holomorphy for every nontrivial representation of the Galois group.
Computational applications include finding primes at which a polynomial has a chosen factorization pattern, testing splitting behavior, and obtaining small prime ideals for calculations in algebraic number theory. For a polynomial over , the factorization pattern modulo an unramified prime reflects the cycle structure of its Frobenius element acting on the roots. Chebotarev supplies the distribution law; a suitable GRH supplies much stronger bounds on the size of a prime realizing the pattern.
Numerical verification and statistical evidence
Numerical work tests GRH for specified conductors and bounded ranges of imaginary height. Such a test can be mathematically rigorous, but only within its stated range. There are infinitely many primitive characters, and each corresponding L-function has infinitely many nontrivial zeros. No finite list of verified zeros settles the full conjecture.
For a primitive character, the functional equation permits a phase-adjusted version of that is real-valued for real . Locating sign changes can therefore identify critical-line zeros. Efficient evaluation uses approximate functional equations, which express L-values through finite sums with controlled remainders. Methods involving the fast Fourier transform accelerate the computation of many related character sums or L-values.
A list of sign changes is not by itself a complete verification. It can miss zeros of even multiplicity, zeros between sample points, or zeros away from the line. Rigorous computations combine numerical error bounds with an independent count of all zeros in a region. The argument principle and methods related to Turing’s method provide such counts. When the certified number of zeros found on the line equals the total count in the region, the finite-range verification is complete.
A distinct kind of evidence comes from the statistics of zero spacings. The mean vertical spacing of zeros near height is approximately
After rescaling by this local density, high zeros of suitable L-functions exhibit correlations connected with random matrix theory. Hugh Montgomery’s work on pair correlation for zeta zeros in the 1970s, together with the connection identified by Freeman Dyson, initiated an influential part of this comparison.
For families of L-functions, the behavior of low-lying zeros depends on the symmetry type of the family. Quadratic Dirichlet families are associated with symplectic statistics, while other families lead to unitary or orthogonal models. This is not the same question as the local statistics of high zeros of one fixed function. Distinguishing these regimes is necessary when interpreting numerical histograms or theoretical predictions.
Statistical agreement is evidence for a broader model of L-functions, not a proof of the critical-line assertion. A sparse collection of hypothetical off-line zeros might evade both finite computations and many aggregate tests. Likewise, GRH does not assert that all zeros are simple, and it allows a zero at the central point . Location, multiplicity, central vanishing, and spacing statistics are separate questions.
Geometric analogues and possible proof mechanisms
The strongest proven analogues arise over finite fields. Let be a smooth projective geometrically connected curve of genus over , and put
Its zeta function is
where is a polynomial of degree . Writing
the Riemann hypothesis for curves states that
for every .
André Weil proved this result for curves in the 1940s. It gives the point-counting bound
The square-root term is the geometric counterpart of the square-root errors predicted by GRH for prime counting. Pierre Deligne’s proof in 1974 of the Riemann-hypothesis part of the Weil conjectures extended the relevant absolute-value statements to smooth projective varieties of arbitrary dimension over finite fields.
The mechanism is geometric. Frobenius acts on cohomology groups, and point counts are expressed through traces of its powers. The reciprocal zeros of zeta functions become eigenvalues of these actions. Geometry and cohomological methods then constrain their absolute values. With the substitution , the condition becomes the critical line .
For number fields, no comparably complete geometric construction is known that yields Dirichlet GRH. The analogy nevertheless gives a concrete model of what a proof might accomplish: reinterpret analytically defined zeros as spectral data and derive their location from structural properties. It also explains why the existence of a functional equation alone is far weaker than the desired conclusion; the geometric proof requires additional positivity and cohomological information.
Another proposed mechanism is the Hilbert–Pólya idea. It seeks a suitable self-adjoint operator whose spectral values encode the imaginary parts of nontrivial zeros. The spectral theorem would force such spectral values to be real. If a correspondence identified every nontrivial zero with a point in that spectrum, the critical-line conclusion would follow. Constructing a natural operator with this complete correspondence is the unresolved part, rather than the elementary spectral implication.
Other approaches use explicit formulas, positivity criteria, trace formulas, and links with automorphic forms. Each seeks additional structure beyond analytic continuation, Euler products, and functional equations. A proof for the zeta function alone would not automatically establish the assertion for every character, and a proof for Dirichlet L-functions would not automatically settle all automorphic analogues. The different conjectural families share a pattern, but their arithmetic inputs and known analytic properties remain distinct.