Numerical Methods: Partial Differential Equations

K. Naveesh Reddy

c Copyright CNEP

May 6, 2011

General form of linear,Second-order PDEs for two independent variables:

2 2 2
A ∂∂ 2 ux + B ∂x∂y
∂ u
+ C ∂∂ 2uy + D ∂u
∂x
+ E ∂u
∂y
+ F = 0.

where A,B,C,D,E and F are functions of x and y or constants.Based on the val-

ues of the coefficients of the second-derivative terms-A,B,C above equation can be
classified into one of the three categories.

1.B 2 − 4AC < 0, Elliptic

Ex:-Laplace Equation(steady state with two spatial dimensions)
∂2T 2
∂2x
+ ∂∂ 2Ty = 0

2.B 2 − 4AC = 0 , Parabolic

Ex:-Heat Conduction Equation(time variable with one spatial dimension)
∂T 2
∂t
= k ∂∂ 2Tx

3.B 2 − 4AC > 0 , Hyperbolic

Ex:-Wave Equation(time variable with one spatial dimension)
∂2y 2

∂2x
= c12 ∂∂ 2yt

1
Finite Difference:Elliptic Equations:

Laplace Equations:
• The Laplace Equation can be used to model a variety of problems involving the
potential of an unknown variable.
• Steady state, boundary value problem.
• Ex:-A thin rectangular heated plate.
Consider a thin rectangular plate of thickness ∆z.At steady state, the flow of heat
into the element over a unit time period ∆t must equal to the flow out.(see fig1)

Figure 1: A thin rectangular heated plate

q(x)∆y ∆z ∆t + q(y)∆x ∆z ∆t = q(x + ∆x) ∆y ∆z ∆t + q(y + ∆y) ∆x ∆z ∆t

rearranging, divide by ∆x ∆y ∆z ∆t and taking the limit results in a PDE rep-

resenting the conservation of energy for the plate.
∂q ∂q
∂x
+ ∂y
=0

since temperature boundary conditions are known. the above expression must be
reformulated in terms of temperature using ”Fourier’s law of heat conduction”,which
can be represented as

qi = -kρC ∂T ∂i
(1)
where
qi is heat flux in the direction of the i dimension.
ρ is density of the material.
C is heat capacity of the material.
k 0 = kρC is the coefficient of thermal conductivity.

substituting it in above equation we get Laplace’s Equation(steady state with two

spatial dimensions)
∂2T ∂2T
∂2x
+ ∂2y
=0

2
Note that for the case where there are sources or sinks of heat which is given by
a function f(x,y)then the corresponding equation is referred to as the poisson’s equa-
tion.

∂2T ∂2T
∂2x
+ ∂2y
= f (x, y)

Solution Technique:

Laplacian Difference Equation:

• Treating the plate as a grid of discrete points.(see fig2)

• Replace derivatives by centered finite-divided-difference formulas.

Figure 2: A grid used for the finite difference solution of Laplacian equation

∂2T Ti+1,j −2Ti,j +Ti−1,j

∂2x
= ∆x2

∂2T Ti,j+1 −2Ti,j +Ti,j−1

∂2y
= ∆y 2

substituting in laplace equation,we get

Ti+1,j −2Ti,j +Ti−1,j Ti,j+1 −2Ti,j +Ti,j−1
∆x2
+ ∆y 2
=0

for ∆ x = ∆ y

Ti+1,j − 4Ti,j + Ti−1,j + Ti,j+1 + Ti,j−1 = 0

this relation holds good for all interior points on the plate,is referred to as the Lapla-
cian difference equation.

Boundary conditions:

3
• The temperature at the boundary is set to a fixed value-Dirichlet Boundary condi-
tion.
• Derivative boundary condition(derivative is given)-Neumann Boundary condition.
for ex:-(see fig3)
Dirichlet Boundary conditions are:- Ti,0 = T0 ,Ti,3 = T1 ,T0,j = T2 ,T3,j = T3

Figure 3: A heated plate with Dirichlet Boundary conditions

and corresponding difference equations are:-

T2,1 − 4T1,1 + T2 + T1,2 + T0 = 0

T3 − 4T2,1 + T1,1 + T2,2 + T0 = 0

T2,2 − 4T1,2 + T2 + T1 + T1,1 = 0

T3 − 4T2,2 + T1,2 + T1 + T2,1 = 0

4 equations and 4 unknowns can be easily solved

The Liebmann Method

• n-by-n grid involves n2 linear algebraic equations.

• For larger sized grids,a significant number of terms will be zero.

• To such sparse systems,full matrix eliminations methods waste great amounts of

computer memory storing these zeros.

• the most commonly employed approach is gauss-seidel,which when applied to PDEs

is also referred to as Liebmann’s Method.

4
 Ti+1,j +Ti−1,j +Ti,j+1 +Ti,j−1
Ti,j = 4

solve iteratively for j=1 to n and for i = 1 to m. since system is diagonally dominant,it
will converge on a stable solution.

Overrelaxation:-
Overrelaxation is sometimes employed to accelerate the rate of convergence by ap-
plying the following formula after each iteration.

