You are on page 1of 6

Available online at www.sciencedirect.

com

ScienceDirect
Journal of Approximation Theory 200 (2015) 221226
www.elsevier.com/locate/jat

Full length article

Finite Hilbert transforms


Dang Vu Giang
Hanoi Institute of Mathematics, Vietnam Academy of Science and Technology, 18 Hoang Quoc Viet,
10307 Hanoi, Viet Nam
Received 12 September 2013; received in revised form 14 July 2015; accepted 10 August 2015
Available online 18 August 2015
Communicated by Vilmos Totik

Abstract
Several interesting formulas concerning finite Hilbert transform and logarithmic integrals are proved
with application determining equilibrium measures.
c 2015 Elsevier Inc. All rights reserved.

MSC: primary 42A20-38; secondary 44A15


Keywords: Hilbert transform; Complex Hardy spaces; Boundary functions; Equilibrium measures

1. Complex Hardy spaces and boundary functions


Recall that the Hilbert transform H f = f of a function f L p (R) (1 p < ) is defined
by letting

1
f (t)

H f (x) = f (x) = (p.v.)


dt.

x
t
We use both notations H f and f for the Hilbert transform of a function f . For example, the
Hilbert transform of the characteristic function (a,b) of the interval (a, b) is

x a
1
.
(a,b) (x) = ln

x b
E-mail address: dangvugiang@yahoo.com.
http://dx.doi.org/10.1016/j.jat.2015.08.001
c 2015 Elsevier Inc. All rights reserved.
0021-9045/

222

D.V. Giang / Journal of Approximation Theory 200 (2015) 221226

To compute the Hilbert transform of several functions we define the complex Hardy spaces
H p (C+ ) where C+ = {z C : Im (z) > 0} and 1 p < . More exactly, H p (C+ ) [4] if
is analytic in C+ and

p
p := sup
| (x + i y)| p d x < .
y>0

It is well known that if H p (C+ ) then for almost every x R there is lim y0 (x + i y) =:
f (x)+i f (x), where f, f L p (R) if 1 < p < . Note that f (x) = Re (x + i0) for f (x) =
Im (x + i0). Therefore, the Hilbert transform is bounded on L p (R) for 1 < p < [4] and
H (H f ) = f for every f L p (R) with 1 < p < . Moreover,


f (x) g (x) d x =
f (x) g (x) d x

for f L p (R) and g L q (R) with 1 < p < and


1

1
p

1
q

= 1. Replace g by (a,b) we have

b
x a
dx =
f (x) ln
f (x) d x
x b
a

for every f L p (R). For a compactly supported function f L p (R) we can define the
logarithmic integral

1
1
F (b) =
d x.
f (x) ln
|x b|

Then
1
F (b) F (a) =

b
x a

dx =
f (x) ln
f (x) d x.
x b
a

Hence, F is locally absolutely continuous with weak derivative f. Specially, we have


Theorem 1. If a function f L p ( p > 1) is supported in a set E of finite disjoint compact
intervals and the logarithmic integral of f is constant in E then f = 0 in E.
Let H p (C+ ) and Hq (C+ ) with 1p + q1 1. Then Hr (C+ ) with
we have

H f g + fg = fg f g with f L p (R) and g L q (R) .

1
r

1
p

1
q

so

(1.1)

Finally, let a1 < a2 < < a2 ,


E=

a2k1 , a2k
k=1

and

K (x) =

x aj .

j=1

Then K (x) 0 if and only if x E. Let

(1)k |K (x)| if x a2k1 , a2k


g (x) = g E (x) =
0
otherwise.

(1.2)

D.V. Giang / Journal of Approximation Theory 200 (2015) 221226

223

Note that
z k1
(z) =
H p (C+ )
K (z)
Here,

for k = 1, 2, . . . , and p (1, 2) .

K (z) z as z . Moreover,

(x + i0) =

(1)m x k1

K (x)

if x (a2m , a2m+1 )

i x k1

g E (x)

if x E.

