Dipartimento di Ingegneria

Università degli Studi di Napoli “Parthenope”

Numerical models for

engineering
Lesson 12

Finite Difference for

1-D Initial Value Problems
A.A. 2014/2015 NUMERICAL MODELS FOR ENGINEERING PROF. N. MASSAROTTI 1
 Outline

Discretization of the heat conduction transient problems;

Methods for discretization of temporal derivative:

Explicit method;

Implicit method;

Crank-Nicholson.

Stability of numerical solution;

Matlab implementation;

Conclusions.

A.A. 2014/2015 NUMERICAL MODELS FOR ENGINEERING PROF. N. MASSAROTTI 2

 Transient problems
For one dimensional transient problems, the unknown is
T(x,q), a function of space and time
∂T ∂ 2T
= a
2
+ uɺ ′′′ (θ ) 0 ≤ x ≤ L; θ > 0
∂θ ∂x

Initial Cond. T (x , 0) = Tb 0≤x ≤L

T0 TL
Dirichlet BC T (0, θ) = T0 θ>0

Dirichlet BC T (L, θ ) = TL θ > 0

x
L

A.A. 2014/2015 NUMERICAL MODELS FOR ENGINEERING PROF. N. MASSAROTTI 3

 Transient problems

We already know how to discretize the Right Hand Side

(RHS) in space using finite differences:
1 ∂T ∂2T
= + uɺ ′′′ 0 ≤ x ≤ L; θ > 0
a ∂θ ∂x 2

n +1 n
∂2T Ti −1 − 2Ti + Ti +1
∂T T − T ≃
2
∂θ
≃
∆θ ∂x ∆x 2

It is easy to discretize the LHS in the same way,

considering the time interval θn+1-θn=∆θ
The problem now is how to calculate these terms (?)

A.A. 2014/2015 NUMERICAL MODELS FOR ENGINEERING PROF. N. MASSAROTTI 4

 Transient problems

Depending on the time instant in which the heat flux is

calculated:
T n +1 − T n
 i   Ti − 1 − 2Ti + Ti + 1 ??
i  
  = a  
 ∆θ   ∆x 2

It is clear that the LHS refers to the node for which the
equation is discretized in space
The RHS can be calculated at different instants of time:
θn+1 or θn (these will be referred to as instants “n” and
“n+1”);

A.A. 2014/2015 NUMERICAL MODELS FOR ENGINEERING PROF. N. MASSAROTTI 5

 Approximation of Physical Problem

From the physical point of view, this corresponds to:

T(θ) n +1
Tn+1 (Qɺ net ) i

∂T  T n +1 − T n 
=  
∂θ  ∆θ 
Tn
n
(Qɺ n et ) i
θn θn+1 θ
∆θ

The temporal approximation of the balance quantities;

T n +1 − T n 
( ρcV )i  i  ??
 = (Qɺ net ) i
i

 ∆θ 

A.A. 2014/2015 NUMERICAL MODELS FOR ENGINEERING PROF. N. MASSAROTTI 6

 Transient problems
The calculation of the RHS characterizes the numerical method:

if the RHS is calculated at time θn, the “explicit method” is obtained:

 T n +1 − T n 
( ρcV )i  i i  = (Qɺ ) n
 i
 ∆θ

if the RHS is calculated at time θn+1, the “implicit method” is obtained:

 T n +1 − T n 
( ρcV )i  i i  = (Qɺ ) n +1
 i
 ∆θ 
if the RHS is calculated as mean of values at θn and θn+1, the “Crank
Nicholson method” is obtained:
 T n +1 − T n  1 
( ρcV )i  i i  = ⋅  (Qɺ ) n + (Qɺ ) n +1 
 
 2 
∆θ i i


A.A. 2014/2015 NUMERICAL MODELS FOR ENGINEERING PROF. N. MASSAROTTI 7

 Explicit method
In the “explicit method”, without considering heat
generation:
 Ti n +1 − Ti n  ɺn ɺ n ɺ n
T0
( ρ cV )i   = QiNET = Qi −1,i − Qi ,i +1 TL
 ∆θ 
Discretizing the fluxes with central
i-1 i i+1 M
difference approximation, we obtain:
n n
ρ cA∆x n+1 n  kA   kA 
∆θ
( Ti − Ti ) =  (Ti −1 − Ti )  +  (Ti +1 − Ti ) 
 ∆x   ∆x 
∆x
where A refers to section crossed by the heat flux.
The RHS is the discrete formulation of steady state problem. We can write:
a∆θ Ti −n1 − 2Ti n + Ti +n1
(T i
n +1
− Ti ) =
n

