Journal of Number Theory 119 (2006) 284296

www.elsevier.com/locate/jnt
On twin primes associated with
the Hawkins random sieve
H.M. Bui

, J.P. Keating
School of Mathematics, University of Bristol, Bristol, BS8 1TW, UK
Received 13 July 2005; revised 27 July 2005
Available online 4 January 2006
Communicated by J. Brian Conrey
Abstract
We establish an asymptotic formula for the number of k-difference twin primes associated with the
Hawkins random sieve, which is a probabilistic model of the Eratosthenes sieve. The formula for k = 1
was obtained by M.C. Wunderlich [A probabilistic setting for prime number theory, Acta Arith. 26 (1974)
5981]. We here extend this to k 2 and generalize it to all l-tuples of Hawkins primes.
2005 Published by Elsevier Inc.
1. Introduction
The random sieve was introduced by Hawkins [4,5] as follows. Let S
1
= {2, 3, 4, 5, . . .}. Put
P
1
= minS
1
. Every element of the set S
1
\ {P
1
} is then sieved out, independently of the others,
with probability 1/P
1
, and S
2
is the set of the surviving elements. In general, at the nth step,
dene P
n
= minS
n
. We then use 1/P
n
as the probability with which to delete the numbers in
S
n
\ {P
n
}. The set remaining is denoted by S
n+1
. The Hawkins sieve is essentially a probabilistic
analogue of the sieve of Eratosthenes. The sequences {P
1
, P
2
, . . . , P
n
, . . .} of Hawkins primes
mimic the primes in the sense that their statistical distribution is expected to be like that of the
primes. The primes themselves correspond to one realization of the process.
A great deal is known about the Hawkins primes. For instance, the analogues of the prime
number theorem [5,6,9], Mertens theorem [6,9] and the Riemann hypothesis [7,8] are true with
*
Corresponding author.
E-mail address: hm.bui@bristol.ac.uk (H.M. Bui).
0022-314X/$ see front matter 2005 Published by Elsevier Inc.
doi:10.1016/j.jnt.2005.11.015
H.M. Bui, J.P. Keating / Journal of Number Theory 119 (2006) 284296 285
probability 1. We here concern ourselves with the density of k-difference Hawkins twin primes
and its generalization to other l-tuples.
Instead of a sequence of probability spaces, as considered by Hawkins, Wunderlich [9] sim-
plied the process in a single probability space. Let X be the space of all sequences of integers
greater than 1, i.e., X consists of all nite and innite sequences. The class of all sets of those
sequences is . For X, we denote by
n
the set of elements of which are less than n, i.e.,

n
= {2, 3, 4, . . . , n 1} and
n
= \
n
.
Denition 1. An element E is called an elementary set if there exists a sequence
{a
1
, a
2
, . . . , a
k
} X and an integer n > a
k
such that E consists of all the sequences such
that
n
= {a
1
, a
2
, . . . , a
k
}. E is denoted by {a
1
, a
2
, . . . , a
k
; n}, and if k = 0, E = { ; n} is the set
of all sequences whose elements are not less than n.
The probability function is now dened recursively on the class of elementary sets.
Denition 2. Dene a non-negative real-valued function on the class of elementary sets as
follows:
(i) ({ ; 2}) =1,
(ii) ({a
1
, . . . , a
k
, n; n +1}) =

k
i=1
(1
1
a
i
)({a
1
, . . . , a
k
; n}),
(iii) ({a
1
, . . . , a
k
; n +1}) =(1

k
i=1
(1
1
a
i
))({a
1
, . . . , a
k
; n}).
For any X, the analogue of the k-difference twin prime counting function is dened as

