IJMMS 25:5 (2001) 331343

PII. S016117120100391X
http://ijmms.hindawi.com
Hindawi Publishing Corp.
LOCAL MAXIMA OF A RANDOM ALGEBRAIC POLYNOMIAL
K. FARAHMAND and P. HANNIGAN
(Received 22 September 1999)
Abstract. We present a useful formula for the expected number of maxima of a normal
process (t) that occur belowa level u. In the derivation we assume chiey that (t),

(t),
and

(t) have, with probability one, continuous 1 dimensional distributions and expected
values of zero. The formula referred to above is then used to nd the expected number of
maxima below the level u for the random algebraic polynomial. This result highlights the
very pronounced dierence in the behaviour of the random algebraic polynomial on the
interval (1, 1) compared with the intervals (, 1) and (1, ). It is also shown that the
number of maxima below the zero level is no longer O(logn) on the intervals (, 1)
and (1, ).
2000 Mathematics Subject Classication. Primary 60H99, 26C99.
1. Introduction. A signicant amount has been written concerning the mean num-
ber of crossings of a xed level, by both stationary and nonstationary normal pro-
cesses. Farahmand [5] has given a formula for the expected number of maxima, below
a level u, for a normal process under certain assumptions. The initial interest in the
stationary case can be traced back to Rice [13], and was subsequently investigated
by Ito [7] and Ylvisarer [15]. These works concentrated on level crossings and have
been reviewed in the comprehensive book by Cramr and Leadbetter [2]. In this book,
Cramr and Leadbetter have also given a formula for the expected number of local
maxima. In addition, their method enabled them to nd the distribution function for
the height of a local maximum. Leadbetter [9], in his treatment of the nonstationary
case, gives a result for the mean number of level crossings by a normal process. In his
work Leadbetter assumes that the random process has continuous sample functions,
with probability one.
In what follows we consider (t) to be a real valued normal process. We assume
(t), and its rst and second derivatives

(t) and

(t), posses, with probability one,

continuous one dimensional distributions, such that the mean number of crossings of
any level by (t), and the zero level by

(t) are nite. In addition, we assume that the

mean of (t),

(t), and

(t) are all zero. (t) has a local maximum at t =t

i
, if

(t)
has a down crossing of the level zero at t
i
. The local maxima which are of interest here,
are those that occur when (t) is also below the level u. The total number of down
crossings of the level zero by

(t) in (, ) is dened as M(, ), and these occur at

the points < t
1
< t
2
< < t
M(,)
< . We dene M
u
(, ) as the number of zero
down crossings by

(t), where 0 i M(, ) and (t

i
) u. The corollary presented
below is a corollary to Theorem 1.1 in [5]. The corollary is proved in Section 2.
332 K. FARAHMAND AND P. HANNIGAN
Corollary 1.1. For any random process (t) satisfying these conditions we denote
the variances of (t),

(t), and

(t) by A
2
A
2
(t), B
2
B
2
(t), and C
2
C
2
(t),
respectively, and let D D(t), E E(t), and F F(t) represent cov{(t),