(m = 0, 1, . . . , , a0 = and a2+1 = ). In fact, it follows from the computation of the


positive harmonic argument of K (z). More exactly, Arg K (x) = (2 j) for x (a j , a j+1 )
and j = 0, 1, . . . , 2. Now using the fact that f (x) = Re (x + i0) for f (x) = Im (x + i0).
We have
1

m x k1

(1)

dy
K (x)
=

g E (y) x y

if x (a2m , a2m+1 )

y k1

if x E.

For example, let (z) = 1/ (z a)(z b), we have

(x a) (x b)

i
(x + i0) =

(x a) (b x)


(x a) (x b)

if x > b
if a < x < b
if x < a

so we have

(x a) (x b)
1
dy
0
=

(y a) (b y) x y


(x a) (x b)

More generally, let E = [b, a] [a, b] and


b2 x 2 x 2 a 2

g (x) =
b2 x 2 x 2 a 2

if x [b, a]
if x [a, b]
otherwise.

if x > b
if a < x < b
if x < a.

(1.3)

224

D.V. Giang / Journal of Approximation Theory 200 (2015) 221226

Then

1 b 2
1
2
2
2
g (x) =
b y
y a

a
xy
x

2
2

a +b

x 2 a 2 x 2 b2
x 2

a 2 + b2
= x2
+
x 2 a 2 x 2 b2

2 + b2

2
x
2

1
dy
+y
if |x| > b
if |x| < a

(1.4)

if a < |x| < b.

2. Finite Hilbert transforms and inversion


Now we are interested in compactly supported positive functions and their Hilbert transforms.
More exactly, if f is supported in the interval [a, b], the finite Hilbert transform of f is given by
b
the Cauchy principal value integral H f (s) = (1/ ) a (s t)1 f (t) dt for real s. By complex
variable arguments we have the inversion formula [3]
b

b
H f (s)
1
f (t) =
(s a)(b s)ds +
f (s)ds
(t a)(b t) a s t
a
for f L p (R) with p > 1. Now let E be a finite union of intervals and assume that f is
supported in E. We are interested in the inversion formula of the Hilbert transform of f . Using
the formula (1.1) after Theorem 1 with g = g E (defined in (1.2)) we have

g E (y) f (y)
g E (y) f (y)
1
1

H f g E + g E f , x =
dy +
dy
E
xy
E
xy

1
gE (x) gE (y)
=
f (y) dy + g E (x) f (x)
E
xy

1
g E (y) f (y)
+
dy
E
xy
= g E (x) f (x) f (x) g E (x) if x E.
Therefore, we have
Theorem 2. The inversion formula

1
g E (x) g E (y)
g E (y) f (y)
f (x) =
f (y) dy +
dy
g E (x)
xy
yx
E
E
holds for f L p supported in E with p > 1 and x E.
It is pretty standard to prove that g E on E is a polynomial of degree (using orthonormal
polynomials on E with respect to the equilibrium measure of E). Hence, the first term

g E (x) g E (y)
f (y) dy
xy
E

D.V. Giang / Journal of Approximation Theory 200 (2015) 221226

225

is a polynomial of degree 1 which is determined uniquely by the first moments of f . It


follows at once from Theorem 2 that if f L p supported in E and f = 0 on E then

1
g E (x) g E (y)
(x)
f (x) =
f (y) dy =
,
g E (x) E
xy
g E (x)
where is a polynomial of degree less than . On the other hand, at the end of Section 1, we have
seen that every function of this form (supported
in the set E) has Hilbert transform vanishing in

E. Specially for = 1, we have g E = |K | so by Theorem 2 we have the density


1
f (t) |K (t)|
f (x) =
f (y) dy +
dt .
(2.1)
tx
|K (x)|
E
E
Now we make another inversion formula which is more applicable. Recall that the equilibrium
measure of a compact set E is the only solution of the energy optimization problem

