Sie sind auf Seite 1von 32

NUMERICAL SOLUTION OF ORDINARY DIFFERENTIAL EQUATIONS

INITIAL VALUE PROBLEMS

Prof. A.A. Adesina, ChSE, UNSW

GENERAL CONSIDERATIONS
The modelling of unsteady-state processes often lead to ODEs with time as the independent variable. In other cases, the independent variable may be a spatial co-ordinate as in the steady-state treatment of tubular or columnal systems such as plug flow and trickle-bed reactors, packed adsorption and absorption towers or even tubular heat exchangers. In general, for an nth order ODE, n conditions are needed to completely solve the problem. If ALL these n specifications are provided at the same value of the independent variable, say, t or x, for all dependent variables, then the ODE is said to be an INITIAL VALUE PROBLEM. While this may imply that the value of t or x is supposed to be the starting point in the system (e.g. t=0 or x=0), there is no mathematical incongruity if the value of t or x is the endpoint or some other physically admissible point in the independent variable range.
Prof. A.A. Adesina, ChSE, UNSW 2

METHODS FOR 1ST ORDER NONLINEAR ODE - Euler methods


The general 1st order nonlinear ODE may be written; dy = f (x, y) dx w ith th e in itia l c o n d itio n th a t

(7 .1 )

at x = x0

y = y0

OR

y ( x0 ) = y0

The Euler method, one of the earliest techniques for solving ODE stipulates that, if we can represent the LHS of Eqn (7.1) by its first forward finite difference, then at any position, i, we have,

yi = yi+1 yi h2 D2 yi h3D3 yi yi = hDyi + + + ........ 2 6

(7.2)

but from our previous discussion on finite difference operators, we can write, (7.3)
3

Prof. A.A. Adesina, ChSE, UNSW

Euler methods
Euler assumed a truncation of the series in Eqn (7.3) after the 1st term, thus,

y i = hD y i + O ( h 2 ) y i +1 = y i + hD y i + O ( h 2 )

(7.4) (7 .5 )

so that upon substitution into Eqn (7.2), w e receive,


In fact, replacing the term, Dyi with f from Eqn (7.1), we get the so-called explicit (forward) Euler method, written,

yi +1 = yi + hf ( xi , yi ) + O ( h 2 )

(7.6)

It is evident that this method is only marginally accurate since the error is of order h2.
Prof. A.A. Adesina, ChSE, UNSW 4

Euler methods
If we use the backward finite difference approximation, we can write,

yi = yi +1 yi = yi +1

(7.7)

so that, in terms of backward differences, we have, yi +1 = yi + yi +1 (7.8)


But from earlier notes, we know that

h 2 D 2 yi +1 h 3 D 3 yi +1 yi +1 = hDyi +1 + ............. 2 6 so that combining with Eqns (7.1) & (7.2) we secure, yi +1 = yi + hf ( xi +1 , yi +1 ) + O ( h 2 )
Prof. A.A. Adesina, ChSE, UNSW

(7.9)

(7.10)
5

Euler methods
Eqn (7.10) is commonly called the implicit Euler method because it involves the calculation of the function, f, at the unknown value of yi+1. We observe that the error in Eqn (7.10) is also of the order of h2. In principle, implicit equations cannot be solved individually but must be set up as sets simultaneous algebraic equations. If these are linear, then we can use any of the techniques we have encountered before (say, generalised inverse matrix approach) to deliver a solution. However, if they are nonlinear, we may use the Jacobian method for solving the set of nonlinear algebraic equations. Regardless, the computation may be time-consuming depending on how many integration steps we need to do. Euler simplified this problem by using Eqn (7.6) to predict yi+1 and then employing this predicted value to get a corrected estimate from Eqn (7.10).
Prof. A.A. Adesina, ChSE, UNSW 6

Euler methods
Eqns (7.6) and (7.10) constitute the Euler predictor-corrector pair and is more accurate because it has an error of order h3 (which can be readily shown by adding Eqns (7.3) & (7.9)). Thus, the Euler Predictor-Corrector method is:

( yi+1)Pr = yi +hf (xi , yi ) +O(h2)


h 2 y = y + f ( x , y ) + f ( x , y ) + O ( h ) ( i+1)Cor i [ i i i+1 i+1 ] 2

(7.11a) (7.11b)

As a result, the Euler Predictor-Corrector method is preferred over the explicit or implicit method for the purpose of computation.

