Chapter IX The Integral Transform Methods IX.

4 The Hankel Transform November 25, 2018 803

IX.4 The Hankel Transform

IX.4.1 The Hankel Transform – Definition and Operational Properties 804

IX.4.2 Instantaneous Heat Source 807
IX.4.3 Heat Equation with Instantaneous Line Source 808
IX.4.4 Solution of PDE with Application of Two Integral Transforms 816
(Fourier and Hankel Transforms) – Diffusion of Light in Semi-Infinite Space
IX.4.5 Finite Hankel Transform
IX.4.5.1 Circular Domain (FHT-1) 820
IX.4.5.2 Development of Velocity Profile in a Pipe under a Pressure Gradient
IX.4.5.3 Annular Domain (FHT-2) 827
IX.4.5.4 Viscous Fluid Flow between Two Cylinders
IX.4.7.8 Some additional notes on Hankel Transform 842
804 Chapter IX The Integral Transform Methods IX.4 The Hankel Transform November 25, 2018

IX.4.1 HANKEL TRANSFORM In the previous sections, we considered application of the Fourier
and the Laplace transforms to the solution of PDE’s in the
Cartesian coordinate system. Application of these transforms
eliminated the corresponding derivatives from the differential
equation. In the cylindrical coordinate system, the Laplacian
includes the term
1 ∂  ∂u 
r 
r ∂r  ∂r 
We want to construct an integral transform which eliminates this
term. Formally, such transform can be constructed from the two-
dimensional Fourier transform with transition to polar
coordinates and application of the integral representation of the
Bessel functions [Debnah]. Then the transform of the function
f (r ) of the polar variable r ∈ [0 , ∞ ) is defined as

∞
Hν { f=
( r )} f=
ν (λ ) ∫ rJν ( λ r ) f ( r ) dr
Jν ( λ r ) is a Bessel function
0
of the 1st kind of the order ν .
It is called the Hankel Transform of order ν .
It is the bounded solution of
the Bessel Equation The inverse Hankel transform of order ν is defined as
(
r 2 y ′′ + ry ′ + λ 2 r 2 −ν 2 y =
0)
for r ∈ [0,∞ ) ∞
{ }
H ν−1 fν (λ ) = f (r ) = ∫ λJν (λr ) fν (λ )dλ
0
J0 ( x )

J1 ( x )
J2 ( x) The conditions for existence of the Hankel transform are similar
to the conditions of existence of the Fourier transform, which are
usually satisfied for the functions which describe the physical
models in engineering.

The Hankel transform and its inverse are linear operators.

The physical domain where we are supposed to apply the Hankel
transform is the infinite 2-dimensional plane with polar
coordinates (r ,θ ) . The r-coordinate term of the Laplacian

1 ∂  ∂u  1 ∂u ∂ 2 u
r =  +
r ∂r  ∂r  r ∂r ∂r 2

is the object for the Hankel transform.

Similar to the previous models, we also assume in this case that
the function u and its derivative have radial decaying at infinity:
∂u
ru r →∞ = 0 , r =0
∂r r →∞

Operational Properties To study the operational properties, consider application of the

Hankel transform of order ν to the expression

1 ∂  ∂u  ν 2
r − u
r ∂r  ∂r  r 2
where the additional to Laplacian the second term defnes the
order of the applied Hankel transform.
 Chapter IX The Integral Transform Methods IX.4 The Hankel Transform November 25, 2018 805

 1 ∂  ∂u  ν 2  ∞
 1 ∂  ∂u  ν 2 
Hν  r − 2 u = ∫ rJν (λr ) r  − 2 u dr
 r ∂r  ∂r  r  0  r ∂r  ∂r  r 
∞
∂  ∂u  ∞
ν 2 
= ∫ Jν (λr )  r dr − ∫ rJν (λr ) 2 u dr
0 ∂r  ∂r  0 r 
∞
 ∂u 
∞
ν 2 
= ∫ Jν (λr )d  r  − ∫ rJν ( λ r )  2 u dr
0  ∂r  0 r 
∞
 ∂u   ∞
∂u ∞
ν 2 
=  r  Jν (λr ) − ∫ r d [Jν (λr )] − ∫ rJν (λr ) 2 u dr
 ∂r   0 0 ∂r 0 r 

[Jν (λr )]′dr − ∫ rJν (λr )ν 2 u dr

∞ ∞ 2
∂u
= −∫ r
0 ∂r 0 r 
∞
′
∞
ν 2 
= − ∫ r [Jν (λr )] du − ∫ rJν (λr ) 2 u dr
0 0 r 

{ } { }
∞
′ ∞
′ ∞
ν 2 
= ru  Jν ( λ r )  + ∫ ud r  Jν ( λ r )  − ∫ rJν ( λ r )  2 u  dr
From self-adjoint form of
0 0 0 r 

{ }
the Bessel Equation: ∞ ′ ∞
ν 2 
′
 ν2
 = ∫ u r  Jν ( λ r )  dr − ∫ rJν ( λ r )  2 u  dr
( ry ′ )′ +  − + λ 2 r  y =0 0 0 r 
 r 
write : ∞
ν 2  ∞
ν 2 
= ∫ ur  2 − λ2 Jν (λr )dr − ∫ rJν (λr ) 2 u dr
 ν2 
( ry ′ )′ = r  − 2 + λ 2  y 0 r  0 r 
 r 
∞ ∞
ν 2  ∞
ν 2 
Then for y = Jν ( λ r ) : = −λ2 ∫ r Jν (λr )udr + ∫ rJν (λr ) 2 u dr − ∫ rJν (λr ) 2 u dr
0 0 r  0 r 
 ν2 
r ) ′ r  λ 2 − 2  Jν ( λ r )
 rJν′ ( λ=
 r  ∞
= −λ2 ∫ r Jν (λr )udr
0

= −λ2 uν (λ )

Therefore,

  ν2   ∞
 1 ∂  ∂u  ν 2 
Hν  ∇ 2 − 2  u ( r )  = ∫ rJν (λr ) r ∂r  r ∂r  − r 2 u dr = −λ2 uν (λ )
  r   0  

0th order and 1st order Hankel Transforms In particular, the zero order Hankel transform eliminates the r-
term of the Laplacian

∞ ∞
 1 ∂  ∂u    1 ∂  ∂u  
H0  r  = ∫ rJ 0 ( λ r )  r ∂r  r ∂r   dr = −λ2 u0 (λ ) u0 (λ ) = ∫ rJ 0 ( λ r ) u ( r )dr 0th order
 r ∂r  ∂r   0   0
∞ ∞
 1 ∂  ∂u  u   1 ∂  ∂u  u 
H1   r ∂r  − 2  =∫ rJ 1 ( λ r )  r ∂r  r ∂r  − 2  dr = −λ u1 ( λ )
2
u1 ( λ ) = ∫ rJ 1 ( λ r ) u ( r )dr 1st order
 r ∂r   r  0    r  0

In the future, we will use mostly the zero order Hankel transform
(index 0 in the transformed function is usually omitted).
806 Chapter IX The Integral Transform Methods IX.4 The Hankel Transform November 25, 2018

EXAMPLE TRANSIENT HEAT TRANSFER Cooling of bell-shaped temperature profile

Consider the axisymmetric case of the Heat Equation in
cylindrical coordinates
1 ∂  ∂u  ∂u
∇2u ≡ r  = a2 r ∈ [0, ∞ ) t > 0
r ∂r  ∂r  ∂t

u0 ( r ) = e − br
2

u (=
r,0 ) u=
0 (r )
2
e − br (initial condition)

There are no boundary conditions but we assume that the

function u and its derivative has radial decaying at infinity:
infinite metal ∂u
ru r →∞ = 0 r =0
plate ∂r r →∞
Physically it can be treated as a heated to bell shaped temperature
profile infinite plate thermoinsulated from both sides.
Heat is conducted along the plate.
Thermal diffusivity of the plate is a 2 .

1) Transformed equation Apply zero order Hankel transform

∞
u (λ ,t ) = ∫ rJ 0 (λr )udr
0
to the equation
∂u
−λ 2 u =a2
∂t
and to initial condition (see Table 17.1 Exponential functions)
λ2
∞ ∞ −
e 4b
u (λ ,0 ) ==
∫0 ( ) ∫0 ( )
2
rJ 0 λ r u0 dr =rJ 0 λ r e − br
dr
2b
That is the 1 order linear ODE
st

∂u λ 2
+ u= 0
∂t a 2
The solution is
−λ2
t
u ( λ ,t ) = u ( λ ,0 ) e a
2

λ2
− −λ2  1 t 
e 4b a2 t 1 − 4b + a2  λ
2

= = e e
2b 2b

2) Inverse transform Solution of the problem (use table)

∞
u (r ,t ) = ∫ λ J 0 ( λ r ) u ( λ ,t ) d λ
0
∞  1 t 
1 − 4b + a2  λ
2

= ∫ λ J0 ( λ r ) e dλ
0
2b
r2
−
 1 t 
1 1 4 + 2 
= ⋅ e  4b a 
2b  1 t 
2 ⋅ + 2 
 4b a 
u ( r,t )
 
1  r2 
u (r ,t ) = ⋅ exp  − 
4bt  1 + 4t 
1+ 2
a  b a2 
 Chapter IX The Integral Transform Methods IX.4 The Hankel Transform November 25, 2018 807

IX.4.2 INSTANTENEOUS ENERGY SOURCE We consider the volumetric heat source S (r ,t ) which releases its
entire energy spontaneously at the time t = t0 at the points r = r0
[Ozisik, HC, p.220].
S ( r,t )=
g
δ ( r − r0 ) δ ( t − t0 )
W ⋅ s 
g In the axisymmetric case (does not depend on θ ) of cylindrical

2π r  m  coordinates and no dependence on the z coordinate (the same
source strength for any plane perpendicular to the z axis), this is a
cylindrical surface heat source which can be represented with
the help of the Dirac delta function
g W 
S (r ,t ) = δ (r − r0 )δ (t − t0 )  m3 
2πr  
W ⋅ s 
where g  m  is the source strength per unit length
 
g W ⋅ s 
2πr  m 2  is the source strength per unit area
 
r [m] the variable r appears in the
denominator as a scale factor of the
transformation to polar coordinates
1
δ (r − r0 )   defines the radius of the source r = r0
m
1
δ (t − t0 )   defines the moment of time t = t0
s
W ⋅ s  Integrated over the entire time-space domain (in the plane
g
 m 
 perpendicular to the z coordinate), it should yield the strength of
g the heat source (conservation of energy):
S ( r,t )
= δ ( r ) δ ( t − t0 ) g [W ⋅ s ]
2π r ∞ 2π ∞ ∞ 2π ∞
 g 
( r,t )
S=
g
δ ( r ) δ ( z − z0 ) δ ( t − t0 ) ∫ ∫ ∫ S (r ,t )rdrdθdt =∫ ∫ ∫  2πr δ (r − r0 )δ (t − t0 ) rdrdθdt
2π r 0 0 0 0 0 0
∞ 2π ∞
g
=
2π ∫ ∫ ∫ δ (r − r0 )δ (t − t0 )drdθdt
0 0 0
∞ 2π
g
=
2π ∫ ∫ δ (t − t0 )dθdt
0 0
2π
g
=
2π ∫ dθ
0

g
= [θ ]02π
2π
=g
g [W ⋅ s ]
In the 3-dimensional case, the heat source is a circular ring in
g
S ( r,t ) = δ ( r − r0 ) δ ( z − z0 ) δ ( t − t0 ) the plane z = z0 defined by
2π r
g
S (r ,t ) = δ (r − r0 )δ (z − z0 )δ (t − t0 )
2πr
An instantaneous source is used for determination of the Green
function for the non-homogeneous PDE. The heat source for HE
equation is defined by
S ( r,t )
k
 W 
where k   is a thermal conductivity of the medium and
m⋅K 
W 
S  3  is a volumetric heat generation rate.
m 
808 Chapter IX The Integral Transform Methods IX.4 The Hankel Transform November 25, 2018

IX.4.3 HEAT EQUATION WITH INSTANTENEOUS LINE SOURCE

r = r0 = 0 Consider the axisymmetric case of the Heat Equation in
cylindrical coordinates with the instantaneous line source located
at r = r0 = 0 (line source in 3-D)

1 ∂  ∂u  2 ∂u g
∇ 2u ≡ r  = a − δ (r )δ (t − t0 ) r ∈ [0,∞ ) t > 0
r ∂r  ∂r  ∂t 2πrk

u (r ,0 ) = u0 (r ) = 0 (initial condition)

