arXiv:math/0612195v1 [math.PR] 7 Dec 2006
AN ARITHMETIC MODEL FOR THE TOTAL DISORDER
PROCESS
C.P. HUGHES, A. NIKEGHBALI, AND M. YOR
Abstract. We prove a multidimensional extension of Selberg’s central
limit theorem for the logarithm of the Riemann zeta function on the
critical line. The limit is a totally disordered process, whose coordinates
are all independent and Gaussian.
1. Introduction
A classical result of Selberg [10] (see also Laurinčikas, [6]) states that
the classical continuous determination of the logarithm of the Riemann zeta
function is asymptotically normally distributed, in the sense that if Γ is a
regular Borel measurable subset of C,


Z 2T 
Z

1
log ζ( 2 + it)
1
2
2
1
∈ Γ dt =
e−(x +y )/2 dx dy
11 q
lim

 1 log log T
T →∞ T T
2π Γ
2
where 11 is the indicator function, and regular means that the boundary of
Γ has zero Lebesgue measure.
If we let
λ
Lλ (N, u) :=
log ζ( 12 + iueN )
√
log N
then Selberg’s result implies that
Z 2
11 {Lλ (N, u) ∈ Γ} du = P{Gλ ∈ Γ}
lim
N →∞ 1
(2)
(1)
Gλ + iGλ
is a complex-valued Gaussian random variable with
where Gλ =
(1)
(2)
mean zero and variance λ/2, i.e.: Gλ and Gλ are independent, centered,
(2)
(1)
and E[(Gλ )2 ] = E[(Gλ )2 ] = λ/2.
It is now a natural question, at least from a probabilistic standpoint, to
look for an asymptotic distribution for (Lλ1 (N, ·), . . . , Lλk (N, ·)), for different λi ’s.
Date: 1 December 2006.
2000 Mathematics Subject Classification. 60F05, 60G15, 11M06.
Key words and phrases. Total disorder process, convergence in distribution, central
limit theorem, Riemann zeta function.
1
2
C.P. HUGHES, A. NIKEGHBALI, AND M. YOR
Theorem 1. For λ1 > λ2 > · · · > λk > 0, and for every (Γi , i ≤ k) regular,
Z 2
k
Y
P Gλj ∈ Γj .
11 {Lλ1 (N, u) ∈ Γ1 , . . . , Lλk (N, u) ∈ Γk } du =
lim
N →∞ 1
j=1
(1)
(2)
(1)
Dλ = Dλ + iDλ , λ > 0 is a totally disordered
(1)
complex-valued Gaussian process, meaning that Dλ , λ > 0 and
(2)
Dλ , λ > 0 are two independent Gaussian processes all of whose coordiWe now note that if
(1)
(2)
nates are independent with E[(Dλ )2 ] = E[(Dλ )2 ] = λ/2, then the quantity
on the right hand side of (1) is
P {Dλ1 ∈ Γ1 , . . . , Dλk ∈ Γk } .
Theorem 1 is an attempt to move from the deterministic set up of the
Riemann zeta function, and the “static” central limit theorem of Selberg
into a more “dynamic” probabilistic world, where a process appears in the
limit. However, this process is quite wild. In the next section, we comment
about it, and some of its occurrences in random matrix theory. Finally, in
the third section we prove Theorem 1 using the method of moments.
Remark. Our methods apply equally well to any L-function from the Selberg
class, but for concreteness and for the sake of simplicity we only state here
the result for the Riemann zeta function.
2. Some remarks on total disorder process
2.1. Non-measurability of the total disorder process. The total disorder process is a “wild” process; indeed there is no measurable process
(λ, ω) 7→ D̃λ (ω) which would be a modification of (Dλ , λ ≥ 0), i.e. P{D̃λ =
Dλ ) = 1 for all λ. Indeed, if so, we would get (use Fubini)
Z b
e λ dλ = 0 a.s.,
D
a
hence
eλ = 0
D
dλdP,
which is absurd (for some further discussion on the total disorder process,
see page 37 of [7]).
2.2. The total disorder process in random matrix theory. The total
disorder process has already been observed asymptotically in random matrix
theory, although in a different guise. Let ZU (θ) = det(I − U e−iθ ) be the
AN ARITHMETIC MODEL FOR THE TOTAL DISORDER PROCESS
3
characteristic polynomial of an N × N unitary matrix U chosen with Haar
measure, then Hughes, Keating and O’Connell [4] prove that
log Z (θ)
q U
1
2 log N
weakly converges to X(θ)+ iY (θ), where X(θ), Y (θ) are independent Gaussian processes with covariance structure
(
1 if θ1 = θ2
E [X(θ1 )X(θ2 )] = E [Y (θ1 )Y (θ2 )] =
0 otherwise
This was used to provide an explanation for the covariance structure of
CU (s, t), the number of eigenangles of U that lie in the interval (s, t), found
earlier by Wieand [12, 13]. A separate explanation was given by Diaconis
and Evans [2]. Let
(t − s)N/2π
eU (s, t) := CU (s, t) −
.
C
1√
log
N
π
Wieand proves that for fixed s, t, if the matrices U are chosen with Haar
eU (s, t) converges in distribution, as
measure from the unitary group, then C
N → ∞, to a standard normal random variable. In fact she goes much fureU (s, t) to a certain Gaussian process
ther by proving weak convergence of C
C(s, t).
Theorem 2 (Wieand). For −π < s < t ≤ π, the finite dimensional distrieU (s, t) converge as N → ∞ to those of a centered
butions of the process C
Gaussian process C(s, t) with covariance structure
E C(s, t)C(s′ , t′ ) =