X,X+k
(x; ) =#{j x: j and j +k }.
Wunderlich [9] showed that
X,X+1
(x) x/(logx)
2
almost surely, which is an analogue of
Hardy and Littlewoods famous conjecture concerning the distribution of the twin primes [3]. The
absence of the twin prime constant factor here is due to the drawback of the probabilistic setting
of the random sieve that it contains little arithmetical information about the primes. Though the
result is not unexpected, it is not easy to establish, as it is, for example, in Cramers model [1],
where every number n is independently deleted with probability 1/logn. In Section 2, we follow
the lines of Wunderlich [9] and extend the result to k =2,
Theorem 1. Almost surely

X,X+2
(x)
x
(logx)
2
.
Theorem 1 requires rather more work than [9, Theorem 4], but the idea is similar and straight-
forward. Nevertheless, it is clear from the proof for k =2 that as k increases, the calculations will
become extremely complicated, and the proof for the general case using Wunderlichs method
is likely to be extremely messy. In Section 3, we therefore develop a different approach and
establish the following theorem.
286 H.M. Bui, J.P. Keating / Journal of Number Theory 119 (2006) 284296
Theorem 2. Almost surely, for any xed integer k, as x ,

X,X+k
(x)
x
(logx)
2
.
As we note in Section 3, our approach extends straightforwardly to l-tuples of Hawkins primes
to yield:
Theorem 3. Let 0 <k
1
< k
2
< < k
l1
and denote by
X,X+k
1
,...,X+k
l1
(x; ) the number of
mx such that the set {m, m+k
1
, . . . , m+k
l1
} . Then as x , almost surely

X,X+k
1
,...,X+k
l1
(x)
x
(logx)
l
.
An immediate corollary of this theorem is:
Corollary 1. For any positive integers d, l, and l 2, almost surely, as x ,

X,X+d,...,X+(l1)d
(x)
x
(logx)
l
,
which is reminiscent of a recent theorem of Green and Tao [2] on the existence of arbitrarily long
arithmetic progressions in the primes, proved using powerful techniques from analytic number
theory, combinatorics and ergodic theory.
2. Proof of Theorem 1
We begin the proof by stating a lemma from [9].
Lemma 1. For r, s, t non-negative integers, r t , dene
M
k
=1 +
k2

j=1
1
j
.
Then
n

k=2
k
s
M
r
k
=
n
s+1
(s +1)M
r
n
+
c(1, r)n
s+1
M
r+1
n
+ +
c(t r 1, r)n
s+1
M
t 1
n
+O
_
n
s+1
M
t
n
_
=
n
s+1
(s +1)M
r
n
+
t r1

j=1
c(j, r)
n
s+1
(s +1)M
r+j
n
+O
_
n
s+1
M
t
n
_
,
where c(j, r) =r(r +1) (r +j 1)/(s +1)
j+1
.
As in [9], we dene
y
m
() =

j<m, j
_
1
1
j
_
.
H.M. Bui, J.P. Keating / Journal of Number Theory 119 (2006) 284296 287
Then P(m ) =y
m
(), and if we let C
n
be the set of all sequences containing n,
(C
n
) =E(y
n
) =
_
y
n
d.
Wunderlich then obtained the asymptotic formula for the kth moment of y
n
, which is an analogue
of Mertens theorem,
E
_
y
k
m
_
=
1
M
k
m
+O
_
1
M
k+2
m
_
.
Some simple calculations give
P(m , m+2 ) =
_
1
1
m
__
y
2
m
()
1
m+1
_
1
1
m
_
y
3
m
()
_
.
Dene the auxiliary function

X,X+2
(x; ) =

mx, m
m+2
_
y
m
()
1
m+1
_
1
1
m
_
y
2
m
()
_
1
.
In what follows, we write

(x; ) for

X,X+2
(x; ), and if f : R R, we dene the usual
difference operator applied to f by f (m) :=f (m+1) f (m). We have

(m; ) =
_
_
y
m+1
()
1
m+2
_
1
1
m+1
_
y
2
m+1
()
_
1
if m+1 , m+3 ,
0 otherwise.
(1)
Hence
E
_

