PRIME TUPLES AND SIEGEL ZEROS

THOMAS WRIGHT
arXiv:2111.14054v5 [math.NT] 19 Oct 2023

Abstract. Under the assumption of infinitely many Siegel zeroes s with

Re(s) > 1 − (log1q)R for a sufficiently large value of R, we prove that there
exist infinitely many m-tuples of primes that are ≪ e1.9828m apart. This
”improves” (in some sense) on the bounds of Maynard-Tao, Baker-Irving, and
Polymath 8b, who found bounds of e3.815m unconditionally and me2m as-
suming the Elliott-Halberstam conjecture; it also generalizes a 1983 result of
Heath-Brown that states that infinitely many Siegel zeroes would imply in-
finitely many twin primes. Under this assumption of Siegel zeroes, we also
improve the upper bounds for the gaps between prime triples, quadruples,
quintuples, and sextuples beyond the bounds found via Elliott-Halberstam.

1. Introduction
One of the most exciting recent breakthroughs in number theory has been the
much-celebrated progress toward the twin prime and Polignac conjectures. The
current wave of excitement was sparked by Zhang [Zh], who proved in 2013 that

H1 ≤ 70, 000, 000,

where Hm is the smallest number such that there are infinitely many intervals of
length Hm containing at least m + 1 primes. Maynard and Tao [Ma] were later
able to establish a slightly different method that would eventually (with the help
of Polymath 8b [Po]) reduce this bound to

H1 ≤ 246.

Moreover, Maynard and Tao (as well as Baker and Irving [BI], Stadlmann [St], and
the Polymath project) were able to bound Hm for larger values of m as well; they
gave bounds for m = 2 to 5, as well as a general upper bound for Hm . Those results
have been iteratively improved from the initial work of Maynard and Tao - we list
here the best current results, as found in [St]:

H2 ≤ 396, 516,
H3 ≤ 24, 407, 016,
H4 ≤ 1, 391, 051, 532,
H5 ≤ 77, 510, 685, 234,
Hm ≪ e3.8075m .

There has also been a great deal of work done on these results under the assump-
tion of certain conjectures. In particular, if one assumes the Elliott-Halberstam
conjecture, one can find even better results (the bounds below are listed in [Po],
1
2 T. WRIGHT

though the H1 bound first appears in [Ma]):

H1 ≤ 12,
H2 ≤ 270,
H3 ≤ 52, 116,
H4 ≤ 474, 266,
H5 ≤ 4, 137, 854,
Hm ≪ me2m
If one assumes the generalized Elliott-Halberstam conjecture, one can make even
further progress on the bounds on the first two cases (see [Po]), finding that
H1 ≤ 6,
H2 ≤ 252;
though there is of yet no progress on cases with four or more primes.
Note that the conjectures mentioned above assume that primes generally behave
well. The Elliott-Halberstam conjecture assumes that the primes up to x behave
nicely mod q (on average) for all q < xθ as long as θ < 1; the generalized Elliott-
Halberstam conjecture goes even further, assuming that all sequences that obey
certain conditions behave nicely mod q (on average) for q < xθ with θ < 1.

2. Introduction: Siegel Zeroes

There are, however, other countervailing conjectures that assume that primes
do not behave nicely in certain circumstances. One of the most famous is the
assumption of Landau-Siegel zeroes, which is explained as follows.
For a character χq modulo q, we define as usual
∞
X χq (n)
L(s, χq ) = .
n=1
ns
In 1918, Landau [La] proved the following:
Theorem. (Landau 1918) There exists a constant c such that for any q and any
real-valued character χq mod q, there exists at most one s = σ + it for which
L(s, χq ) = 0 and
c
Re(s) > 1 − .
log q(|t| + 2)
.
Such a zero, if it is to exist, is known as a Siegel zero, a Landau-Siegel zero, or an
exceptional zero, and the associated character is known as an exceptional character.
It is known that if such a zero exists, it must be real (i.e. t = 0) and the character
must be real-valued as well.
We note that in our definition above (and indeed in the literature about the
topic), the definition of a Siegel zero is a bit ill-defined in that it depends on the
choice of a constant c in the equation above. Generally, at the beginning of a paper
assuming the existence of Siegel zeroes, the authors must declare which specific
definition of Siegel zero they will require for their theorem; we will do so here
shortly.
 PRIME TUPLES AND SIEGEL ZEROS 3

Before we do that, however, we recall the two theorems that give us the best
known upper bounds for Siegel zeroes. The first is the best effective bound; the
second is the best ineffective bound:

Theorem. (Landau 1918) There exists an effectively computable constant C such

that for any q and any real-valued character χq mod q, a Siegel zero β must satisfy
C
β <1− 1 .
q log2 q
2

Theorem. (Siegel 1935) For any ϵ there exists a constant C(ϵ) such that

β < 1 − C(ϵ)q −ϵ .