(t)},
cov{(t),

(t)}, and cov{

(t),

(t)}, respectively. Then the expected number of max-

ima below the level u, where E{(t)} =E{

(t)} =E{

(t)} =0, is given by

EM
u
(, ) =(2)
1
_

d
_
2e

2a
_
1

_
2au
_
dt
(2)
1
_

d
_
2e

2a
_
1
_
b
2
4ac
_
1/2

_
u

b
2
2c
_
exp
_

aeu
2
c
_
dt
(1.1)
for |

| =A
2
B
2
C
2
A
2
F
2
B
2
E
2
C
2
D
2
+2DEF,
a =
B
2
C
2
F
2
2|

|
, b =
DF B
2
E
|

|
, c =
A
2
B
2
D
2
2|

|
, d =

1/2
, e =c
b
2
4a
.
(1.2)
The above corollary is useful for determining the total expected number of maxima
of a normal process and also for nding the expected number of maxima of a random
polynomial, in both cases these maxima occur below a level u. The result derived by
Farahmand [5] for the random trigonometric polynomial is particularly suited to this
treatment.
In this paper, we concentrate on the random algebraic polynomial
T(x) =
n1
_
j=0
a
j
()x
j
. (1.3)
We assume a
0
(), a
1
(), . . . , a
n1
() is a sequence of independent, normally dis-
tributed random variables dened on a probability space (, A, Pr), each having mean
zero and variance one. Let M
u
(, ) be the number of maxima of the polynomial T(x)
in the interval (, ) below the level u, and let EM
u
(, ) be its mathematical expecta-
tion. We list a small number of relevant results for random polynomials below; a more
complete reference can be found in the book by Bharucha-Reid and Sambandham [1].
Littlewood and Oord [10, 11] initiated the study of random algebraic polynomials
and Oord [12] continues this work in related elds. Kac [8] found the leading be-
haviour and an error term for the expected number of zero crossings of the random
polynomial (1.3). Farahmand [4] found the expected number of K level crossings in
the interval (1, 1) is asymptotic to (1/)log(n/K
2
), and in the intervals (, 1)
and (1, ) is asymptotic to (1/2)logn, provided K
2
/n 0 as n . When K
2
/n
tends to a nonzero constant Farahmand found the expected number of crossings to be
asymptotic to (1/)logn for the interval (, ). Wilkins [14] in recent times using
a delicate method reduced the error term in Kacs result signicantly. Das [3] consid-
ered the polynomial (1.3) with the same assumptions on the random coecients. In
this work, Das showed that the expected number of local maxima in the whole real
line is asymptotic to (

3+1)logn/2. Farahmand and Hannigan [6] showed that if

the random coecients have nonzero mean, , then the expected number of maxima
is asymptotically half that of the Das result, this result remains valid when 0 and
LOCAL MAXIMA OF A RANDOM ALGEBRAIC POLYNOMIAL 333
. The expected number of maxima below a level u, for a random trigonometric
polynomial with the same random coecients as (1.3), was considered previously by
Farahmand [5]. He showed that only about an eighth of the maxima appear below the
level zero, or any level u such that u
2
/n0 as n.
In Theorem 1.2, we show that asymptotically all the maxima on the interval (1, 1)
occur below the level u, where u

n. Also, we see that asymptotically around a fth

of the maxima occur below the level zero, or any level u such that u/
_
logn 0 as
n . On the intervals (, 1) and (1, ) the result is very dierent. In this case,
asymptotically all the maxima on the intervals occur belowthe level uexp(n/logn).
A seemingly strange result at rst is that the number of maxima below the level zero,
or any level u such that u/n
k
0 as n, is negligible compared with the number
of maxima in all. We will show in the proof of Theorem 1.3 that k can be any positive
constant.
In Section 3, we will prove the following two theorems.
Theorem 1.2. If the coecients of T(x) in (1.3) are independent, normally dis-
tributed random variables with mean zero and variance one, then for all suciently
large n, and u u
n
dened below, the mathematical expectation of the number of
maxima below the level u satises
EM
u
(1, 0) =EM
u
(0, 1)
_
31
_
logn
8
, where u=o
__
logn
_
,
EM
u
(1, 0) =EM
u
(0, 1)

3logn
4
, where u

n.
(1.4)
Theorem 1.3. For the same polynomial T(x) dened above, and for all suciently
large n, and uu
n
dened below, then
EM
u
(, 1) =EM
u
(1, ) =o(logn), where u=o
_
n
k
_
,
EM
u
(, 1) =EM
u
(1, )
logn
4
, where uexp
_
n
logn
_
.
(1.5)
The second result in Theorem 1.2 is not surprising when we consider the results,
outlined earlier, from Farahmand [4]. From Kac [8] and Das [3] we know that for the
intervals (, 1) and (1, ) the expected number of both zero crossings and turning
points are asymptotically equal. This suggests that asymptotically all the maxima are
above the level zero; this partially explains the rst result in Theorem 1.3.
2. Local maxima of a normal process. Farahmand proved in Theorem 1.1 of [5]
that
EM
u
(, ) =
_

_
u

_
0

|z|p
t
(x, 0, z)dzdxdt, (2.1)
where p
t
(x, y, z) denotes the 3-dimensional normal density function for (t),

(t),
and

(t). The 3-dimensional density function is based on the covariance matrix

_
=
_
_
_
_
A
2
D E
D B
2
F
E F C
2
_
_
_
_
(2.2)
334 K. FARAHMAND AND P. HANNIGAN
in which
A
2
=var
_
(t)
_
, B
2
=var
_