1





−1
1
2


1

−

2


0
if s = s′ , t = t′
if s = t′ , t = s′
if s = s′ or if t = t′ but not both
if s = t′ or if t = s′ but not both
otherwise
A similar process result had previously been found by Costin and Lebowitz
[1] for GUE matrices, and Soshnikov [11] considers a process result for counting the number of eigenangles in an interval with a given minimum displacement. The surprising thing about these correlations is that they imply that
even if an interval I contains more than the average number of eigenangles,
then any subset of I not sharing a common endpoint with I will usually
still contain its average number. Also, no matter how close two intervals I
and J are, unless they share an endpoint, then CI and CJ (with obvious
notations) are independent.
4
C.P. HUGHES, A. NIKEGHBALI, AND M. YOR
Of course, the results of Wieand and of Hughes, Keating and O’Connell
are strongly related, because
log ZU (s)
eU (s, t) = Im√log ZU (t) − Im√
.
C
log N
log N
3. Proof of theorem 1
We will prove Theorem 1 via the method of moments. The following
lemma will play an essential role in our argument.
Lemma 3. A complex random variable Z = X + iY has moments
(
m n
n!2n σ 2n if m = n
E Z Z =
0
otherwise
(2)
if and only if X and Y are independent and distributed according to the
normal law with mean zero and variance σ 2 .
Proof. Let Z = X + iY where X and Y are independent, centered normal
random variables with variance σ 2 . Consider the joint moment generating
function of Z and Z: for (α, β) ∈ C2 ,
i
i
h
h
E eαZ eβZ = E e(α+β)X+i(α−β)Y
2 σ 2 /2−(α−β)2 σ 2 /2
= e(α+β)
= e2αβσ
=
∞
X
2
n!2n σ 2n
n=0
αn β n
n! n!
which is the two-variable moment generating function of (2). Conversely,
assume that (2) holds for the joint moments of Z and Z. Then working up
the above chain of equalities proves that Z = X + iY where X and Y are
independent, centered gaussians with variance σ 2 .
From Lemma 3, if one can show that for any positive integers k and any
integers m1 , . . . , mk ; n1 , . . . , nk , and if for any λ1 > · · · > λk
#
" k
k
Y
Y m
nℓ
ℓ
=
nℓ !λnℓ ℓ δmℓ ,nℓ
E
Dλℓ Dλℓ
ℓ=1
ℓ=1
then one may conclude that (Dλ , λ > 0) is a centered complex-valued Gaussian totally disordered process with covariance structure
(
λi if λi = λj
E Dλi Dλj =
0
otherwise
and E Dλi Dλj = 0 for all λi , λj .
AN ARITHMETIC MODEL FOR THE TOTAL DISORDER PROCESS
5
Therefore, Theorem 1 is a consequence of the following
Theorem 4. Let
λ
log ζ( 12 + iueN )
√
.
Lλ (N, u) =
log N
If λ1 < λ2 < · · · < λk are fixed, then
Z 2Y
k
k
Y
nj
n
mj
Lλj (N, u) Lλj (N, u) du =
nj !λj j δ(mj , nj )
lim
N →∞ 1
j=1
j=1
where δ(mj , nj ) = 1 if mj = nj and zero otherwise.
We need the following theorem of Selberg, [10].
Theorem 5 (Selberg). If n is a positive integer, 0 < a < 1, and T a/n ≤
x ≤ T 1/n , then
2n
Z 2T −it
X
p 1
4n An
log ζ( 1 + it) −
dt
=
O
n
e
√
2
T T p p≤x
for some constant A which depends upon a.
We also need to calculate the moments of certain prime sums.