There are no boundary conditions but we assume that the

function u and its derivative has radial decaying at infinity:
∂u
ru r →∞ = 0 r =0
∂r r →∞
1) Transformed equation Apply zero order Hankel transform
∞
u (λ ,t ) = ∫ rJ 0 (λr )udr
0
to the equation and initial condition
∞
∂u g
− λ2 u = a 2 − δ (t − t0 ) u (λ ,0 ) = ∫ rJ 0 (λr )u0 dr = 0
∂t 2πk 0
this is the 1st order linear ODE
∂u λ2 g
+ u= δ (t − t0 )
∂t a 2 2πa 2 k
The solution by variation of parameter
− λ2 − λ2 t λ2
t t t
 g 
u = u (λ ,0 )e a2
+e a2
∫e a2
 2πa 2 k δ (t − t0 )dt
0  
g − λ2 λ2
S ( r,t )
= δ ( r ) δ ( t − t0 )
g t t
2π r
e a e a H (t − t0 )
2 2 0
= 2
2πka
2
−λ
g (t −t0 )
= 2
e a2
H (t − t0 )
2πka
2) Inverse transform
∞
u (r ,t ) = ∫ λ J 0 ( λ r ) u ( λ ,t ) d λ
0

∞  g
−λ2
( t − t0 ) 
= ∫ λ J0 ( λ r )  ea H ( t − t0 )  d λ
2

0  2π ka 2

∞ −λ2
g ( t − t0 )
= H ( t − t0 ) ∫ λ J 0 ( λ r ) e a dλ
2

2π ka 2 0
− a 2r 2
g a2
= H (t − t ) e 4 (t −t0 )
2(t − t0 )
0
2πka 2
− a 2r 2
g H (t − t0 ) 4 (t −t0 )
= e
4πk (t − t0 )
Hence the Heaviside step function provides a zero temperature
for t < t0 , the solution can be rewritten in the form
 0 t ≤ t0

Solution for u0 = 0 and r0 = 0 (H-1.mws) u ( r,t ) =  g
− a2 r 2
4 ( t − t0 )
 e t > t0
 4π k ( t − t0 )
 Chapter IX The Integral Transform Methods IX.4 The Hankel Transform November 25, 2018 809

Cylindrical heat source 3) Case r = r0 ≠ 0

1 ∂  ∂u  2 ∂u g
∇ 2u ≡ r  = a − δ (r − r0 )δ (t − t0 )
r ∂r  ∂r  ∂t 2πrk

∂u g
− λ2 u = a 2 − δ (t − t0 )J 0 (λr0 )
∂t 2πk

− λ2 − λ2 t λ2
t t t
 g 
u = u (λ ,0 )e a2
+e a2
∫e a2
 2πa 2 k δ (t − t0 )J 0 (λr0 )dt
0  
− λ2 λ2
g t t
e a e a J 0 (λr0 )H (t − t0 )
2 2 0
= 2
2πka
2
−λ
g (t −t0 )
= 2
e a2
J 0 (λr0 )H (t − t0 )
2πka

∞
u (r ,t ) = ∫ λJν (λr )u (λ ,t )dλ
0

∞  g − λ2
(t −t0 ) 
= ∫ λJν (λr ) e a2
J 0 (λ r0 )H (t − t 0 ) dλ
0  2πka 2 
 
2
−λ
g ∞ (t −t0 )
= H (t − t 0 )∫ λ J ν (λ r )J 0 (λ r0 )e a2
dλ
2πka 2 0

 a 2 rr0 
(
− a 2 r 2 + r02 )
g a2
integration in [Ozisik, p.161, eqn(3-151)] = H (t − t0 ) I0  e
4 (t −t0 )

2πka 2
2(t − t0 )  2(t − t0 ) 

 a 2 rr0 
(
− a 2 r 2 + r02 )
g
= I0  e
4 (t −t0 )
H (t − t0 )
4πk (t − t0 )  2(t − t0 ) 

 a 2 rr0 
(
− a 2 r 2 + r02 )
g
Solution for u0 ≠ 0 u (r ,t ) = I0  e
4 (t −t0 )
H (t − t0 )
4πk (t − t0 )  2(t − t0 ) 

This is the Green function for the Heat Equation in cylindrical

coordinates.
810 Chapter IX The Integral Transform Methods IX.4 The Hankel Transform November 25, 2018

H-1.mws Hankel transform solution of Heat Equation with instanteneous point source r0 = 0
> restart;
> with(plots):
> it:=int(exp(-lambda^2*(t-t0)/a^2)*
BesselJ(0,lambda*r)*lambda,lambda=0..infinity):
> u(r,t):=g*it/2/Pi/k;
 a 2 r 2 
 − 
2  4 ( t − t0 ) 
1 ga e
u( r, t ) :=
4 ( t − t0 ) π k
> g:=1;a:=2;t0:=1;k:=1;
g := 1
a := 2
t0 := 1
k := 1

> u2(r,t):=subs(r=-r,u(r,t)):
> u1(r):=subs(t=1.01,u2(r,t)):u2(r):=subs(t=1.05,u2(r,t)):
u3(r):=subs(t=1.1,u2(r,t)):
> plot({u1(r),u2(r),u3(r)},r=-0.5..0.5,axes=boxed,color=black);

t = 1.01

t = 1.05
t = 1.1

> animate(u(r,t),r=-0.5..0.5,t=1.002..1.2,frames=300);

for animation of
temperature profile
see web site

> animate3d([r,theta,u(r,t)],r=0..0.4,theta=0..2*Pi,t=1.004..1.2,
frames=200,style=wireframe,coords=cylindrical);

for animation of
temperature profile
see web site
 Chapter IX The Integral Transform Methods IX.4 The Hankel Transform November 25, 2018 811

Heat Equation with the instantantenous cylindrical source with r0 = 1

> restart;
> g:=1;a:=2;t0:=0;r0:=1;k:=1;
g := 1
a := 2
t0 := 0
r0 := 1
k := 1

> u(r,t):=g/4/Pi/(t-t0)*BesselI(0,a^2*r0*r/2/(t-t0))*
exp(-(r^2+r0^2)/4/(t-t0)*a^2);
 r 2 + 1 
2 r   − t 

BesselI  0 , e
1  t 
u( r, t ) :=
4 πt

> u2(r,t):=subs(r=-r,u(r,t)):
> u1(r):=subs(t=0.01,u2(r,t)):u2(r):=subs(t=0.25,u2(r,t)):
u3(r):=subs(t=2.0,u2(r,t)):
> plot({u1(r),u2(r),u3(r)},r=-4..4,axes=boxed,color=black);

t = 0.01

t = 0.25
t = 2.0

> with(plots):
> animate(u(r,t),r=-5..5,t=0.002..3,frames=300);

for animation of
temperature profile
see web site

> animate3d([r,theta,u(r,t)],r=0..2.5,theta=0..2*Pi,t=0.004..3,
frames=200,style=wireframe,coords=cylindrical,color=black);

for animation of
temperature profile
see web site

H-1-3.gif
812 Chapter IX The Integral Transform Methods IX.4 The Hankel Transform November 25, 2018

IX.4.5 The Wave Equation (vibration of axisymmetric infinite membrane)

1 ∂  ∂u  1 ∂ 2 u
∇ 2u ≡ r  = r ∈ [0 , ∞ ) t > 0
r ∂r  ∂r  a 2 ∂t 2

i.c. u (r ,0 ) = u0 (r )

∂u (r ,0 )
= u1 (r )
∂t

1) Transformed equation Apply zero order Hankel transform

∞
u (λ ,t ) = ∫ rJ 0 (λr )u (r ,t )dr
0
to the equation and initial condition

∞
1 ∂ 2u
− λ2 u = u (λ ,0 ) = ∫ rJ 0 (λr )u0 dr = u0
a 2 ∂t 2 0
∞
∂u
(λ ,0 ) = ∫ rJ 0 (λr )u1dr = u1
∂t 0
this is the 2nd order linear ODE

∂ 2u
+ a 2 λ2 u = 0
∂t 2

General solution

u = c1 cos(aλt ) + c2 sin(aλt )

u ′ = −aλc1 sin(aλt ) + aλc2 cos(aλt )

Initial conditions:

u (λ ,0 ) = u0 = c1

∂u
(λ ,0 ) = u1 = aλc2 ⇒ c2 = u1
∂t aλ

Solution:

u1
u = u0 cos(aλt ) + sin(aλt )
aλ

2) Inverse transform:

Solution of the problem is given by the inverse Hankel transform

∞
1∞
u (r ,t ) = ∫ λJ 0 (λr )u0 (λ ) cos(aλt )dλ + λJ 0 (λr )u1 (λ ) sin(aλt )dλ
0 a ∫0
Chapter IX The Integral Transform Methods IX.4 The Hankel Transform November 25, 2018 813

3) Maple example (Hankel-1.mws):

Let the membrane have initially the bell shape and u1 ( r ) = 0 :

1 A
u0 (r ) = or u0 ( r ) =
2 2
r +1 r +s

The Hankel transform of this function is

e −λ e − sλ
u0 (λ ) = u0 ( λ ) = A
λ λ

Assume also zero initial velocity of the membrane

u1 (r ) = 0

Then the solution of this problem is given by the following integral

∞
u (r ,t ) = ∫ J 0 (λr )e −λ cos(aλt )dλ
0

[I.S.Gradstein, I.M.Ryzhik, Table of Integrals #6.751.3]

∞
 2
 ( 2
) 2
b + c 2 − a 2 + 4 a 2b 2 + b 2 + c 2 − a 2 
 
∫ J 0 (cλ )e cos(aλ )dλ =
− bλ

0 2 2
2 b + c − a + 4a b (
2 2 2 2
)
identify: a = at b = −1 c = r

Then the solution becomes:

1

( 2 2 2 2
)
2 2 2
 1 + r − a t + 4 a t + 1 + r − a t 
2 22

u (r ,t ) = =
( 2
2 1 + r 2 − a 2t 2 + 4 a 2t 2)
814 Chapter IX The Integral Transform Methods IX.4 The Hankel Transform November 25, 2018

Hankel-1.mws Wave equation - vibration of the infinite membrane

> restart;
> with(plots):
Initial condition:
> u0(r):=1/sqrt(r^2+1);
1
u0 ( r ) :=
2
r +1

> cylinderplot([r,theta,u0(r)],r=0..6,theta=0..2*Pi);

Hankel transform of initial condition:

> ub0(lambda):=simplify(factor(int(r*BesselJ(0,lambda*r)*u0(r),
r=0..infinity)));
( −λ )
e
ub0 ( λ ) :=
λ

> v:=2;
v := 2

Solution - inverse Hankel transform:

> u(r,t):=int(lambda*BesselJ(0,lambda*r)*ub0(lambda)*
cos(v*lambda*t),lambda=0..5);
5

u( r, t ) := ⌠
( −λ )
 BesselJ ( 0 , λ r ) e
 cos ( 2 λ t ) d λ
⌡0

Table of integrals [Gradshtein, Ryzhik #6.751.3]

> b:=-1;a:=-v*t;c:=r;
b := -1
a := −2 t
c := r

> u(r,t):=sqrt(sqrt((b^2+c^2-a^2)^2+4*a^2*b^2)+b^2+c^2-a^2)/
sqrt((b^2+c^2-a^2)^2+4*a^2*b^2)/sqrt(2);

1 + 2 r 2 + 8 t 2 + r 4 − 8 r 2 t 2 + 16 t 4 + 1 + r 2 − 4 t 2 2
u( r, t ) :=
2 2 4 2 2 4
2 1 + 2 r + 8 t + r − 8 r t + 16 t

> u2(r,t):=subs(r=-r,u(r,t)):
> u1(r):=subs(t=0.01,u2(r,t)):u2(r):=subs(t=2.0,u2(r,t)):
u3(r):=subs(t=5.0,u2(r,t)):u4(r):=subs(t=10.0,u2(r,t)):
> plot({u1(r),u2(r),u3(r),u4(r)},r=-30..30,axes=boxed,color=black);

t = 0.01

t = 2.0
t = 5.0
t = 10.0
Chapter IX The Integral Transform Methods IX.4 The Hankel Transform November 25, 2018 815

> animate({u2(r,t),u(r,t)},r=-40..40,t=0..20,frames=200);

for animation
see web site

> animate3d([r,theta,u(r,t)],r=0..40,theta=0..2*Pi,t=0..20,
frames=100,style=wireframe,coords=cylindrical);

for animation
see web site

> animate3d([r,theta,u(r,t)],r=0..40,theta=0..2*Pi,t=0..20,
frames=100,style=patchnogrid,coords=cylindrical);

for animation
see web site
816 Chapter IX The Integral Transform Methods IX.4 The Hankel Transform November 25, 2018

IX.4.6 Solution of PDE with application of two integral transforms (Fourier and Hankel transforms)

Diffusion of light (heat) in semi-infinite space Consider a problem of diffusion of light in the half-space z ≥ 0
with instantaneous point source
Q
S (r ,t ) = δ (r )δ (z − z0 )δ (t )
2πr
Since the problem has the angular symmetry in the xy-plane, it is
reasonable to consider it in the cylindrical coordinates. In the
purely scattering medium, the diffusion approximation of light
propagation is described by the following equation:

