Copyright © Sandip Mazumder

Lecture 5: What we learnt The Ohio State University

 Review of Linear ODEs

 Characteristic or Eigenvalue problem
 Two-dimensional steady state heat conduction—
separation of variables

1
 Copyright © Sandip Mazumder
Lecture 6: What we will learn The Ohio State University

 Two-dimensional steady state heat conduction—

separation of variables in Cartesian coordinates
 Two-dimensional steady state heat conduction
o Principle of superposition
o Cylindrical coordinates

2
 Copyright © Sandip Mazumder
Two-dimensional Steady State Heat Conduction
The Ohio State University

Next, we will develop analytical methods for solving the 2D steady state
heat conduction equation.
Assumptions: heat conduction only, no heat generation (we will look at
this later), constant properties (k), 2D, steady state.
y To 2 2
∂ T ∂ T
GE: ∇ 2T = 2
+ 2
= 0
∂x ∂y
T (0, y ) = T0
To L TB
T ( L, y ) = TB
BCs:
T ( x,0) = T0
To x T ( x, L) = T0
Q: Is the system homogeneous?
A: No. The GE is homogeneous, but the BCs are not.
Q: Is there a transformation that will make the BCs homogeneous?
A: Yes, but not all of them. φ = T – To will make 3 of the BCs homog. 3
 Copyright © Sandip Mazumder
Two-dimensional Steady State Heat Conduction
The Ohio State University

Regardless, let us proceed with the idea of using φ = T – To since it

makes every BC homogeneous, except one.
φ = T − To ⇒ ∇ 2φ = ∇ 2T
y 0 ∂ 2
φ ∂ 2
φ
GE: ∇ 2φ= 2
+ 2
= 0 H
∂x ∂y
φ (0, y ) = 0 H
0 L φB
BCs:
φ ( L, y ) = TB − To = φB N-H
φ ( x,0) = 0 H

0 x φ ( x, L ) = 0 H

Another way of attaining the same goal (of homogenizing to the best
possible extent) could be to use a non-dimensional temperature:
T − To
θ= ⇒ (TB − To ) ∇ 2θ =
∇ 2T
TB − To 4
 Copyright © Sandip Mazumder
Two-dimensional Steady State Heat Conduction
The Ohio State University

Proceeding with the non-dimensional temperature:

y 0 ∂ 2
θ ∂ 2
θ
GE: ∇ θ =
2
2
+ 2
= 0 H
∂x ∂y
θ (0, y ) = 0 H
0 L 1
θ ( L, y ) = 1 N-H
BCs:
θ ( x,0) = 0 H
x θ ( x, L ) = 0 H
0
Note: the objective is to homogenize the system, not necessarily non-
dimensionalize it.
In this case, the non-dimensional variable was chosen such that it
homogenizes the system, as well.

Non-dimensionalization sometimes helps simplify the mathematical steps.

Let us proceed with the non-dimensional system shown above. 5

 Copyright © Sandip Mazumder
Two-dimensional Steady State Heat Conduction
The Ohio State University

Separation of Variables
Next, we write θ as a product of two separate functions:
=θ θ ( x=
, y ) X ( x) ⋅ Y ( y )
Note: this may not always work out mathematically. In order for it to work
out, a certain set of rules must be obeyed strictly. For now, we will just
proceed, and visit those rules later.
∂2 ∂2
Substitute into GE: 2
( XY ) + 2 ( XY ) =
0
∂x ∂y
d2X d 2Y 1 d2X 1 d 2Y
Simplify: Y 2
+X 2 = 0 2
= −
dx dy X dx Y dy 2
Since X = X(x) and Y = Y(y), the LHS is a function of x only, while the
RHS is a function of y only.
1 d2X 1 d 2Y
The only way they can be equal is if: 2
=
− 2
constant
=
X
 dx Y dy
  6
f ( x) f ( y)
 Copyright © Sandip Mazumder
Two-dimensional Steady State Heat Conduction
The Ohio State University

The next big question is if the constant should be positive or negative.

