Bombieri-Vinogradov Theorem

• Bomieri-Vinogradov Theorem is an averaged prime-number theorem stating that primes are equidistributed among reduced residue classes for moduli up to the square-root scale, providing a detailed upper-level distribution.
• The theorem describes prime distributions for moduli Q up to x^(1/2) with logarithmic factors, crucial for modular arithmetic.
• By providing a basis for distributions of primes in congruence classes, this theorem is crucial in sieve theory for obtaining asymptotics for prime functions in the set

The Bombieri–Vinogradov theorem is an averaged prime-number theorem for arithmetic progressions. It asserts that primes are equidistributed among reduced residue classes for moduli approaching the square-root scale of the main parameter, although comparable estimates are not known uniformly for every individual modulus. In a standard von Mangoldt formulation, with

$$\psi(x;q,a)=\sum_{\substack{n\le x\ n\equiv a\pmod q}}\Lambda(n), \qquad \psi(x)=\sum_{n\le x}\Lambda(n),$$

one has, for every fixed $A>0$,

$$\sum_{q\le Q}\max_{(a,q)=1} \left|\psi(x;q,a)-\frac{x}{\varphi(q)}\right| \ll_A \frac{x}{(\log x)^A}$$

when

$Q\le \frac{x^{1/2}}{(\log x)^{B(A)}}.$

Here $\Lambda$ is the von Mangoldt function and $\varphi$ is Euler’s totient function. The theorem has level of distribution $1/2$, up to logarithmic factors. It is an averaged assertion over moduli, not a pointwise theorem for each $q$.

1. Classical formulation and analytic meaning

For a reduced residue class $a\bmod q$, the prime-number-theorem prediction is

$\psi(x;q,a)\sim \frac{x}{\varphi(q)}.$

The Bombieri–Vinogradov theorem measures the cumulative discrepancy

$A>0$0

or, in formulations adapted to uniformity in the endpoint,

$A>0$1

A typical statement is

$A>0$2

for

$A>0$3

Equivalent formulations use dyadic intervals $A>0$4, where $A>0$5, or the prime-counting function

$A>0$6

with the expected main term $A>0$7 or $A>0$8.

The exponent $A>0$9 is called the Bombieri–Vinogradov level of distribution. It does not mean that the prime-number theorem is known individually for all $$\sum_{q\le Q}\max_{(a,q)=1} \left|\psi(x;q,a)-\frac{x}{\varphi(q)}\right| \ll_A \frac{x}{(\log x)^A}$$0. Rather, the aggregate error over this family of moduli is smaller than every prescribed fixed power of $$\sum_{q\le Q}\max_{(a,q)=1} \left|\psi(x;q,a)-\frac{x}{\varphi(q)}\right| \ll_A \frac{x}{(\log x)^A}$$1.

The theorem is a central input in sieve theory, including bounded-gap results, shifted-prime problems, almost-prime problems, and asymptotic formulas involving primes in families of congruence classes.

2. Character decomposition and the proof architecture

For $$\sum_{q\le Q}\max_{(a,q)=1} \left|\psi(x;q,a)-\frac{x}{\varphi(q)}\right| \ll_A \frac{x}{(\log x)^A}$$2, character orthogonality gives

$$\sum_{q\le Q}\max_{(a,q)=1} \left|\psi(x;q,a)-\frac{x}{\varphi(q)}\right| \ll_A \frac{x}{(\log x)^A}$$3

where

$$\sum_{q\le Q}\max_{(a,q)=1} \left|\psi(x;q,a)-\frac{x}{\varphi(q)}\right| \ll_A \frac{x}{(\log x)^A}$$4

After subtracting the principal-character contribution, the progression discrepancy is reduced to averaged estimates for twisted von Mangoldt sums.

The standard proof uses Vaughan’s identity, which decomposes $$\sum_{q\le Q}\max_{(a,q)=1} \left|\psi(x;q,a)-\frac{x}{\varphi(q)}\right| \ll_A \frac{x}{(\log x)^A}$$5 into four components,

$$\sum_{q\le Q}\max_{(a,q)=1} \left|\psi(x;q,a)-\frac{x}{\varphi(q)}\right| \ll_A \frac{x}{(\log x)^A}$$6

The resulting sums have different structures:

• Small-variable terms are treated by elementary estimates.
• Type I sums are essentially linear character sums.
• Type II sums are bilinear sums involving two variables of comparable size.
• The fourth Vaughan term contains a bilinear structure with Möbius-weighted coefficients.

