COMMUNICATIONS IN NUMERICAL METHODS IN ENGINEERING
Commun. Numer. Meth. Engng, 14, 809±819 (1998)
AN ANALYSIS OF FACTORS THAT GOVERN
THE MINIMUM TIME STEP SIZE TO BE
USED IN THE FINITE ELEMENT ANALYSIS
OF DIFFUSION PROBLEMS
HYWEL RHYS THOMAS* AND ZHENGMING ZHOU
Division of Civil Engineering, School of Engineering, University of Wales, Cardi, PO Box 925, Cardi CF2 1YF, UK
SUMMARY
The solution of the equation system of the ®nite element method for diusion problems is presented in the
format of a sequence. The characteristics of the numerical results can thus be known through an analysis of
this `theoretical expression' of the solution. The characteristics of the sequence are analysed to yield
expressions for the minimum time step size. # 1998 John Wiley & Sons, Ltd.
KEY WORDS ®nite element analysis; diusion problems; minimum time step size
1. INTRODUCTION
Minimum time step sizes for one-dimensional two-noded elements have been derived by a
number of researchers.1±5 Some of these methods are dicult to extend to obtain minimum time
step sizes for other types of elements, especially multidimensional elements. Thomas and Zhou6
have now proposed a rigorous approach to obtain the minimum time step size. Limitations of
space, however, prevent the full inclusion of the mathematical formulation employed. To over-
come this de®ciency and provide a more complete presentation of the method proposed for use
by other research workers, a detailed description of the mathematics will be addressed in this
paper. In particular, a theoretical expression of the ®nite element solution of a one-dimensional
diusion problem is presented in terms of a sequence. The product formula of the sequence is
obtained so that the theoretical expression of the solution is explicit. The analysis of the sequence
leads to the establishment of the minimum time step size.
2. SOLUTION OF FINITE ELEMENT EQUATION OF ONE-DIMENSIONAL
DIFFUSION PROBLEMS
A one-dimensional classical diusion problem can be expressed via a heat conduction equation,
as follows:
@T @2 T
c k 2 1
@t @x
* Correspondence to: H. R. Thomas, School of Engineering, University of Wales, Cardi, PO Box 925, Cardi
CF2 1YF, UK
CCC 1069±8299/98/090809±11$17.50 Received 22 May 1997
# 1998 John Wiley & Sons, Ltd. Accepted 25 February 1998
810 H. R. THOMAS AND Z. ZHOU
where c is the speci®c heat capacity, k is the coecient of thermal conductivity, T is the temp-
erature, t is the time and x is the co-ordinate.
The corresponding ®nite element equation may be expressed as
K T C T_ f 2
where
Z
K kHNi HNj dx dy 3
O
Z
C cNi N j dx dy 4
O
and
Z
f Ni p dG 5
G
where O and G are the analysed domain and the domain boundary, respectively, p is the ¯ux on
the boundary and Ni and Nj are the shape functions.
Employing a backward dierence time-stepping algorithm for the temporal discretization,
equation (2) may be written as
p
p Cp m1 C m m1
K T T f 6
Dt Dt
where Dt is the time step size, the superscript m is the time level and p is the iteration number.
Consider the one-dimensional problem divided into (n±1) elements as shown in Figure 1. Each
element has the same size, material properties and shape functions. A ®xed value boundary
condition is applied at node 1. The ¯ux and initial value is assumed to be zero without losing
generality.