1 ∂ϕ 1 1 ∂  ∂ϕ  1 ∂ 2ϕ Q δ (r )
− r − = δ (z )δ (t )
c ∂t d r ∂r  ∂r  d ∂z 2
2π r
instanteneous
point source The boundary condition describes partial reflection of light from
the surface and partial escape of the light through the surface
 ∂ϕ 
− d ∂z + hϕ  = 0
  z =0
Initial condition:
ϕ t =0 = 0
where ϕ (r , z , t ) fluency describes the radiation field
d = 3(1 − g )µ diffusion coefficient
µ scattering coefficient
g anisotropy factor
c speed of light
Q source power
1 1− R
h=
2 1+ R
R surface reflectivity (optical property)
h
H=
d

Determine what type of integral transform to use:

The Laplacian of the equation includes derivatives w.r.t. two
variables:
z ∈ [0, ∞ ) (Robin b.c) ⇒ Fourier standard transform
r ∈ [0, ∞ ) ⇒ Hankel transform

∞
Fourier standard transform ϕ (r , ω , t ) = ∫ ϕ (r , z , t )K (ω , z )dz
0
∞
ϕ (r , z , t ) = ∫ ϕ (z , ω , t )K (ω , z )dω
0

ω cos ωz + H sin ωz
K (ω , z ) =
N (ω )

N (ω ) =
π
2
(ω 2
+ H2 ) 1
=
2 1
N 2 π ω2 + H 2

∞
Hankel transform ϕ (λ , ω , t ) = ∫ rJ 0 (λr )ϕ (r , ω , t )dr
0
∞
ϕ (r , ω , t ) = ∫ λJ 0 (λr )ϕ (λ , ω , t )dλ
0
 Chapter IX The Integral Transform Methods IX.4 The Hankel Transform November 25, 2018 817

Apply integral transforms to the equation and initial condition:

∞
ω
1) Fourier Transform ∫ δ (z )K (ω , z )dz = N (ω )
0

1 ∂ϕ 1 1 ∂  ∂ϕ  ω 2 Q ω δ (r )
− r + ϕ = δ (t )
c ∂t d r ∂r  ∂r  d 2π N (ω ) r

∞
δ (r )
2) Hankel Transform ∫ rJ 0 (λr ) r
dρ = 1
0

1 ∂ϕ λ 2 + ω 2 Q ω
transformed equation + ϕ = δ (t ) ϕ =0
c ∂t d 2π N (ω ) t =0

−(λ +ω ) t −(λ +ω ) τ
c τ c
cQ ω 2 2 2 2

solution by variation of parameter ϕ = e d

∫ e d
δ (τ )dτ
2π N (ω ) 0

−(λ +ω ) t
c
cQ ω 2 2

= e d
2π N (ω )

∞ c c
cQ ω −ω 2 t −λ2 t
3) Inverse Hankel Transform ϕ = e d ∫ λJ 0 (λr )e d dλ
2π N (ω ) 0
c −d
cQ ω −ω 2
t d 4ct r 2

= e d
e
2π N (ω ) 2ct
−d c
Qd 4 ct r ω 2
−ω 2
t
= e e d
4πt N (ω )

−d c
Qd 4 ct r 2 ∞ ω −ω 2 d t
4) Inverse Fourier Transform ϕ =
4πt
e ∫ N (ω ) e K (ω , z )dω
0
−d c
Qd 4ct r 2 ∞ ω ω cos ωz + H sin ωz −ω 2 d t
=
4πt
e ∫ N (ω ) N (ω )
e dω
0
−d c
Qd 4 ct r 2 ∞ ω 2 cos ωz + Hω sin ωz −ω 2 d t
=
2π 2t
e ∫ ω2 + H 2
e dω
0

−d ∞ c
Qd 4 ct r 2
ω 2 cos ωz + Hω sin ωz −ω 2
t
Solution ϕ (r , z , t ) = e ∫ e d
dω
2π 2t 0 ω2 + H 2

−d
Qd 4 ct r 2 ∞
(ω 2
+ H 2 − H 2 cos ωz + Hω sin ωz −ω) 2 c
t
=
2π 2t
e ∫ ω2 + H 2
e d
dω
0
−d c
Qd 4 ct r 2 ∞
 H 2 cos ωz Hω sin ωz  −ω 2
t
=
2π 2t
e ∫ cos ω z −
ω2 + H 2
+ 2
ω + H2 
e
d
dω
0

−d
Qd 4 ct r 2
∞ 2 c
−ω t
∞
2  cos ωz 
c
−ω t
∞
 ω sin ωz  −ω d t 
c 2 2

= e  ∫ [cosωz ]e d dω − H ∫  2 2 
e d dω + H ∫  2 2 
e dω 
2π 2 t  0 0 ω + H  0 ω + H  
 1 πd − z d  ω sin ωz  −ω d t 
2
−d ∞ c ∞ c
Qd 4 ct r 2  cosωz  −ω d t 2 2

= e  e 4 ct − H 2 ∫  2 2 
e dω + H ∫  2 2 
e dω 
2π 2 t  2 ct 0 ω + H  0 ω + H  
818 Chapter IX The Integral Transform Methods IX.4 The Hankel Transform November 25, 2018

SI-7b.mws Diffusion of Radiation in the semi-infinite space from instantaneous point source

> restart;with(plots):

Fluency:
> phi:=Q*d/2/Pi^2/t*exp(-d*r^2/4/c/t)*(1/2*sqrt(Pi*d/c/t)*exp(-
z^2*d/4/c/t)-H^2*int((cos(omega*z))/(omega^2+H^2)*exp(-
omega^2*c*t/d),omega=0..25)+int((H*omega*sin(omega*z))/
(omega^2+H^2)*exp(omega^2*c*t/d),omega=0..25));

25 2 25
  
 − ω c t   ω 2 c t  
2
 − 1 /4 d r   2
 − 1 /4 z d 
⌠ ⌠  −  
1   d   d  
1  c t   πd  c t  2  cos ( ω z ) e  H ω sin( ω z ) e  
φ := Qde 
2 e −H  dω +
 d ω 
 ( π2 t )
2  ct  ω2 + H2  2
ω +H 2

 
 
 
 ⌡0 ⌡0 
 

> d:=1;c:=300000000;Q:=1;H:=10;G:=10000000;

d := 1
c := 300000000
Q := 1
H := 10
G := 10000000

> f1:=subs(t=0.000000001,phi)/G:
> f2:=subs(r=-r,f1):
> plot3d({f1,f2},r=-2..2,z=0..2,axes=boxed);

distribution of fluency
in the medium
at t=10 -9

Reflection – fluency at the surface z=0 - light exiting the layer:

> R(r):=subs(z=0,f1):
> RR(r):=subs(r=-r,R(r)):
> plot({RR(r),R(r)},r=-2..2);

t=10 -9
Chapter IX The Integral Transform Methods IX.4 The Hankel Transform November 25, 2018 819

> densityplot({f1,f2},r=-2..2,z=0..2,axes=boxed,style=patchnogrid);

time = 10 -10

> densityplot({f1,f2},r=-2..2,z=0..2,style=patchnogrid);

time = 10 -9

> R1:=subs(z=0,phi)/G:
> R2:=subs(r=-r,R1):
> animate({R1,R2},r=-2..2,t=0.0000000001..0.000000001,frames=100);

for animation of
reflection profile
see web site
820 Chapter IX The Integral Transform Methods IX.4 The Hankel Transform November 25, 2018

IX.4.7 FINITE HANKEL TRANSFORM

IX.4.7.1 Circular Domain (FHT-1) In this section, we will construct the integral transform, which eliminates
derivatives with respect to polar coordinate r.
Start with the auxiliary Sturm-Liouville problem to generate
eigenfunctions. Consider two parametric Bessel equations of order ν in
the finite circular domain
( )
r 2 u ′′ + ru ′ + λ 2 r 2 − ν 2 u = 0 r ∈ [0 , r1 ] (BE)
r Rewrite the BE equation in self-adjoint form
0 r1
 2

(ru′)′ +  − ν + λ2 r u = 0
 r 
According to SLT, this equation with homogeneous boundary condition at
r = r1 has infinitely many solutions (eigenfunctions) u = Jν (λn r ) where
eigenvalues λn are the roots of the corresponding characteristic equation
(table CD). The set of eigenfunctions {Jν (λn r )} is orthogonal on
r ∈ [0 , r1 ] with the weight function r and can be used for Fourier-Hankel
expansion of the function u:
∞
u (r ) = ∑ cn Jν (λn r )
n =1
where
r1 r1

∫ u (r )Jν (λn r )rdr ∫ u (r )Jν (λn r )rdr

cn = 0
= 0
Jν (λn r )
r1 2

∫ J (λn r )rdr
2
ν
0
If we rearrange the coefficients in the expansion
 r1 
∞ ∫
u (r )Jν (λn r )rdr 
 J (λ r ) =  u (r )J (λ r )rdr  Jν (λn r )
∞ r1
u (r ) = ∑  0 ∑ ∫
Jν (λn r )  Jν (λn r )
ν
n =1 
2  ν n 0
n =1 
n 2

 
 
then the integral transform pair of order ν can be defined as

r1
Finite Hankel Transform (FHT) u n = ∫ u (r )Jν (λn r )rdr
0

∞
Jν (λn r )
Inverse Finite Hankel Transform u (r ) = ∑ u n
Jν (λn r )
2
n =1

with the inverse transform in the from of an infinite series.

Because functions Jν (λn r ) are solutions of SLP, they are solutions of BE
′  ν2 2 
 rJν′ ( λn r )  +  − r + λn r  Jν ( λn r ) =0 or rewrite it as
 
′ ν 2 
 rJν′ ( λn r=
)  − λn2 r  Jν ( λn r )
 r 
and satisfy one of the homogeneous boundary conditions:
I) Dirichlet Jν (λn r1 ) = 0
II) Neumann Jν′ (λn r1 ) = 0
h
 ν 
k  − Jν +1 ( λn r1 ) λn + Jν ( λn r1 )  + h ? III) Robin kJν′ (λn r1 ) + hJν (λn r1 ) = 0 or Jν′ (λn r1 ) = − Jν (λn r1 )
 r  k
 Chapter IX The Integral Transform Methods IX.4 The Hankel Transform November 25, 2018 821

Operational properties of FHT Let some PDE include terms with the polar coordinate r

1 ∂  ∂u  ν 2
r  − u
r ∂r  ∂r  r 2

which are the objects for elimination with FHT. Apply FHT of order ν

I  1 ∂  ∂u  ν 
2
Hν( )   r  − 2 u
 r ∂r  ∂r  r 

r1
 1 ∂  ∂u  ν 2 
= ∫  r ∂r  r ∂r  − r 2 u Jν ( λn r ) rdr
0  

r1 r1
 1 ∂  ∂u  ν 2 
= ∫  r Jν (λn r )rdr − ∫  2 u Jν (λn r )rdr
0  r ∂r  ∂r   0 r 

r1 r
 ∂u  1 ν 
2
= ∫ Jν (λn r )d  r  − ∫  2 u Jν (λn r )rdr
0  ∂r  0  r 

r1 r r
 ∂u  
1
∂u 1
ν 2 
=  r  Jν (λn r ) − ∫ r Jν′ (λn r )dr − ∫  2 u Jν (λn r )rdr
 ∂r   0 0 ∂r 0 r 

r1 r
 ∂u   1
ν 2 
=  r  Jν (λn r ) − ∫ rJν′ (λn r )du − ∫  2 u Jν (λn r )rdr
 ∂r   r =r1 0 0 r 

r1 r
 ∂u   ′ 1
ν 2 
=  r  Jν (λn r ) − [rJν′ (λn r )u ]01 + ∫ u[rJν′ (λn r )] dr − ∫  2 u Jν (λn r )rdr
r

 ∂r   r =r1 0 0 r 

r r
 ∂u    ν 2  1
ν 2 1

=  r  Jν ( λn r )  −  rJν′ ( λn r ) u  r = r + ∫ u  − λn2 r  Jν ( λn r ) dr − ∫  2 u Jν ( λn r ) rdr
 ∂r   r =r 1
0   r  1
 0 r 

r r r
 ∂u   ν 2 
1 1
ν 2  1

=  r  Jν ( λn r )  −  rJν′ ( λn r ) u  r = r − λn2 ∫ uJν ( λn r ) rdr + ∫  2 u Jν ( λn r ) rdr − ∫  2 u Jν ( λn r ) rdr

 ∂r   r =r 1
0
1
0 r  0 r 

∂u
= r1 Jν ( λn r1 ) − r1 Jν′ ( λn r1 ) u ( r1 ) − λn2 un
∂r r = r1
822 Chapter IX The Integral Transform Methods IX.4 The Hankel Transform November 25, 2018

For the particular type of boundary conditions, the result can be specified
as following:

I) Dirichlet Jν (λn r1 ) = 0
u ( r1 ) = f1 ( t ) u (r1 ) = f 1 (t )