The fundamental analytic input is the multiplicative large sieve: $$\sum_{q\le Q}\max_{(a,q)=1} \left|\psi(x;q,a)-\frac{x}{\varphi(q)}\right| \ll_A \frac{x}{(\log x)^A}$$7 Here the star denotes primitive characters. Cauchy–Schwarz, dyadic decomposition, Perron-type Fourier inversion, and estimates for divisor coefficients convert this second-moment inequality into bounds for the Type I and Type II components.

The transition from primitive to imprimitive characters is essential. If $$\sum_{q\le Q}\max_{(a,q)=1} \left|\psi(x;q,a)-\frac{x}{\varphi(q)}\right| \ll_A \frac{x}{(\log x)^A}$$8 modulo $$\sum_{q\le Q}\max_{(a,q)=1} \left|\psi(x;q,a)-\frac{x}{\varphi(q)}\right| \ll_A \frac{x}{(\log x)^A}$$9 is induced by a primitive character $Q\le \frac{x^{1/2}}{(\log x)^{B(A)}}.$0, then

$Q\le \frac{x^{1/2}}{(\log x)^{B(A)}}.$1

which is bounded by $Q\le \frac{x^{1/2}}{(\log x)^{B(A)}}.$2. Multiplicity estimates for the number of induced characters and partial summation then transfer primitive-character estimates to progressions.

The classical argument uses the Siegel–Walfisz theorem for small moduli. Its constants are ineffective because of the possible existence of a Siegel zero, a real zero of a Dirichlet $Q\le \frac{x^{1/2}}{(\log x)^{B(A)}}.$3-function extremely close to $Q\le \frac{x^{1/2}}{(\log x)^{B(A)}}.$4. Consequently, the standard Bombieri–Vinogradov theorem has an ineffective implied constant.

3. Quantitative refinements and effective variants

Several refinements improve the logarithmic losses or modify the family of moduli.

Dress, Iwaniec, and Tenenbaum obtained a formulation in which the relevant loss is $Q\le \frac{x^{1/2}}{(\log x)^{B(A)}}.$5. Sedunova improved this to $Q\le \frac{x^{1/2}}{(\log x)^{B(A)}}.$6 by using a weighted Vaughan identity and an estimate of Barban, Vehov, and Graham for smoothly truncated Möbius divisor sums. Her central primitive-character estimate is

$Q\le \frac{x^{1/2}}{(\log x)^{B(A)}}.$7

The improvement is logarithmic; it does not increase the level $Q\le \frac{x^{1/2}}{(\log x)^{B(A)}}.$8.

Akbary and Hambrook established a fully effective variant for moduli whose least prime divisor is large. If $Q\le \frac{x^{1/2}}{(\log x)^{B(A)}}.$9 denotes the least prime factor of $\Lambda$0, then, for $\Lambda$1, they prove an estimate over moduli satisfying

$\Lambda$2

The restriction removes moduli with small prime factors, which are the source of the exceptional-character difficulty. Their explicit constants include

$\Lambda$3

Sedunova improved the corresponding logarithmic factor from $\Lambda$4 to $\Lambda$5, with

$\Lambda$6

where

$\Lambda$7

These results are explicit but restricted: they do not make the full classical theorem effective over all moduli.

A later effective sieve framework shows that many classical sieve consequences can nevertheless be made effective without changing their leading asymptotic term. The method isolates the possible exceptional modulus and deletes its prime divisors from the sieve decomposition. For standard applications, it yields, for example, the effective upper bound

$\Lambda$8

where

$\Lambda$9

is the twin-prime constant. This does not make the standalone Bombieri–Vinogradov theorem effective; it makes a broad class of sieve-theoretic applications effective.

4. Extensions to sparse and structured sequences

The Bombieri–Vinogradov method extends beyond the ordinary prime sequence, but the attainable level depends strongly on the structure of the sequence.

Smooth numbers

For

$\varphi$0

define

$\varphi$1

Harper proved Bombieri–Vinogradov and Barban–Davenport–Halberstam analogues for $\varphi$2-smooth numbers in the range

$\varphi$3

The expected progression main term is

$\varphi$4

His Bombieri–Vinogradov estimate holds for

$\varphi$5

while the corresponding mean-square theorem permits

$\varphi$6

The relevant parameter is

$\varphi$7

When $\varphi$8 is small, smooth numbers are sparse, and a direct large-sieve argument does not detect sufficient sparsity. Harper combines zero-density estimates with factorization of smooth numbers and a direct large-sieve treatment of large conductors. Possible exceptional zeros are less damaging because the smooth Euler product is exponentiated. The individual character-sum estimates contain decay of the form

$\varphi$9

The result extends the earlier range

$1/2$0

to the polylogarithmic range $1/2$1 (Harper, 2012).