T new i,j = λT new i,j + (1 − λ)T old i,j , 1 < λ < 2

the percent relative errors(a )i,j

is given as:- 
 T new −T old i,j 
(a )i,j = Ti,j new
i,j
 < s

where s is stopping criterion.

Secondary Variable
For the heated plate
• a primary variable is Temperature(T).
• a secondary variable is the rate of heat flux across the plate’s surface(qi ).

qi = −kρC ∂T
∂i

Ti+1,j −Ti−1,j
qx = −k 0 2∆x

Ti+1,j −Ti−1,j
qy =−k 0 2∆y

the resultant heat flux can be computed from these two quantities by
p
qn = qx2 + qy2
where the direction of qn is given by

θ = arctan( qqxy ), qx > 0

θ = arctan( qqxy ) + π, qx < 0

Derivative Boundary Conditions

5
Derivative Boundary Conditions,in which derivative is specified-Neumann Boundary
condition.
for ex:-
For the heated plate,where the edge is insulated i.e heat flux is zero.
fig depicts a node(-1,j) lying outside the plate.(see fig4)

Figure 4: A Boundary node (0, j) on the left edge of heated plate

T1,j − 4T0,j + T−1,j + T0,j+1 + T0,j−1 = 0

Imaginary point (-1,j) lying outside the plate is required for this equation.it serves as
the vehicle for incorporating the derivative boundary condition into the problem.

∂T ∼
=
T1,j −T−1,j
∂x 2∆x

T−1,j = T1,j − 2∆x ∂T

∂x
substitute it we get

2T1,j − 4T0,j − 2∆x ∂T

∂x
+ T0,j+1 + T0,j−1 = 0

Irregular Boundaries

• many systems have irregular boundaries.(see fig5)

• we use weight functions α, β.

6
 Figure 5: A grid of heated plate with an irregular shaped Boundary

( ∂T ) ∼ T1,j −Ti−1,j
∂x i−1,j = α∆x

( ∂T ) ∼ T1,j −Ti,j−1
∂y i,j−1 = β∆y

• simple but important.

• Becomes difficult to handle when imposing Neumann’s condition on irregular bound-

aries.

Control volume Approach

• also called as volume integral approach.

• it is used for the system having unequal grid spacing,made of different materi-
als and mixed boundary conditions.(see fig6)

• rather than approximating the PDE at a point,the approximation is applied to

a volume surrounding the point.

• For orthogonal grid,the volume is formed by the perpendicular lines through the
midpoint of each line joining adjacent nodes.

7
Figure 6: A heated plate with unequal grid spacing,two materials,and mixed boundary conditions

A steady-state heat balance for the volume(Energy conservation)(see fig7).

Figure 7: A control volume for node A with arrows indicating heat transfer through the boundaries

q1 h2 t − q4 h2 t + q2 h2 t + q3 h4 t − qh h4 t = 0

where
q1 = −k10 ∂T
∂x
, q2 = −k20 ∂T
∂y
, q3 = −k20 ∂T
∂y
, q4 = −k20 ∂T
∂x
, qh = hc (TA − T∞ )

q1 = −k10 TA −T
h
B
, q2 = −k10 TA −T
h
D
, q3 = −k20 TA −T
h
D
, q4 = −k20 TF(−T
h
)
A
,
4

k20 TF −T
h
A −TB
− k10 TA2h −TD
− k10 TA2h −TD
− k20 TA4h − hc TA −T
4
∞
=0

8
Finite Difference:Parabolic Equations

• Parabolic equations are employed to characterize time-variable problems.

• Ex:-A long,thin insulated rod-one dimensional case.

The amount of heat stored in the element over a unit time period ∆t.

Input - output = Accumulation

q(x)∆y ∆z ∆t - q(x + ∆x) ∆y ∆z ∆t = ρC ∆x ∆y ∆z ∆t

rearranging, divide by ∆x ∆y ∆z ∆t and taking the limit results in a PDE rep-

resenting the conservation of energy for the thin insulated rod.

− ∂x
∂q
=ρC ∂T
∂t

Substituting Fourier’s law of heat conduction in above equation, results in Heat Con-
duction Equation
2
k ∂∂ 2Tx = ∂T
∂t

Explicit Method

• we approximate the spatial derivative at time level l.

• centered finite-divided-difference formulas.

l −2T l +T l
Ti+1
∂2T i i−1
∂2x
= ∆x2

∂T Til+1 −Til
∂t
= ∆t

l represent the time step.

substituting in Heat Conduction Equation,we get

l −2T l +T l
Ti+1 i i−1 Til+1 −Til
k ∆x2
= ∆t

9
Til+1 = Til + λ(Ti+1
l
− 2Til + Ti−1
l
)
∆t
where λ = k ∆x 2

Boundary conditions:

• The temperature at the boundary is set to a fixed value-Dirichlet Boundary condi-

tion.

