Primes in Short Intervals: The Guth–Maynard Breakthrough Explained

The Guth–Maynard breakthrough moves the prime number theorem into shorter intervals. For every fixed ε > 0, it gives the expected asymptotic number of primes in intervals whose length is at least x^17/30+ε, once x is large enough. The earlier uniform threshold was x^7/12+ε.

This is a statement about prime density. It says much more than “there is a prime somewhere in the interval.” Near a large number x, an interval of length y should contain about y / log x primes. Guth and Maynard proved that this prediction remains valid for a shorter range of y than was previously known without assuming the Riemann Hypothesis.

The work first appeared as a preprint in May 2024. The revised paper by Larry Guth and James Maynard was published in the Annals of Mathematics in 2026.

What a short interval means

A short interval has the form

where x is large and y is smaller than x. The number of primes inside it is

The prime number theorem says that primes near x have average spacing close to log x. That leads to the prediction

If y is comparable with x, this prediction is relatively accessible. The proof becomes harder as y shrinks. Local fluctuations matter more, and the error terms tied to zeros of the Riemann zeta function become harder to control.

Why the interval length is written as x^θ

Number theorists often set y = x^θ. The exponent θ provides a clean scale:

• θ = 1 gives an interval whose length is comparable with x.
• θ < 1 gives an interval that becomes small relative to x.
• A smaller θ gives a stronger short-interval theorem.

The movement from 7/12 to 17/30 may look small in decimal form. It grows into a large multiplicative saving when x becomes very large.

The theorem in precise form

For every fixed ε > 0, Guth and Maynard prove that when

the prime count satisfies

The first term, y / log x, is the predicted count. The second term is an error that becomes small relative to the main term as x grows.

Why the ε cannot be removed

The notation x^17/30+ε means that the result holds for every fixed exponent slightly larger than 17/30. The value of ε may be chosen as small as desired, but the theorem does not set ε = 0.

The paper also describes the scale as x^17/30+o(1). This records the limiting exponent while allowing a term that tends to zero. For a general audience, 17/30 + ε is usually easier to interpret.

Every interval and almost every interval

The paper contains two short-interval results. They answer different questions and should not be merged into one claim.

The uniform theorem allows no bad starting points once the unstated size threshold has been crossed. The almost-all theorem permits an exceptional set, so it reaches much shorter intervals:

For all but a quantitatively controlled set of integers x in [X, 2X], the interval [x, x + y] then contains the expected asymptotic number of primes.

Prime existence is a weaker question

Results about primes in short intervals can make three different claims:

1. An interval contains at least one prime.
2. An interval contains at least a stated number of primes.
3. An interval contains the predicted asymptotic number of primes.

Guth–Maynard belongs to the third category. It controls the whole prime count, up to a relatively small error.

Why the exponent 0.525 does not replace 17/30

A well-known result of Baker, Harman and Pintz reaches intervals of length roughly x^0.525. Since 0.525 < 17/30, those intervals are shorter. The conclusion is also weaker: it supplies a lower-bound or existence-type result rather than the full asymptotic formula y / log x.

The barrier left by Huxley

The former uniform exponent 7/12 came from work by Martin Huxley in 1972. For more than five decades, the same exponent marked the best unconditional range for a prime number theorem valid in every short interval.

Guth and Maynard lower it to

The difference is exactly 1/60. This gives a simple ratio between the old and new interval lengths:

At x = 10⁶⁰, the new scale is ten times shorter. At x = 10¹²⁰, it is one hundred times shorter.

Why zeta zeros enter a problem about primes

The distribution of primes is linked to the Riemann zeta function. A convenient weighted counting function is

where the von Mangoldt function Λ(n) places a logarithmic weight on primes and prime powers. An explicit formula expresses this count as a main term together with contributions from the nontrivial zeros of the zeta function.

Zeros whose real part lies close to 1 can create large errors. The Riemann Hypothesis would place every nontrivial zero on the line with real part 1/2, but that remains unproved. A different approach counts how many zeros may lie to the right of a chosen vertical line.

The zero-density estimate behind 17/30

Let N(σ, T) count zeta zeros ρ satisfying

A zero-density estimate bounds this quantity. Guth and Maynard obtain, after combining their new estimate with an earlier bound of Ingham,

The constant 30/13 ≈ 2.3077 improves the former constant 12/5 = 2.4 used in this setting. The smaller exponent allows fewer zeros in the dangerous right-hand region of the critical strip. Their total contribution to the short-interval error can then be controlled for a smaller y.