(t)
_
, C
2
=var
_

(t)
_
,
D =cov
_
(t),

(t)
_
, E =cov
_
(t),

(t)
_
, F =cov
_

(t),

(t)
_
,
(2.3)
and the determinant of the covariance matrix, denoted |

|, is

=A
2
B
2
C
2
A
2
F
2
B
2
E
2
C
2
D
2
+2DEF. (2.4)
We dene p
t
(x, 0, z), not in terms of the variables above, but with the more convenient
notation
a =
B
2
C
2
F
2
2|

|
, b =
DF B
2
E
|

|
, c =
A
2
B
2
D
2
2|

|
, d =

1/2
. (2.5)
That is,
p
t
(x, 0, z) =(2)
3/2
dexp
_
ax
2
bxzcz
2
_
. (2.6)
Combining (2.1) and (2.6) we nd that
EM
u
(, ) =
_

_
u

_
0

(2)
3/2
dexp
_
ax
2
bxzcz
2
_
dzdxdt. (2.7)
By splitting the exponential element in (2.7) into two parts, one of which isolates x,
we have
EM
u
(, ) =(2)
3/2
_

d
_
0

zexp
_

_
c
b
2
4a
_
z
2
_

_
u

exp
_
_
_

_
2ax+bz/

2a
_
2
2
_
_
_
dxdzdt.
(2.8)
It is possible now to rewrite the portion of (2.8) referring to x, in terms of v; that is,
v =

2ax+bz/

2a. It also simplies our analysis if we let

e =c
b
2
4a
. (2.9)
Therefore,
EM
u
(, ) =(2)
3/2
_

2a
_
0

zexp
_
ez
2
_

2au+bz/

2a

exp
_

v
2
2
_
dvdzdt.
(2.10)
At this point we interchange the order of the integration in (2.10) with respect to the
variables v and z, thus
EM
u
(, ) =(2)
3/2
_

2a
_

2au

exp
_

v
2
2
_

_
0

2a(v

2au)/b
zexp
_
ez
2
_
dzdvdt.
(2.11)
LOCAL MAXIMA OF A RANDOM ALGEBRAIC POLYNOMIAL 335
We can integrate with respect to z in (2.11) to nd that
EM
u
(, ) =(2)
3/2
_

d
_
2e

2a
_
1
_
_
_

2au

exp
_

v
2
2
_
dv

2au

exp
_

v
2
2

2ae
_
v

2au
_
2
b
2
_
dv
_
_
dt.
(2.12)
After a little algebraic manipulation we can separate the exponential in the second
integral of (2.12) into two parts, that is
exp
_

v
2
2

2ae
_
v

2au
_
2
b
2
_
=exp
_

_
__
b
2
+4ae
_
/b
2
_
1/2
v
_
32a
3
e
2
/
_
b
4
+4ab
2
e
__
1/2
u
_
2
2
_

_
exp
_

aeu
2
c
_
.
(2.13)
Using (2.13) and integration by substitution, together with our knowledge of the prob-
ability distribution function (x) we can rewrite (2.12) as
EM
u
(, ) =(2)
1
_

d
_
2e

2a
_
1

_
2au
_
dt
(2)
1
_

d
_
2e

2a
_
1
_
b
2
4ac
_
1/2

_
u

b
2
2c
_
exp
_

aeu
2
c
_
dt.
(2.14)
This completes the proof of the corollary.
3. Proof of theorems for the random algebraic polynomial. In order to nd EM
u
(, ) for the random algebraic polynomial (1.3), we principally use the result in
Corollary 1.1. We also use a result due to Das [3].
Applying Corollary 1.1 to the polynomial (1.3), we nd
EM
u
(, ) =(2)
1
_

d
_
2e

2a
_
1

_
2au
_
dx
(2)
1
_

d
_
2e

2a
_
1
_
b
2
4ac
_
1/2

_
u

b
2
2c
_
exp
_

aeu
2
c
_
dx
=
_

I
1
(x)dx
_

I
2
(x)dx,
(3.1)
where
A
2
=var{T(x)}, B
2
=var
_
T

(x)
_
, C
2
=var
_
T

(x)
_
,
D =cov
_
T(x), T

(x)
_
, E =cov
_
T(x), T

(x)
_
, F =cov
_
T

(x), T

(x)
_
,
(3.2)
and

=A
2
B
2
C
2
A
2
F
2
B
2
E
2
C
2
D
2
+2DEF,
a =
B
2
C
2
F
2
2|

|
, b =
DF B
2
E
|

|
, c =
A
2
B
2
D
2
2|

|
, d =

1/2
.
(3.3)
336 K. FARAHMAND AND P. HANNIGAN
Since the coecients of T(x) in (1.3) are independent, normally distributed random
variables with mean zero and variance one, it is a trivial task to show that
A
2
=
n1
_
j=0
x
2j
, B
2
=
n1
_
j=0
j
2
x
2j2
, C
2
=
n1
_
j=0
j
2
(j 1)
2
x
2j4
,
D =
n1
_
j=0
jx
2j1
, E =
n1
_
j=0
j(j 1)x
2j2
, F =
n1
_
j=0
j
2
(j 1)x
2j3
.
(3.4)
By repeated dierentiation of the geometric sum
n1
_
j=0
x
2j
=
1x
2n
1x
2
, (3.5)
we obtain
A
2
=
x
2n
1
x
2
1
,
B
2
=
n
2
x
2n2
x
2
1