∂u
r = r1 Jν ( λn r1 ) − r1 Jν′ ( λn r1 ) u ( r1 ) − λn2 un
0 r1 ∂r r = r1

−r1 Jν′ ( λn r1 ) f1 ( t ) − λn2 un

=
 ν 
−r1λn  − Jν +1 ( λn r1 ) +
= Jν ( λn r1 )  f1 ( t ) − λn2 un
 λn r1 

= r1λn Jν +1 ( λn r1 ) f1 ( t ) − λn2 un

or if the other recurrence formula is used, then

 ν 
−r1λn  Jν −1 ( λn r1 ) −
= Jν ( λn r1 )  f1 ( t ) − λn2 un
 λn r1 

−r1λn Jν −1 ( λn r1 ) f1 ( t ) − λn2 un
=

III) Robin kJν′ (λn r1 ) + hJν (λn r1 ) = 0 or

h
Jν′ (λn r1 ) = − Jν (λn r1 )
k
∂u ( r1 )
k + hu ( r1 ) =
f1 ( t ) ∂u (r1 )
∂r k + hu (r1 ) = f 1 (t )
∂r
r
∂u
0 r1 = r1 Jν ( λn r1 ) − r1 Jν′ ( λn r1 ) u ( r1 ) − λn2 un
∂r r = r1

∂u h
= r1 Jν ( λn r1 ) + r1 Jν ( λn r1 ) u ( r1 ) − λn2 un
∂r r = r1 k
 ∂u h 
r1 
= + u ( r1 )  Jν ( λn r1 ) − λn2 un
 ∂r r = r k 1 

f1 ( t )
= r1 Jν ( λn r1 ) − λn2 un
k
Chapter IX The Integral Transform Methods IX.4 The Hankel Transform November 25, 2018 823

∂u ( r1 ) ∂u (r1 )
= f1 ( t ) II) Neumann = f 1 (t )
∂r ∂r
r
0 r1
Characteristic equation (see Chapter VII, p.507):

Jν′ (λn r1 ) = 0

ν
Roots of − λJ ν +1 (λL ) + J ν (λL ) = 0 , when ν >0
L

positive roots of J 1 (λL ) = 0 , and λ0 = 0 when ν =0

yn ( x ) = Jν ( λn x )

L2  ν2  2  2 L2 
 Jν (λn L )
Nν2,n = 1 − 2 2   N 0 ,0 = 
2  λn L  2 
  

ν =0 Eigenfunctions: R0 = 1
Rn ( r ) = J 0 ( λn r )

Eigenvalues: λ0 = 0
λn are positive roots of characteristic equation:
J 1 ( λ r1 ) = 0
r1
2 r12
Squared Norm: R=
0 ∫=
rdr
2
0
r1
2 r12 2
Rn ( r )
= J 02 ( λn r ) rdr
∫= J 1 ( λn r1 )
0 2

I  1 ∂  ∂u  ν 
2
Hν( )   r  − 2 u
 r ∂r  ∂r  r 
∂u
= r1 Jν ( λn r1 ) − r1 Jν′ ( λn r1 ) u ( r1 ) − λn2 un
∂r r = r1

I  1 ∂  ∂u  ν 
2
Hν( )   r  − 2 u  = r1 Jν (λn r1 ) f 1 (t ) − λ u n
2
n = 1,2,...
 r ∂r  ∂r  r 
824 Chapter IX The Integral Transform Methods IX.4 The Hankel Transform November 25, 2018

The Finite Hankel Transform – II of zero order IX.4.7.1 Circular Domain (FHT-1) RENAME λ ↔ µ

The case ν = 0 , λ0 = 0, λn > 0, n = 1,2,... λn are the roots of the equation J 1 ( λ r1 ) = 0

Characteristic equation J 1 ( λn r1 ) = 0

Eigenfunctions y0 ( r ) = 1

yn ( r ) = J 0 ( λn r )

2 r12
Squared norm y0 ( r ) =
2

2 r12 2
yn ( r ) = J 0 ( λn r1 )
2

r1

FHT u0 = ∫ u ( r ) rdr For u = const , u0 = ur1

0
r1
J 1 ( λn r1 )
un = ∫ u ( r ) J 0 ( λn r ) rdr un = u =0
0 λn

u0 ∞ Jν ( λn r )
Inverse FHT =u (r ) 2
+ ∑ un 2
J 0 ( λ0 r ) n =1 Jν ( λn r )

2 2 ∞ J 0 ( λn r )
(r )
u=
r1
u + 2
2 0
r1
∑u n =1
n
J 02 ( λn r1 )

II  1 ∂  ∂u  02 
Operational Property Hν( =0)  r =  − 2 u r1 Jν =0 ( λn r1 ) f1 ( t ) − λn2 un n = 1,2,...
 r ∂r  ∂r  r 

II  1 ∂  ∂u  02 
Hν( =0)   r  2 u
− = r1 f1 ( t ) n=0
 r ∂r  ∂r  r 

DIDIER / 00 Hajduk Finite Hankel example

 Chapter IX The Integral Transform Methods IX.4 The Hankel Transform November 25, 2018 825

IX.4.7.2 Development of velocity profile in a pipe under a pressure gradient

Consider the viscous fluid which is initially at rest inside the long
circular cylinder, and then set in motion by a spontaneously imposed
difference between the pressures at the ends of the pipe. Let G be the
axial pressure gradient, then axial velocity u (r , t ) is described by the
equation [Batchelor]:

∂u v ∂  ∂u  G
= r + r ∈ [0 , r1 ]
∂t r ∂r  ∂r  ρ

with the initial condition (initially fluid is at rest)

u (r ,0 ) = 0
and the boundary condition
u (r , t ) = 0 t >0 Dirichlet

1) Integral transform The form of integral transform is determined

by the domain (finite) and the boundary condition (Dirichlet), the order of
FHT is determined by the first term in the r.h.s. of the equation - order 0.
According to (Bessel Functions, section 12, example 3):

Eigenvalues λ n are positive roots of J 0 (λr1 ) = 0

2 r12
Eigenfunctions X n (r ) = J 0 (λn r ) J 0 (λn r ) = J 12 (λn r1 )
2
r1
u n (t ) = ∫ u (r ,t )J 0 (λn r )rdr
0

G  rJ (λ r )  G r1 J 1 (λn r1 )
r1 r1
G
Sn = ∫ J 0 ( λn r ) rdr =  1 n  =
0 ρ ρ  λn  0 ρ λn

2) Transformed equation Operational property I of FHT-I yields

∂u n G r1 J 1 (λn r1 )
= −νλ2n u n +
∂t ρ λn
∂u n r G J 1 (λn r1 )
+ νλ2n u n = 1
∂t ρ λn
Solution by variation of parameter
un = 1
r G J 1 (λn r1 ) 1
ρ λn 2
νλn
( 2
1 − e −νλnt )
3) Inverse transform Solution of IBVP:

∞
J 0 (λn r )
u (r , t ) = ∑ un
J 0 (λn r )
2
n =1

= ∑
(
2G ∞ 1 − e −νλnt J 0 (λn r )
2
)
νρr1 n=1 λ3n J 1 (λn r1 )

4) Steady state
1 J 0 (λn r )
G 2
( 2G
)
∞
u s (r ) = r1 − r 2 = ∑ λ3
4νρ νρr1 n =1 n
J 1 (λn r1 )

6) Maple Example (pipe-01.mws)

826 Chapter IX The Integral Transform Methods IX.4 The Hankel Transform November 25, 2018

pipe-01.mws Starting Flow in a Pipe (Batchelor, p.193) Solution with the Finite Hankel Transform FHT-1
> restart;with(plots):
> nu:=0.1;G:=3;rho:=2;r1:=0.5;
ν := 0.1
G := 3
ρ := 2
r1 := 0.5

> w(x):=BesselJ(0,x*r1);plot(w(x),x=0..15);
w( x ) := BesselJ( 0, 0.5 x )

λ1 λ2

Eigenvalues:
> lambda:=array(1..50);
λ := array( 1 .. 50, [ ] )

> n:=1: for m from 1 to 100 do z:=fsolve(w(x)=0,x=m*1..(m+1)*1): if

type(z,float) then lambda[n]:=z: n:=n+1 fi od:
> for i to 2 do lambda[i] od;
4.809651115
11.04015622

> N:=n-1;
N := 5

> n:='n':i:='i':x:='x':x:='z':
Eigenfunctions:
> u[n](r,t):=2*G/r1/nu/rho*BesselJ(0,lambda[n]*r)/
lambda[n]^3/BesselJ(1,lambda[n]*r1)*(1-exp(-nu*lambda[n]^2*t));
2 
  −0.1 λ t  

n 

60.00000000 BesselJ( 0, λn r )  1 − e  
un( r, t ) := 3
λn BesselJ( 1, 0.5 λn )

Solution:
> u(r,t):=sum(u[n](r,t),n=1..N):
> u0:=subs(t=0.1,u(r,t)):
> u1:=subs(t=0.5,u(r,t)):
> u2:=subs(t=1,u(r,t)):
> u3:=subs(t=1.5,u(r,t)):
Steady state solution (from Batchelor):
> f:=G*(r1^2-r^2)/4/nu/rho;
f := 0.9375000000 − 3.750000000 r 2
> plot({f,u0,u1,u2,u3},r=0..r1,color=black);

u ( r,t )
steady state
t=1.0 for animation of the
velocity profile
t=0.5 development
see web site
 Chapter IX The Integral Transform Methods IX.4 The Hankel Transform November 25, 2018 827

IX.4.7.3 Viscous flow inside of the rotating cylinder

Consider the motion of viscous fluid inside of the circular cylinder

generated by rotation of cylinder around its axis. If the motion of fluid
remains purely rotational then the axial component of velocity is equal to
zero and the development of the angular velocity is described by the
equation

∂u ∂ 1 ∂  
=ν  (ru ) = ν ∂  1 ∂  r ∂u  − 12 u r ∈ [0 , r1 ]
∂t ∂r  r ∂r  ∂r  r ∂r  ∂r  r 
with initial condition (initially fluid is at rest)
u (r ,0 ) = 0
and boundary condition
u (r , t ) = ωr1 t > 0 Dirichlet ω = cons tan t

1) Integral transform The form of integral transform is determined

by the domain (finite) and the boundary condition (Dirichlet), the order of
FHT is determined by the second term in the r.h.s. of equation - order 1.
According to (Bessel Functions, section 12, example 3):

Eigenvalues λ n are positive roots of J 1 (λr1 ) = 0

2 r12
Eigenfunctions X n (r ) = J 1 (λ n r ) J 1 (λ n r ) = J 02 (λ n r1 )
2
r1
FHT of u (r , t ) u n (t ) = ∫ u (r , t )J 1 (λ n r )rdr
0
2) Transformed equation Operational property I of FHT yields
∂u n
∂t
[
= ν − λ n r1 J 0 (λ n r1 )ωr1 − λ 2n u n ]
∂u n
+ νλ 2n u n = −νωλ n r12 J 0 (λ n r1 )
∂t
solution by variation of parameter
J (λ r )
u n (t ) = ωr12 0 n e −νλn t − 1
λn
( 2
)
3) Inverse transform provides a solution of the problem

u (r , t )
∞
= ∑ un
J 1 (λn r )
= 2ω ∑
(e
∞ −νλ2nt
)
− 1 J 1 (λn r )
n =1 J 1 (λn r )
2
n =1 λn J 0 (λn r1 )

1 J 1 ( λn r ) e −νλn t J 1 ( λn r )
2
∞ ∞
= −2ω ∑ +2ω ∑
n =1 λn J 0 ( λn r1 ) n =1 λn J 0 ( λn r1 )

m2 
4) Example r = 0.1 [m] , ν = 10 −6   (water ), ω = 10
 s 

5) Steady state Steady state solution – fully developed flow does not
depend on time, equation is reduced to Cauchy-Euler ODE, solution is
u s (r ) = ωr
Steady state is obtained from series solution as a limit when t → ∞

1 n
∞
u (r ) = lim 2ω ∑
(e ∞
−νλn2 t
)
− 1 J (λ r )
= −2ω ∑
1 J 1 ( λn r )
J (λ r ) λn J 0 ( λn r1 )
s
=
t →∞
n 1=
n 0 n 1 n 1 λ
828 Chapter IX The Integral Transform Methods IX.4 The Hankel Transform November 25, 2018

pipe-02.mws Viscous flow inside of rotating cylinder Solution with Finite Hankel Transform FHT-1
> restart;with(plots):
> nu:=0.1;omega:=3;r1:=0.5;
ν := 0.1
ω := 3
r1 := 0.5

> w(x):=BesselJ(1,x*r1);plot(w(x),x=0..15);
w( x ) := BesselJ( 1, 0.5 x )

λ1

Eigenvalues:
> lambda:=array(1..250);
λ := array( 1 .. 250, [ ] )

> n:=1: for m from 1 to 50 do z:=fsolve(w(x)=0,x=m*5..(m+1)*5):

if type(z,float) then lambda[n]:=z: n:=n+1 fi od:
> for i to 2 do lambda[i] od;
7.663411940
14.03117334

> N:=n-1;
N := 40

> n:='n':i:='i':x:='x':x:='z':
Eigenfunctions:
> u[n](r,t):=2*omega*BesselJ(1,lambda[n]*r)/lambda[n]/
BesselJ(0,lambda[n]*r1)*(exp(-nu*lambda[n]^2*t)-1);
  −0.1 λn 2 t  
6 BesselJ( 1, λn r )  e  
− 1 
un( r, t ) :=
λn BesselJ( 0, 0.5 λn )