(m+1)
_
E
_

(m)
_
=
_
1
1
m+1
_
E(y
m+1
).
Thus
E
_

(n)
_
=
n1

m=2
_
1
1
m+1
_
E(y
m+1
) =
n

m=3
_
1
1
m
__
1
M
m
+O
_
1
M
3
m
__
=
n
M
n
+
n
M
2
n
+O
_
n
M
3
n
_
. (2)
We now wish to estimate the variance of

(n). It is easy to see from (1) that
E
_

2
(m)
_
=2
_
1
1
m+1
_
E
_
y
m+1

(m)
_
+O(1). (3)
288 H.M. Bui, J.P. Keating / Journal of Number Theory 119 (2006) 284296
It is necessary to nd another recursion for y
i
m+1
()

(m; ). We have
y
i
m+2
()

(m+1; )
=

_
1
1
m+1
_
i
y
i
m+1
_

(m; ) +
_
y
m+1
()
1
m+2
_
1
1
m+1
_
y
2
m+1
()
_
1
_
if m+1 , m+3 ,
_
1
1
m+1
_
i
y
i
m+1

(m; ) if m+1 , m+3 / ,

y
i
m+1

(m; ) if m+1 / .
Since

P(m+1 , m+3 ) =
_
1
1
m+1
__
y
2
m+1
()
1
m+2
_
1
1
m+1
_
y
3
m+1
()
_
,
P(m+1 , m+3 / ) =y
m+1
()
_
1
1
m+1
__
y
2
m+1
()
1
m+2
_
1
1
m+1
_
y
3
m+1
()
_
,
P(m+1 / ) =1 y
m+1
(),
we easily obtain
E
_
y
i
m+1

(m)
_
=
_
1
1
m+1
_
i+1
E
_
y
i+1
m+1
_

_
1
_
1
1
m+1
_
i
_
E
_
y
i+1
m+1

(m)
_
. (4)
Taking i =3, summing from 1 to n 1, and using E(y
4
k+1

(k)) =O(kE(y
4
k+1
)), we have
E
_
y
3
n+1

(n)
_
=O
_
n

m=2
E
_
y
4
m
_
_
=O
_
n

m=2
1
M
4
m
_
=O
_
n
M
4
n
_
.
Letting i =2 in (4),
E
_
y
2
m+2

(m+1)
_
E
_
y
2
m+1

(m)
_
=
_
1
1
m+1
_
3
E
_
y
3
m+1
_

_
1
_
1
1
m+1
_
2
_
E
_
y
3
m+1

(m)
_
.
Summing from 1 to n 1, we obtain
E
_
y
2
n+1

(n)
_
=
n

m=2
_
1
1
m
_
3
E
_
y
3
m
_
+O
_
n

m=2
1
M
4
m
_
=
n

m=2
1
M
3
m
+O
_
n

m=2
1
M
4
m
_
=
n
M
3
n
+O
_
n
M
4
n
_
.
We are now ready to nd E(y
m+1

(m)). Letting i =1 in (4),

E
_
y
m+2

(m+1)
_
E
_
y
m+1

(m)
_
=
_
1
1
m+1
_
2
E
_
y
2
m+1
_

1
m+1
E
_
y
2
m+1

(m)
_
.
H.M. Bui, J.P. Keating / Journal of Number Theory 119 (2006) 284296 289
So
E
_
y
n+1

(n)
_
=
n

m=2
_
1
1
m
_
2
E
_
y
2
m
_

m=2
1
m
E
_
y
2
m

(m1)
_
=
n

m=2
1
M
2
m

m=2
1
M
3
m
+O
_
n

m=2
1
M
4
m
_
=
n
M
2
n
+
n
M
3
n
+O
_
n
M
4
n
_
.
Substituting this into (3), we have
E
_