1
I () =
ln
d (x) d (t) min
|x t|
where is running in the set of Borel probability measures supported in E. The density function
E of the equilibrium measure of E is [2]
E (x) =

1 |1 (x)|
1 1 (x)

=

g E (x)
|K (x)|

where 1 (x) = x 1 + = (t 1 )(t 2 ) (t 1 ) is that unique polynomial satisfying


a2 j+1
1 (x)
dx = 0

|K (x)|
a2 j
for j = 1, 2, . . . , 1. On the other hand, the Hilbert transform of the density function E is
zero in E. In fact, it follows directly from (1.3). The density function E itself is in L q for any
q <
2 if E is afinite union of compact intervals. Let g0 = E and we try to use the formula
H f g0 + fg0 = fg0 f g0 . Because g0 L q for any q < 2 we should assume that f L p
with p > 2. On
the other
hand, f g0 is identically 0, because f is supported on E and g0 = 0 on
E. Hence, H fg0 , x = f (x) g0 (x) for x E. Therefore, we have
Theorem 3. Let f L p (R) for some p > 2. If f is supported in E then

1
f (y) E (y)
dy for a.e. x E,
f (x) =
E (x) E
yx
where E denotes the density function of the equilibrium measure of E.
Remark. The assumption p > 2 is very essential. Otherwise, the density function E itself
does not satisfy this inversion formula. Moreover, if we take g(x) = (x)/g E (x) for x E and
g(x) = 0 for x E we also have g(x)

= 0 for x E. Here, denotes a polynomial of degree


< (the number of holes of E). Therefore, if f is supported in E and f L p for some p > 2
then

g E (x)
(y) f (y)
f (x) =
dy for a.e. x E.
(x) E g E (y)(y x)

226

D.V. Giang / Journal of Approximation Theory 200 (2015) 221226

Theorem 4. Let f L p (R) for some p > 4. If f is supported in E then

2
2
| f (y)| E (y) dy =
f (y) E (y) dy
E
E

2
2
| f (y)| E (y) ydy =
f (y) E (y) ydy
E

where E denotes the density function of the equilibrium measure of E.


Proof. Without loss of generality we assume that f is real valued. Then

2
2
f (y)2 f (y)2 E (y) dy
f (y) | f (y)| E (y) dy =
E
E

H f f, y E (y) dy
=2
E

= 2
f (y) f (y) H E (y) dy = 0
R

(the Hilbert transform of E is identically 0 on E) and the first identity is proved. For the second
one, note that the Hilbert transform of x E (x) is identically 1 on E so

2
f (y)2 f (y)2 E (y) ydy
f (y) | f (y)|2 E (y) ydy =
E
E

H f f, y y E (y) dy
=2
E

= 2
f (y) f (y) H (y E (y)) dy
R

2
f (y) f (y) dy = 0.
=
R
The proof is now complete.
Remark. This theorem is proved in [1] in very special case where f is continuous and E =
[b, b].
Acknowledgment
Deepest appreciation is extended towards the NAFOSTED (the National Foundation for
Science and Technology Development in Vietnam) (25/2015/101/HTN) for the financial
support.
References
[1] M. Rosenblum, J. Rovnyak, Two theorems on finite Hilbert transforms, J. Math. Anal. Appl. 48 (1974) 708720.
[2] E.B. Saff, V. Totik, Logarithmic Potentials with External Fields, in: Grundlehren Math. Wiss., vol. 316, Springer,
Berlin, 1997.
[3] F.G. Tricomi, Integral Equations, Dover Publications, Inc., New York, 1985, p. viii+238. Reprint of the 1957 original.
[4] A. Zygmund, Trigonometric Series, vol. I, II, third ed., Cambridge Mathematical Library, Cambridge University
Press, Cambridge, 2002, p. xii. With a foreword by Robert A. Fefferman, vol. I: xiv+383 pp; vol. II: viii+364 pp.

You might also like