Solution:
> u(r,t):=sum(u[n](r,t),n=1..N):
> u0:=subs(t=0.01,u(r,t)):
> u1:=subs(t=0.05,u(r,t)):
> u2:=subs(t=0.1,u(r,t)):
> u3:=subs(t=1,u(r,t)):
Steady state solution (from Batchelor):
> f:=omega*r;fn:=-r*omega;
f := 3 r
fn := −3 r

> plot({f,u0,u1,u2,u3},r=0..r1,color=black);

steady state

t = 0.05 for animation of the

t = 0.1 t = 0.05 velocity profile
development
see web site
 Chapter IX The Integral Transform Methods IX.4 The Hankel Transform November 25, 2018 829

IX.4.7.4 Annular Domain (FHT-2)

Auxiliary Sturm-Liouville Problem. Consider a two-parameter Bessel

equation of order ν in the annular domain
( )
r 2 u ′′ + ru ′ + λ 2 r 2 − ν 2 u = 0 r ∈ ( r1 ,r2 ) (BE)
with homogeneous boundary conditions:
 ∂u 
− k1 ∂r + h1u  =0
  r =r1
 ∂u 
k 2 ∂r + h2 u  =0
  r =r2

According to SLT, this regular boundary value problem has infinitely

many solutions (eigenfunctions)

X n (r ) = c1,n Jν (λn r ) + c2 ,nYν (λn r )

where eigenvalues λn are the roots of the corresponding characteristic

equation (table CD).
The set of eigenfunctions {X n (r )} is orthogonal on r ∈ [r1 , r2 ] with the
weight function r and can be used for Fourier-Hankel expansion of the
function u:
∞
u (r ) = ∑ an X n (r )
n =1
where
r2 r2

∫ u (r )X n (r )rdr ∫ u (r )X n (r )rdr
r1 r1
an = =
X n (r )
r2 2

∫ X n (r )rdr
2

r1

If we rearrange the coefficients in the expansion

 r2 
 ∫ u (r )X n (r )rdr 
 X (r ) =  u (r )X (r )rdr  X n (r )
∞ ∞ r2
u (r ) = ∑  1 ∑ ∫
r
 n 
( )  X n (r )
2 n 2
n =1
 X n r  
n =1  r1

 
then the integral transform pair of order ν can be defined as

r2
Finite Hankel Transform (FHT-2) u n = ∫ u (r )X n (r )rdr
r1

∞
X n (r )
Inverse Finite Hankel Transform u (r ) = ∑ u n
X n (r )
2
n =1

with inverse transform in the form of an infinite series.

830 Chapter IX The Integral Transform Methods IX.4 The Hankel Transform November 25, 2018

Because the eigenfunctions X n (r ) are solutions of SLP, they are solutions

of BE
 2

[rX n′ (r )]′ +  − ν + λ2 r  X n (r ) = 0 or rewrite it as
 r 
  2
[rX n′ (r )]′ =  ν
− λ2 r  X n (r )
 r 
and satisfy one of the combinations of the homogeneous boundary
conditions:
I) Dirichlet X n (r1 ) = 0 X n (r2 ) = 0
II) Neumann X ν′ (r1 ) = 0 X ν′ (r2 ) = 0
III) Robin kX n′ (r1 ) + hX n (r1 ) = 0 kX n′ (r2 ) + hX n (r2 ) = 0

There are 9 possible combinations of these boundary conditions.

Operational properties of FHT-2 Consider the application of FHT-2 to the following expression

1 ∂  ∂u  ν 2
r  − u
r ∂r  ∂r  r 2

Apply FHT-2 of order ν

r2
 1 ∂  ∂u  ν 2 
u  X n ( r ) rdr
∫  r ∂r  r ∂r  − r
2
r1 

r2 r2
 1 ∂  ∂u  ν 2 
= ∫  r  X n (r )rdr − ∫  2 u  X n (r )rdr
r1  r ∂r  ∂r   r1  r 
r2 r
 ∂u  2 ν 
2
= ∫ X n (r )d  r  − ∫  2 u  X n (r )rdr
r1  ∂r  r1  r 
rr2 r
 ∂u  
2
∂u 2
ν 2 
=  r  X n (r ) − ∫ r X n′ (r )dr − ∫  2 u  X n (r )rdr
 ∂r   r1 r1 ∂r r1  r 
rr2 r
 ∂u  
2 2
ν 2 
=  r  X n (r ) − ∫ rX n′ (r )du − ∫  2 u  X n (r )rdr
 ∂r   r1 r1 r1  r 
r r2 r
 ∂u  
2
′ 2
ν 2 
=  r  X n (r ) − [rX n′ (r )u ]r12 + ∫ u[rX n′ (r )] dr − ∫  2 u  X n (r )rdr
r

 ∂r   r1 r1 r1  r 
r r2
 ∂u  
2 r2
 ν 2   ν 2 
=  r  X n (r ) − [rX n′ (r )u ]r12 + ∫ u  − λ2 r  X n (r )dr − ∫  2 u  X n (r )rdr
r

 ∂r   r1 r1  r   r1  r 
r r2 r r
 ∂u  
2 2
ν 2  2
ν 2 
=  r  X n (r ) − [rX n′ (r )u ]r12 − λ2 ∫ u X n (r )rdr + ∫  2 u  X n (r )rdr − ∫  2 u  X n (r )rdr
r

 ∂r   r1 r1 r1  r  r1  r 
r2
 ∂u  
=  r  X n (r ) − [rX n′ (r )u ]r12 − λ2 u n
r

 ∂r   r1

For the particular type of boundary conditions, this result can be specified.
The complete solution of the Sturm-Liouville problem in the annular
domain is presented in the table” SLP in annular domain”. Here, consider
derivation of operational properties of FHT-2 for three types of boundary
conditions:
 Chapter IX The Integral Transform Methods IX.4 The Hankel Transform November 25, 2018 831

h1
Robin-Robin − k1 X n′ (r1 ) + h1 X n (r1 ) = 0 ⇒ X n′ (r1 ) = X n (r1 )
k1
−h2
k 2 X n′ (r2 ) + h2 X n (r2 ) = 0 ⇒ X n′ ( r2 ) = X n ( r2 )
k2
∂u ( r1 )
−k1 + h1u ( r1 ) =
f1 ( t )
∂r
∂u ( r2 )
k2 + h2 u ( r2 ) =
f2 (t )
∂r
 ∂u    ∂u  
=  r  X n (r ) −  r  X n (r ) − [rX n′ (r )u ]r2 + [rX n′ (r )u ]r1 − λ2 u n
 ∂r   r2  ∂r   r1
 ∂u    ∂u    h   h 
=  r  X n (r ) −  r  X n (r ) − − r 2 X n (r )u  + r 1 X n (r )u  − λ2 u n
 ∂r   r2  ∂r   r1  k 2  r2  k1  r1
 ∂u (r2 )   ∂u (r1 )  h h
 X n (r2 ) −  r1  X n (r1 ) + r2 2 X n (r2 )u (r2 ) + r1 1 X n (r1 )u (r1 ) − λ u n
2
=  r2
 ∂r   ∂r  k2 k1
 ∂u (r1 ) h1   ∂u (r2 ) h2 
= r1 X n (r1 )− + u (r1 ) + r2 X n (r2 ) + u (r2 ) − λ2 u n
 ∂r k1   ∂r k2 

r1 r
= X n ( r1 ) f1 ( t ) + 2 X n ( r2 ) f 2 ( t ) − λn2 un (☼)
k1 k2

∂u (r1 )
Neumann-Dirichlet X n′ (r1 ) = 0 = f 1 (t )
∂r
X n (r2 ) = 0 u ( r2 ) = f 2 ( t )

 ∂u    ∂u  
=  r  X n (r ) −  r  X n (r ) − [rX n′ (r )u ]r2 + [rX n′ (r )u ]r1 − λ2 u n
 ∂r   r2  ∂r   r1
 ∂u  
= −  r  X n (r ) − [rX n′ (r )u ]r2 − λ2 u n
 ∂r   r1
 ∂u (r1 ) 
 X n (r1 ) − r2 X n′ (r2 )u (r2 ) − λ u n
2
= − r1
 ∂r 

−r1 X n ( r1 ) f1 ( t ) − r2 X n′ ( r2 ) f 2 ( t ) − λn2 un
= (☼☼)

Dirichlet-Neumann X n (r1 ) = 0 u (r1 ) = f 1 (t )

∂u (r2 )
X n′ (r2 ) = 0 = f 2 (t )
∂r
 ∂u  
=  r  X n (r ) + [rX n′ (r )u ]r1 − λ2 u n
 ∂r   r2
 ∂u (r2 ) 
 X n (r2 ) + r1 X n′ (r1 )u (r1 ) − λ u n
2
=  r2
 ∂r 

= r1 X n′ ( r1 ) f1 ( t ) + r2 X n ( r2 ) f 2 ( t ) − λn2 un (☼☼☼)
832 Chapter IX The Integral Transform Methods IX.4 The Hankel Transform November 25, 2018

r1
Robin-Dirichlet = X n′ ( r1 ) f1 ( t ) − r2 X n ( r2 ) f 2 ( t ) − λn2 un (☺)
h1

∂u (r1 ) ∂u (r1 )
Neumann-Robin = f 1 (t ) = f 1 (t )
∂r ∂r

∂u ( r2 ) ∂u ( r2 ) h2 f2 (t )
k2 + h2 u ( r2 ) =
f2 (t ) ⇒ + u ( r2 ) =
∂r ∂r k2 k2

X n′ (r1 ) = 0
−h2
k 2 X n′ (r2 ) + h2 X n (r2 ) = 0 ⇒ X n′ ( r2 ) = X n ( r2 )
k2

Operational property:

r2
 1 ∂  ∂u  ν 2 
∫  r ∂r  r ∂r  − r 2 u  X n ( r ) rdr
r1  

r2
 ∂u  
=  r  X n (r ) − [rX n′ (r )u ]r12 − λ2 u n
r

 ∂r   r1

∂u ∂u
= r2 X n ( r2 ) − r1 X n ( r1 ) − r2 X n′ ( r2 ) u ( r2 ) + r1 X n′ ( r1 ) u ( r1 ) − λ 2 un
∂r r2 ∂r r1

∂u ∂u h2
= r2 X n ( r2 ) − r1 X n ( r1 ) + r2 X n ( r2 ) u ( r2 ) + r1 X n′ ( r1 ) u ( r1 ) − λ 2 un
∂r r2 ∂r r1 k2

 ∂u h2  ∂u
r2 
= + u ( r2 )  X n ( r2 ) − r1 X n ( r1 ) − λ 2 un
 ∂r r2 k2  ∂r r1

f2 (t )
= r2 X n ( r2 ) − r1 f1 ( t ) X n ( r1 ) − λ 2 un
k2

f2 (t )
= r2 X n ( r2 ) − r1 f1 ( t ) X n ( r1 ) − λn2 un (☺☺)
k2
 Chapter IX The Integral Transform Methods IX.4 The Hankel Transform November 25, 2018 833

∂u (r1 )
Neumann- Neumann X n′ (r1 ) = 0 = f 1 (t )
∂r
∂u (r2 )
X n′ (r2 ) = 0 = f 2 (t )
∂r

 ∂u    ∂u  
=  r  X n (r ) −  r  X n (r ) − [rX n′ (r )u ]r2 + [rX n′ (r )u ]r1 − λ2 u n
 ∂r   r2  ∂r   r1
 ∂u    ∂u  
=  r  X n ( r )  −  r  X n ( r )  − λ 2 un
 ∂r   r2  ∂r   r1
 ∂u ( r2 )   ∂u ( r1 ) 
 X n ( r2 ) −  r1  X n ( r1 ) − λ un
2
=  r2
 ∂r   ∂r 

−r1 X n ( r1 ) f1 ( t ) + r2 X n ( r2 ) f 2 ( t ) − λn2 un
= (☼ N-N)

Dirichlet-Neumann X n (r1 ) = 0 u (r1 ) = f 1 (t )

X n ( r2 ) = 0 u ( r2 ) = f 2 ( t )

[solution of Badger and Butler, 2017]

= r1 X n′ ( r1 ) f1 ( t ) − r2 X n′ ( r2 ) f 2 ( t ) − λn2 un (☼ D-D)
834 Chapter IX The Integral Transform Methods IX.4 The Hankel Transform November 25, 2018

IX.4.7.5 Viscous fluid flow between two cylinders

∂u  1 ∂  ∂u  1   ∂ 2 u 1 ∂u 1 
=ν   r  − 2 u =ν  2 + − u
∂t  r ∂r  ∂r  r   ∂r r ∂r r 2 
initial condition:
 r − r2 
u (r ,0 ) = u0 (r ) = ω 2 r2  1 − 
 r1 − r2 
boundary conditions:
∂u (r1 )
= f 1 (t ) = c Neumann
∂r
u 2 (r2 ) = f 2 (t ) = ω 2 r2 Dirichlet