2
(n)
_
=
n

m=2
2
_
1
1
m
__
m
M
2
m
+
m
M
3
m
+O
_
m
M
4
m
__
+O(n)
=2
_
n
2
2M
2
n
+
n
2
2M
3
n
+O
_
n
2
M
4
n
__
+2
_
n
2
2M
3
n
+O
_
n
2
M
4
n
__
=
n
2
M
2
n
+
2n
2
M
3
n
+O
_
n
2
M
4
n
_
. (5)
From (2) and (5) we deduce that

E(

(n)) =
n
M
n
+
n
M
2
n
+O
_
n
M
3
n
_
,
Var(

(n)) =O
_
n
2
M
4
n
_
.
Theorem 2 in [9] then implies that

(n)
n
logn
.
Now we dene
r
m
() =
_
1 if m , m+2 ,
0 otherwise.
Then

X,X+2
(n; ) =

mn
r
m
()
=

mn
r
m
()
y
m
()
1
m+1
_
1
1
m
_
y
2
m
()
_
y
m
()
1
m+1
_
1
1
m
_
y
2
m
()
_
=

mn
a
m
()b
m
(),
where
a
m
() =
r
m
()
y
m
()
1
m+1
_
1
1
m
_
y
2
m
()
290 H.M. Bui, J.P. Keating / Journal of Number Theory 119 (2006) 284296
and
b
m
() =y
m
()
1
m+1
_
1
1
m
_
y
2
m
().
Let A
0
() =0 and A
m
() =

m
j=1
a
j
(). Using Abel summation,

X,X+2
(n; ) =

mn
a
m
()b
m
() =A
n
()b
n
()

m<n
A
m
()
_
b
m+1
() b
m
()
_
.
Since A
n
() =

(n; ), A
n
b
n
(n/logn)(1/logn) n/(logn)
2
. The result follows if we can
show that

m<n
A
m
|b
m+1
b
m
| =o
_
n
(logn)
2
_
.
Firstly,
A
m
() =

jm, j
j+2
1
y
j
()
1
j+1
_
1
1
j
_
y
2
j
()

jm, j
j+2
1
y
j
()
1
3
y
j
()
=
3
2

jm, j
j+2
1
y
j
()
,
which is O(m/logm) from [9, Theorem 4].
Secondly,

b
m+1
() b
m
()

_
y
m+1
()
1
m+2
_
1
1
m+1
_
y
2
m+1
()
_

_
y
m
()
1
m+1
_
1
1
m
_
y
2
m
()
_

.
Since
y
m+1
() =
__
1
1
m
_
y
m
() =
m1
m
y
m
() if m ,
y
m
() otherwise,
we obtain

b
m+1
() b
m
()

=
_

1
m
y
m
() +
3(m1)
m(m+1)(m+2)
y
2
m
()

if m ,

m2
m(m+1)(m+2)
y
2
m
()

otherwise.
So

b
m+1
() b
m
()

_
1
m
y
m
() if m ,
1
m(m+1)
y
2
m
() otherwise.
H.M. Bui, J.P. Keating / Journal of Number Theory 119 (2006) 284296 291
Hence

m<n
A
m
()

b
m+1
() b
m
()

m<n
m
A
m
()

b
m+1
() b
m
()

m<n
m/
A
m
()

b
m+1
() b
m
()

m<n
m
1
m
A
m
()y
m
() +

m<n
m/
1
m(m+1)
A
m
()y
2
m
().
Thus

m<n
A
m
()

b
m+1
() b
m
()

=O
_

m<n
m
1
(logm)
2
_
+O
_

m<n
m/
1
m(logm)
3
_
=O
_

m<n
m
1
(logm)
2
_
+O(1),
which is easily seen to be O(n/(logn)
3
) from the analogue of prime number theorem for
Hawkins random sieve. The result follows.
3. Proof of Theorems 2 and 3
In this section, we take m[i
1
, i
2
, . . . , i
l
] , where i
1
< i
2
< < i
l
, to mean m +
{i
1
, i
2
, . . . , i
l
} , and m+h / for all h [i
1
, i
l
] \ {i
1
, i
2
, . . . , i
l
}.
Lemma 2. Given a non-negative integer l and 0 =i
0
<i
1
<i
2
< <i
l
<i
l+1
=k, dene
T
[0,i
1
,i
2
,...,i
l
,k]
(n) :=
mn