T (t = 0)∀i
Ti=0 ∀t; Ti=N ∀t;

• Derivative boundary condition(derivative is given)-Neumann Boundary condition.

qi=0,n = −k 0 ( ∂T )
∂x i=0,n

where
T0 −T−1
( ∂T ) =
∂x i=0 ∆x

Tn+1 −Tn
( ∂T )
∂x i=n
= ∆x

where T0 andTn+1 are fictitious temperature that can be determined once q is known.

Convergence and stability:

• Convergence means that as ∆x and ∆t approach zero.The results of the finite-

difference technique approach the true solution.

• Stability means that errors at any stage of the computation are not amplified but
are attenuated as the computation progresses.
∆t
λ = k ∆x2

k is given,∆x and ∆t are chosen.

the Explicit method is convergent and stable if λ ≤ 12 .

10
•λ≤ 1
2
- errors don’t grow,but oscillate.

•λ≤ 1
4
- ensures that solution will not oscillate.

•λ≤ 1
6
- tends to minimize truncation error.

A simple Implicit Method:

Implicit method overcome the two difficulties of explicit method such as

• Problems related to stability,computational burden may be large to attain accept-

able accuracy.

• they exclude information that has a bearing on the solution.

• In explicit method,we approximate the spatial derivative at time level l.

• In implicit method,we approximate the spatial derivative at time level l+1.

l+1
∂2T Ti+1 −2Til+1 +Ti−1
l+1

∂2x
= ∆x 2

∂T T l+1 −T l
∂t
= i ∆t i

substituting in Heat Conduction Equation,it can be expressed as

Til = −λTi+1
l+1
+ (1 + 2λ)Til+1 − λTi−1
l+1

∆t
where λ = k ∆x 2

this equation applies to all but for the first and last nodes,which must be modi-
fied to reflect the boundry condition.

• The sysytem obtained is tridiagonal,we can utilize the extremely efficient solu-
tion algorithms that are available for tridiagonal systems.

• The implicit method described is stable and convergent.

• It can be also used to obtain steady state solutions for elliptic equations.

11
The Crank-Nicolson Method:

• it is an alternative implicit scheme that is second order accurate in both space

and time.

• difference approximations are developed at the midpoint of the time increment

i.e l+ 12 .

l+1
∂2T Ti+1 −2Til+1 +Ti−1
l+1 l −2T l +T l
Ti+1
∂2x
= 12 [ ∆x 2 + i
∆x2
i−1
]

∂T T l+1 −T l
∂t
= i ∆t i

substituting in Heat Conduction Equation and rearranging,it can be expressed as

−λTi+1
l+1
+ 2(1 + λ)Til+1 − λTi−1
l+1 l
= λTi+1 + 2(1 − λ)Til + λTi−1
l

∆t
where λ = k ∆x 2

• It is often used for one spatial dimensions.

• It can be also used for unequally spaced meshes.

Parabolic equations in two spatial dimensions:

• Heat Conduction Equation in two spatial dimensions is of the form.

∂T 2 ∂2T
∂t
= k( ∂∂ 2Tx + ∂2y
)

The ADI scheme:

• ADI-Alternating Direction Implicit.

• It is used for solving Parabolic equations in two spatial dimensions using tridi-
agonal matrices.

12
• In this method each time increment is executed in two steps.

• Square grid(∆x = ∆y).

for the first step,the spatial derivative x is at time level l , the spatial derivative
y is at time level l+ 12 and the time derivative from l to l+ 21 .

l+ 1 l+ 1 l+ 1 l+ 1
Ti,j 2 −Ti,j
l Tl −2T l +T l 2 −2T
Ti,j+1 2 2
i,j +Ti,j−1
∆t = k[ i+1,j ∆xi,j2 i−1,j + ∆y 2 ]
2

substituting in two spatial dimension Heat Conduction Equation and rearranging,it

can be expressed as

l+ 1 l+ 1 l+ 1
l
λTi+1,j + 2(1 − λ)Ti,j
l l
+ λTi−1,j = −λTi,j+1
2
+ 2(1 + λ)Ti,j 2 − λTi,j−1
2

∆t
where λ = k 2∆x 2

for the second step,the spatial derivative x is at time level l+1 , the spatial derivative
y is at time level l+ 12 and the time derivative from l+ 21 to l+1.

l+1 l+ 1 l+ 1 l+ 1 l+ 1
Ti,j −Ti,j 2 T l+1 −2Ti,j
l+1 l+1
+Ti−1,j 2 −2T
Ti,j+1 2 2
i,j +Ti,j−1
∆t = k[ i+1,j ∆x 2 + ∆y 2 ]
2

substituting in two spatial dimension Heat Conduction Equation and rearranging,it

can be expressed as

l+ 1 l+ 1 l+ 1
−λTi+1,j
l+1 l+1
+ 2(1 + λ)Ti,j − λTi−1,j
l+1
= λTi,j+1
2
+ 2(1 − λ)Ti,j 2 + λTi,j−1
2

∆t
where λ = k 2∆x 2

• The ADI method results in tridiagonal system if it is applied along the dimen-
sion that is implicit.thus on the first step, it is applied along the y dimension and on
the second step, it is applied along the x dimension.

13