The assembled equation system may be written in the following form:
AT 1 BT 2 AT 3 0 7 1
AT 2 BT 3 AT 4 0 7 2
AT 3 BT 4 AT 5 0 7 3
AT nÿ2 BT nÿ1 AT n 0 7 n ÿ 2
B
AT nÿ1 T 0 7 n ÿ 1
2 n
where
Lc k
A ÿ 8 1
6Dt L
# 1998 John Wiley & Sons, Ltd. Commun. Numer. Meth. Engng, 14, 809±819 (1998)
MINIMUM TIME STEP SIZE 811
Figure 1. One-dimensional element arrangement
and
Lc K
B2 8 2
3Dt L
Now T2 may be eliminated from equations (7(1)) and (7(2)) to yield
AT 1 X 3 T 3 Y 3 T 4 0 9
where
B2
X3 ÿ A 10
A
and
Y 3 ÿB 11
Similarly, T3 may be eliminated from equations (7(3)) and (9) to yield
AT 1 X 4 T 4 Y 4 T 5 0 12
where
B
X4 ÿ X Y3 13
A 3
and
Y 4 ÿX 3 14
# 1998 John Wiley & Sons, Ltd. Commun. Numer. Meth. Engng, 14, 809±819 (1998)
812 H. R. THOMAS AND Z. ZHOU
It can be seen that Ti may be eliminated to yield
AT 1 X i1 T i1 Y i1 T i2 0 15
where
X i DX iÿ1 ÿ X iÿ2 16
Y i ÿX iÿ1 17
where
B
Dÿ 18
A
and i is a positive integer greater than 2 and smaller then n 7 1.
This elimination procedure can be carried on up to equation (7(n 7 2)) to yield
AT 1 X nÿ1 T nÿ1 Y nÿ1 T n 0 19
and equation (7(n 7 1)) remains as
B
AT nÿ1 T 0 20
2 n
Rewriting the set of equations after elimination yields:
AT 1 BT 2 AT 3 0 21 1
AT 1 X 3 T 3 Y 3 T 4 0 21 2
AT 1 X 4 T 4 Y 4 T 5 0 21 3
AT 1 X i1 T i1 Y i1 T i1 0 21 i
AT 1 X nÿ1 T nÿ1 Y nÿ1 T n 0 21 n ÿ 2
B
AT nÿ1 T 0 21 n ÿ 1
2 n
Solving equations (21(n 7 1)) and (21(n 7 2)) yields
ÿA
Tn T1 22
B
Xn X nÿ1
2A
# 1998 John Wiley & Sons, Ltd. Commun. Numer. Meth. Engng, 14, 809±819 (1998)
MINIMUM TIME STEP SIZE 813
Setting
ÿA
Zn 23
B
Xn X
2A nÿ1
yields
T n Zn T 1 24
The temperature at the (n 7 1)th node can be obtained by back substitution of solution (24) into
equation (21(n 7 2)). It is expressed as
ÿA ÿY nÿ1
T nÿ1 Z T1
X nÿ1 X nÿ1 n
ÿA X nÿ2 25
T1
X nÿ1 X nÿ1 Zn
Znÿ1 T 1
where
ÿA X
Z nÿ1 nÿ2 Z 26
X nÿ1 X nÿ1 n
This back substitution can be repeated until the temperatures at all the nodes are obtained. It is
found that the solution of the equation can be expressed by
T iÿ1 Z iÿ1 T 1 27
where the term Zi71 is given by the expression
ÿA X
Ziÿ1 iÿ2 Z 28
X iÿ1 X iÿ1 i
i.e. a recurrence formula.
3. PRODUCT FORMULA OF THE SEQUENCE X
The ®rst two terms for the sequence X are obtained from equation (21(1)):
X 1 ÿA
29
X2 B
It may be assumed that the product formula of the sequence X is
i
X i ca 30
where c and a are constants.