2nx
2n
_
x
2
1
_
2
+
_
1+x
2
__
x
2n
1
_
_
x
2
1
_
3
,
C
2
=
n
4
x
2n4
x
2
1

6n
3
x
2n2
2n
3
x
2n4
_
x
2
1
_
2
+
13n
2
x
2n
2n
2
x
2n2
+n
2
x
2n4
_
x
2
1
_
3
+
12x
2n
_
x
2
+1
_
_
x
2
1
_
4
+
_
x
2n
1
__
4x
4
+16x
2
+4
_
_
x
2
1
_
5
,
D =
nx
2n1
x
2
1

x
_
x
2n
1
_
_
x
2
1
_
2
,
E =
n
2
x
2n2
x
2
1

3nx
2n
nx
2n2
_
x
2
1
_
2
+
2x
2
_
x
2n
1
_
_
x
2
1
_
3
,
F =
n
3
x
2n3
x
2
1

n
2
x
2n3
_
4x
2
1
_
_
x
2
1
_
2
+
nx
2n1
(5x
2
+1)
_
x
2
1
_
3

2x
2n+3
+4x
2n+1
2x
3
4x
_
x
2
1
_
4
.
(3.6)
Looking at the properties of the randomcoecients in (1.3) it is clear that EM
u
(b, a)
= EM
u
(a, b). Obviously, we need only nd the leading behaviour for EM
u
(0, 1) and
EM
u
(1, ) to prove the two theorems.
To this end we look initially at the interval 0 <x <1, where
n
=n

, and
is any positive value, satisfying
=
1loglogn
k
logn
. (3.7)
The positive constant k is critical to these theorems and we will return to dene
a set of qualifying k values later. From (3.6), and since for all n suciently large,
LOCAL MAXIMA OF A RANDOM ALGEBRAIC POLYNOMIAL 337
x
2n
<(1n

)
2n
<exp(2n
1
) we have
A
2
=
_
1x
2
_
1
_
1+O
_
exp
_
2n
1
___
,
B
2
=
_
1+x
2
__
1x
2
_
3
_
1+O
_
n
2
exp
_
2n
1
___
,
C
2
=
_
4+16x
2
+4x
4
__
1x
2
_
5
_
1+O
_
n
4
exp
_
2n
1
___
,
D =x
_
1x
2
_
2
_
1+O
_
nexp
_
2n
1
___
,
E =2x
2
_
1x
2
_
3
_
1+O
_
n
2
exp
_
2n
1
___
,
F =
_
4x+2x
3
__
1x
2
_
4
_
1+O
_
n
3
exp
_
2n
1
___
.
(3.8)
Using (3.8) and the denitions we have for a, b, c, d, e, and |

| it is a reasonably
simple task to show that
a =
_
1x
6
__
1+O
_
n
4
exp
_
2n
1
___
2
,
b =
x
2
_
1x
2
_
3
_
1+O
_
n
4
exp
_
2n
1
___
2
,
c =
_
1x
2
_
5
_
1+O
_
n
4
exp
_
2n
1
___
8
,
d =
_
1x
2
_
9/2
_
1+O
_
n
4
exp
_
2n
1
___
2
,
e =
_
1x
2
_
5
_
1x
4
__
1x
6
_
1
_
1+O
_
n
4
exp
_
2n
1
___
8
.
(3.9)
Employing our denition of in (3.7) we can show
exp
_
2n
1
_
=exp
_
2logn
k
_
=n
2k
, (3.10)
for all suciently large n, all terms inside O{ } in (3.8) and (3.9) will tend to zero if k
is any number greater than 2. Using (3.1) and (3.9) we have that
_
1
0
I
1
(x)dx
_
1n