Of course, mathematicians do not expect such a zero to exist for any q or χq ;

in fact, the generalized Riemann hypothesis predicts all non-trivial zeroes will have
Re(s) = 21 . Nevertheless, we cannot yet rule out the existence of Siegel zeroes.
In fact, the world of Siegel zeros has been described as “an ‘illusory’ or ‘ghostly’
parallel universe in analytic number theory that should not actually exist, but is
surprisingly self-consistent and to date proven to be impossible to banish from the
realm of possibility1.”
We reiterate that if such an s and χq exist with χq being a primitive character
mod q, χq must also be a real-valued character mod q. This means that there
must exist a fundamental discriminant M |q such that χq (n) is derived from the
M


Kronecker symbol n .
Surprisingly, the existence of a Siegel zero often yields unusually nice results,
including results related to prime gaps. Perhaps the nicest example of such a result
is the following, which appears in [HB]:

Theorem. (Heath-Brown 1983) Let χq be a character mod q, and let β be such

that L(β, χq ) = 0 with
1
β >1− .
3 log q
If there are infinitely many χq for which such a β exists then there are infinitely
many twin primes.

In particular, Heath-Brown found that for any such q, the number of twin primes
(p, p + 2) with q 250 ≤ p ≤ q 500 is as conjectured in the Hardy-Littlewood conjec-
tures2.
Unfortunately, Heath-Brown’s methods do not appear to generalize to tuples
beyond pairs. However, with the new results on prime tuples, it seems natural to
ask whether Siegel zeroes might be of help in the more general prime tuple case.
This is the motivating question of our current paper.

1This quote appears in Terence Tao’s blog article “Heath-Brown’s theorem on prime twins and
Siegel zeroes,” published 26 August 2015.
1
2These bounds on q have since been widened by Tao and Teräväinen to q 20.5+ϵ ≤ p ≤ q η 2 ,
log q
where η = 1−β
[TT].
4 T. WRIGHT

3. Main Result
We assume a somewhat stronger condition on the exceptionality of the Siegel
zeroes to prove the following:
Main Theorem 1. Fix an A > 2, and let r = 554, 401. Assume that there
are infinitely many D and χD such that for each D, there exists a real sD with
L(sD , χD ) = 0 and
1
sD > 1 − .
(log D)rr +A
Then for any m ≥ 1,
Hm ≪ e1.9828m .
For smaller values of m, we also find the following:
Main Theorem 2. With the same assumptions about Siegel zeroes as given in the
previous theorem, we have
H2 ≤ 264,
H3 ≤ 49, 192,
H4 ≤ 439, 812,
H5 ≤ 3, 775, 860.
We note that all of these are better results than the assumption of Elliott-
Halberstam, although our H2 result is still a lesser bound than the one given by
generalized Elliott-Halberstam. In the case of m = 1, our methods give H1 ≤ 12,
which, while on par with Elliott-Halberstam, is obviously majorized by Heath-
Brown’s result and hence is not further discussed here.

4. Methods
The key result we will use will be one of Friedlander and Iwaniec [FI03], who
proved the following:
Theorem 4.1. (Friedlander-Iwaniec 1992) Let χD be a real character mod D. Let
233
x > Dr with r = 554, 401, let q < x 462 , and let (a, q) = 1. Then
   
π(x) aD  r
(1) π(x, q, a) = 1 − χD + O L(1, χD )(log x)r ,
ϕ(q) (q, D)
where the constant is absolute and computable.
233
We note here that while Friedlander and Iwaniec claim this theorem for q < x 462 ,
their work actually proves this theorem for
58(1− 1 )
r
q < x 115 = x.5043469... ,
which is a slightly better bound (since 233/462 = .504329...). Friedlander and
Iwaniec simply chose the above to simplify the statements of their main theorems;
however, this additional leeway will slightly improve our results3.
Now, Theorem 4.1 is true for any choice of χD or D, but it is only nontrivial
when L(1, χD ) is small, which means that the character must be exceptional. Note
3See [FI03], Page 2049 for more information; Friedlander’s and Iwaniec’s simplification of
results occurs between (5.11) and Proposition 5.2. Friedlander and Iwaniec themselves note that
they have shrunk the bounds slightly to make the result easier to state.
 PRIME TUPLES AND SIEGEL ZEROS 5