Before answering this question, let us examine the homogeneous BCs
Recall that=θ θ ( x=
, y ) X ( x) ⋅ Y ( y )
θ (0, y ) =0 ⇒ X (0) ⋅ Y ( y ) =
0 ⇒ X (0) = 0
Therefore: θ ( x,0) = 0 ⇒ X ( x) ⋅ Y (0) =0 ⇒ Y (0) = 0
θ ( x, L ) =
0 ⇒ X ( x) ⋅ Y ( L) =
0 ⇒ Y ( L) = 0
Note that the three homogeneous BCs also separated out easily (just
like the GE) and has given us 3 BCs—one for X and 2 for Y.
Suppose we choose a negative constant. This gives us
1 d 2Y d 2Y
− =
− λ 2
− λ 2
0
Y=
Y dy 2 dy 2

We already know from our previous discussion of EV problems, that this

particular GE with the two highlighted BCs above (in red) can only
produce a trivial solution. This implies that choosing a negative constant
is ill-advised. 7
 Copyright © Sandip Mazumder
Two-dimensional Steady State Heat Conduction
The Ohio State University

1 d 2Y
Next, we try a positive constant: − = + λ 2

Y dy 2
d 2Y Y (0) = 0
GE: 2
+λ Y =
2
0 BCs:
dy Y ( L) = 0
We already know that this is an EV problem, and has a non-trivial solution:
nπ
Yn ( y ) C1n sin( =
= y ), n 0,1, 2,..., ∞
L
Now that we know that the constant must be positive, let us consider
solution to the other part of the problem.
1 d2X 1 d 2Y
=
− =
+ λ 2

X dx 2 Y dy 2
d2X
GE: 2
− λ 2
0
X= BC: X (0) = 0
dx 8
 Copyright © Sandip Mazumder
Two-dimensional Steady State Heat Conduction
The Ohio State University

We have already looked at the solution of this ODE before:

) D1 exp(λ x) + D2 exp(−λ x)
X ( x=
Apply BC: X (0) =+ 0 ⇒ D2 =
D1 D2 = − D1
[
X ( x) D1 exp(λ x) − exp(−λ x)
Therefore, = ]
 nπ nπ 
Plug in the Evs: X n ( =
x) D1n exp( x) − exp(− x)  , =
n 0,1, 2,..., ∞
 L L 

Now that we have both parts of the solution, let us assemble θ :

 nπ nπ  nπ
θ n ( x, y )= X n ( x) ⋅ Yn ( y )= D1n exp( x) − exp(− x)  ⋅ C1n sin( y )
 L L  L
 nπ nπ  nπ
= Bn exp( x) − exp(− x)  ⋅ sin( y)
 L L  L
Note: (1) we have infinite many solutions, and (2) we still have an
undetermined coefficient. 9
 Copyright © Sandip Mazumder
Two-dimensional Steady State Heat Conduction
The Ohio State University

Recall that while we have used the three homogeneous BCs, we still
have not used the non-homogeneous BC.
Q: Why did we not try to derive a BC for X from the n-h BC?
A: Because a n-h BC, by definition, is also non-separable.

θ ( L, y ) = 1 X ( L) ⋅ Y ( y ) =
1
There are infinite many combinations that can satisfy this equation. In
other words, it is not separable uniquely, as opposed to the other BCs.

Since we have one undetermined coefficient, and one remaining BC, it

makes sense now to use it to determine that coefficient. Therefore,
nπ nπ
( L, y ) Bn [ exp(nπ ) − exp(− nπ ) ] ⋅ sin( =
θn= y) 1 An sin( y) = 1
 L L
An
The above equation cannot be satisfied for all values of y no matter
what the coefficient An is.
10
 Copyright © Sandip Mazumder
Two-dimensional Steady State Heat Conduction
The Ohio State University

Recall that we discussed earlier that a linear combination of the

eigenfunctions also satisfies the GE+BCs.
In other words, not only does θn satisfy the GE + 3 homogeneous BCs, a
linear combination of θn does so, as well.
Let us now use such a solution (instead of a one-term solution) in the hope
that this solution might satisfy the non-homogeneous BC.
Therefore,
∞ ∞
 nπ nπ  nπ
y ) ∑ X n ( x) ⋅ Yn (=
θ ( x,= y ) ∑ Bn exp( x) − exp(− x)  ⋅ sin( y)
n 0= n 0  L L  L
Now apply the inhomogeneous BC:
∞
nπ ∞
nπ
θ ( L, y ) ∑ Bn [ exp(nπ ) − exp(−nπ )] ⋅ sin( y )
= ∑ An sin( y) = 1
n=0 L n=0 L
From the theory of Fourier series, we know that any continuous function
can be represented by a Fourier series. Therefore, the above equation has
potential. 11
 Copyright © Sandip Mazumder
Two-dimensional Steady State Heat Conduction
The Ohio State University