1) Integral transform FHT-2 of order ν = 1

λn are positive roots of the equation:

 1   1 
− J 2 (λr1 ) + J 1 (λr1 )Y1 (λr1 ) − − Y2 (λr1 ) + Y1 (λr1 ) J 1 (λr1 ) = 0
 λr1   λr1 
Eigenfunctions:
J (λ r ) Y (λ r )
X n (r ) = 1 n − 1 n
J 1 (λn r2 ) Y1 (λn r2 )
Transformed initial condition:
r2
u0 ,n = ∫ u0 (r )X n (r )rdr
r1

2) Transformed equation Use equation (☼☼):

∂u n
∂t
[
= ν − r1 X n (r1 ) f 1 (t ) − r2 X n′ (r2 ) f 2 (t ) − λ2 u n ]
∂u n
Note: here, ν is a viscosity + νλ2 u n = −ν [r1 X n (r1 ) f 1 (t ) + r2 X n′ (r2 ) f 2 (t )] = ν Qn
∂t
Notation: Qn (t ) = [r1 X n (r1 ) f 1 (t ) + r2 X n′ (r2 ) f 2 (t )]
Solution by variation of parameter
t
u n (t ) = u0 ,n e −νλnt − νe −νλnt ∫ Qn (τ )eνλnτ dτ
2 2 2

0
t
−νλ2nt
(assume Qn (=
t ) Q=
2 2
= u0 ,n e − νQn e −νλnt ∫ eνλnτ dτ n const )
0
2
= u0 ,n e −νλnt −
νQn −νλ t νλ τ
νλ2n
e e
2
n
[ ] 2
n
t
0

2
= u0 ,n e −νλnt −
Qn
λ 2
2
[
e −νλnt eνλnt − 1
2
]
n

 Q  2 Q
=  u0 ,n + 2n e −νλ t − 2n
n

λn  λn
 
3) Inverse transform

∞ 
Q  Q  X n (r )
u (r ) = ∑  u0 ,n + 2n e −νλ t − 2n 
2
n


n =1  λn  λn  X n (r ) 2

4) Steady state solution Euler-Cauchy equation

1  r r (ω 2 − f 1 ) Xn (r )
)x Qn
2 2
(
∞
u s (r ) = 
1 2
+ r12 f 1 + r22ω 2 =∑
r + r12 
1
2
x 
2
n =1 λn Xn (r )
2
 Chapter IX The Integral Transform Methods IX.4 The Hankel Transform November 25, 2018 835

FHT-2-01.mws Annular domain : Neuman-Dirichlet boundary conditions

> restart;with(plots):
> u0:=omega2*L2*(1-(x-L2)/(L1-L2));
x − L2 
u0 := ω2 L2  1 − 
 L1 − L2 

> f1:=1;f2:=omega2*L2;
f1 := 1
f2 := ω2 L2

> omega1:=2;omega2:=5;v:=2;
ω1 := 2
ω2 := 5
v := 2

> L1:=2;L2:=3;
L1 := 2
L2 := 3

> nu:=1;
ν := 1

Characteristic equation (Table SLP in annular domain):

> a11:=-x*BesselJ(nu+1,x*L1)+nu/L1*BesselJ(nu,x*L1);
1
a11 := −x BesselJ( 2, 2 x ) + BesselJ( 1, 2 x )
2

> a12:=-x*BesselY(nu+1,x*L1)+nu/L1*BesselY(nu,x*L1);
1
a12 := −x BesselY( 2, 2 x ) + BesselY( 1, 2 x )
2
> a21:=BesselJ(nu,x*L2);
a21 := BesselJ( 1, 3 x )

> a22:=BesselY(nu,x*L2);
a22 := BesselY( 1, 3 x )

> w(x):=a11*a22-a12*a21;
1
w( x ) :=  −x BesselJ( 2, 2 x ) + BesselJ( 1, 2 x )  BesselY( 1, 3 x )
 2 
1
−  −x BesselY( 2, 2 x ) + BesselY( 1, 2 x )  BesselJ( 1, 3 x )

 2 
> plot(w(x),x=0..10);

Eigenvalues:

> lambda:=array(1..200);
λ := array( 1 .. 200, [ ] )

> n:=1: for m from 1 to 150 do y:=fsolve(w(x)=0,x=m/2..(m+1)/2): if

type(y,float) then lambda[n]:=y: n:=n+1 fi od:
> for i to 4 do lambda[i] od;
836 Chapter IX The Integral Transform Methods IX.4 The Hankel Transform November 25, 2018

1.756990510
4.778245758

> N:=n-1;
N := 24
>n:='n':i:='i':m:='m':y:='y':x:='x':

Eigenfunctions:

> a11[n]:=subs(x=lambda[n],a11):
> a21[n]:=subs(x=lambda[n],a21):
> a12[n]:=subs(x=lambda[n],a12):
> a22[n]:=subs(x=lambda[n],a22):
> X[n](x):=BesselJ(nu,lambda[n]*x)/a21[n]-
BesselY(nu,lambda[n]*x)/a22[n];

BesselJ( 1, λn x ) BesselY( 1, λn x )
Xn( x ) := −
BesselJ( 1, 3 λn ) BesselY( 1, 3 λn )

Derivative of eigenfunction, and the values of eigenfunction and its derivative at the boundaries:
> XP[n](x):=diff(X[n](x),x):
> X[n](L1):=subs(x=L1,X[n](x)):
> XP[n](L2):=subs(x=L2,XP[n](x)):

Norm:
> NN[n]:=int(x*X[n](x)^2,x=L1..L2):
Transformed initial condition:
> u0t[n]:=int(u0*X[n](x)*x,x=L1..L2):
> Q[n]:=L1*X[n](L1)*f1+L2*XP[n](L2)*f2:

Fourier-Bessel series:
> u[n](x,t):=((u0t[n]+Q[n]/lambda[n]^2)*exp(-v*lambda[n]^2*t)-
Q[n]/lambda[n]^2)*X[n](x)/NN[n]:
> u(x,t):=sum(u[n](x,t),n=1..N):

Steady State Solution:

> c1:=(f2-f1*L2)/(L2/L1^2+1/L2):c2:=(f1/L2+f2/L1^2)/(L2/L1^2+1/L2)
:us(x):=c1/x+c2*x:
> animate({u(x,t),us(x),u0},x=L1..L2,t=0..1,axes=boxed,frames=100);

steady state

for animation of
the the velocity profile
initial velocity see web site
 Chapter IX The Integral Transform Methods IX.4 The Hankel Transform November 25, 2018 837

IX.4.7.6 Transient heat transfer through the cylindrical wall (thermal shock) Vedat Suat Erturk 2015

impulse boundary flux 1 ∂  ∂u  1 ∂u

r  = ν =0
∂u ( r1 ,t ) r ∂r  ∂r  α ∂t
−k q0δ ( t )
=
∂r

∂ 2 u 1 ∂u 1 ∂u
+ = r1 < r < r2 , t > 0
∂r 2 r ∂r α ∂t

u ( r,0 ) = 0 r1 ≤ r ≤ r2
0
r1 f1 ( t )
 
r2 ∂u ( r1 ,t ) ∂u ( r1 ,t ) q0
−k q0δ ( t )
= ⇒ = − δ (t )
r ∂r ∂r k

f2 (t )
∂u ( r2 ) ∂u ( r2 )  h
k + h2 u ( r2 ) =
0 ⇒ + H 2 u ( r2 ) =0 , H2 = 2
∂r ∂r k

1) Finite Hankel Transform of order ν = 0 (see Neumann-Robin, Chapter 3, p.38)

r2

un ( t ) = ∫ u ( r,t ) X n ( r ) rdr
r1

Inverse Finite Hankel Transform

∞ X n (r )
u ( r,t ) = ∑ un ( t ) 2
n =1 X n (r )

where eigenfunctions and squared norm are

J 0 ( λn x ) Y0 ( λn x )
X νn =0 ( x ) = −
−λn J 1 ( λn r2 ) + H 2 Jν ( λn r2 ) −λnY1 ( λn r2 ) + H 2Y0 ( λn r2 )

r2
2
X n ( r ) = ∫ X n2 ( r ) rdr
r1

and eigenvalues are positive roots of equation:

λ J 1 ( λ r1 )  −λY1 ( λ r2 ) + H 2Y0 ( λ r2 ) 
−λY1 ( λ r1 )  −λ J 1 ( λ r2 ) + H 2 Jν ( λ r2 )  =
0

λ1 λ2 λ3

2) Hankel transform of equation (use operational property ☺☺, p.93):

838 Chapter IX The Integral Transform Methods IX.4 The Hankel Transform November 25, 2018

f2 (t ) 1 ∂u
r2 X n ( r2 ) − r1 f1 ( t ) X n ( r1 ) − λn2 un =
=n
k2 α ∂t

∂un ( t ) q
α r1 0 δ ( t ) X n ( r1 )
+ αλn2 un ( t ) = initial condition: un ( 0 ) = 0
∂t k

3) Apply Laplace transform

∞
q
sU n ( s ) + αλ 2U n ( s ) =
α r1 0 X n ( r1 ) where U n ( s ) = ∫ un ( t ) e − st dt
k 0

q0 1
U n ( s ) = α r1 X n ( r1 )
k s + αλn2

4) Apply Inverse Laplace tranform

q0
un ( t ) = α r1 X n ( r1 ) e −αλn
2

5) Inverse Hankel Transform – solution of the IBVP:

q0 ∞ X n (r )
u ( r,t ) = α r1 ∑ X ( r ) e αλ
2
− n
n 1 2
k n =1 X n (r )

6) Example (paper of Yun et al, 2009 ASME JPVT)

=r1 1.0,
= r2 1.5,
= k 0.111,
= α 0.06=
, h2 0.00677988,
= q0 36.0

t = 0.001

t = 0.005

t =0 t = 0.01

t = 0.05

t = 0.4

Simple way to obtain the other solution for verification.

 Chapter IX The Integral Transform Methods IX.4 The Hankel Transform November 25, 2018 839

Cylindrical wall can be approximated by a plane wall.

If the values of t are not big then the influence of other boundary is negligible.
Problem can be reduced to following:

∂ 2 u( x,t ) 1 ∂u( x,t )

= for x > 0 , and 0 < t < not too big
∂x 2 kd ∂t

u( =
x,0 ) 0, x ≥ 0
∂u( 0,t )
−K q0δ ( t )
= Neumann
∂x

Apply Laplace Transform:

∂ 2 U( x,s) 1
= sU ( x,s )
∂x 2 kd

∂U( 0,s)
−K q0
= Neumann
∂x

Solve:
s s
− x + x
( x,s ) c1e
U=
kd
+ c2 e
kd
c2 = 0 for solution to be bounded
s
− x
U ( x,s ) = c1e
kd

s
∂ s − x
U ( x,s ) = −c1
kd
e
∂x kd

s
∂ s − ⋅0
q0 kd
U ( 0,s )= c1 K
kd
−K e = q0 ⇒ c1 =
∂x kd K s

s
q kd − x
U ( x,s ) = 0
kd
e
K s

Apply inverse Laplace transform to get a solution:

x2
−
kd e 4 kd t
u ( x,t ) L=
= −1
{U ( x,s )} K
q0
πt

Shift solution to x = a

Temperature profile for near future prediction:

( x − a )2
−
kd e 4 kd t
v ( x,t ) = q0 for x ≥ a , and 0 < t < not too big
K πt
840 Chapter IX The Integral Transform Methods IX.4 The Hankel Transform November 25, 2018

This solution yields nice smooth curves, solution is exact, it provides very good prediction for time for which
temperature profiles do not reach the second boundary, and the layer can be considered semi-infinite.
This solution is good for comparison with the solution in the finite layer.
Comparison of this solution with the solution by Hankel transform (48 terms in summation):

For t=0.005 t=0.05

Solution are almost identical.

It makes me absolutely confident that both solutions are correct.
For shorter time t<0.005, oscillations in Hankel solution are noticible.