∆x ∆x
i = 1,..., M − 1

A.A. 2014/2015 NUMERICAL MODELS FOR ENGINEERING PROF. N. MASSAROTTI 8

 Explicit method
For the explicit method ( Forward Euler)
T n +1 − T n
 i
 Ti −1 − 2Ti + Ti +1 n T n
− 2T n
+ T n
i  = a   = a i −1 i i +1
 
 ∆θ   ∆x 2  ∆x 2

the RHS has to be considered a known term, as the

unknown is the temperature at time n+1!
For each point, the algebraic equation to be solved is:

a ∆θ
Tin + 1 = Tin +
∆x 2
(Tin−1 − 2Tin + Tin+ 1 )

Which can be solved starting from the initial condition

(t=0 → values all known on the domain)!
A.A. 2014/2015 NUMERICAL MODELS FOR ENGINEERING PROF. N. MASSAROTTI 9
 Explicit method

We can write:
 2a∆θ  n a∆θ
Ti n+1 = 1 −
 ∆x 2
⋅
 i

T +
∆x 2
⋅ ( T n
i −1 + Ti +1 )
n
i = 1,..., M − 1

Defining a local Fourier number, Fo:

a∆θ
Foi =
∆x 2

Tin +1 = ( 1 − 2Foi ) ⋅ Tin + Foi ⋅ (Tin−1 + Tin+1 ) i = 1,..., M − 1

For an uniform spatial grid, the local Fourier number assumes a

single value in all domain.

A.A. 2014/2015 NUMERICAL MODELS FOR ENGINEERING PROF. N. MASSAROTTI 10

 Explicit method
Introducing a matrix form for M=5:
 T2n +1 − T2n 
 
∆ θ
1 0 0   n +1   2 −1 0  T2  T1n 
n

0 1 0   T3 − T3  = − a  −1 2 −1 T n  + a ⋅ ∆θ  
n

   ∆θ  ∆x 2    3  ∆x 2 0

0 0 1  n +1   0 −1 2  T4n  T n 
 T4 − T4n   5
 
 ∆θ 
⇓
 T n +1 − T n  n n
[M ]  = − [ K ]{T } + {b}
 ∆θ 

so we can write:
{T }
n +1
( −1
) n −1
= 1 − [ M ] ⋅ ∆θ [ K ] {T } + [ M ] ∆θ {b}
n

A.A. 2014/2015 NUMERICAL MODELS FOR ENGINEERING PROF. N. MASSAROTTI 11

 Explicit method: Stability
The stability of “explicit method” depends on the
growth factor (1-2Foi). So it requires small time steps to
keep the error in the result bounded, in fact:
Ti n
Ti n +1
= (1 − 2 ⋅ Foi ) ⋅ Ti + Foi ⋅ (T + T
n n
i −1
n
i +1 ) Ti−n1 Ti+n1

i-1 i i+1
After ∆θ
This condition corresponds to a limitation on the primary
n +1 n +1
coefficient on RHS; this coefficient should assume a positive Ti−1 Ti n +1 Ti+1
value to achieve a stable solution with explicit method:
1 ∆x 2 i-1 i i+1
1 − 2 ⋅ Foi ≥ 0 ⇔ Foi ≤ ⇔ ∆θ ≤
2 2a
Therefore, only ∆x is fixed.

A.A. 2014/2015 NUMERICAL MODELS FOR ENGINEERING PROF. N. MASSAROTTI 12

 Explicit method: Stability
The implementation of explicit method  T1n +1 = c1 ⋅ T1n + ...
requires to write a system of M equations, T2n +1 = ... + c2 ⋅ T2n + ...

where M is the number of nodes. The   ⋮
stability condition requires that all primary T n +1 = ... + c ⋅ T n + ...
 i i i
coefficient should be positive.  ⋮
 n +1 n
ci ≥ 0 i = 1,..., M  T
 M = ... + cM ⋅ TM

To respect stability condition for all equations it is necessary to define:

min ( ci ) = clim ite

i =1,..., M

Imposing that this limit value should be positive, stability condition

will be respected for all equations.