It is important to understand that the right BC could have been any

function of y [let’s say, g(y)], rather than a constant. In other words, we
could have a temperature distribution on that face, rather than a single
value. In such a scenario,
∞
nπ
∑
n=0
An sin(
L
y) = g ( y) This is also a perfectly valid equation.
∞
nπ
Let us get back to the problem at hand. How do we solve ∑ An sin( y) = 1 ?
n=0 L
mπ
Multiply both sides of the equation by sin( y) :
L
mπ ∞
nπ mπ ∞
nπ mπ mπ
sin( y )∑ An sin( y ) = sin( y) ∑ An sin( y )sin( y ) = sin( y)
L n=0 L L n=0 L L L
Integrate from 0 to L:
L L
∞
nπ mπ mπ
∑
n=0
An ∫ sin(
L
y )sin(
L
y ) dy = ∫ sin(
L
y ) dy
0 0 12
 Copyright © Sandip Mazumder
Two-dimensional Steady State Heat Conduction
The Ohio State University

Let us first examine the integral on the LHS of the equation. It turns out
this integral is equal to zero unless n = m. Thus,
Principle of
L
 0 if n ≠ m Orthogonality
nπ mπ L
I LHS ∫0=sin( y )sin( y ) dy  2 mπ
L L  ∫ sin ( L y )dy if n = m
0
L L
2 mπ mπ
Therefore, Am ∫ sin ( y ) dy = ∫ sin( y ) dy
0
L 0
L
L L
mπ L  mπ  L
∫0 sin( L y) dy =
I RHS = −  cos( y) = [1 − cos(mπ )]
mπ  L  0 mπ
L L
mπ 1  2mπ 
∫0 sin ( L y= ∫
2
I=
LHS ) dy 1− cos( y )  dy
20 L 
L
1 L 2mπ  1 L
=  y− sin( y ) = [ L − 0 − (0 − 0)=
]
2  2mπ L 0 2 2 13
 Copyright © Sandip Mazumder
Two-dimensional Steady State Heat Conduction
The Ohio State University

I RHS 2[1 − cos(mπ )] 2[1 − (−1) m ]

Therefore, Am =
= =
I LHS mπ mπ
Am 2[1 − cos(mπ )]
It follows that Bm =
[exp(mπ ) − exp(−mπ )] mπ [exp(mπ ) − exp(−mπ )]
Note that m is just a dummy index in the above equation, and it can be
readily switched to n (or any other notation) without any repercussion.

As the last step, assemble the final solution:

∞
2[1 − cos(nπ )]  nπ nπ  nπ
θ ( x, y ) ∑ 
n = 0 nπ [ exp( nπ ) − exp( − nπ ) ] 
exp( x) − exp(−
L
x)  ⋅ sin(
L  L
y)

Noting that sinh(x) =[exp(x)-exp(-x)]/2, we may alternatively write:

∞
2[1 − cos(nπ )] nπ nπ
θ ( x, y ) ∑
n = 0 nπ sinh( nπ )
sinh( x) ⋅ sin(
L L
y)
14
 Copyright © Sandip Mazumder
Two-dimensional Steady State Heat Conduction
The Ohio State University

To generate numerical answers, we still need to write a computer program

to calculate the infinite series.
Note that as n gets large, the denominator quickly becomes very large,
and the series converges rapidly.
General practice is to compare the current term with the current sum, and
terminate the calculation if the current term is a certain small fraction of
the current sum.
Sample solution:
Note: Heat flux vector is always locally
perpendicular to the isotherm.

15
 Copyright © Sandip Mazumder
Two-dimensional Steady State Heat Conduction
The Ohio State University

Now that we have completed the derivation of the solution, let us re-visit
some aspects of the derivation.
Criteria that must be obeyed for Separation of Variables to apply:
1. PDE must be linear.
2. ALL equations (GE + BC + IC combination), EXCEPT ONE, must be
homogeneous.
3. All equations cannot be homogeneous. This will only produce a trivial
solution, no matter what we do.
Note: If all 3 of the above criteria are not satisfied, do not attempt SoV.