# 1998 John Wiley & Sons, Ltd. Commun. Numer. Meth. Engng, 14, 809±819 (1998)
814 H. R. THOMAS AND Z. ZHOU
Therefore
iÿ1
X iÿ1 ca 31
and
X iÿ2 caiÿ2 32
Equations (30), (31) and (32) may be substituted into recurrence formula (16), yielding
i iÿ1 iÿ2
ca Dca ÿ ca 33
Equation (33) may be reduced to
2
a Da ÿ 1 34
The solution of equation (34) is
p
D+ D2 ÿ 4
a 35
2
Equation (35) shows that the recurrence formula (16) has two dierent solutions. For con-
venience, these two solutions are renamed as:
p
D D2 ÿ 4
a 36
2
p
D ÿ D2 ÿ 4
b 37
2
The general solution of the recurrence formula (16) is therefore
i i
X i c1 a c2 b 38
The ®rst and second terms are known according to equation (29) and can be used to determine
the coecients in the general solution (38). Substituting equation (38) into equation (29) yields
ÿA c1 a c2 b 39
2 2
B c1 a c2 b 40
Equation (39) is multiplied by b to give
ÿAb c1 ab c2 b2 41
Equation (40) is subtracted from equation (41) to get
2
B Ab c1 a ÿ c1 ab 42
# 1998 John Wiley & Sons, Ltd. Commun. Numer. Meth. Engng, 14, 809±819 (1998)
MINIMUM TIME STEP SIZE 815
Making use of equations (18), (36) and (37), the left-hand side of equation (42) becomes
B Ab ÿAa 43
and the right-hand side of equation (42) becomes
c1 a2 ÿ c1 ab c1 a a ÿ b 44
Hence
ÿA
c1 45
aÿb
Similarly it can be shown that
A
c2 46
aÿb
The product formula for the sequence X is
i
a ÿ bi
X i ÿA 47
aÿb
4. SOME CHARACTERISTICS OF THE SOLUTION
According to physics, a heating boundary should obviously not result in cooling inside the body.
Furthermore the temperature gradient within the body should yield the highest temperatures
nearest the heating source and lower temperatures away from the source. In this Section, it is
attempted to prove if
ÿA 5 0 48
the solution of equation set (7) would not violate the laws of physics.
In equation (47), absolute a is greater than absolute b, and a is always greater than zero,
therefore
Xi 5 0 49
To prove that the general solution (27) is greater than zero, ®rst, solution (22) of node n needs to
be proved to be greater than zero. Noticing that the numerator in equation (22) is greater than
zero according to (48), only the denominator needs to be con®rmed to be greater than zero.
By utilizing (47), the denominator can be calculated as
nÿ1 B nÿ1 B
a Aa ÿ b Ab ÿ
B 2 2
Xn X nÿ1 50
2A aÿb
# 1998 John Wiley & Sons, Ltd. Commun. Numer. Meth. Engng, 14, 809±819 (1998)
816 H. R. THOMAS AND Z. ZHOU
Making use of equations (18), (26) and (37), it can be calculated that
B
Aa ÿ aÿb 51
2
and
B
Ab ÿ ÿ a ÿ b 52
2
Thereby
B nÿ1 nÿ1
Xn X a ÿb
2A nÿ1
50
So
Zn 5 0 54
and
Tn 5 0 55
Every term of Zn71 is not less than zero according to (48), (49) and (54); hence
Z nÿ1 5 0 56
Therefore from recurrence formula (28) it is known that
Zi 5 0 57
It is then known from equation (27) that
Ti 5 0 58
This indicates that the heating boundary does not cause any cooling inside. The following
paragraphs will prove that if condition (48) is satis®ed
T iÿ1 5 T i 59
Calculating the derivative of equation (47),
0 aiÿ1 ÿ biÿ1
X i ÿAi
aÿb
iX iÿ1
Inequality (49) can be substituted into (60) to give
0
Xi 4 0 61
# 1998 John Wiley & Sons, Ltd. Commun. Numer. Meth. Engng, 14, 809±819 (1998)
MINIMUM TIME STEP SIZE 817
Therefore X is a monotonously increasing sequence, i.e.