A.A. 2014/2015 NUMERICAL MODELS FOR ENGINEERING PROF. N. MASSAROTTI 13

 Implicit method
Implicit method is obtained by calculating the RHS at time n+1:

 n +1 − T n
 Ti   Ti − 1 − 2Ti + Ti + 1 n + 1 Tin−+1 1 − 2Tin + 1 + Tin++1 1
i  
  = a  
= a
 ∆θ   ∆x 2 ∆x 2

and again, as the unknown is the temperature at time n+1, this

value cannot be calculated explicitly!
For each point, the algebraic equation to be solved is:

T in + 1 − T in a
∆θ
=
2
(T i n−+1 1 − 2T i n + 1 + T i n++1 1 )
∆x

Which still needs the known initial condition (t=0)!

A.A. 2014/2015 NUMERICAL MODELS FOR ENGINEERING PROF. N. MASSAROTTI 14

 Implicit method
For a case study without generation terms, we can write the implicit
method as follows:

 T n +1 − T n  T0
( ρcV )i  i i
 = Qɺ i
n +1
= Qɺ in−+1,1i − Qɺ in,i++11
 ∆θ 
TL

Discretizing the fluxes with central 0 i-1 i i+1 M

difference approximation, we obtain:
ρcA∆x n +1  kA kA  n +1
∆θ
(Ti − Tin ) =  (T − Ti ) + (Ti +1 − Ti ) 
 ∆x i −1 ∆x  ∆x
Moving known terms from LHS to RHS, we obtain:

( 1 + 2 ⋅ Foi ) ⋅ Tin +1 − Foi ⋅ (Tin−+1 1 + Tin++1 1 ) = Tin i = 1,..., M − 1

A.A. 2014/2015 NUMERICAL MODELS FOR ENGINEERING PROF. N. MASSAROTTI 15

 Implicit method:
Introducing a matrix form:


T n + 1 − T n 
 n +1 n +1
 
 { } { }
M    = −  K  T + b
  ∆θ 

 

⇓
n +1 n n +1
(  M  + ∆θ  K  ){T } =  M  {T } + ∆θ {b }
     
posing:

A = M  + K 
     
we get:
n +1 −1 n −1 n +1
{T } =  A   M  {T } +  A  ∆θ {b }

A.A. 2014/2015 NUMERICAL MODELS FOR ENGINEERING PROF. N. MASSAROTTI 16

 Crank-Nicholson method
The basic idea of semi-implicit scheme is to calculate the RHS as
a mean of values at times n and n+1:

 T n +1 − T n 
( ρcV )i  i i
 = ξ ⋅ Qɺ i
n +1
+ ( 1 − ξ ) ⋅ Qɺ in with 0<ξ<1
 ∆θ 

previous equation can be rewritten as follows:

(Tin +1 − Tin ) = Fo  ξ ⋅ (Ti −1 − 2Ti + Ti +1 )n +1 + ( 1 − ξ ) ⋅ (Ti −1 − 2Ti + Ti +1 )n 



re-organizing the terms of the equation, we get
n +1
( 1 + 2 ⋅ ξ ⋅ Foi ) ⋅ Tin +1 − ξ ⋅ Foi ⋅ (Ti −1 + Ti +1 ) =
 1 − 2 ⋅ ( 1 − ξ ) ⋅ Foi  ⋅ Tin + ( 1 − ξ ) ⋅ Foi ⋅ (Ti −1 + Ti +1 )n i = 1,..., M − 1
 

for ξ=1/2 the classical Crank-Nicholson method is obtained, while for ξ=1 and
ξ=0, the implicit method and the explicit method are obtained, respectively.
A.A. 2014/2015 NUMERICAL MODELS FOR ENGINEERING PROF. N. MASSAROTTI 17
 Analysis of stability
By considering a general spatial discretization with M node and by
introducing the matrix [K]:
 1 − 2 ⋅ ( 1 − ξ ) ⋅ Fo ( 1 − ξ ) ⋅ Fo 
 