Where 17/30 comes from

The connection between the density constant and the interval exponent is visible in the identity

The proof must also balance truncation and zero-free-region errors, so this identity is not the entire argument. It does show why 17/30 is tied to the new 30/13 density bound rather than being an isolated numerical choice.

Dirichlet polynomials and their large values

A Dirichlet polynomial in the paper has the form

Each term has magnitude controlled by its coefficient, while its angle changes with t log n. For many values of t, cancellation keeps the sum moderate. At special values, many terms align well enough to make |D(t)| large.

The central analytic question is:

The hardest range for several prime-distribution problems occurs near V = N^3/4. Earlier mean-value and large-value estimates met at a common obstruction there. Guth and Maynard improve the count in this range.

In the zero-density application, the older methods produce a bound of about T^3/5+o(1) in the limiting case. The new estimate lowers it to about T^13/25+o(1).

Structured and unstructured large-value sets

Let W be a separated set of points where |D(t)| is large. The pattern inside W matters.

Its additive energy counts approximate relations of the form

A set with high additive energy contains many such relations. It has strong additive organization, as an arithmetic progression does. Earlier methods already gave useful control in that case.

Low-energy sets were harder. Guth and Maynard developed estimates suited to that regime. The two cases then fit together:

• High energy: existing difference-set estimates limit how often large values can occur.
• Low energy: the new method extracts cancellation that earlier arguments did not capture.

This division prevents the large-value set from escaping through either type of behavior.

Matrices, singular values and preserved cancellation

The values of a smoothed Dirichlet polynomial on W can be encoded in a matrix whose entries look like

Large values of the polynomial force the largest singular value of this matrix to be large. Guth and Maynard estimate it through traces of powers of MM*, with the cubic trace playing a central role.

They then apply Poisson summation. A standard route would simplify the resulting Fourier integrals early by stationary phase. Their argument keeps those integrals unsimplified for longer. This preserves separation among variables and exposes cancellation that would otherwise be hidden inside coupled coefficients.

How the analytic steps produce a prime theorem

1. Bound the large values

The new estimate limits the number of separated t-values at which a Dirichlet polynomial can be unusually large, especially near the difficult N^3/4 scale.

2. Convert that bound into zero density

Zero-detection methods associate zeta zeros to large values of suitable Dirichlet polynomials. Better large-value control gives a smaller upper bound for N(σ, T).

3. Insert the density bound into the explicit formula

The contribution of zeros to the prime-counting error is summed according to their real parts and heights. The new density exponent reduces this total.

4. Balance the remaining errors

The truncation height and the interval length are chosen so that every error remains smaller than the main term. This balance works when y ≥ x^17/30+ε.

Compare the old and new interval scales

These numbers compare exponents only. They do not establish that the asymptotic theorem is numerically accurate at the displayed values of x.

What the theorem does not settle

It does not prove the Riemann Hypothesis

The proof improves a bound on how many zeta zeros may lie in parts of the critical strip. It does not place every nontrivial zero on the critical line.

It does not reach the square-root scale

The exponent 17/30 ≈ 0.5667 remains above 1/2. Under the Riemann Hypothesis, standard explicit-formula reasoning points toward uniform asymptotic results for intervals a little longer than x^1/2. Reaching or passing that scale without such an assumption remains open.

It is not an algorithm for locating the next prime

The theorem is asymptotic and does not state an explicit starting value. It cannot replace direct computation when the task is to test a particular integer or search a finite interval. For concrete examples, the Prime Number Checker can verify individual candidates, while the theorem describes the long-range density pattern behind such calculations.

It does not solve prime gaps

Short-interval prime counts and gaps between consecutive primes are related, yet they ask different questions. A theorem that gives the correct average count throughout an interval is not automatically the sharpest theorem about the largest possible gap.

Why the method may travel further

The lasting mathematical value lies in the large-value estimate itself. Similar Dirichlet polynomials occur in zero-density problems, primes in arithmetic settings, and other questions involving multiplicative functions.

The method also leaves a clear technical obstacle. Its energy estimates use the pointwise coefficient condition |b_n| ≤ 1. Extending comparable bounds to broader coefficient classes could open further applications.

Open thresholds after Guth–Maynard

The next uniform target is any reduction below 17/30. Reaching 1/2 would require sharper control of zeta zeros or a different way to limit their contribution. The wide gap between the uniform exponent 17/30 and the almost-all exponent 2/15 also shows how much harder it is to rule out every exceptional starting point.

Questions about the result