Lemma 6. Given k a positive integer, let λ1 > · · · > λk > 0. Let
1
P (λ, n) = P (λ, n; k, N, u) = √
log N
X
“ λ
N
p≤exp 40kn
Nλ
p−iue
√
p
”
For any non-negative integers m1 , . . . , mk and n1 , . . . , nk ,
Z 2Y
k
k
Y
nj
P (λj , mj )mj P (λj , nj ) du =
lim
(nj )! (λj )nj δ(mj , nj )
N →∞ 1
j=1
(3)
j=1
(Note that for the sake of simplicity the variable u does not appear explicitly in (3) and in some expressions below)
Proof. We wish to expand out
k
Y
P (λj , mj )mj
j=1
as a multiple sum over primes. It is exceedingly complicated. We will
introduce the following notation: For j = 1, . . . , k, let
pj = pj,1 , . . . , pj,mj
6
C.P. HUGHES, A. NIKEGHBALI, AND M. YOR
and let
N λj
Pj = P (λj , mj ) = pj : pj,ℓ is prime , pj,ℓ ≤ exp
40kmj
Hence
k
Y
P (λj , mj )mj = (log N )−(m1 +···+mk )/2
j=1
×
X
p1 ∈P1 ,...,pk ∈Pk
Similarly, we let
P
λj
exp −iu kj=1 eN log(pj,1 . . . pj,mj )
qQ
Qmj
k
j=1
ℓj =1 pj,ℓj
qj = qj,1 , . . . , qj,nj
and let
Qj ≡ P (λj , nj ) =
and so
k
Y
nj
P (λj , nj )
λj N
qj : qj,ℓ is prime , qj,ℓ ≤ exp
40knj
= (log N )−(n1 +···+nk )/2
j=1
×
Finally, let p =
Sk
X
q1 ∈Q1 ,...,qk ∈Qk
Sk
and let
k
X
qj,1 . . . qj,nj
λj
log
exp N
F (p, q) =
pj,1 . . . pj,mj
j=1 pj
and q =
P
λj
exp iu kj=1 eN log(qj,1 . . . qj,nj )
qQ
Qnj
k
j=1
ℓj =1 qj,ℓj
j=1 qj ,
j=1
Therefore,
k
Y
nj
P (λj , mj )mj P (λj , nj )
= (log N )−(m1 +···+mk +n1 +···+nk )/2
j=1
×
X
p1 ∈P1 ,...,pk ∈Pk
q1 ∈Q1 ,...,qk ∈Qk
exp (iuF (p, q))
Q
Q
mj
nj
p
q
j,ℓ
j,ℓ
j
j
j=1
ℓj =1
ℓj =1
r
Qk
(4)
We divide the sum up into two parts, depending on whether F (p, q) equals
zero or not. The terms where the sum vanishes we call diagonal terms;
the other terms are off-diagonal. The proof of the lemma will follow from
showing that the off-diagonal terms do not contribute in the large-N limit,
and using a simple combinatorial enumeration of the diagonal terms, along
with the prime number theorem, to estimate the diagonal terms.
AN ARITHMETIC MODEL FOR THE TOTAL DISORDER PROCESS
7
3.1. The diagonal terms. We will see below in §3.2 that since λ1 > · · · >
λk , then for sufficiently large N , the only way for F (p, q) = 0 is if
qj,1 . . . qj,nj
λj
=0
log
exp N
pj,1 . . . pj,mj
for each j = 1, . . . , k separately. Thus the diagonal terms are those contained
in the sets
(
)
mj
nj
Y
Y
Dj := (pj , qj ) : pj ∈ Pj , qj ∈ Qj ,
pj,ℓ =
qj,ℓ
(5)
ℓ=1
ℓ=1
Since pj,ℓ and qj,ℓ are both prime, the set Dj is empty unless mj = nj . Under
such an assumption, the diagonal terms in (4) are


k
Y
X
X
1
1


=
Qk Qnj
q
.
.
.
q
j,1
j,n
q
j
j,ℓ
j
j=1
ℓ =1
(p1 ,q1 )∈D1 ,...,(pk ,qk )∈Dk
j=1
j
(pj ,qj )∈Dj
If qj,1 , . . . , qj,nj are distinct primes, then there are (nj )! ways of choosing
pj,1 , . . . , pj,nj such that the products are equal. (If the qj,ℓ are not distinct,
then the result is similar, but with a different combinatorial factor, and the
result is at least a couple of logarithms smaller). Hence
X
X
1
1
1
= (nj )!
1 + O( 2 )
qj,1 . . . qj,nj
qj,1 . . . qj,nj
log N
qj ∈Qj
(pj ,qj )∈Dj

nj

X
1
1


= (nj )! 
1 + O( 2 )