Steps in Separation of Variable:

1. Separate GE and the three homogeneous BCs into two parts. Write
down these two parts clearly.
2. Eigenvalue problem must be set up in the direction (variable) where
both BCs are homogeneous.
3. Satisfy the final non-homogeneous BC with a Fourier sum. Apply
principle of orthogonality to determine coefficients of Fourier series. 16
 Copyright © Sandip Mazumder
2D Heat Conduction: Superposition
The Ohio State University

In the previous lecture we learnt how to use Separation of Variables to

solve steady state heat conduction problems.
One of the criteria that has to be met in order to apply Separation of
Variables is:
“ALL equations (GE + BC + IC combination), EXCEPT ONE, must be
homogeneous.”
What if this criterion is not satisfied?

y TT ∂ 2
T ∂ 2
T
GE: ∇ 2T = 2
+ 2
= 0
∂x ∂y
T (0, y ) = T0
To L TR
BCs:
T ( L, y ) = TR
T ( x,0) = T0
x T ( x, L) = TT
To
17
 Copyright © Sandip Mazumder
2D Heat Conduction: Superposition
The Ohio State University

Based on our earlier discussion, use θ = T – To to homogenize system.

y θT ∂ 2
θ ∂ 2
θ
GE: ∇ θ=
2
2
+ 2= 0
∂x ∂y
θ (0, y ) = 0 H
0 L θR θ ( L, y ) = TR − To = θ R N-H
BCs:
θ ( x,0) = 0 H
θ ( x, L) = TT − To = θT N-H
0 x
Thus, this system does not satisfy the criterion for SoV!

To overcome this barrier, we apply Superposition.

CAUTION: Superposition is only applicable to a linear system (which was
one of the requirements for SoV to begin with).

Superposition: θ
= ( x, y ) M ( x, y ) + D ( x, y ) 18
 Copyright © Sandip Mazumder
2D Heat Conduction: Superposition
The Ohio State University

Next, we substitute the superposition into our governing system.

GE: ∇ 2θ = ∇ 2 M + ∇ 2 D = 0 ∇2 M =
∇2 D =
0

θ (0, y ) = 0 ⇒ M (0, y ) + D(0, y ) = 0

θ ( L, y ) = TR − To = θ R ⇒ M ( L, y ) + D( L, y ) = θ R
BCs:
θ ( x,0) = 0 ⇒ M ( x,0) + D( x,0) = 0
θ ( x, L) = TT − To = θT ⇒ M ( x, L) + D( x, L) = θT

M-Problem D-Problem
GE:
2
∇ M= 0 ∇2 D = 0
M (0, y ) = 0 D(0, y ) = 0
M ( L, y ) = θ R D ( L, y ) = 0
BCs: D( x,0) = 0
M ( x,0) = 0
M ( x, L ) = 0 D ( x, L ) = θT 19
 Copyright © Sandip Mazumder
2D Heat Conduction: Superposition
The Ohio State University
θT 0 θT

Graphically: 0 θR = 0 θR + 0 0

0 0 0
Note now that both the M-problem and the D-problem satisfy all criteria
that need to be satisfied for the application of SoV.
As a matter of fact, the M-problem is the one we solved in class in the
previous lecture (with the exception that we are not using a non-
dimensional temperature here). Thus,
∞
2[1 − cos(nπ )] nπ nπ
M ( x, y ) θ R ∑ sinh( x) ⋅ sin( y)
n = 0 nπ sinh( nπ ) L L
The solution to the D-problem is identical in form to the solution of the
M-problem, except that we switch x & y, and replace θR by θT. Thus,
∞
2[1 − cos(nπ )] nπ nπ
D ( x, y ) θT ∑ sinh( y ) ⋅ sin( x)
n = 0 nπ sinh( nπ ) L L 20
 Copyright © Sandip Mazumder
2D Heat Conduction: Cylindrical Coordinates
The Ohio State University

Next, we will consider some steady state heat conduction problems in

2D cylindrical coordinates, and apply SoV to these problems.
Problem 1 2 1 ∂  ∂T  ∂ 2T
=
GE: ∇T r = + 2 0
r TT r ∂r  ∂r  ∂z
ro
T (0, r ) = T0
To To T ( L, r ) = T0
BCs: ∂T
z =0
∂r ( z ,0)
Homogenize system using θ = T – T0
T ( z , ro ) = TT
1 ∂  ∂θ  ∂ 2
θ
=
GE: ∇θ
2
r = + 2 0
r ∂r  ∂r  ∂z Note:
 All conditions of SoV met
∂θ
= θ (0, r ) 0,= 0  BCs in z direction are both
BCs: ∂r ( z ,0) homogeneous
θ ( L, r ) = 0, θ ( z , ro ) = TT − T0 = θT 21
 Copyright © Sandip Mazumder
2D Heat Conduction: Cylindrical Coordinates
The Ohio State University