that if s is a Siegel zero then L(1, χD ) ≪ (1 − Re(s)) log2 D by mean value theorem
(since L′ (s, χD ) ≪ log2 D on the interval between the Siegel zero and 1), so the
hypothesis on Siegel zeroes in the Main Theorem is sufficient to prove that the error
term here is less than the main term.
In their paper, Friedlander and Iwaniec highlight the fact that if q and D are
relatively prime, one can find equidistribution of primes in arithmetic progressions
233 1
mod q as long as q < x 462 , which is a better result than the x 2 −ϵ that is achievable
under GRH. In our case, however, the more interesting result is what happens when
D|q and χD (a) = −1:
Corollary 4.2. Fix A and r as in Main Theorem 1, and assume that the conditions
on Siegel zeroes in that theorem are satisfied as well. Let D|q and χD (a) = −1. If
q < x 115 (1− r ) then
58 1

 
2π(x) π(x)
π(x, q, a) = +O
ϕ(q) ϕ(q)(log x)A−2
In other words, the number of primes is double what one would normally expect
in a congruence class. This is the key insight that will allow our main theorems to
be proven.
In order to explain this insight, let us define θ′ to be the level of distribution of
the primes, which is to say the supremum of the exponents t for which
X π(x) x
max π(x, q, a) − ≪
q≤xt
(a,q)=1 ϕ(q) logB x
for every B > 0.
In [Ma] and [Po], the authors define a quantity Mk as the supremum of the
quotient of two integrals (see (3)-(5) below for a more specific definition). For
practical purposes, however, this quantity gives information about the ratio between
S2 , the sieved sum over the primes, and S1 , the sum of the sieve weights themselves.
The work of [Ma] establishes that for us to find tuples of m + 1 primes in intervals
of length k, we require
2m
Mk > ′ .
θ
In our case, let θ denote the largest value for which (1) holds for every q ≤ xθ . If we
choose all of our primes from a congruence class a modulo D for which χ(a) = −1
then by the corollary above, we will have double the expected number of primes;
thus, the size of S2 is also doubled, and hence we only require
m
Mk > .
θ
While our θ isn’t as large as θ′ would be under Elliott-Halberstam (which allows
any θ′ < 1), the excess of primes has the same effect as would doubling the value of
θ′ . Since our θ is also greater than 1/2 according to Theorem 4.1, this means that
our results actually surpass those assuming Elliott-Halberstam.

5. Remarks
Obviously, no one has made conjectures as to what the largest possible value for θ
might be for (1) or Corollary 4.2 to hold, since most mathematicians do not believe
there are Siegel zeroes at all. In the paper of Friedlander and Iwaniec, the authors
58
encounter an obstruction at θ = 115 (1 − 1r ) from the error term and then another
6 T. WRIGHT

one at θ = 2/3 from the main term. The authors remark that while overcoming
the first bound might be tractable, the second one seems a bit more difficult to
improve. The reason for this is that the first bound is derived from a paper on
ternary divisor sums [FI85] where the main result has since been improved; the
improved version is not yet known to be amenable to the methods of [FI03], but
it is not hard to imagine that one could find a way to apply the improved result.
The second bound, however, comes from the famous Weil bound for Kloosterman
sums; as such, any improvement over 2/3 would likely require either a method of
estimating the main term that avoids using this bound or else an improvement on
the Weil bound itself.
On the other hand, conjectures about primes in arithmetic progressions (such
as Montgomery’s conjecture on arithmetic progressions [Mo1], [Mo2] or Elliott-
Halberstam) say that the equidistribution of primes in classes should apply modulo
′
q < xθ for every θ′ < 1. Let’s imagine for a moment that one could make a similar
conjecture about non-equidistribution of primes mod q in the presence of a Siegel
zero; in other words, assume that (1) and Corollary 4.2 could be proven for q < xθ
for every θ < 1, and assume that there are infinitely many Siegel zeroes as required
by the Main Theorems. Then one would find the following4 bounds for Hm :

H1 = 2
H2 ≤ 12,
H4 ≤ 270,
H6 ≤ 52, 116,
′
Hm ≪ e(1+ϵ )m ,
where the last line holds for any ϵ′ > 0.
The result for H1 is notable here, since it would be a resolution of the twin prime
conjecture. Note that this result for H1 does not require θ to go all the way up to
1; if the distribution held for θ = .722, we could recover the twin primes result of
Heath-Brown [HB].