„
« q
log N
λ
q≤exp
N j
40knj
nj N λj
1
= (nj )! log
+ O(1)
1 + O( 2 )
40knj
log N
1
)
(6)
= (nj )!(λj log N )nj 1 + O(
log N
Hence the diagonal contribution to (4) is


k
Y
 (nj )!(λj )nj δ(mj , nj ) 1 + O(
j=1
1
)
log N
which is the right-hand side of (3) in the large-N limit. (The constant
implicit in the O-term depends on mj , nj , λj and k, but these are all constants). Hence the proof of the lemma will be complete if we can show
there is no contribution to (3) from the off-diagonal terms, the terms where
F (p, q) 6= 0.
8
C.P. HUGHES, A. NIKEGHBALI, AND M. YOR
3.2. The off-diagonal terms. Now we show that the non-diagonal terms
of (4) do not contribute to (3) in the limit. Upon integrating (4) for u
between 1 and 2, we obtain
X′
exp (i2F (p, q)) − exp (iF (p, q))
r
(7)
Q
Qk Qmj
n
j
p1 ∈P1 ,...,pk ∈Pk iF (p, q)
j=1
ℓj =1 pj,ℓj
ℓj =1 qj,ℓj
q1 ∈Q1 ,...,qk ∈Qk
P′
denotes that we are summing only over the non-diagonal terms,
where
those terms where F (p, q) 6= 0.
Recall that, without loss of generality, we assumed λ1 > · · · > λk . Assume
q1,1 . . . q1,n1
log
6= 0
(8)
p1,1 . . . p1,m1
λ λ N 1
N 1
and
q
≤
exp
Since p1,ℓ ≤ exp 40km
1,ℓ
40kn1 , we have
1
q . . . q
1
1
1,1
1,n1 λ1
λ1 exp N
log p1,1 . . . p1,m > 2 exp (1 − 40k )N
1
which follows from the fact that if m, n are positive integers, and m 6= n,
then | log(m/n)| > 1/(2 min(m, n)). Furthermore, for j > 1, for pj ∈ Pj
and qj ∈ Qj , then for sufficiently large N ,
q . . . q
j,1
j,nj
λj ≤ 1 exp N λj N λj < exp 2N λj
exp N
log
pj,1 . . . pj,mj 40k
and so we can conclude that if (8) holds,
X
k
1
exp 2N λj
N λ1 −
|F (p, q)| > exp
1−
40k
j=2
1
> exp
1−
N λ1
20k
for sufficiently large N , since λj < λ1 for all j > 1.
The contribution of such terms to (7) is clearly bounded by
X
1
1
r
Q
1
λ
Qmj
Qk
exp (1 − 20k )N 1 p ∈P ,...,p ∈P
nj
1
1
k
k
p
q
j=1
ℓj =1 j,ℓj
ℓj =1 j,ℓj
q1 ∈Q1 ,...,qk ∈Qk