Multiplicative functions

For a multiplicative function $1/2$2, write

$1/2$3

Granville and Shao study the class

$1/2$4

where

$1/2$5

For general $1/2$6, a prime-strength theorem cannot hold without additional hypotheses. Two obstructions occur:

1. Small-conductor correlations: $1/2$7 may correlate strongly with a nonprincipal character of small conductor.
2. Large-prime behavior: the values $1/2$8 for primes near $1/2$9 can be chosen adversarially even when $q$0 satisfies Siegel–Walfisz estimates.

After removing finitely many exceptional characters, the authors obtain Bombieri–Vinogradov-type estimates below $q$1. For functions supported on $q$2-smooth numbers, arbitrary logarithmic savings are recovered in an appropriate range. For a fixed residue class $q$3, they obtain a beyond-$q$4 result up to

$q$5

The exponent $q$6 arises from bilinear Kloosterman-fraction estimates of Bettin and Chandee; it does not give a theorem with a maximum over residue classes (Granville et al., 2017).

Products of primes

For the indicator

$q$7

where $q$8 counts prime factors with multiplicity, a Bombieri–Vinogradov-type theorem with additive twists has been proved for

$q$9

The averaged expression is

$a\bmod q$0

The main term is not simply the total sum divided by $a\bmod q$1, because $a\bmod q$2 has a nontrivial correlation with $a\bmod q$3. The proof constructs a Vaughan-type Type I/Type II decomposition rather than applying Vaughan’s identity literally (Bazin, 16 Jul 2026).

5. Number-field, automorphic, and geometric analogues

Number-field variants replace rational primes and ordinary residue classes by prime ideals and ray classes.

For a finite Galois extension $a\bmod q$4, a conjugacy class $a\bmod q$5 defines a Chebotarev condition. A short-interval Bombieri–Vinogradov theorem gives, for suitable $a\bmod q$6,

$a\bmod q$7

with $a\bmod q$8, subject to

$a\bmod q$9

The field $\psi(x;q,a)\sim \frac{x}{\varphi(q)}.$0 is the fixed field of the largest abelian subgroup $\psi(x;q,a)\sim \frac{x}{\varphi(q)}.$1 meeting $\psi(x;q,a)\sim \frac{x}{\varphi(q)}.$2. The proof reduces the nonabelian Frobenius condition to families of Hecke characters and applies Hecke $\psi(x;q,a)\sim \frac{x}{\varphi(q)}.$3-function zero-density estimates (Thorner, 2014).

For primes in a narrow ideal class and a shrinking sector in a finite Galois extension $\psi(x;q,a)\sim \frac{x}{\varphi(q)}.$4, one obtains Bombieri–Vinogradov estimates in short intervals. If the sector has radius $\psi(x;q,a)\sim \frac{x}{\varphi(q)}.$5, its density contributes a factor

$\psi(x;q,a)\sim \frac{x}{\varphi(q)}.$6

to the main term. The permissible range satisfies

$\psi(x;q,a)\sim \frac{x}{\varphi(q)}.$7

where $\psi(x;q,a)\sim \frac{x}{\varphi(q)}.$8. The proof uses Hecke characters, Selberg majorants and minorants, Mellin transforms, zero-free regions, and zero-density estimates (Khale et al., 2020).

For real quadratic fields of class number one, prime elements can be restricted simultaneously by genuine congruence conditions in $\psi(x;q,a)\sim \frac{x}{\varphi(q)}.$9 and by sectors in the Minkowski embedding. The resulting estimate has the form

$A>0$00

The genuine level of distribution is $A>0$01. The loss from the classical $A>0$02 level is attributed to the current Hecke large-sieve estimate (Baier et al., 3 Sep 2025).

For a self-dual cuspidal automorphic representation $A>0$03 of $A>0$04 over a number field $A>0$05, the prime-ideal coefficients $A>0$06 satisfy a Bombieri–Vinogradov theorem in narrow ray classes. If

$A>0$07

then the level is

$A>0$08

Thus the level is essentially $A>0$09 for $A>0$10, and $A>0$11 for $A>0$12. The proof uses Rankin–Selberg positivity, a generalized Vaughan decomposition, automorphic large sieves, and an unconditional Siegel-type lower bound for quadratic twists (Jiang et al., 2021).

6. Well-factorable weights, exceptional moduli, and structured moduli

The square-root barrier can be surpassed when the averaging weights have additional factorization.