Prof. A.A. Adesina, ChSE, UNSW

Runge-Kutta Methods
The Runge-Kutta methods are arguably the most widely used integration techniques for solving ODEs with initial conditions (IVP-ODEs). Although the formal derivation for each method is similar to the Euler method in the sense that we take advantage of particular approximations of the Taylor series expansion of the differential operator (cf. Lecture notes on Finite Difference operators as recalled in Eqns (7.3) & (7.9)), we will NOT provide formal derivations here but simply summarise the necessary equations to be used for each RK method as displayed in the Tables below.

Prof. A.A. Adesina, ChSE, UNSW

Table 1: Summary of Runge-Kutta integration formulas (2nd-4th order) Order Formula for yi+1 of the RK Method Second 1 y i +1 = y i + ( k1 + k 2 ) 2 Third
yi +1 1 = y i + ( k1 + 4 k 2 + k 3 ) 6

Formula for RK constants


k 1 = hf ( x i , y i ) k 2 = hf ( x i + h , y i + k 1 ) k 1 = hf ( x i , y i )

Order of the error invloved O(h3) O(h4)


k1 )

Fourth

yi +1

1 = y i + ( k1 + 2 k 2 + 2 k 3 + k 4 ) 6

h k k 2 = hf x i + , y i + 1 2 2 k 3 = hf ( x i + h , y i + 2 k 2 k 1 = hf ( x i , y i )

O(h5)

h k k 2 = hf x i + , y i + 1 2 2 h k k 3 = hf x i + , y i + 2 2 2 k 4 = hf ( x i + h , y i + k 3 )

Prof. A.A. Adesina, ChSE, UNSW

Table 2: Runge-Kutta-Gill integration formulas Order Formula for yi+1 Formula for RK constants of the RK Method Runge k1 = hf ( xi , yi ) 1 1 1 = + + + + + y y k k k k 2 1 2 1 i +1 i 1 2 3 4 Kutta6 2 2 k = hf x + h , y + k1 Gill i 2 2 i 2
h x + i 2, k3 = hf k k y + 2 1 1 + 2 2 2 i 2 2 xi + h, k4 = hf k k 3 2 y + 2+ 2 i 2 2

Order of the error invloved O(h5)

The Runge-Kutta-Gill method is the most widely used 4th order method and the constants are selected to reduce the amount of storage required in the solution of a large number of simultaneous 1st order ODEs. We will discuss application of RK methods to set of ODEs later.
Prof. A.A. Adesina, ChSE, UNSW 10

Table 3: Summary of Runge-Kutta integration formulas (5th order)

Order Formula for yi+1 Formula for RK constants of the RK Method Fifth k = hf ( xi , yi ) 1 yi+1 = yi + (7k1 + 32k3 + 12k4 + 32k5 + 7k6 ) 1 90 h k k2 = hf xi + , yi + 1 2 2
h 3k k k3 = hf xi + , yi + 1 + 2 2 16 16 k h k4 = hf xi + , yi + 3 2 2 3h 3k 6k 9k k5 = hf xi + , yi 2 + 3 + 4 4 16 16 16 k 4k 6k 12k 8k k6 = hf xi + h, yi + 1 + 2 + 3 4 + 5 7 7 7 7 7

Order of the error invloved O(h6)

Prof. A.A. Adesina, ChSE, UNSW

11

Comments on Runge-Kutta method selection


In general, the 4th order RK method is more commonly used because it has a good accuracy for most engineering situations. However, for stiff ODEs, it may be necessary to reduce the step-size, h, between successive integration steps. What is the criterion for implementing a step-size reduction to avoid error propagation through the integration procedure? Collatz (1960) has recommended that if,

k3 k2 0 .1 k 2 k1

(7 .1 2 )

After each integration step, then the step-size, h, should be decreased (to probably half its current value).
Prof. A.A. Adesina, ChSE, UNSW 12

Illustrative example
Let us find the solution to the IVP

dy = y + 2x 1 (E1.1) dx with y(0)=1.0 over the interval 0 x 1 using 4th order RK method
In this case, clearly,

f ( x, y) = y + 2 x 1
We divide the interval into 10 steps so that x0=0.0, x1=0.1..x10=1.0 with h=0.1. Thus, y(x00=1.0. Using the iterative formulas given on Table 1, we get,
Prof. A.A. Adesina, ChSE, UNSW 13

Results of the 4th order Runge-Kutta methods for the Illustrative Example xi
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