m[0,i
1
,i
2
,...,i
l
,k]
1.
Then T
[0,i
1
,i
2
,...,i
l
,k]
(n) n/(logn)
l+2
almost surely.
Proof. We simply write T (n) for T
[0,i
1
,i
2
,...,i
l
,k]
(n). Let A
n
be the event n and B
n
be the
complement of A
n
, i.e., B
n
=A
c
n
. We then have
P
m
=P
_
(m+1)[0, i
1
, i
2
, . . . , i
l
, k]
_
=P(A
m+k+1
B
m+k
. . . B
m+2+i
l
A
m+1+i
l
. . . B
m+2
A
m+1
).
292 H.M. Bui, J.P. Keating / Journal of Number Theory 119 (2006) 284296
By the chain rule
P
m
=P(A
m+1
)
P(B
m+2
|A
m+1
) . . . P(B
m+i
1
|B
m+i
1
1
. . . B
m+2
A
m+1
)
P(A
m+1+i
1
|B
m+i
1
. . . B
m+2
A
m+1
)

P(A
m+k+1
|B
m+k
. . . B
m+2+i
l
A
m+1+i
l
. . . B
m+2
A
m+1
)
=y
m+1

_
1
_
1
1
m+1
_
y
m+1
_
i
1
1

__
1
1
m+1
_
y
m+1
_

__
1
1
m+1 +i
l
__
1
1
m+1 +i
l1
_

_
1
1
m+1
_
y
m+1
_
,
or, in short,
P
m
=
l

j=0
_
1
1
m+1 +i
j
_
l+1j
y
l+2
m+1
l

j=0
_
1
lj

h=0
_
1
1
m+1 +i
h
_
y
m+1
_
i
l+1j
i
lj
1
.
Since
l

j=0
_
1
lj

h=0
_
1
1
m+1 +i
h
_
y
m+1
_
i
l+1j
i
lj
1
=1
l

j=0
(i
l+1j
i
lj
1)y
m+1
+O
_
y
m+1
m
_
+O
_
y
2
m+1
_
=1 (k l 1)y
m+1
+O
_
y
m+1
m
_
+O
_
y
2
m+1
_
,
we have
P
m
=y
l+2
m+1
(k l 1)y
l+3
m+1
+O
_
y
l+3
m+1
m
_
+O
_
y
l+4
m+1
_
.
From the denition of T (m),
T (m+1) T (m) =
_
1 if (m+1)[0, i
1
, i
2
, . . . , i
l
, k] ,
0 otherwise.
H.M. Bui, J.P. Keating / Journal of Number Theory 119 (2006) 284296 293
Hence
E
_
T (m+1)
_
E
_
T (m)
_
=E
_
y
l+2
m+1
_
(k l 1)E
_
y
l+3
m+1
_
+O
_
E(y
l+3
m+1
)
m
_
+O
_
E
_
y
l+4
m+1
__
=
1
M
l+2
m+1

k l 1
M
l+3
m+1
+O
_
1
M
l+4
m+1
_
+O
_
1
mM
l+3
m+1
_
=
1
M
l+2
m+1

k l 1
M
l+3
m+1
+O
_
1
M
l+4
m+1
_
.
Summing from 1 to n 1 yields
E
_
T (n)
_
=
n
M
l+2
n
+
(l +2)n
M
l+3
n