Apply SoV: θ ( z=
, r ) Z ( z ) ⋅ R(r )
Z d  dR  d 2Z
Substitute in GE: r +R 2 = 0
r dr  dr  dz
1 d  dR  1 d 2Z
r =− 2
constant
=
rR dr  dr  Z dz
We have already noted that BCs in the z direction are homogeneous.
Therefore, the EV problem needs to be set up in the z direction. Thus,
1 d  dR  1 d 2Z d 2Z
 r  =
− +
= λ 2
2
+λ 2
Z =0
rR dr  dr  Z dz 2 dz
θ (0, r ) =0 ⇒ Z (0) ⋅ R (r ) =0 ⇒ Z (0) = 0
Now separate out the θ ( L, r ) =
0 ⇒ Z ( L) ⋅ R (r ) =0 ⇒ Z ( L) =0
3 homogeneous BCs:
∂θ
0 ⇒ Z ( z ) ⋅ R′(0) =
= 0 ⇒ R′(0) =
0
∂r ( z ,0) 22
 Copyright © Sandip Mazumder
2D Heat Conduction: Cylindrical Coordinates
The Ohio State University

Write down the two ODE systems separately:

d 2Z d  dR 
+λ 2
Z =0  r  − λ 2
0
rR =
dz 2
dr  dr 
Z (0) = 0 EV Problem
R′(0) = 0
Z ( L) = 0

Solution to EV problem is already known from previous discussions:

nπ
n ( z)
Z= Cn sin(λn z ), =
λn = , n 0,1, 2,..., ∞
L
Other ODE (in R) is solved using a series solution method known as
the Frobenius’ method. Solution is as follows:
Rn (r ) D1n I 0 (λn r ) + D2 n K 0 (λn r )
=
I 0 = modified Bessel function of first kind and order 0
where
K 0 = modified Bessel function of second kind and order 0 23
 Copyright © Sandip Mazumder
2D Heat Conduction: Cylindrical Coordinates
The Ohio State University

Functions we have dealt with previously, such as exp(), sin() etc. are all
so-called power series. For example, the exponential function may be
written as 2 n
x ∞
x
exp( x) =1 + x + + ... = ∑
2! n=0 n!

Likewise, the Bessel function is just another power series. For example
∞
x2n
I 0 ( x) = ∑ 2 n 2
n=0 2 ( n !)
In order to apply the two BCs, we must know how the two Bessel
functions behave.

I 0 ( x) K 0 ( x)
Zero
slope

1
3 x x 24
 Copyright © Sandip Mazumder
2D Heat Conduction: Cylindrical Coordinates
The Ohio State University

=
Solution: Rn (r ) D1n I 0 (λn r ) + D2 n K 0 (λn r )
According to the plots on the previous slide, K0 goes to infinity as r
approaches 0. Therefore D2n must be equal to zero for the solution to
be bounded (Note: temperature cannot be infinite). Thus,
Rn (r ) = D1n I 0 (λn r )
The BC R′(0) = 0 Is automatically satisfied by the solution.
θ n ( z , r ) Cn sin(λn z ) ⋅ D1n I 0 (λn r )
We now combine the two solutions:=

Combine the two constants, and construct Fourier series (linear

combination) as explained earlier:
∞
θ ( z, r ) ∑B
n=0
n sin(λn z ) ⋅ I 0 (λn r )

∞ ∞
Apply the final non-
homogeneous=
=
BC: θ ( z , r0 )
n 0
n 0=
n 0 n
n 0
∑B I (λ=
r )sin(λ z ) ∑=
A sin(λ z )
n n
25
θT
 Copyright © Sandip Mazumder
2D Heat Conduction: Cylindrical Coordinates
The Ohio State University

We have already solved this equation by applying the principle of

orthogonality, and the solution is (from Lecture 5):
2[1 − cos(nπ )] 2[1 − (−1) n ]
An = θT θT
nπ nπ
2[1 − cos(nπ )]
Therefore, Bn = θT
nπ I 0 (nπ r0 / L)

Our final solution may then be written as:

∞
2[1 − cos(nπ )] nπ nπ
θ ( z ,= T0 θT ∑
r ) T ( r , z ) −= sin( z ) ⋅ I 0 ( r )
n = 0 nπ I 0 ( nπ r0 / L ) L L

The solution form is similar to the solution form in Cartesian

coordinates, except that the hyperbolic sine function has been
replaced by the Bessel function.

26