!!! This solution at t0=0.005 (or some other, for example at t0=0.0001) can be used as an initial condition for
Numerical (finite-difference) solution of simpler problem:

∂ 2 u 1 ∂u 1 ∂u
+ = r1 < r < r2 , t > 0
∂r 2 r ∂r α ∂t

( r − r1 )2
−
4α t0
α e
u ( r,0 ) = q0 r1 ≤ r ≤ r2
k πt

∂u ( r1 ,t )
=0
∂r
∂u ( r2 )
k + h2 u ( r2 ) =
0 no impulse!
∂r
 Chapter IX The Integral Transform Methods IX.4 The Hankel Transform November 25, 2018 841

IX.4.7.7 Simple example of the Transient heat transfer through the cylindrical wall (modifiedIX.7.4.6) Feb, 2015

Heat equation in cylindrical coordinates:

insulated wall 1 ∂  ∂u  1 ∂u
r  = ν =0
∂u ( r1 ,t ) r ∂r  ∂r  α ∂t
=0
∂r

Consider the following initial boundary value problem:

∂ 2 u 1 ∂u 1 ∂u
+ = r1 < r < r2 , t > 0
∂r 2 r ∂r α ∂t

Uniform initial temperature:

u ( r,0 ) = 1 r1 ≤ r ≤ r2
0
r1 Boundary r = r1 is thermoinsulated:
r2 f1 ( t )
∂u ( r1 ,t ) 
r = 0
∂r
Boundary r = r2 is exposed to convective heat transfer with
surroundings at 0 temperature:
f2 (t )
∂u ( r2 ) ∂u ( r2 )  h
k + h2 u ( r2 ) =
0 ⇒ + H 2 u ( r2 ) =0 , H2 = 2
∂r ∂r k

1) Finite Hankel Transform of order ν = 0 (see Neumann-Robin, Chapter 3, p.38)

r2

un ( t ) = ∫ u ( r,t ) X n ( r ) rdr
r1

Inverse Finite Hankel Transform

∞ X n (r )
u ( r,t ) = ∑ un ( t ) 2
n =1 X n (r )
where eigenfunctions (Finite Hankel of 0 order) are
J 0 ( λn x ) Y0 ( λn x )
X νn =0 ( x ) = −
−λn J 1 ( λn r2 ) + H 2 Jν ( λn r2 ) −λnY1 ( λn r2 ) + H 2Y0 ( λn r2 )
The squared norm of eigenfunctions is
r2
2 2
X n0 ( r ) = ∫  X n0 ( r )  rdr
r1

Eigenvalues are positive roots of equation:

λ J 1 ( λ r1 )  −λY1 ( λ r2 ) + H 2Y0 ( λ r2 )  −λY1 ( λ r1 )  −λ J 1 ( λ r2 ) + H 2 Jν ( λ r2 )  =

0

λ1 λ2 λ3
842 Chapter IX The Integral Transform Methods IX.4 The Hankel Transform November 25, 2018

2) Take Finite Hankel transform of equation (use operational property ☺☺, p.93):

f2 (t ) 1 ∂u
r2 X n ( r2 ) − r1 f1 ( t ) X n ( r1 ) − λn2 un =n
k2 α ∂t

∂un ( t ) r2 r2

+ αλn2 un ( t ) =
0 initial condition: un ( 0=
) ∫ u0 ( r,0 ) X n ( r ) rdr= ∫ 1 ⋅ X ( r ) rdr
n
∂t r1 r1

3) Apply Laplace transform to equation

∞
sU n ( s ) − un ( 0 ) + αλ 2U n ( s ) =
0 where Laplace transform is defined by U n ( s ) = ∫ un ( t ) e − st dt
0

1
U n ( s ) = un ( 0 )
s + αλn2

4) Apply Inverse Laplace transform

un ( t ) = un ( 0 ) e −αλn
2

5) Inverse Hankel Transform – solution of the IBVP:

∞ X n (r )
u ( r,t ) = ∑ un ( 0 ) ⋅
2

2
⋅ e −αλn
n =1 X n (r )

6) Example Cooling of shaurma (my favorite food on Taksim square)

Maple: SF-AD-6-0 Heat Transfer Problem N-R finite cylinder 02. Mws

T, o C
t =0 s
t = 600 s

t = 6000 s

t = 60000 s

t = 600000 s
Chapter IX The Integral Transform Methods IX.4 The Hankel Transform November 25, 2018 843

Boundary value problems of heat conduction

Özışık, M. Necati
Scranton: International Textbook Co., 1968
844 Chapter IX The Integral Transform Methods IX.4 The Hankel Transform November 25, 2018

Hotdog Problem

The Heat Equation:

r
1 ∂  ∂u  1 ∂u 0 r1
r  = , 0 ≤ r < r1
r ∂r  ∂r  α ∂t

I Boiling II Cooling

0 ≤ t ≤ t1 t > t1

Ta

Tsat

Initial condition:

u I (=
r, 0 ) u=
0 (r) T0 ( r, 0 ) u=
u II = 0 (r)
II
u I ( r, t1 )

Boundary conditions:
∂u ( r1 )
u ( r1 ) = Tsat k + hu ( r1 ) =
hTa
∂r

1) What time is needed for the hotdog to be completely cooked?

2) What time is needed for the hotdog to be cool enough not to be burned?

3) Sketch the temperature field of the middle cross-section as a function of time.

4) Sketch the temperature of the center point of the middle cross-section as a function of time.
 Chapter IX The Integral Transform Methods IX.4 The Hankel Transform November 25, 2018 845

IX.4.7.8 Some additional notes on HANKEL TRANSFORM Feb, 2015

The Hankel transform of order ν of the function f (r ) of the polar variable r ∈ [0 , ∞ ) is defined as

∞
Hν { f (r )} = fν (λ ) = ∫ rJν (λr ) f (r )dr
0

The inverse Hankel transform of order ν is defined as

∞
{ }
H ν−1 fν (λ ) = f (r ) = ∫ λJν (λr ) fν (λ )dλ
0

Calculation of the Hankel transforms is performed by direct application of the definition. The main difficulty will be
integration. Therefore, the special tables of integrals involving Bessel functions should be applied. Integration can be
performed with the help of Maple or any other computer software.
But the simplest is to have a table of Hankel transforms similar to 17.1 and 17.2.

Examples:

1
1) Calculate 0th order Hankel transform of f ( r ) =
r

Apply definition:

∞ ∞ ∞
1 1 1 1
f0 ( λ ) ≡ H 0  = ∫ rJ ( λ r ) r dr= ∫ J ( λ r ) dr= J0 ( λ r ) d ( λ r =
)
λ ∫0
0 0
r  λ
0 0

1

1
2) Calculate inverse of the 0th order Hankel transform of f0 ( λ ) = to reconstruct function f ( r ) .
λ

Apply definition of the inverse transform:

∞ ∞ ∞
1 1 1 1
f (r ) ∫0 λ J 0 ( λ=
r ) dλ (λr ) dλ
∫ J 0= J 0 ( λ=
r ) d ( rλ )
r ∫0
= H 0−1  
=
λ  λ 0
r

1 0 < r < R
3) Calculate 0th order Hankel transform of f ( r ) =1 − H( r − R ) = Heaviside function
0 r>R
∞
f0 ( λ ) ≡ H 0 {1 − H ( r =
− R )} ∫ rJ ( λ r ) 1 − H ( r − R ) dr
0
0
R R
1 1 R 1
= ∫ rJ 0 ( λ r ) dr
=
0
 rJ 1 ( λ r ) 
λ r ∫=0 
d=
λ
rJ 1 ( λ r )  r =0
=
λ
RJ 1 ( λ R )
846 Chapter IX The Integral Transform Methods IX.4 The Hankel Transform November 25, 2018

EXAMPLE TRANSIENT HEAT TRANSFER Cooling of the cylindrical-shaped temperature profile

Consider the axisymmetric case of the Heat Equation in
initial temperature distrubution cylindrical coordinates
of the thin infinite metal plate
1 ∂  ∂u  ∂u
∇2u ≡ r  = a2 r ∈ [0, ∞ ) t > 0
heated to temperature h i r ∂r  ∂r  ∂t
nside a circle of raius R Initial condition (for r<R temperature is T=h, for r>R, T=0)
and 0 outside of the circle h 0 < r < R
u ( r,0 ) =u0 ( r ) =h ⋅ 1 − H ( r − R )  =
0 r>R
u0 ( r ) =h ⋅ 1 − H ( r − R )  There are no boundary conditions but we assume that the
h function u and its derivative has radial decaying at infinity:
∂u
ru r →∞ = 0 r =0
infinite metal ∂r r →∞
plate Interpretation: infinite plate thermoinsulated on both sides.
Heat is conducted along the plate.
a Thermal diffusivity of the plate is a 2 .
R r
1) Transformed equation Apply zero order Hankel transform
∞
u (λ ,t ) = ∫ rJ 0 (λr )udr
0
to the equation
∂u
−λ 2 u = a2
∂t
and to initial condition (see Table 17.1 Exponential functions)
∞ R
hR
u (λ ,0 ) = ∫ rJ
= 0 ( λ r ) u0 dr h=∫ rJ 0 ( λ r )dr J1 ( λ R )
0 0
λ
That is the 1 order linear ODE
st

∂u λ 2
+ u= 0
∂t a 2
The solution is
−λ2
t
u ( λ ,t ) = u ( λ ,0 ) e a
2

−λ2
hR t
= J1 ( λ R ) e a
2

λ
u ( r,t ) =R 0.5,
= h 1.0,
= a 2.0
2) Inverse transform Solution of the problem
∞

u0 ( r ) =a ⋅ 1 − H ( r − R ) 
u (r ,t ) = ∫ λ J 0 ( λ r ) u ( λ ,t ) d λ
0
∞ −λ2
hR t
= ∫ λ J0 ( λ r ) J1 ( λ R ) e a d λ
2

u ( r,t = 0.25 ) λ
0
∞ −λ2
t
= hR ∫ J 0 ( λ r ) J 1 ( λ R ) e a d λ
2

r 0

∞ −λ2
t
u (r ,t ) = hR ∫ J 0 ( λ r ) J 1 ( λ R ) e a d λ
2

HANKEL.mws Maple numerical integration:

> u:=h*R*evalf(Int(BesselJ(0,r*x)*BesselJ(1,R*x)*exp(-x^2/a^2*t),x=0..infinity));
 Chapter IX The Integral Transform Methods IX.4 The Hankel Transform November 25, 2018 847

Example Thransient heat transfer through the cylindrical wall [problem in Guzii] March 12, 2015 - not finished

Heat equation in cylindrical coordinates:

1 ∂  ∂u  1 ∂u
r  = ν =0
r ∂r  ∂r  α ∂t

Consider the following initial boundary value problem:

∂ 2 u 1 ∂u 1 ∂u
+ = r1 < r < r2 , t > 0
∂r 2 r ∂r α ∂t

prescribed flux
Uniform initial temperature:
∂u ( r2 ,t )
k
∂r
= q (t ) u ( r,0 ) = 0 r1 ≤ r ≤ r2
0
r1 Boundary r = r1 is thermoinsulated:
r2 f1 ( t )
∂u ( r1 ,t ) 
r = 0
∂r
Boundary r = r2 , prescribed flux:
2 (t )
f 
∂u ( r2 ) ∂u ( r2 ) 1
k = q (t ) ⇒ = q (t )
∂r ∂r k

1) Finite Hankel Transform of order ν = 0 (see Chapter 3, p.38?) Neumann-Neumann 0 term?

r2

un ( t ) = ∫ u ( r,t ) X n ( r ) rdr
r1

Inverse Finite Hankel Transform

∞ X n (r )
??? 0 term u ( r,t ) = ∑ un ( t ) 2
n =1 X n (r )
where eigenfunctions (Finite Hankel of 0 order) are
J 0 ( λn x ) Y0 ( λn x )
??? X νn =0 ( x ) = −
−λn J 1 ( λn r2 ) + H 2 Jν ( λn r2 ) −λnY1 ( λn r2 ) + H 2Y0 ( λn r2 )
The squared norm of eigenfunctions is
r2
2 2
??? X 0
n (r ) = ∫  X n0 ( r )  rdr
r1

Eigenvalues are positive roots of equation:

??? λ J 1 ( λ r1 )  −λY1 ( λ r2 ) + H 2Y0 ( λ r2 )  −λY1 ( λ r1 )  −λ J 1 ( λ r2 ) + H 2 Jν ( λ r2 )  =

0

λ1 λ2 λ3
 848 Chapter IX The Integral Transform Methods IX.4 The Hankel Transform November 25, 2018

2) Take Finite Hankel transform of equation (use operational property ☺☺, p.93):

−r1 X n ( r1 ) f1 ( t ) + r2 X n ( r2 ) f 2 ( t ) − λn2 un
= (☼ N-N)

q (t ) 1 ∂u
−r1 X n ( r1 ) f1 ( t ) + r2 X n ( r2 ) − λn2 un =n
k α ∂t

∂un ( t ) q (t ) r2

+ αλn2 un ( t ) =
α r2 X n ( r2 ) condition: un ( 0 )
initial= u ( r,0 ) X ( r ) rdr
∫= 0 n 0
∂t k r1

3) Apply Laplace transform to equation

Q (s)
α r2 X n ( r2 )
sU n ( s ) − un ( 0 ) + αλ 2U n ( s ) = where Laplace transform is defined by
k
∞
U n ( s ) = ∫ un ( t ) e − st dt
0

α 1
Un (s) = r2 X n ( r2 ) Q (s)
k s + αλn2

4) Apply Inverse Laplace transform (use convolution theorem):

t t
α α
un ( t ) r2 X n ( r2 ) ∫ e −αλn τ q ( t − τ )dτ r2 X n ( r2 ) ∫ e −αλn (t −τ ) q (τ )dτ
2 2
= =
k 0
k 0

Example: 1 − H (t − b )
q (t ) =

t
α
un ( t ) = r2 X n ( r2 ) ∫ e
−αλn2 ( t −τ )
q (τ )dτ
k 0
b τ =b
α α
r2 X n ( r2 ) ∫ e
−αλn2 ( t −τ )
− 2 r2 X n ( r2 ) ∫ eαλn (τ −t ) d (αλn2 (τ − t ) )
2
= dτ =
k 0 αλn k τ =0