6. Outline
The proof, for the most part, will follow the framework of Maynard and Tao.
The differences are as follows:
1.) Obviously, instead of invoking Bombieri-Vinogradov or Elliott-Halberstam,
we invoke the result of Friedlander and Iwaniec. Since we have an explicit error
term mod q for each q, we are not required to do any fancy Cauchy-Scwhartz
machinations; we can simply plug in an estimate for π(x, q, a) and then sum this
term up over all allowable choices of q.
2.) If we wish to ensure that the number of primes is double what Dirichlet’s
theorem would predict, we must shift the prime tuple so as to put all of the terms
in classes a mod q for which χD (a) = −1 and D|q. Since D is small relative to x,
this causes little issue in our analysis. This shifting is covered in section 7.
From here, the proof mostly follows the ideas described in [Ma], albeit with
W a bit bigger than in the original paper. We note that [Ka] and [BFM] outline
4The numbers on this list come from [Po], Table 3, where the numbers chosen are simply the
lowest k for which Mk > m + 1.
 PRIME TUPLES AND SIEGEL ZEROS 7

the process for dealing with a larger W , and the effect on the final calculations is
minimal.

7. Shifting the Primes

To begin, let us choose an admissible tuple
T = {n + h1 , · · · , n + hk }.
In order to use the Friedlander-Iwaniec result, we must first shift our prime tuple
so that all of the terms are in a good class for the character modulo D.
We know that an exceptional zero indicates that the character must be real-
valued and hence must arise from a Kronecker character. As such, the character
can only take the values of −1, 1, and 0. We will deal only with the case where χD
is a primitive character; in other words,
 
χD (−1)D
χD (n) = .
n
We do not need to consider the case when χD is imprimitive, for if it is, it must
arise from a primitive character χC for some C|D. If that were true, L(s, χC ) would
have the same Siegel zero with
1 1
Re(s) > 1 − >1− .
(log D)rr +A (log C)rr +A
So we may simply take C to be our initial D and proceed as normal.
First, let g be the largest prime divisor of D, and let
(2) D = gD′ .
Since χD is primitive, we know that D must be square-free except possibly for a
power of 2 that is at most 23 . In particular, this means that g ≫ log D.
Now, our character must split up multiplicatively, so χD (n) = χg (n)χD′ (n),
where χg (n) and χD′ (n) must both take values in {−1, 1, 0}.
In this section, our goal will be to show the following:
Lemma 7.1. For a fixed k, let D be such that log log log(D) ≫ k. Then there
exists an l ≤ D such that for all i,
χD (l + hi ) = −1.
Proof. First, we know that since our tuple is admissible, there exists an n′ such
that (n′ + hi , D) = 1 for each i.
Now, consider the expression
k
Y
(1 − χD (x + n′ + hi )) .
i=1
Clearly, this expression is always non-negative, but it will only equal a non-zero
value when all of the χD (x + n′ + hi ) are either -1 or 0.
So for D′ and g as defined above in (2), let x = D′ y and consider the sum
g Y
X k
H= (1 − χD (D′ y + n′ + hi )) ,
y=1 i=1

Note that since g is prime and (n′ + hi , gD′ ) = 1 for all i, χD (D′ y + n′ + hi ) can
only equal 0 if g|D′ y + n′ + hi ; for each choice of i, this will occur exactly once
8 T. WRIGHT

as y ranges from 1 to g. So there are exactly k choices of y where any of the

χD (D′ y + n′ + hi ) = 0. For each of these choices of y,
k
Y
0≤ (1 − χD (D′ y + n′ + hi )) ≤ 2k−1 .
i=1

So if H > k2 k−1
then there must exist a tuple where all of the χD (D′ y + n′ + hi ) =
−1, which is as required.
Now, we can rewrite H as
g Y
X k
H= (1 − χD′ (D′ y + n′ + hi )χg (D′ y + n′ + hi ))
y=1 i=1
g Y
X k
= (1 − χD′ (n′ + hi )χg (D′ y + n′ + hi ))
y=1 i=1

For each i, let us write ϵi = χD′ (n′ + hi ). Again, these ϵi are all ±1 since (D′ , n′ +
hi ) = 1. So
g Y
X k
H= (ϵi − χg (D′ y + n′ + hi ))
y=1 i=1

Let us multiply out the product. So

g Y k
!  g 2k −1   
X XX Y
H= ϵi +   ϵi  χg (Qj (y))
y=1 i=1 y=1 j=1 i∈Sj

where the Qj (y) are polynomials of degree ≤ k that are square-free when viewed
over Fg , and the Sj are distinct subsets of {1, 2, · · · , k}. By triangle inequality,
k
 