0
_
1+x
2
+x
4
_
1/2
_
1x
4
_
1

_
_
1x
6
_
1/2
u
_
dx. (3.11)
Obviously,

1
_
1n

1(logn)
1
_
1+x
2
+x
4
_
1/2
_
1x
4
_
1

_
_
1x
6
_
1/2
u
_
dx

_
1
0
I
1
(x)dx

1
_
1(logn)
1
0
_
1+x
2
+x
4
_
1/2
_
1x
4
_
1
dx
+
1
_
1n

1(logn)
1
_
1+x
2
+x
4
_
1/2
_
1x
4
_
1

_
_
1x
6
_
1/2
u
_
dx.
(3.12)
Concentrating on the integral common to both sides of the composite inequality
(3.12) and using the fact that {(1x
6
)
1/2
u} is a decreasing function on the interval
338 K. FARAHMAND AND P. HANNIGAN
(1(logn)
1
, 1n

), it can be shown that

_
_
6
n

_
1/2
u
_
_
1n

1(logn)
1
_
1+x
2
+x
4
_
1/2
_
1x
4
_
1
dx

1
_
1n

1(logn)
1
_
1+x
2
+x
4
_
1/2
_
1x
4
_
1

_
_
1x
6
_
1/2
u
_
dx

_
_
6
logn
_
1/2
u
_
_
1n

1(logn)
1
_
1+x
2
+x
4
_
1/2
_
1x
4
_
1
dx.
(3.13)
For some
n
, such that 0 <
n
<1, and lim
n

n
=0, we can see that
_

__
1
0
x
2
_
1x
4
_
1
dx
1
_
1
0
_
1+x
2
+x
4
_
1/2
_
1x
4
_
1
dx

__
1
0
_
1x
4
_
1
dx.
(3.14)
Using partial fraction expansions in (3.14) we can quickly show that

1
_
1
0
_
1+x
2
+x
4
_
1/2
_
1x
4
_
1
dx

3log(1/)
4
. (3.15)
Hence, using (3.11), (3.12), (3.13), (3.14), and (3.15) it becomes clear that
_
1
0
I
1
(x)dx

3logn
8
, where u=o
__
logn
_
, (3.16)
_
1
0
I
1
(x)dx

3logn
4
, where u

f, (3.17)
if f is any function of n such that it tends to innity as n tends to innity. The
condition on u in (3.17) can be greatly simplied, leading to a more meaningful result,
by increasing the level u marginally, to yield
_
1
0
I
1
(x)dx

3logn
4
, where u

n. (3.18)
At this juncture we turn our attention to the integral of I
2
(x) and we employ many of
the techniques previously used. Using (3.1) and (3.9),
_
1
0
I
2
(x)dx
1
_
1n

0
x
2
_
1x
4
_
1

_
x
2
_
1x
2
_
1/2
u
_
exp
_

_
1x
4
_
u
2
2
_
dx.
(3.19)
Obviously,

1
_
1n

1(logn)
1
x
2
_
1x
4
_
1

_
x
2
_
1x
2
_
1/2
u
_
exp
_

_
1x
4
_
u
2
2
_
dx

_
1
0
I
2
(x)dx

1
_
1(logn)
1
0
x
2
_
1x
4
_
1
exp
_

_
1x
4
_
u
2
2
_
dx
+
1
_
1n

1(logn)
1
x
2
_
1x
4
_
1

_
x
2
_
1x
2
_
1/2
u
_
exp
_

_
1x
4
_
u
2
2
_
dx.
(3.20)
LOCAL MAXIMA OF A RANDOM ALGEBRAIC POLYNOMIAL 339
As in the integral of I
1
(x), the integral common to both sides of the double inequality
(3.20), is critical to our proof. Since x
2
(1x
2
)
1/2
has a maximumat (1/3)
1/3
, it is clear
that {x
2
(1x
2
)
1/2
u} is decreasing on the interval (1(logn)
1
, 1n

). It is also
evident that exp{(1x
4
)u
2
/2} is an increasing function for x (0, 1). Combining
our knowledge of {x
2
(1x
2
)
1/2
u} and exp{(1x
4
)u
2
/2} it is clear that

_
_
2
n

_
1/2
u
_
exp
_

2u
2
logn
__
1n

1(logn)
1
x
2
_
1x
4
_
1
dx

1
_
1n

1(logn)
1
x
2
_
1x
4
_
1

_
x
2
_
1x
2
_
1/2
u
_
exp
_

_
1x
4
_
u
2
2
_
dx

_
_
2
logn
_
1/2
u
_
exp
_

2u
2
n

__
1n

1(logn)
1
x
2
_
1x
4
_
1
dx.
(3.21)
Simply by integrating the terms in (3.21) we nd
(4)
1