R-K
1.0 1.01034164 1.04280472 1.09971619 1.18364811 1.29744053 1.44423485 1.62750244 1.85107803 2.11920166 2.43655777

R-K-G
1.0 1.01034164 1.04280472 1.09971619 1.18364811 1.29744148 1.44423580 1.62750340 1.85107994 2.11920357 2.43656063

Exact value of yi
1.0 1.01034260 1.04280663 1.09971809 1.18365002 1.29744339 1.44423676 1.62750530 1.85108185 2.11920643 2.43656349

Prof. A.A. Adesina, ChSE, UNSW

14

Predictor-Corrector Methods
The Runge-Kutta methods are single-step methods and are good for solving IVP because they self-starting. However, they may be plagued with instabilities for highly stiff ODEs which require small step-size to overcome numerical instabilities. Predictor-Corrector methods are preferred for such situations because they deliver more stable results. They are, however, nonself-starting and are regarded as multi-step techniques. The most common PC methods are summarised below.

Prof. A.A. Adesina, ChSE, UNSW

15

Table 4: Summary of commonly used Predictor-Corrector Methods


Type of Predictor Equation PC method Euler ( yi+1)Pr = yi + hfi Milnes 4th order AdamsMoulton method Milnes 6th order
4h ( yi+1)Pr = yi3 + 3 ( 2 fi fi1 + 2 fi2 ) h y = y + ( i+1)Pr i 24( 55 fi 59 fi1 + 37 fi2 9 fi3 )

Corrector Equation

Error involved
O(h2)

h = + y y ( i+1)Cor i 2( fi+1 + fi )

O(h5) h ( yi+1)Cor = yi1 + 3( fi+1 + 4 fi + fi1) O(h5) h ( yi+1)Cor = yi + 24( 9 fi+1 +19 fi 5 fi1 + fi2 )

3h11 fi 14 fi1 + 26 fi2 y y = + ( i+1)Pr i5 10 14 f +11f i3 i4

2h 7 fi+1 + 32 fi +12 fi1 y y = + ( i+1)Cor i 45 +32 f + 7 f i2 i3

O(h7)

Note that: fi = f(xi, yi); fi+1 = f(xi+1, yi+1) etc, in the Table above
Prof. A.A. Adesina, ChSE, UNSW 16

Predictor-Corrector Methods
How many times do we iterate on the Corrector equation? Typically, we employ a termination criterion to determine when to stop. For example, we can use:

yi+1,m yi+1,m1 yi+1,m1

m = 1,2,....

(7.12)

where m indicates the number of iterations performed so far on yi+1 using the Corrector Equation and is the tolerance factor, a small positive number, typically, 10-4 or smaller.

Prof. A.A. Adesina, ChSE, UNSW

17

Worked Example
Use the Milnes 4th order PC method to solve

dy = 0.1( x + y )0.8 (E1) dx with y(1.0)=3.0 over the range x=1.0 to 2.0
Solution

f ( x , y ) = 0.1( x + y )0.8
Assume h=0.1, then let us calculate the 1st three points (y1,y2,y3) by the single-step method of 4th order R-K technique since the PC is NOT self-starting. Recall that we need, yi-3 to use the Milnes P-C method.
Prof. A.A. Adesina, ChSE, UNSW 18

Starting with the 4th order RK method and using y(1) =3.0; we have, x0=1.0 and y0=3.0

1st integration step


k 1 = h f ( x 0 , y 0 ) = 0.1[0.1(1 + 3 ) 0.8 ] = 0.0303143 k 2 = h f ( x 0 + h / 2, y 0 + k 1 / 2 ) = 0.1[0.1(1.05 + 3.01516 ) 0.8 ] = 0.0307087 k 3 = h f ( x 0 + h / 2, y 0 + k 2 / 2 ) = 0.0307099 k 4 = h f ( x 0 + h , y 0 + k 3 ) = 0.0311043 1 k + 2k2 + 2k3 + k4 = 0.0307093 6 1 y 1 = y 0 + y = 3.0 307093 y = an d x 1 = 1 + 0.1 = 1 . 1

2nd integration step


k 1 = h f ( x 1 , y 1 ) = 0.1[0.1(4.13071) 0.8 ] = 0.0311042 k 2 = h f ( x 1 + h / 2, y 1 + k 1 / 2 ) = 0.0314985 k 3 = h f ( x 1 + h / 2, y 1 + k 2 / 2 ) = 0.0314997 k 4 = h f ( x 1 + h , y 1 + k 3 ) = 0.0311043 1 k + 2k2 + 2k3 + k4 = 0.0314991 6 1 y 2 = y 1 + y = 3.03 149 9 1 y = an d x 2 = 1.1 + 0. 1 = 1 . 2