X i 4 X iÿ1 62
It is known that
B 4 ÿ 2A 63
The dierence of two consecutive terms of the sequence X can be obtained using the recurrence
formula (16):
X i ÿ X iÿ1 D ÿ 1X iÿ1 ÿ X iÿ2
64
BA
ÿ X iÿ1 ÿ X iÿ2
A
Substituting inequality (63) into equation (64) yields
X i ÿ X iÿ1 4 X iÿ1 ÿ X iÿ2 65
Equation (23) can be transformed, using equation (16), into
ÿA
Zn 66
B
ÿ X ÿ X nÿ2
2A nÿ1
Inequality (63) may be substituted into equation (66), yielding
ÿA
Zn 4 67
X nÿ1 ÿ X nÿ2
The dierence of the last two terms of sequence Z is
ÿA X
Z nÿ1 ÿ Zn nÿ2 Z ÿ Z n
X nÿ1 X nÿ1 n
ÿA X nÿ2
ÿ 1 Zn 68
X nÿ1 X nÿ1
Inequality (67) may be substituted into equation (68) to yield
ÿA X nÿ2 ÿA
Z nÿ1 ÿ Z n 5 ÿ1
X nÿ1 X nÿ1 X nÿ1 ÿ X nÿ2
0
hence
Z nÿ1 5 Z n 70
# 1998 John Wiley & Sons, Ltd. Commun. Numer. Meth. Engng, 14, 809±819 (1998)
818 H. R. THOMAS AND Z. ZHOU
Utilizing inequalities (67) and (65) yields
ÿA X
Z nÿ1 nÿ2 Z
X nÿ1 X nÿ1 n
ÿA
5 71
X nÿ1 ÿ X nÿ2
ÿA
5
X nÿ2 ÿ X nÿ3
The recurrence formula (28) is used to calculate the dierence
ÿA X
Znÿ2 ÿ Z nÿ1 nÿ3 Z nÿ1 ÿ Z nÿ1
X nÿ2 X nÿ2
72
ÿA X nÿ3
ÿ 1 Z nÿ1
X nÿ2 X nÿ2
Substituting inequality (71) into equation (72) yields
Z nÿ2 ÿ Z nÿ1 4 0 73
Repeating the above process it can be shown that
Z iÿ1 ÿ Z i 4 0 74
and therefore
Z iÿ1 5 Z i 75
Substituting inequality (75) into equation (27) yields
T iÿ1 5 T i 76
This indicates the correct temperature gradient within the heated body.
From equations (22), (27) and (47), it is known that A has to be a non-positive number
otherwise the solution of equation set (7) will violate the laws of physics. Hence inequality (48) is
a sucient and necessary condition for a physically reasonable solution.
Substituting equation (8(1)) into inequality (48) results in
L2 c
Dt 5 77
6k
The right-hand side of the inequality is the minimum value of the time step size.
# 1998 John Wiley & Sons, Ltd. Commun. Numer. Meth. Engng, 14, 809±819 (1998)
MINIMUM TIME STEP SIZE 819
5. CONCLUSION
The solution of the ®nite element method of a one-dimensional diusion problem has been
derived theoretically. The solution was expressed in terms of a sequence. The product formula of
the sequence has been obtained mathematically. Some characteristics of the sequence were
analysed, leading to the establishment of a minimum time step size.
REFERENCES
1. P. A. Vermeer and A. Verruijt, `An accuracy condition for consolidation by ®nite elements', Int. j. numer.
anal. methods geomech., 5, 1±14 (1981).
2. E. Rank, C. Katz and H. Werner, `On the importance of the discrete maximum principle in transient
analysis using ®nite element methods', Int. j. numer. methods eng., 19, 1771±1782 (1983).
3. V. Murti, S. Valliappan and N. Khalili-Naghadeh, `Time step constraints in ®nite element analysis of the
Poisson type equation', Comput. Struct., 31, 269±273 (1989).
4. L. Simoni and B. A. Schre¯er, `A staggered F. E. solution for water and gas ¯ow in deforming porous
media', Commun. appl. numer. methods, 7, 213±223 (1991).
5. E. Turska and B. A. Schre¯er, `On convergence conditions of partitioned solution procedures for
consolidation problems', Comput. Methods Appl. Mech. Eng., 106, 51±63 (1993).
6. H. R. Thomas and Z. Zhou, `Minimum time step size for diusion problem in FEM analysis', Int. j.
numer. methods eng., 40, 3865±3880 (1997).
# 1998 John Wiley & Sons, Ltd. Commun. Numer. Meth. Engng, 14, 809±819 (1998)