_
_
2
n

_
1/2
u
_
exp
_

2u
2
logn
+loglogn
_

1
_
1n

1(logn)
1
x
2
_
1x
4
_
1

_
x
2
_
1x
2
_
1/2
u
_
exp
_

_
1x
4
_
u
2
2
_
dx
(4)
1

_
_
2
logn
_
1/2
u
_
exp
_

2u
2
logn
+loglogn
_
.
(3.22)
Hence, using (3.20) and (3.22) it is clear that
_
1
0
I
2
(x)dx
logn
8
, where u=o
__
logn
_
, (3.23)
_
1
0
I
2
(x)dx 0, where u

n. (3.24)
From (3.16), (3.18), (3.23), and (3.24) we have proved Theorem 1.2 on the intervals
(1+, 0) and (0, 1). We will use a result due to Das [3] to complete the proof of
the theorem. From (3.18) and (3.24) it is clear that
EM
u
(0, 1)

3logn
4
, where u

n. (3.25)
This is the same asymptotic value that Das [3] obtained for the number of maxima on
the interval (0, 1). Therefore, EM
u
(1, 1) =o(logn), irrespective of the level u and
the theorem is proved.
In Theorem 1.3, we want to evaluate EM
u
(1, ) asymptotically. It becomes apparent
later that we need only nd EM
u
(1+, ) (where is dened below), and it proves
convenient to make the substitution y =1/x. If we let 1+ =(1)
1
, where =n

as before. It will be necessary to dene a set of qualifying k values for (3.7). We make
340 K. FARAHMAND AND P. HANNIGAN
the substitution in (3.6) to nd
A
2
=y
2n+2
_
1y
2
_
1
_
1+O
_
exp
_
2n
1
___
,
B
2
=n
2
y
2n+4
_
1y
2
_
1
_
12n
1
_
1y
2
_
1
+
_
1+y
2
_
n
2
_
1y
2
_
2
+O
_
n
2+2
exp
_
2n
1
___
,
C
2
=n
4
y
2n+6
_
1y
2
_
1
_
1+
_
6+2y
2
_
n
1
_
1y
2
_
1
+
_
132y
2
+y
4
_
n
2
_
1y
2
_
2
12
_
1+y
2
_
n
3
_
1y
2
_
3
+
_
4+16y
2
+4y
4
_
n
4
_
1y
2
_
4
+O
_
n
4+4
exp
_
2n
1
___
,
D =ny
2n+3
_
1y
2
_
1
_
1n
1
_
1y
2
_
1
+O
_
n
1+
exp
_
2n
1
___
,
E =n
2
y
2n+4
_
1y
2
_
1
_
1+
_
3+y
2
_
n
1
_
1y
2
_
1
+2n
2
_
1y
2
_
2
+O
_
n
2+2
exp
_
2n
1
___
,
F =n
3
y
2n+5
_
1y
2
_
1
_
1
_
4y
2
_
n
1
_
1y
2
_
1
+
_
5+y
2
_
n
2
_
1y
2
_
2

_
2+4y
2
_
n
3
_
1y
2
_
3
+O
_
n
3+3
exp
_
2n
1
___
.
(3.26)
Using (3.26) and our denition for a, b, c, d, e, and |

| it can be shown that

a =n
4
_
1y
2
_
5
y
2n
_
8y
6
_
1
_
1+
_
62y
2
_
n
1
_
1y
2
_
1
+
_
13+10y
2
+y
4
_
n
2
_
1y
2
_
2
+
_
1212y
2
_
n
3
_
1y
2
_
3
+
_
4+4y
2
+4y
4
_
n
4
_
1y
2
_
4
+O
_
exp
_
2n
1
___
,
b =n
2
y
2n2
_
1y
2
_
5
_
4y
6
_
1
_
1+
_
3y
2
_
n
1
_
1y
2
_
1
+2n
2
_
1y
2
_
2
+O
_
exp
_
2n
1
___
,
c =y
2n4
_
1y
2
_
5
_
8y
6
_
1
_
1+O
_
exp
_
2n
1
___
,
d =y
3n6
_
1y
2
_
9/2
_
2y
3
_
1
_
1+O
_
exp
_
2n
1
___
,
e =y
2n4
_
1y
2
_
3
_
2n
2
y
2
_
1
_
1+O
_
exp
_
2n
1
___
.
(3.27)
Since we are using the denition of given in (3.7), it is clear that k can be any positive
constant in this case.
Employing (3.1) and (3.27) it is clear that
_