X g Y k 2X −1 Y Xg
H≥ ϵi −  ϵi  χg (Qj (y))
y=1 i=1 j=1 i∈Sj y=1
k
g
X −1
2X g
X
≥ 1 − χg (Qj (y))
y=1 j=1 y=1

The first absolute value is clearly equal to g; for the second, we can use the Weil
bound for character sums over Fg , finding that
g
X √
χg (Qj (y)) ≤ (k − 1) g.
y=1

So
√
H ≥ g − k2k−1 g.
This is clearly larger than k2k−1 for large values of g. So there must exist a y ′ ≤ g
such that χD (D′ y ′ + n′ + hi ) = −1 for all i. We then set
l ≡ D ′ y ′ + n′ (mod D)
to complete the theorem. □
 PRIME TUPLES AND SIEGEL ZEROS 9

8. The Maynard-Tao Sieve

Now, we proceed as in the work of Maynard-Tao. First, in [Ma], the author
introduces a parameter W that eliminates the concern for small primes. We do
something similar here.
θ
Fix a large L > r. For a given D, let X = DL , and let R = X 2 −ϵ for some
58
θ < 115 (1 − 1r ) and ϵ > L1 . We then define
Y
V = p,
2<p<log log log X
p∤D

and let

W = DV.
Because our tuple is admissible, we know that there exists a congruence class
v mod V such that all of the terms in the tuple are coprime to V . So for l as in
Lemma 7.1, define v0 such that

v0 ≡ v (mod V )
v0 ≡ l (mod D)
As is standard, for a well-chosen function λ, we let
 2
X X
S1 =  λd1 ,··· ,dk  ,
n≡v0 (mod W ) di |n+hi
X≤n≤2X

and
 2
X k
X X
S2 = 1P (n + hi )  λd1 ,··· ,dk  .
n≡v0 (mod W ) i=1 di |n+hi
X≤n≤2X

This is essentially the sum that was considered in [Ma]. The only difference
here is that W is of size X ϵ for some ϵ < 1r . However, such alterations have been
addressed in [Ka], [BFM], and elsewhere; the only difference is a slight change in
the allowable level of distribution.
Obviously, it will be helpful if we have a definition for λ. To this end, let Sk be
the set of all piecewise differentiable functions F : [0, ∞)k → R where the support
of F lies on the simplex
k
X
Rk = {(t1 , ..., tk ) : ti ≤ 1}
i=1

For our purposes, we will choose functions F ∈ Sk and then let

 
log d1 log dk
λd1 ,··· ,dk = µ(d1 ) · · · µ(dk )F ,··· , .
log R log R
We require a lemma about the behavior of such functions; this lemma is the crux of
Maynard and Tao’s work, since this lemma allows one to turn the search for prime
10 T. WRIGHT

gaps into a search for the best possible F . Define

ϕ(W )
B= log R,
W
and let
Z ∞ Z ∞
(3) Ik (F ) = ... F (t1 , ..., tk )2 dt1 ...dtk ,
0 0
Z ∞ Z ∞ 
(m) 2
(4) Jk (F ) = ... F (t1 , ..., tk ) dtm dt1 ...dtm−1 dtm+1 dtk .
0 0
In practice (in [Po] and elsewhere), the F are always taken to be symmetric; thus,
(m)
the Jk (F ) are independent of m, and hence we will often drop the superscript.
So we have the following:
Lemma 8.1. Let F , λ, W , and X be as described above. Then
X λd1 ,··· ,dk λe1 ,··· ,ek
= (1 + o(1))B −k Ik (F )
[dj , ej ]
d1 ,···dk ,e1 ,··· ,ek
[d1 ,e1 ],··· ,[dk ,ek ],W coprime

and
X λd1 ,··· ,dk−1 ,1 λe1 ,··· ,ek−1 ,1
= (1 + o(1))B 1−k Jk (F ).
ϕ([dj , ej ])
d1 ,···dk−1 ,e1 ,··· ,ek−1
[d1 ,e1 ],··· ,[dk ,ek ],W coprime

Proof. This is essentially the evaluation of (5.2) and (5.18) of [Ma]. That paper
only deals with very small W ; however, the method applies easily to larger W , and
others have done exactly this - see e.g. (5.4) and (5.20) of [Ka]. □

9. The Sum S
From here, we evaluate the two sums. Define Sk to be the set of all functions F
that satisfy the criteria listed above, and define
Let
Pk (m)
m=1 Jk (F )
(5) Mk = sup .
F ∈Sk Ik (F )
Then we have the following:
Theorem 9.1. For a given k, choose F ∈ Sk . Let S1 and S2 be as above, and let
D, X, and R be as defined at the beginning of Section 8. Then