k
X
1
1
1
λ1
λ
λ
1
j
≤ exp −(1 −
)N
N  ≤ exp −(1 −
)N
exp 
20k
40k
10k
j=1
once more using the fact that
k
X
1
1
N λj ≤
N λ1
40k
20k
j=2
for sufficiently large N .
AN ARITHMETIC MODEL FOR THE TOTAL DISORDER PROCESS
9
Therefore, as N tends to infinity, we see that terms which satisfy (8)
vanish. Thus, for a non-zero result in the limit, we must have
q1,1 . . . q1,n1
log
=0
p1,1 . . . p1,m1
That is, we must have (p1 , q1 ) ∈ D1 , where D1 is defined in (5).
The terms which might possibly contribute to (7) are
X′
exp (i2F (p, q)) − exp (iF (p, q))
r
Q
Qk Qmj
nj
(p1 ,q2 )∈D1
iF (p, q)
j=1
ℓj =1 pj,ℓj
ℓj =1 qj,ℓj
p2 ∈P2 ,...,pk ∈Pk
q2 ∈Q2 ,...,qk ∈Qk
The same argument as above, shows that the terms with
q2,1 . . . q2,n2
6= 0
log
p2,1 . . . p2,m2
contribute
1
1
)N λ2
exp (1 − 20k
X
(p1 ,q2 )∈D1
p2 ∈P2 ,...,pk ∈Pk
q2 ∈Q2 ,...,qk ∈Qk
1
)N λ2
≤ exp −(1 −
10k
r
1
Qk
j=1
Q
mj
X
(p1 ,q2 )∈D1
ℓj =1 pj,ℓj
q Q
m1
Q
nj
ℓj =1 qj,ℓj
1
Qn1
ℓ=1 p1,ℓ
ℓ=1 q1,ℓ
If (p1 , q2 ) ∈ D1 , then (6) shows that
X
1
nj
q Q
≪ (log N )
Qn1
m1
(p1 ,q2 )∈D1
ℓ=1 q1,ℓ
ℓ=1 p1,ℓ
1
which is negligible compared to exp −(1 − 10k
)N λ2 . Hence, finally, these
terms do not contribute.
Repeating the argument for λj , j = 3, 4, . . . , k we see that any term with
qj,1 . . . qj,nj
log
6= 0
pj,1 . . . pj,mj
has a vanishing contribution to the large-N limit. Therefore, the main term
must come from those terms for which
qj,1 . . . qj,nj
=0
log
pj,1 . . . pj,mj
for all j. Such terms are the diagonal terms, and their contribution has been
calculated above. This completes the proof of Lemma 6.
Proof of Theorem 4. Recall that
λ
Lλ (N, u) =
log ζ( 12 + iueN )
√
.
log N
10
C.P. HUGHES, A. NIKEGHBALI, AND M. YOR
and
1
P (λ, n; k, N, u) = √
log N
X
Nλ
„
N λℓ
p≤exp 40km
ℓ
p−iue
√
p
«
Let
ǫ(λ, n) = ǫ(λ, n; k, N, u)
= Lλ (N, u) − P (λ, n; k, N, u)
so that, if we write T = exp(N λj ), then changing variables to t = T u,
Z
2
|ǫ(λj , mj )|2kmj du
1
=
1
1
km
j
T
(log N )
Z
2T
T
2kmj
−it
X
p log ζ( 1 + it) −
dt
√ 2
p
1/40kmℓ
p≤T
(kmj )4kmj eAkmj
(9)
=O
(log N )kmj
by Theorem 5. Since the mj are fixed, this tends to zero as N → ∞.
Consider
Z 2 Y
k
Y
k
n
n
j
j
m
m
P (λj , mj ) j P (λj , nj ) du
Lλj (N, u) j Lλj (N, u) −
1 j=1
j=1
(10)
Writing Lλj (N, u) in terms of P (λj , mj ) and ǫ(λj , mj ), we see that the term
inside the modulus signs equals
k X
Y
mj
nj
nj −βj
βj
P (λj , mj )mj −αj ǫ(λj , mj )αj P (λj , nj )
ǫ(λj , nj )
αj
βj
0≤α1 ≤m1 ,...,0≤αk ≤mk j=1
0≤β1P
≤n1 ,...,0≤βk ≤nk
αj +βj ≥1
The integral of this in (10) is clearly bounded by



k Y
X
mj
nj
0≤α1 ≤m1 ,...,0≤αk ≤mk
0≤β1P
≤n1 ,...,0≤βk ≤nk
αj +βj ≥1
×
Z
k
2Y
1 j=1