1+
I
1
(x)dx (2)
1
_
1
0
y
2
_
1y
2
_
1

_
n
2
_
1y
2
_
5/2
y
n3
u
2
_
dy. (3.28)
It can easily be shown that { } in (3.28) is an increasing function on the interval
(0, 1). In much the same way as we proved Theorem 1.2, we see that
LOCAL MAXIMA OF A RANDOM ALGEBRAIC POLYNOMIAL 341
(2)
1
_
1n

1(logn)
1
y
2
_
1y
2
_
1

_
n
2
_
1y
2
_
5/2
y
n3
u
2
_
dy

1+
I
1
(x)dx
(2)
1
_
1(logn)
1
0
y
2
_
1y
2
_
1
dy
+(2)
1
_
1n

1(logn)
1
y
2
_
1y
2
_
1

_
n
2
_
1y
2
_
5/2
y
n3
u
2
_
dy.
(3.29)
As in Theorem 1.2, we concentrate on the integral in the interval (1(logn)
1
, 1n

)
and use the fact that { } is an increasing function
(2)
1

_
2

2n
2
exp(n/logn)u
(logn)
5/2
__
1n

1(logn)
1
y
2
_
1y
2
_
1
dy
(2)
1
_
1n

1(logn)
1
y
2
_
1y
2
_
1

_
n
2
_
1y
2
_
5/2
y
n3
u
2
_
dy
(2)
1

_
2

2n
2
exp
_
n
1
_
u
n
5/2
__
1n

1(logn)
1
y
2
_
1y
2
_
1
dy.
(3.30)
Hence, using (3.29), (3.30), and a little simple integration it is clear that
_

1+
I
1
(x)dx
logn
8
, where u=o
_
n
k
_
, (3.31)
_

1+
I
1
(x)dx
logn
4
, where u
exp(n/logn)(logn)
5/2
f
n
2
, (3.32)
if f is any function which tends to innity as n tends to innity.
The only signicant part to prove now is the integral of I
2
(x) on the interval
((1+)
1
, ). From (3.1) and (3.27) we can show
_

1+
I
2
(x)dx (2)
1
_
1
0
y
2
_
1y
2
_
1
exp
_

n
2
_
1y
2
_
3
y
2n2
u
2
2
_

_
n
2
_
1y
2
_
5/2
y
n3
u
2
_
dy.
(3.33)
By calculus methods we can prove that exp{ } in (3.33) is a decreasing function on the
interval (0, 1). The { } element in (3.33) is identical to the { } in (3.29). In much
the same way as before
(2)
1
_
1n

1(logn)
1
y
2
_
1y
2
_
1
exp
_

n
2
_
1y
2
_
3
y
2n2
u
2
2
_

_
n
2
_
1y
2
_
5/2
y
n3
u
2
_
dy

1+
I
2
(x)dx
(2)
1
_
1(logn)
1
0
y
2
_
1y
2
_
1
exp
_

n
2
_
1y
2
_
3
y
2n2
u
2
2
_
dy (3.34)
+(2)
1
_
1n

1(logn)
1
y
2
_
1y
2
_
1
exp
_

n
2
_
1y
2
_
3
y
2n2
u
2
2
_

_
n
2
_
1y
2
_
5/2
y
n3
u
2
_
dy.
342 K. FARAHMAND AND P. HANNIGAN
Also,
(2)
1
exp
_