X −k
S1 = (1 + o(1)) B Ik (F )
W
and
k
X X (m)
S2 = (2 + o(1)) B 1−k Jk (F ),
ϕ(W ) log X m=1
where the o(1)-term goes to zero as D → ∞.
log R
Since log X = θ2 , S2 can be rewritten as
k
X −k X (m)
S2 = (θ + o(1)) B Jk (F ).
W m=1
 PRIME TUPLES AND SIEGEL ZEROS 11

This means that if

S = S2 − ρS1
then
k
!
X (m) X −k
S= (θ + o(1)) Jk (F ) − ρIk (F ) B .
m=1
W

The proof here is similar to the proofs in [Ma] or in any number of papers that
are based on [Ma]. We split the proof into two parts, one for each sum, below.

10. The Sum S1

First, we evaluate
 2
X X
S1 =  λd1 ,··· ,dk  .
n≡v0 (mod W ) di |n+hi
X≤n≤2X

The usual trick is to expand out the square and reverse the order of summation,
giving
X′ X
S1 = λd1 ,··· ,dk λe1 ,··· ,ek 1.
d1 ,··· ,dk n≡v0 (mod W ),X≤n≤2X
e1 ,··· ,ek [di ,ei ]|n+hi

where the prime indicates that the sum requires the [d1 , e1 ], · · · , [dk , ek ], W to be
pairwise coprime. The di and ei must all be coprime to W (by choice of v0 );
moreover, ([di , ei ], [dj , ej ]) = 1 since Q a prime p|hi − hj implies p|W . We can then
replace the inside
Q sum with X/(W [di , ei ]) + O(1). Since λ ≪ 1 and is only
supported on di ≤ R, we have
 
X′ X
λd1 ,··· ,dk λe1 ,··· ,ek Q + O(1)
W [di , ei ]
d1 ,··· ,dk
e1 ,··· ,ek
 
 X′ X
= λd1 ,··· ,dk λe1 ,··· ,ek  + O(R).
Q 
W [di , ei ]
d1 ,··· ,dk
e1 ,··· ,ek

Hence the error term is negligible.

For the main term, we pull out the X and W and find
 
X  X ′ λd1 ,··· ,dk λe1 ,··· ,ek 
Q
W [di , ei ]
 
d1 ,··· ,dk
e1 ,··· ,ek

We then invoke Lemma 8.1, giving us

X −k
S1 = (1 + o(1)) B Ik (F ).
W
12 T. WRIGHT

11. The Sum S2

Recall first that by Corollary 4.2, if D|q and χ(a) = −1 we have
 
2π(X) π(X)
π(x, q, a) = +O .
ϕ(q) ϕ(q) logA−2 (x)
Now, for S2 , let us write
 2
X X
S2m = 1P (n + hm )  λd1 ,··· ,dk  .
n≡v0 (mod W ) di |n+hi
X≤n≤2X

Again, we expand out the square and reverse the order of summation, giving
X′ X
= λd1 ,··· ,dk λe1 ,··· ,ek 1P (n + hm ).
d1 ,··· ,dk−1 ,1 n≡v0 (mod W ),n∼X
e1 ,··· ,ek−1 ,1 [di ,ei ]|n+hi

Evaluating the inside sum gives

   !
(m)
X′ 1 2π(X)
S2 = λd1 ,··· ,dk−1 λe1 ,··· ,ek−1 1+O .
logA X
Qk
d1 ,··· ,dk ϕ(W ) i=1 ϕ([di , ei ])
e1 ,··· ,ek

Pulling the W ’s and X’s out front gives

  X
2π(X) ′ λd1 ,··· ,dk λe1 ,··· ,ek
(1 + o(1)) Qk−1
ϕ(W ) i=1 ϕ([di , ei ])
d1 ,··· ,dk−1
e1 ,··· ,ek−1

As in the evaluation of S1 , by Lemma 8.1 we can simplify the inner sum to

X
S2m =(2 + o(1)) B 1−k Jkm (F )
ϕ(W ) log X
 
θ X −k m
=(2 + o(1)) B Jk (F ).
2 W

12. Bounds for Hm

From Theorem 9.1
k
!
X X k
S = S2 − ρS1 = (1 + o(1)) (θ − 2ϵ) Jkm (F ) − ρIk (F ) B
m=1
W
Let Mk be as defined in (5). In order to prove that there exist infinitely many
ρ + 1-tuples, it will then suffice to show that
(θ − 2ϵ)Mk > ρ.
The work of Friedlander and Iwaniec allows any
 