(k l 1)n
M
l+3
n
+O
_
n
M
l+4
n
_
=
n
M
l+2
n

(k 2l 3)n
M
l+3
n
+O
_
n
M
l+4
n
_
. (6)
The next step is to estimate the variance of T (n). As in the case of the previous theorem, we
need to establish a recursion for y
i
m+1
T (m). For this we have
y
i
m+2
T (m+1) =

y
i
m+1
T (m) if m+1 / ,
_
1
1
m+1
_
i
y
i
m+1
(T (m) +1) if (m+1)[0, i
1
, i
2
, . . . , i
l
, k] ,
_
1
1
m+1
_
i
y
i
m+1
T (m) otherwise.
Since
_
P(m+1 / ) =1 y
m+1
,
P((m+1)[0, i
1
, i
2
, . . . , i
l
, k] ) =P
m
,
we deduce that
E
_
y
i
m+2
T (m+1)
_
=E
_
y
i
m+1
T (m)(1 y
m+1
)
_
+
_
1
1
m+1
_
i
E
_
y
i
m+1
_
T (m) +1
_
P
m
_
+
_
1
1
m+1
_
i
E
_
y
i
m+1
T (m)(y
m+1
P
m
)
_
,
or, equivalently,
E
_
y
i
m+1
T (m)
_
=
_
1
1
m+1
_
i
E
_
y
i
m+1
P
m
_

_
1
_
1
1
m+1
_
i
_
E
_
y
i+1
m+1
T (m)
_
. (7)
Letting i =l +4, and recalling that
P
m
=y
l+2
m+1
(k l 1)y
l+3
m+1
+O
_
y
l+4
m+1
_
,
294 H.M. Bui, J.P. Keating / Journal of Number Theory 119 (2006) 284296
we obtain
E
_
y
l+4
m+1
T (m)
_
=O
_
E
_
y
2l+6
m+1
__
=O
_
1
M
2l+6
m+1
_
.
So
E
_
y
l+4
n+1
T (n)
_
=O
_
n

m=2
1
M
2l+6
m
_
=O
_
n
M
2l+6
n
_
.
Letting i =l +3 in (7), and summing from 1 to n 1, we have
E
_
y
l+3
n+1
T (n)
_
=
n

m=2
1
M
2l+5
m
+O
_
n

m=2
1
M
2l+6
m
_
=
n
M
2l+5
n
+O
_
n
M
2l+6
n
_
.
Finally, substituting i =l +2 in (7),
E
_
y
l+2
m+1
T (m)
_
=
_
1
1
m+1
_
l+2
E
_
y
l+2
m+1
P
m
_

_
1
_
1
1
m+1
_
l+2
_
E
_
y
l+3
m+1
T (m)
_
=
1
M
2l+4
m+1
(k +1)
1
M
2l+5
m+1
+O
_
1
M
2l+6
m+1
_
.
Thus
E
_
y
l+2
n+1
T (n)
_
=
n

m=2
1
M
2l+4
m
(k +1)
n

m=2
1
M
2l+5
m
+O
_
n

m=2
1
M
2l+6
m
_
=
n
M
2l+4
n
+(2l +4)
n
M
2l+5
n
(k +1)
n
M
2l+5
n
+O
_
n
M
2l+6
n
_
=
n
M
2l+4
n
(k 2l 3)
n
M
2l+5
n
+O
_
n
M
2l+6
n
_
. (8)
Now, back to the variance of T (n),
T
2
(m+1) =
_
T
2
(m) +2T (m) +1 if (m+1)[0, i
1
, i
2
, . . . , i
l
, k] ,
T
2
(m) otherwise.
Therefore
E
_
T
2
(m+1)
_
=E
_
T
2
(m)(1 P
m
)
_
+E
__
T
2
(m) +2T (m) +1
_
P
m
_
=E
_
T
2
(m)
_
+2E
_
T (m)P
m
_
+E(P
m
).
H.M. Bui, J.P. Keating / Journal of Number Theory 119 (2006) 284296 295
And hence
E
_
T
2
(m)
_
=2E
_
y
l+2
m+1
T (m)
_
2(k l 1)E
_
y
l+3
m+1
T (m)
_
+O
_
E
_
y
l+4
m+1
T (m)
__
+O
_
E
_
y
l+2
m+1
__
.
From (8), we obtain
E
_
T
2
(m)
_
=2
_
m
M
2l+4
m
(k 2l 3)
m
M
2l+5
m
_
2(k l 1)
m
M
2l+5
m
+O
_
m
M
2l+6
m
_
=
2m
M
2l+4
m