4n
2
exp
_
2n
1
_
u
2
n
3
_

_
2

2n
2
exp(n/logn)u
(logn)
5/2
_

_
1n

1(logn)
1
y
2
_
1y
2
_
1
dy
(2)
1
_
1n

1(logn)
1
y
2
_
1y
2
_
1
exp
_

n
2
_
1y
2
_
3
y
2n2
u
2
2
_

_
n
2
_
1y
2
_
5/2
y
n3
u
2
_
dy
(2)
1
exp
_

4n
2
exp(2n/logn)u
2
(logn)
3
_

_
2

2n
2
exp
_
n
1
_
u
n
5/2
_

_
1n

1(logn)
1
y
2
_
1y
2
_
1
dy.
(3.35)
We can tidy (3.35) to yield
(4)
1

_
2

2n
2
exp(n/logn)u
(logn)
5/2
_
exp
_

4
_
logn
k
_
2
u
2
n
+2k
+loglogn
_
(2)
1
_
1n

1(logn)
1
y
2
_
1y
2
_
1
exp
_

n
2
_
1y
2
_
3
y
2n2
u
2
2
_

_
n
2
_
1y
2
_
5/2
y
n3
u
2
_
dy
(4)
1

_
2

2
_
logn
k
_
2
u
n
(/2)+k
_
exp
_

4n
2
exp(2n/logn)u
2
(logn)
3
+loglogn
_
.
(3.36)
Clearly
_

1+
I
2
(x)dx
logn
8
, where u=o
_
n
k
_
, (3.37)
_

1+
I
2
(x)dx 0, where u
exp(n/logn)(logn)
5/2
f
n
2
, (3.38)
if f is any function that tends to innity as n . Collecting together (3.31), (3.32),
(3.37), and (3.38) we have proved a marginally more general result than is stated
in Theorem 1.3. By increasing the level u given in (3.32) and (3.38) a little to u
exp(n/logn), we obtain the result in Theorem 1.3 which can be more readily
understood.
References
[1] A. T. Bharucha-Reid and M. Sambandham, Random Polynomials, Probability and Math-
ematical Statistics, Academic Press, Inc., Orlando, Fla., 1986. MR 87m:60118.
Zbl 615.60058.
[2] H. Cramr and M. R. Leadbetter, Stationary and Related Stochastic Processes. Sample Func-
tion Properties and their Applications, John Wiley & Sons, Inc., New York, London,
Sydney, 1967. MR 36#949. Zbl 162.21102.
LOCAL MAXIMA OF A RANDOM ALGEBRAIC POLYNOMIAL 343
[3] M. Das, The average number of maxima of a random algebraic curve, Proc. Cambridge
Philos. Soc. 65 (1969), 741753. MR 39#1026. Zbl 218.60017.
[4] K. Farahmand, On the average number of real roots of a random algebraic equation, Ann.
Probab. 14 (1986), no. 2, 702709. MR 87k:60140. Zbl 609.60074.
[5] , Local maxima of a random trigonometric polynomial, J. Theoret. Probab. 7 (1994),
no. 1, 175185. MR 95b:60059. Zbl 786.60069.
[6] K. Farahmand and P. Hannigan, The expected number of local maxima of a random al-
gebraic polynomial, J. Theoret. Probab. 10 (1997), no. 4, 9911002. MR 99a:60053.
Zbl 897.60055.
[7] K. It, The expected number of zeros of continuous stationary Gaussian processes, J. Math.
Kyoto Univ. 3 (1963/1964), 207216. MR 29#4097. Zbl 139.34102.
[8] M. Kac, On the average number of real roots of a random algebraic equation, Bull. Amer.
Math. Soc. 49 (1943), 314320. MR 4,196d. Zbl 060.28602.
[9] M. R. Leadbetter, On crossings of levels and curves by a wide class of stochastic processes,
Ann. Math. Statist. 37 (1966), 260267. MR 34#8476. Zbl 141.14906.
[10] J. E. Littlewood and A. C. Oord, On the number of real roots of a random algebraic
equation, J. London Math. Soc. 13 (1938), 288295. Zbl 020.13604.
[11] , On the number of real roots of a randomalgebraic equation. II, Proc. Camb. Philos.
Soc. 35 (1939), 133148. Zbl 021.03702.
[12] A. C. Oord, The distribution of the values of an entire function whose coecients are
independent random variables. II, Math. Proc. Cambridge Philos. Soc. 118 (1995),
no. 3, 527542. MR 96m:30046. Zbl 846.60037.
[13] S. O. Rice, Mathematical analysis of random noise, Bell System Tech. J. 24 (1945), 46156.
MR 6,233i. Zbl 063.06487.
[14] J. E. Wilkins, Jr., An asymptotic expansion for the expected number of real zeros of
a random polynomial, Proc. Amer. Math. Soc. 103 (1988), no. 4, 12491258.
MR 90f:60105. Zbl 656.60062.
[15] N. D. Ylvisarer, The expected number of zeros of a stationary Gaussian process, Ann. Math.
Statist. 36 (1965), 10431046. MR 31#1721. Zbl 139.34201.
K. Farahmand: Department of Mathematics, University of Ulster, Jordanstown, Co.
Antrim BT37 0QB, UK
E-mail address: k.farahmand@ulst.ac.uk
P. Hannigan: Department of Mathematics, University of Ulster, Jordanstown, Co.
Antrim BT37 0QB, UK