58 1
θ< 1− = .504346916...
115 r
Moreover, since we can choose L to be as large as we like, we can similarly choose
epsilon to be as small as we like. So we will require
ρ
Mk > .
.504346916
 PRIME TUPLES AND SIEGEL ZEROS 13

From the work of Maynard and Tao [Ma], it is known that

Mk ≥ log k − 2 log log k − 2.
Polymath 8b [Po] improved this bound to
Mk ≥ log k − C
for some constant C.
For our purposes, it does not matter which bound we use, since setting either
ρ
one greater than .504346916 will yield

k ≫ e(1.98276)ρ .
From equation (149) in [Po], the minimum diameter H(k) of a k-tuple can be
bounded by
H(k) ≤ k log k + k log log k − k + o(k),
So
Hm ≪ e1.9828m .
which is Main Theorem 1.

13. Bounds for Hm for m = 3, 4, 5

For small values of m, we can use the ideas of the Polymath 8b [Po] project to
find our bounds for Mk . First, for m = 3, 4, and 5, we can use Theorem 6.7 in the
paper mentioned above:
Polymath 8b, Theorem 6.7. Let k ≥ 2, and let c, T, τ > 0 be parameters. Define
the function g : [0, T ] → R by
1
g(t) :=
c + (k − 1)t
and the quantities
Z T
m2 := g(t)2 dt
0
Z T
1
µ := tg(t)2 dt
m2 0
Z T
2 1
σ := t2 g(t)2 dt − µ2 .
m2 0

Assume the inequalities

kµ ≤ 1 − τ
kµ < 1 − T
kσ 2 < (1 + τ − kµ)2 .
Then one has
k k Z + Z3 + W X + V U
log k − Mk ≤ 2
k−1 k − 1 (1 + τ /2)(1 − (1+τkσ−kµ)2)
14 T. WRIGHT

where Z, Z3 , W, X, V, U are the explicitly computable quantities

! !
1 1+τ r − kµ kσ 2 r2
Z
Z := r log + + dr
τ 1 T 4(r − kµ)2 log r−kµ
T
4kT
Z T
1 t
Z3 := kt log(1 + )g(t)2 dt
m2 0 T
Z T
1 τ
W := log(1 + )g(t)2 dt
m2 0 kt
log k 2
X := c
τ
Z T
c 1
V := g(t)2 dt
m2 0 2c + (k − 1)t
log k 1
Z
(1 + uτ − (k − 1)µ − c)2 + (k − 1)σ 2 du.


U :=
c 0
[α]
The theorem works by finding a lower bound not for Mk but for a variant Mk ,
which restricts Sk to those functions supported on the simplex
k
[α]
X
Rk = {(t1 , ..., tk ) ∈ [0, α]k : ti ≤ 1}.
i=1

Here, the authors set α = T . In some sense, this theorem can be seen to build
on Zhang’s original idea that one can often derive better results by restricting the
analysis to smooth moduli.
[T ]
More specifically, the authors of [Po] seek to maximize Mk by treating it prob-
abilistically, eventually finding (see (114) of [Po])that for any arbitrary r > 1,
Z r−X1 −···−Xk−1 ! [T ]
2 m 2 Mk r
(6) g(t) dt ≤ P (X1 + · · · + Xk ≥ r)
0 k

where the Xi are independent random variables taking values in [0, T ] with probabil-
ity distribution m12 g(t)2 . The values of µ and σ for each variable in this distribution
are calculated as in the statement of the theorem, and one can see that the goal is
then to choose our c, T , and τ that maximizes Mk [T ] in (6) above.
[Po] defines the parameters c, T, τ in terms of new parameters β and η (and our
previously defined k), where
η
c :=
log k
β
T :=
log k
τ = 1 − kµ
[Po] then sets about finding the best choices of β and η to push Mk over the various
magical marks that will guarantee a certain number of primes.
β and η can be chosen arbitrarily. To find optimal choices for β and η, we
performed a Matlab computation where we began with the β and η chosen in [Po]
for the nearest tuple and then let the variables “walk” until we found apparent
maxima. The results are below:
 PRIME TUPLES AND SIEGEL ZEROS 15

Theorem 13.1. Given the following values for β, η, and k, we can find the fol-
lowing for Mk :

(i) β = .973, η = .9650: M5229 ≥ 5.9484

(ii) β = .9433, η = .9786: M38,801 ≥ 7.9310494
(iii) β = .9215, η = .9861: M284,030 ≥ 9.913813

Noting that
3
5.9484 > = 5.94828...
.504346916
4
7.9310494 > = 7.9310487...
.504346916
5
9.9138119 > = 9.9138109...
.504346916
we see that these choices of k are the desired ones.
As in [Po], we will let H(k) denote the minimal diameter hk −h1 of an admissible
k-tuple. Andrew Sutherland was kind enough to compute the minimal tuples5 for
the three numbers above:

H3 ≤ H(5229) ≤ 49, 192,

H4 ≤ H(38, 801) ≤ 439, 812,
H5 ≤ H(284, 030) ≤ 3, 775, 860.