 ( 1 − ξ ) ⋅ Fo 1 − 2 ⋅ ( 1 − ξ ) ⋅ Fo ( 1 − ξ ) ⋅ Fo 
 
 ( 1 − ξ ) ⋅ Fo 1 − 2 ⋅ ( 1 − ξ ) ⋅ Fo ( 1 − ξ ) ⋅ Fo 
[K ] =  

 ( 1 − ξ ) ⋅ Fo 1 − 2 ⋅ ( 1 − ξ ) ⋅ Fo ( 1 − ξ ) ⋅ Fo 

 ( 1 − ξ ) ⋅ Fo 1 − 2 ⋅ ( 1 − ξ ) ⋅ Fo (1 − ξ ) ⋅ Fo 
 
 ( 1 − ξ ) ⋅ Fo 1 − 2 ⋅ ( 1 − ξ ) ⋅ Fo 


and the matrix M:

 1 + 2 ⋅ ξ ⋅ Fo −ξ ⋅ Fo 
 
 −ξ ⋅ Fo 1 + 2 ⋅ ξ ⋅ Fo −ξ ⋅ Fo 
 
 −ξ ⋅ Fo 1 + 2 ⋅ ξ ⋅ Fo −ξ ⋅ Fo 

[M ] =  
−ξ ⋅ Fo 1 + 2 ⋅ ξ ⋅ Fo −ξ ⋅ Fo 
 
 
 −ξ ⋅ Fo 1 + 2 ⋅ ξ ⋅ Fo −ξ ⋅ Fo 
 
 −ξ ⋅ Fo 1 + 2 ⋅ ξ ⋅ Fo 

we can write:
 M  {T }n +1 =  K  {T }n + {b }n
   

A.A. 2014/2015 NUMERICAL MODELS FOR ENGINEERING PROF. N. MASSAROTTI 18

 Analysis of stability
By introducing the global discretization error:
j j j
{e } = { t } − {T }
where tj is the exact solution at time j, we can write:
M 
  ( {t }
n +1
− {e }
n +1
) =  K  ⋅ ( {t } n n
− {e } ) + {b }
⇓
 M  {t }n +1 −  K  ⋅ {t }n + {b }n =  M  {e }n +1 −  K  ⋅ {e }n
       
=0
so we obtain:

{e }
n +1
(
=  K  ⋅  M 
−1
)
⋅ {e }
n

⇓
n
{e }
n +1
(
=  K  ⋅  M 
−1
) ⋅ {e }
0

A.A. 2014/2015 NUMERICAL MODELS FOR ENGINEERING PROF. N. MASSAROTTI 19

 Analysis of stability
By introducing the Euclidian norm:
n
e n +1 = A e0

We can demonstrate that stability conditions are the same of that ones
calculated for the previous case.

Implicit method: UNCONDITIONALLY STABLE;

Explicit method: STABLE for Fo≤0.5;

Crank-Nicholson method: STABLE for Fo≤1.

A.A. 2014/2015 NUMERICAL MODELS FOR ENGINEERING PROF. N. MASSAROTTI 20

 Analysis of stability
To analyze the stability limits of the three methods, it is possible to
consider a grid with two nodes; only one node is at an unknown
temperature.

(1 + 2 ⋅ ξ ⋅ Fo) ⋅ T1n+1 − 2 ⋅ ξ ⋅ Fo ⋅ T0 = 1 − 2 ⋅ ( 1 − ξ ) ⋅ Fo  ⋅ T1n + (1 − ξ ) ⋅ Fo ⋅ T0

by imposing the BC we get:
 1 − 2 ⋅ ( 1 − ξ ) ⋅ Fo  T0=0 qɺ = 0
T1n +1 =   ⋅Tn
1
( 1 + 2 ⋅ ξ ⋅ Fo )
=G
0 1
or: T1n +1 =G ⋅ T1n