A sequence $A>0$13 is well-factorable of level $A>0$14 if, for every factorization $A>0$15, it can be written as

$A>0$16

where $A>0$17 and $A>0$18 are $A>0$19-bounded and supported on $A>0$20 and $A>0$21. Bombieri, Friedlander, and Iwaniec proved a prime-distribution theorem with well-factorable weights at level

$A>0$22

More recent convolution-type results combine well-factorable weights with bilinear and trilinear estimates, incomplete Kloosterman sums, and triply-well-factorable decompositions. The admissible modulus exponent depends on the convolution length and is described by piecewise functions such as $A>0$23, with values at least $A>0$24 and reaching $A>0$25 in some ranges (Yang, 13 Aug 2026).

Structured families of moduli can also yield unusually small exceptional sets. Baker proved that if $A>0$26 is a pairwise coprime set and

$A>0$27

then the number of $A>0$28 for which

$A>0$29

is

$A>0$30

The family may consist of prime powers, and the conclusion is uniform in $A>0$31 and reduced residue classes. The point is not a larger raw modulus range than classical Bombieri–Vinogradov, but a polylogarithmic exceptional set inside a structured family (Baker, 2019).

For prime powers $A>0$32 with $A>0$33, Baier and Pujahari extended the structured-modulus range to

$A>0$34

with at most

$A>0$35

exceptional moduli. The proof exploits the exact identity

$A>0$36

which allows a $A>0$37-adic form of Harman’s sieve to compare progressions modulo $A>0$38 and $A>0$39 (Baier et al., 2021).

Polynomially parametrized moduli provide another form of structure. For

$A>0$40

the moduli are products of rational primes represented as norms of Gaussian primes. A polynomial large sieve, ultimately based on multidimensional Vinogradov mean-value estimates, improves the standard variable range from approximately

$A>0$41

to

$A>0$42

where $A>0$43 reflects the polynomial saving. This is a theorem for rational primes in progressions to specially parametrized rational moduli, not a Bombieri–Vinogradov theorem over $A>0$44 (Halupczok, 2016).

7. Further applications and scope

Bombieri–Vinogradov estimates serve as distribution inputs in a wide range of arithmetic arguments.

For prime values of a primitive positive definite binary quadratic form

$A>0$45

a number-field formulation gives

$A>0$46

for every $A>0$47, under the stated discriminant hypotheses. Here $A>0$48 counts admissible residue classes. This leads, through the Maynard sieve, to bounded gaps between prime values of $A>0$49, including generalizations of bounded gaps between Gaussian primes (Lam, 2018).

For the weighted sequence $A>0$50, where

$A>0$51

and

$A>0$52

an enveloping-sieve argument proves Bombieri–Vinogradov-type distribution for primes of the form $A>0$53 in the polylogarithmic range

$A>0$54

The theorem concerns the representation-weighted sequence $A>0$55, not an unweighted count of distinct represented primes (Dimitrov, 2024).

Nilsequence twists give a different hierarchy of levels. For $A>0$56, with $A>0$57 a bounded-complexity nilsequence, the strongest unrestricted theorem in the cited work has level $A>0$58 when the nilsequence may vary with the modulus, level $A>0$59 for a fixed nilsequence, and level $A>0$60 for fixed residue classes with well-factorable weights. Polynomial phases $A>0$61 are included as nilsequences. Applications include bounded gaps between primes in nil-Bohr sets, distribution of Chen primes in nil-Bohr sets, and Green–Tao-type results for primes in progressions for almost all moduli (Shao et al., 2020).

In the function-field setting $A>0$62, the integer correspondence

$A>0$63

produces a Bombieri–Vinogradov theory at degree level $A>0$64. An induction principle shows that if two arithmetic functions satisfy suitable growth, Siegel–Walfisz, and Bombieri–Vinogradov properties, then their Dirichlet convolution does as well. This yields distribution results for divisor functions and products of irreducible polynomials, including a function-field version of the Titchmarsh divisor problem (Dey et al., 2023).

The theorem’s principal limitations remain structural. The classical result is ineffective because of possible Siegel zeros; effective variants generally exclude small-prime-factor moduli or alter the conclusion. The level $A>0$65 is not a pointwise threshold. Extensions beyond $A>0$66 require additional structure, such as well-factorable weights or a fixed residue class. For sparse sequences, geometric sequences, automorphic coefficients, nilsequence twists, or structured moduli, the attainable level is governed by the relevant large-sieve, zero-density, bilinear, trilinear, and dispersion estimates rather than by the classical theorem alone.