This proves Main Theorem 2 for H3 , H4 , and H5 .

14. Bounds for Hm for m = 2

For small values of m, concrete values for Mk have been calculated by the Poly-
math project. We skip the m = 1 case because it is inferior to Heath-Brown’s
result; if m = 2, Nielsen’s table of results for Mk indicates that

M53 ≥ 3.986213
2
and 3.986213 is greater than .504346916 = 3.96552.... So assuming the existence of
infinitely many strong Siegel zeroes, we have

H2 ≤ 264,

which falls between the H2 found if one assumes EH (270) and that found if one
assumes GEH (252).

5The tuples are currently hosted at

https://math.mit.edu/~drew/admissible_5229_49192.txt,
https://math.mit.edu/~drew/admissible_38802_439820.txt,
https://math.mit.edu/~drew/admissible_284031_3775886.txt.
Note that for the latter two, the tuples are one term larger than for the results stated above;
the bounds we state are then the smallest 38801-tuple and 284030-tuple that can be found on
these lists. Interestingly, the direct calculations of a 38801-tuple and a 284030-tuple ran into
issues where the algorithm fell into a “valley” of sorts and ended with a much larger tuple than
the above, so we revert to the linked lists.
16 T. WRIGHT

15. Acknowledgements
I would like to thank Andrew Sutherland for his help in computing the tuples
and for alerting me to the work of [St], as well as catching a couple of important
discrepancies between [Po] and the Polymath website. I would also like to thank
Beau Christ and Jamie Wright for their technological help. Finally, I would like to
thank the referee for numerous helpful comments and suggestions.

References
[BI] R.C. Baker, A.J. Irving. Bounded intervals containing many primes, Math.
Z. 286 (2017), 821—841.
[BFM] W.D. Banks, T. Freiberg, J. Maynard. On limit points of the sequence of
normalized prime gaps, Proc. Lond. Math. Soc., (3) 113 (2016), 515–539.
[Da] H. Davenport. On character sums in a finite field, Acta Mathematica 71
(1939), 99–121.
[FI03] J. Friedlander and H. Iwaniec, Exceptional characters and prime numbers
in arithmetic progressions, Int. Math. Res. Not. 37 (2003), 2033—2050.
[FI85] J. Friedlander and H. Iwaniec, Incomplete Kloosterman Sums and a Divisor
Problem, Ann. of Math., Second Series, Vol. 121 (2) (1985), 319–344.
[Fo] K. Ford, Large prime gaps and progressions with few primes, Riv. Math.
Univ. Parma , vol. 12 (1) (2021), 41—47.
[HB] D. R. Heath-Brown, Prime twins and Siegel zeros, Proc. London Math. Soc.
(3) 47 (1983), no. 2, 193–224.
[Ka] D.A. Kaptan, A note on small gaps between primes in arithmetic progres-
sions, Acta Arith., 172 (2016), 351—375.
[La] E. Landau, Über die Klassenzahl imaginär-quadratischer Zahlkörper, Nachr.
Ges. Wiss. Göttingen (1918), 285–295.
[Ma] J. Maynard, Small gaps between primes, Ann. of Math. 181 (2015), 383—413.
[Mo1] H. L. Montgomery, Primes in arithmetic progressions, Michigan Math. J. 17
(1970), 33–39.
[Mo2] H. L. Montgomery, Topics in Multipicative Number Theory, Lecture Notes
in Mathematics 227, Springer-Verlag, Berlin (1971).
[Po] D. H. J. Polymath, Variants of the Selberg sieve and bounded gaps between
primes, Research in the Mathematical Sciences 1:12 (2014), 1–83.
[St] Stadlmann, J. On primes in arithmetic progressions and bounded gaps be-
tween many primes, arXiv:2309.00425.
[TT] T. Tao and J. Teräväinen, The Hardy-Littlewood-Chowla conjecture in the
presence of a Siegel zero, preprint, https://arxiv.org/pdf/2109.06291.pdf.
[Zh] Y. Zhang, Bounded gaps between primes, Annals Math 179 (2014), 1121—
1174.