j=1
αj
βj 
nj −βj βj
|P (λj , mj )|mj −αj |ǫ(λj , mj )|αj P (λj , nj )
ǫ(λj , nj ) du
(11)
AN ARITHMETIC MODEL FOR THE TOTAL DISORDER PROCESS
11
A version of the generalized Hölder inequality states that
Z
k
2Y
1 j=1
|Aj ||Bj ||Cj ||Dj | du
≤
k Z
Y
j=1
2
1
2krj
|Aj |
×
1
rj
so long as
+
1
sj
Z
1/(2krj ) Z
du
2
1
2
= 1 and
1
tj
+
1
uj
|Bj |
1/(2ktj ) Z
du
2ktj
|Cj |
1
2ksj
1/(2ksj )
du
2
1
2kuj
|Dj |
du
1/(2kuj )
= 1 for all j = 1, . . . , k.
Choosing rj = mj /(mj − αj ) and sj = mj /αj , and tj = nj /(nj − βj ) and
uj = nj /βj , we see that we may bound the above integral by
k Z
Y
j=1
2
1
2kmj
|P (λj , mj )|
×
Z
1
du
j −αj Z
m2km
j
1
2knj
du
P (λj , nj )
2
From (9), if αj 6= 0,
lim
N →∞
Z
2
1
2
|ǫ(λj , mj )|
nj −βj
2knj
Z
|ǫ(λj , mj )|2kmj du
1
2kmj
du
αj
2km
j
βj
2kn
2knj
j
du
ǫ(λj , nj )
2
αj
2km
j
=0
and from Lemma 6 we have
j −αj
m2km
Z 2
j
|P (λj , mj )|2kmj du
≪1
1
P
Since the sum in (11) is over those αj and βj such that
αj + βj ≥ 1, there
must be at least one j with a non-zero αj or βj . Hence, as N → ∞, all the
terms in (11) tend to zero. The sum is over a finite number of terms, so we
may conclude that
Z 2 Y
k
k
Y
nj
nj mj
mj
P (λj , mj ) P (λj , nj ) du = 0
Lλj (N, u) Lλj (N, u) −
lim
N →∞ 1 j=1
j=1
which implies
Z 2Y
k
k
Y
nj
nj
P (λj , mj )mj P (λj , nj )
Lλj (N, u)mj Lλj (N, u) du = lim
lim
N →∞ 1
j=1
N →∞
j=1
assuming the limits make sense. Therefore Theorem 4 follows from Lemma 6.
12
C.P. HUGHES, A. NIKEGHBALI, AND M. YOR
References
[1] O. Costin and J.L. Lebowitz, “Gaussian fluctuation in random matrices”, Phys. Rev.
Lett. 75 (1995) 69–72
[2] P. Diaconis and S.N. Evans, “Linear functionals of eigenvalues of random matrices”,
Trans. Amer. Math. Soc. 353 (2001) 2615–2633
[3] D. Hejhal, “On Euler products and multi-variate Gaussians”, C.R. Acad. Sci. Paris.
Ser. I. 337 (2003) 223–226
[4] C.P. Hughes, J.P. Keating and N. O’Connell, “On the characteristic polynomial of a
random unitary matrix”, Comm. Math. Phys. 220 (2001) 429–451
[5] J.P. Keating and N.C. Snaith, “Random matrix theory and ζ(1/2 + it)”, Comm.
Math. Phys. 214 (2000) 57–89.
[6] A. Laurinčikas, Limit Theorems for the Riemann Zeta Function (Kluwer Academic
Publishers, 1996)
[7] D. Revuz and M. Yor, Continuous Martingales and Brownian Motion (Springer, 1999)
[8] A. Selberg, “On the remainder in the formula for N (T ), the number of zeros of ζ(s)
in the strip 0 ≤ t ≤ T ”, Avhand. Norske Vid. Akad. Oslo, (1944) 2–27, and also in
Collected Papers, vol I (Springer-Verlag, 1989)
[9] A. Selberg, “Contributions to the theory of the Riemann zeta-function”, Arch. Math.
Natur. B 48, (1946) 89–155, and also in Collected Papers, vol I (Springer-Verlag,
1989)
[10] A. Selberg, “Old and new conjectures and results about a class of Dirichlet series, in
Collected Papers, Vol II (Springer-Verlag, 1991), pp. 47–63
[11] A. Soshnikov, “Level spacings distribution for large random matrices: Gaussian fluctuations”, Ann. of Math. (2) 148 (1998) 573–617
[12] K. Wieand, Eigenvalue Distributions of Random Matrices in the Permutation Group
and Compact Lie Groups, PhD Thesis, (Harvard University, 1998)
[13] K. Wieand, “Eigenvalue distributions of random unitary matrices”, Probability Theory and Related Fields 123 (2), 202-224 (2002)
Department of Mathematics, University of York, Heslington, York, YO10
5DD, U.K.
E-mail address: [email protected]
ETHZ, Departement Mathematik, Rämistrasse 101, HG G16, Zürich 8092,
Switzerland
E-mail address: [email protected]
Laboratoire de Probabilités et Modéles Aléatoires, Université Pierre et
Marie Curie, et C.N.R.S. UMR 7599, 175, rue du Chevaleret, F-75013 Paris,
France