1 b
= 2 r2 X n ( r2 ) eαλn (τ −t ) 
2

λn k  0
1
r2 X n ( r2 ) eαλn (b −t ) − e −αλn t 
2 2
=
λn k
2  
1
r2 X n ( r2 ) eαλn b − 1 ⋅ e −αλn t
2 2
=
λn k
2  

5) Inverse Hankel Transform – solution of the IBVP:

∞ X n (r )
u ( r,t )
= ∑ u (t ) ⋅
n 2
0 term ???
n =1 X n (r )
 Chapter IX The Integral Transform Methods IX.4 The Hankel Transform November 25, 2018 849

6) Example

Maple: SF-AD-6-0 Heat Transfer Problem N-R finite cylinder 02. Mws

T, o C
t =0 s
t = 600 s

t = 6000 s

t = 60000 s

t = 600000 s
850 Chapter IX The Integral Transform Methods IX.4 The Hankel Transform November 25, 2018

4 Neumann-Neumann

Consider BE in the annular domain

( )
x 2 y ′′ + xy ′ + λ 2 x 2 −ν 2 y = 0 , x ∈ (L1 , L2 )

x
with homogeneous boundary conditions:
0 L1 L2

 dy 
 dx  =0 (Neumann)
x =L1

 dy 
 dx  =0 (Neumann)
 x =L2
The general solution is given by

y (x ) = c1 Jν (λx ) + c2Yν (λx )

The derivative of the general solution (use chain rule and differential
identities)
d  ν   ν 
y ( x ) = c1λ  − Jν +1 ( λ x ) + Jν ( λ x )  + c2 λ  −Yν +1 ( λ x ) + Yν ( λ x ) 
dx  λ x   λ x 
Substitute into boundary conditions:

 ν   ν 
x = L1 c1λ  − Jν +1 ( λ L1 ) + Jν ( λ L1 )  − c2 λ  −Yν +1 ( λ L1 ) + Yν ( λ L1 ) 
 λ L1   λ L1 
 ν   ν 
x = L2 c1λ  − Jν +1 ( λ L2 ) + Jν ( λ L2 )  + c2 λ  −Yν +1 ( λ L2 ) + Yν ( λ L2 ) 
 λ L2   λ L2 

Denote:
 ν  
=a11 λ Jν +1 ( λ L1 ) −   Jν ( λ L1 ) 
  L1  
 ν  
=a12 λYν +1 ( λ L1 ) −   Yν ( λ L1 ) 
  L1  
 ν  
 −λ Jν +1 ( λ L2 ) + 
a21 =  Jν ( λ L2 ) 
  L2  
 ν  
a22 = −λYν +1 ( λ L2 ) +   Yν ( λ L2 ) 
  L2  
Then a system for coefficients has the following matrix form:

 a11 a12   c1  0 
a =
 21 a22  c2  0 

A necessary condition for a system to have a non-trivial solution is

a a12 
det  11 =0
a21 a22 

it yields a characteristic equation for values of the parameter λ for which

the BVP has a non-trivial solution:
 Chapter IX The Integral Transform Methods IX.4 The Hankel Transform November 25, 2018 851

 ν    ν  
Equation for eigenvalues λ n : λ Jν +1 ( λ L1 ) −   Jν ( λ L1 )   −λYν +1 ( λ L2 ) +   Yν ( λ L2 ) 
  L1     L2  

 ν    ν  
− λYν +1 ( λ L1 ) −   Yν ( λ L1 )   −λ Jν +1 ( λ L2 ) +   Jν ( λ L2 )  =0
  L1     L2  

The positive roots of this equation provide an infinite set of eigenvalues

λn (note for Neumann boundary conditions, λn = 0 is also an eigenvalue
for ν = 0 ). Then for the determined eigenvalues λn , coefficients c1,n and
c 2 , n can be found from one of the equations of the system (choose the
second one):
a21c1 + a22 c2 = 0
One of the coefficients can be taken as a free parameter, choose
1 1
c1 = , then c2 =
a21 a22
With determined coefficients, solutions of the BVP yn (x )
(eigenfunctions) have the form:

Eigenfunctions: y0 ( x ) = 1 for ν =0
Jν (λn x ) Yν (λn x )
yn (x ) = −
a21,n a22 ,n
Jν ( λn x )
=
 ν  
 −λn Jν +1 ( λn L2 ) +   Jν ( λn L2 ) 
 L
 2 
Yν ( λn x )
−
 ν  
 −λnYν +1 ( λn L2 ) +   Yν ( λn L2 ) 
 L
 2 

The norm of the eigenfunctions is determined by the integral

L2
Nν2,n = ∫ xyn2 (x )dx ,
L1

L2 L2 L2
 x2  L22 − L21
N02,0
= ∫ xyn2 ( x=
) dx
∫=
xdx  = 
L 1 L  2 L 1
2
1

Fourier-Bessel series:
∞ ∞
f ( x ) = ∑ an yn ( x ) and f ( x ) = a0 + ∑ an yn ( x ) for ν = 0
n =1 n =1
L2 L2 L2

∫ xy02 ( x ) f ( x ) dx ∫ xf ( x ) dx ∫ xf ( x ) dx
L1 L1 L1
where
= a0 L2
= =
L
for ν = 0
2
N02,0
∫ xy ( x ) dx ∫ xdx
2
0
L1 L1
L2 L2

∫ xyn ( x ) f ( x ) dx ∫ xyn ( x ) f ( x ) dx
L1 L1
and
= an =
L 2
Nν2,n
∫ xyn ( x ) dx
2

L1
852 Chapter IX The Integral Transform Methods IX.4 The Hankel Transform November 25, 2018

IX.4.7.9 Comments about approximation of the Dirac function for Vedat May 30, 2015

Consider the boundary condition appeared in the Section 6.4.5.5

f1 ( t )
 
∂u ( r1 ,t ) ∂u ( r1 ,t ) q0
−k q0δ ( t )
= ⇒ = − δ (t )
∂r ∂r k

Can this condition be numerically approximated?

In my opinion, the answer is NOT!

The Dirac function δ ( t ) is not a classical function, because it is not defined at t = 0 .

The rigorous definition of the Dirac function can be done only in terms of the Theory of distributions.
In applied mathematics, we use a simplified version of this definition:

Definition of the Dirac function δ ( t ) :

δ ( t ) has a distribution property
h

∫ f ( t ) δ ( t=
) dt f ( 0 ) , ∀h > 0 for any continuous on ( −∞ , ∞ ) function f ( t ) with a finite support
−h

Only in terms of this definition we can make any conclusions about approximation of δ ( t ) .

Other definitions used in practice, for example,

0 t ≠ 0 ∞
δ (t ) =  and ∫ δ ( t ) dt = 1
∞ t = 0 −∞
are NOT mathematically correct – they only help to have some idea about this function.

There are many “approximations” of δ ( t ) which also help to understand the physical meaning of the Dirac function,
but they cannot be used as a numerical approximation.

For example: a step function with the area equal to unity δ h (t )

2h
0 t < −h and t > h

δ h (t ) =  1
 2h −h < t < h

−h h

Then usually you can see that it is written that δ ( t ) = lim δ h ( t )

h →0

But this convergence is not a point-wise because lim δ h ( t ) = ∞ , however δ h ( t ) is defined for any h ≠ 0 .
h →0
 Chapter IX The Integral Transform Methods IX.4 The Hankel Transform November 25, 2018 853

It is correctly to say that δ h ( t ) → δ ( t ) in terms of distributions:

h
lim ∫ f ( t ) δ h ( t ) dt = f ( 0 )
h →0
−h

If you use this “approximation” for boundary condition, it will yield some results. But they will not be more accurate if
you decrease h , as it should be with a proper approximation (actually, I have never tried that).

If you want a correct setting of the problem for numerical (finite-difference) solution it can be properly done in the
following way (already discussed before):

1) u ( r,0 ) = 0

2) Choose any small value t1 (for example, t1 = 0.001 )

3) Then the initial-boundary value problem for numerical solution does not involve the Dirac function:

∂ 2 u 1 ∂u 1 ∂u
+ = r1 < r < r2 , t > t1
∂r 2 r ∂r α ∂t

( r − a )2
−
kd e 4 kd t1
u ( r,t1 ) = q0 r1 ≤ r ≤ r2 initial condition (EXACT!!! See derivation in p. )
K π t1

∂u ( r1 ,t )
=0
∂r

∂u ( r2 )
k + h2 u ( r2 ) =
0
∂r

This approach with small enough value of t1 and small enough step in r to represent the initial profile should yield
an accurate numerical solution.
854 Chapter IX The Integral Transform Methods IX.4 The Hankel Transform November 25, 2018

IX.4.7.10 One-sided Dirac-delta which can be used in B.C. From review of the paper of Lappa [see reviews]

ii) Second, interpretation of the “delta Dirac function” is not consistent (even taken into account
controversial mathematical nature of Dirac function):
“where … δ s is well-known Dirac function (which takes value 1 on the interface and is zero
elsewhere). Though such representation is correct from a mathematical point of view...”
In our opinion, it is not correct. Dirac function is not 1 at interface. Some definitions assign to
Dirac function “ ∞ ” at the interface, but never 1. And more strictly speaking, Dirac function is not
a function at all, but rather a distribution, which possesses distribution property given by equation
(22)
∞

∫ δ ( s )ds = 1 .
−∞
Moreover, this property should be held over any open interval containing point s = 0 :
h

∫ δ ( s )ds = 1 for any h > 0

−h
(does not matter how small).

Therefore, insertion of standard Dirac function to the momentum equation is not appropriate,
1
because Dirac function is “loosly” equal zero outside of the interface, but the interface y = is not
2
1 1
included into the domain of momentum equation, which is − < y < .
2 2
Sometimes, Dirac function is interpreted as a “limit” (not point-by-by limit), which can be called a
limit in distribution sense or a weak limit of the sequence of regular functions, for example, of a
sequence of triangular functions
τh ( y) = ( − y h2 + 1 h ) ⋅  H ( y + h ) − H ( y ) + ( y h2 + 1 h ) ⋅  H ( y ) − H ( y + h ) ,
where H ( y ) is a Heaviside unit-step function. As can be seen, when h → 0 , the value of function
at zero τ h ( 0 ) → ∞ . The distribution property of this function is
h

∫ τ ( y )dy = 1 for any h > 0

−h
h (unit power).

Then, in the limit in distribution sense,

δ ( y ) = limτ h ( y ) (impulse of a unit power)
h →0
Therefore, for Dirac delta function, open neighborhood, which includes points from both sides of
the interface is needed. That is, definitely, not the case for momentum equation, which is valid
only for interior of the domain, and not valid at the boundary and outside of the boundary.
1
So, “spreading” of boundary condition at y = + into the domain with the help of Dirac delta is
2
not possible. In this sense, equations (20),(23-25) is just a symbolic description of what the author
is trying to accomplish.

However, author’s approach, in fact, is correct “spreading” of boundary condition into the
domain. But this “spreading” is performed not with the help of traditional Dirac function, but with
the help of the function which can be defined as
n −1
 1
ψ n (=
y) n y + 
 2
 Chapter IX The Integral Transform Methods IX.4 The Hankel Transform November 25, 2018 855

1 1
which for any n > 1 is defined for all − ≤ y≤+ .
2 2

ψ 10 ( y )

+1 2

∫ ψ ( y )dy = 1
−1 2
n for all n>1

ψ 4 ( y)

ψ 3 ( y)

ψ 2 ( y)

The property of this function is similar to property of Dirac function,

+1 2

∫ ψ ( y )dy = 1
−1 2
n (area under any curve is equal to unity)

however, integration is only over interior side of interface (i.e. over the domain of the governing
equation).
In a limit (in distribution sense), one can introduce an impulse function of a unit power
ψ ( y ) = limψ n ( y )
n →∞
which describes the spread of unit power only inside the domain.
This function formally can be obtained by differentiation of the function φ ( y ) defined by equation
(26) in the paper:
n ′
 1   1
n −1

φ ′ ( y ) = y +   =n  y +  = ψ n ( y)
 2    2
and equation (27) can be rewritten with the help of ψ n ( y ) as
n −1
d2  1
2
−φ ′ ( y ) + C1 =
g2 ( y ) = −n  y +  −ψ n ( y ) + C1 .
+ C1 =
dy  2

Then, equation (27) can be written for the limiting case as

d2
g2 ( y ) =
−ψ ( y ) + C1
dy 2
856 Chapter IX The Integral Transform Methods IX.4 The Hankel Transform November 25, 2018

Application of it as a starting point for a sequence of further steps of obtaining equations (28) and
(29) together with boundary conditions will yield the limiting case of solution directly.

Therefore, the momentum equation should include function ψ ( y ) , which describes

spreading only into the domain, instead of Dirac δ , which describes spreading into the
domain and outside of the domain.

Sequence of functions ψ n ( y ) used, in fact, in the paper, describe correctly approach to the limiting
case.