2(2k 3l 4)m
M
2l+5
m
+O
_
m
M
2l+6
m
_
.
So
E
_
T
2
(n)
_
=
n

m=2
2m
M
2l+4
m
(2k 3l 4)
n

m=2
2m
M
2l+5
m
+O
_
n

m=2
m
M
2l+6
m
_
=
_
n
2
M
2l+4
n
+
(l +2)n
2
M
2l+5
n
_
(2k 3l 4)
n
2
M
2l+5
n
+O
_
n
2
M
2l+6
n
_
=
n
2
M
2l+4
n

(2k 4l 6)n
2
M
2l+5
n
+O
_
n
2
M
2l+6
n
_
. (9)
Combining (6) with (9), we have

E(T (n)) =
n
M
l+2
n

(k2l3)n
M
l+3
n
+O
_
n
M
l+4
n
_
,
Var(T (n)) =O
_
n
2
M
2l+6
n
_
.
Theorem 2 in [9] again yields T (n) n/(logn)
l+2
as asserted.
The proof of Theorem 2 now follows immediately from Lemma 2 by noting that

X,X+k
(x) =
k1

l=0

0<i
1
<i
2
<<i
l
<k
T
[0,i
1
,i
2
,...,i
l
,k]
(x)
=T
[0,k]
(x) +
k1

l=1

0<i
1
<i
2
<<i
l
<k
T
[0,i
1
,i
2
,...,i
l
,k]
(x)
=
_
1 +o(1)
_
x
(logx)
2
+O
_
x
(logx)
3
_
.
Similarly for Theorem 3,

X,X+k
1
,...,X+k
l1
(x) =T
[0,k
1
,k
2
,...,k
l1
]
(x) +O
_
x
(logx)
l+1
_
=
_
1 +o(1)
_
x
(logx)
l
+O
_
x
(logx)
l+1
_
. 2
296 H.M. Bui, J.P. Keating / Journal of Number Theory 119 (2006) 284296
Acknowledgment
J.P.K. is supported by an EPSRC Senior Research Fellowship.
References
[1] H. Cramer, On the order of magnitude of the differences between consecutive prime numbers, Acta Arith. 2 (1937)
2328.
[2] B. Green, T. Tao, The primes contain arbitrarily long arithmetic progressions, math.NT/0404188.
[3] G.H. Hardy, J.E. Littlewood, Some problems in Partitio Numerorum III: On the expression of a number as a sum
of primes, Acta Math. 44 (1923) 170.
[4] D. Hawkins, The random sieve, Math. Mag. 31 (1957/1958) 13.
[5] D. Hawkins, Random sieves II, J. Number Theory 6 (1974) 192200.
[6] C.C. Heyde, On asymptotic behaviour for the Hawkins random sieve, Proc. Amer. Math. Soc. 56 (1976) 277280.
[7] C.C. Heyde, A log log improvement to the Riemann hypothesis for the Hawkins random sieve, Ann. Probab. 6 (5)
(1978) 870875.
[8] W. Neudecker, D. Williams, The Riemann hypothesis for the Hawkins random sieve, Compos. Math. 29 (1974)
197200.
[9] M.C. Wunderlich, A probabilistic setting for prime number theory, Acta Arith. 26 (1974) 5981.
Further reading
[10] W. Neudecker, On twin primes and gaps between successive primes for the Hawkins random sieve, Math. Proc.
Cambridge Philos. Soc. 77 (1975) 365367.
[11] P. Ribenboim, The Book of Prime Number Records, second ed., Springer, New York, 1989.
[12] M.C. Wunderlich, The prime number theorem for the random sequences, J. Number Theory 8 (1976) 369371.