G = 1 − 2 ⋅ Fo ← ξ = 0 Explicit method

1 − Fo ∆x
G = with: ξ = 1 2 Crank − Nicholson method
1 + Fo
1
G = → ⇒ ξ = 1 Implicit method
1 + 2 ⋅ Fo
A.A. 2014/2015 NUMERICAL MODELS FOR ENGINEERING PROF. N. MASSAROTTI 21
 Analysis of stability
By introducing the global discretization error:
ej = t j −T j
where tj is the exact solution at time j, we can write:
t1n +1 − e1n +1 = G ⋅ ( t1n − e1n ) ⇒ t1n +1 − G ⋅ t1n = e1n +1 − G ⋅ e1n
=0
so we obtain:
e1n +1 = G ⋅ e1n
⇓
e1n +1 = G n ⋅ e10

The stability is achieved for:

0 ≤G ≤ 1

A.A. 2014/2015 NUMERICAL MODELS FOR ENGINEERING PROF. N. MASSAROTTI 22

 Analysis of stability
The stability condition can be examined for the different
problems based on:

T1n +1 = λ ⋅ T1n

G E = 1 − 2 ⋅ F o
1 − F o
G S =
1 + F o
1
G I =
1 + 2 ⋅ F o

A.A. 2014/2015 NUMERICAL MODELS FOR ENGINEERING PROF. N. MASSAROTTI 23

 Matlab implementation
Case Study:

∂T ∂ 2T
= a 0 ≤ x ≤ L; θ > 0
∂θ 2
∂x

Initial Cond. T (x , 0) = Tb = 0 0≤x ≤L

T0 TL
Dirichlet BC T(0, θ) = T0 = 20 θ>0

Dirichlet BC T (L, θ) = TL = 0 θ>0

x
L

A.A. 2014/2015 NUMERICAL MODELS FOR ENGINEERING PROF. N. MASSAROTTI 24

 Matlab implementation
Implicit method
Parameters declaration
and vectors and matrix
inizialization

A.A. 2014/2015 NUMERICAL MODELS FOR ENGINEERING PROF. N. MASSAROTTI 25

 Matlab implementation
Implicit method

Algorithm implementation

n +1 n
{T } − {T } n +1 n +1
M
 
 =  K  SFD ⋅ {T } + {b }
∆θ

A.A. 2014/2015 NUMERICAL MODELS FOR ENGINEERING PROF. N. MASSAROTTI 26

 Matlab implementation
Implicit method

Solution plot

A.A. 2014/2015 NUMERICAL MODELS FOR ENGINEERING PROF. N. MASSAROTTI 27

 Matlab implementation
Explicit method

Algorithm implementation

A.A. 2014/2015 NUMERICAL MODELS FOR ENGINEERING PROF. N. MASSAROTTI 28

 Conclusions

Several engineering problems are described by

differential equations;
Problems described by the same type of equations are
analogous and can be solved using the same methods;
It is fundamental to understand the physical meaning of
single terms appearing in the equations;
Analytical solutions for one dimensional problems
boundary value problems is very often available;
For simple problems, the solution should always be
found by hand (you don’t need big computers)!

A.A. 2014/2015 NUMERICAL MODELS FOR ENGINEERING PROF. N. MASSAROTTI 29

 Exercises

Exercise 1: Write a matlab code for the solution of the

problem:

∂T ∂ 2T
= a 0 ≤ x ≤ L; θ > 0
∂θ ∂x 2

Initial Cond. T (x , 0) = Tb 0≤x ≤L T0 qɺ = 0

Dirichlet BC T (0, θ) = T0 θ>0

Neumann BC ∂T (L, θ)
=0 θ>0
∂x L=0.7 m

Use for thermal properties the same values of previous example

A.A. 2014/2015 NUMERICAL MODELS FOR ENGINEERING PROF. N. MASSAROTTI 30
 Matlab implementation
Exercise 2: Write a matlab code for the solution of the
problem:

∂T ∂ 2T
= a 0 ≤ x ≤ L; θ > 0
∂θ 2
∂x

T0 TL
Initial Cond. T (x , 0) = Tb = 0 0≤x ≤L

Dirichlet BC T(0, θ) = T0 = 20 θ>0

x
L
Dirichlet BC T (L, θ) = TL ( θ ) θ>0

Assume a boundary condition function of time.

A.A. 2014/2015 NUMERICAL MODELS FOR ENGINEERING PROF. N. MASSAROTTI 31