Similar repetitive steps give, Prof. A.A. Adesina, ChSE, UNSW


x 3 = 1.3, y 3 = 3.09449

19

At the 4th integration step, we now turn to the 4th order Milnes PC. Hence,
P r : y4 = y0 + f 1 = 0.311042 f 2 = 0.318939 f 3 = 0.326834 y 4 = 3.0 + 0.4 (0.622084 + 0.653668 0.318939 ) = 3.127 58 3 h C or : y 4 = y 3 + ( f 4 + 4 f 3 + f 2 ) 3 f 4 = f ( x 4 , y 4 ) = 0.1(1.4 + 3.12758 ) 0.8 = 0.334728 4h ( 2 f 3 f 2 + 2 f1 ) 3

y 4 ,0 = 3.09449 +

0.1 (1.961 00 ) = 3.1598 6 3 1st iteration step of th e C orrector E q u ation f 4 = f ( x 4 , y 4 ,0 ) = 0.1(1.4 + 3.15986 ) 0.8 = 0.336636 h ( f + 4 f3 + f2 ) 3 4 0. 1 (1.96291) = 3.15992 = 3.09449 + 3 C h eck for con verge n c e y 4 ,1 = y 3 + y 4 ,1 y 4 ,0 y 4 ,0 L et u s assu m e th at = 1 0 -4 3.15992 3.15986 = 1.8988 10 5 10 4 3.15986

h en ce con vegren ce w as m et after th e 1 st iteration so w e can n ow p roceed to th e 5 th in tegration step .

Prof. A.A. Adesina, ChSE, UNSW

20

Solution of higher order ODEs (Solution to Simultaneous 1st order ODEs)


In principle, since it is possible to convert an nth order ODE to a set of n 1st order ODEs. The approach to solving an nth order ODE is therefore exactly the same as solving n 1st order ODEs. We would 1st show how this conversion may be carried out and then proceed with a discussion of the simlutaneous solution of n 1st order ODEs.
dny d n 1 y d n 2 y dy dy d n1 y = ( , ) + ( , ) + .... ( , ) + ( , ) = ( , , ,... ) a x y a x y a x y a x y g x y n1 n 2 1 0 dx n dx n 1 dx n 2 dx dx dx n 1 L et us defi ne new variables as: z1 = y z2 = (7.1 4 ) (7.15) (7.16) dz 1 dy = = f 1 ( x , z 1 , z 2 , ..... z n ) dx dx dz d2y = f 2 ( x , z 1 , z 2 , ..... z n ) z3 = 2 = dx dx 2 M (7.13)

dz n 2 d n 2 y = = f n 2 ( x , z 1 , z 2 , ..... z n ) z n 1 = (7.17) dx dx n 2 dz d n1 y (7.18) z n = n 1 = = f n 1 ( x , z 1 , z 2 , ..... z n ) dx dx n 1 Introducing these equations into the original equation, w e receive dz n d n y dy d n 1 y ( , , ,... g x y = = )= f n ( x , z 1 , z 2 , ..... z n ) dx dx n dx dx n 1 (7.19)

Prof. A.A. Adesina, ChSE, UNSW

21

Solution of higher order ODEs (Solution to Simultaneous 1st order ODEs)


Eqns (7.15) to (7.19) constitute the new n 1st order ODEs we now need to solve. The applicable 4th order Runge-Kutta constants are:

1 zi +1, j = zij + k1 j + 2k2 j + 2k3 j + k4 j 6 k1 j = hf j ( xi , zi1 , zi 2 ,....zin )

j = 1,2,...n j = 1,2,...n j = 1,2,...n j = 1,2,...n j = 1,2, ...n

(7.20) (7.21) (7.22) (7.23) (7.24)


22

k4 j = hf j ( xi + h, zi1 + k31 , zi 2 + k32 ,.....zin + k3n )

k1n h k11 k12 k2 j = hf j xi + , zi1 + , zi 2 + ,....zin + 2 2 2 2 k2n h k21 k22 k3 j = hf j xi + , zi1 + , zi 2 + ,....zin + 2 2 2 2
Prof. A.A. Adesina, ChSE, UNSW

Comments on Runge-Kutta method selection


A system ODE is said to be stiff if the so-called stiffness ratio, SR, exceeds a particular value as indicated below, where SR is defined,

m ax Re( i ) SR = m in R e( i )

(7.25)

Where is are the eigenvalues of the ODE system and max |Re(i)| is the real part of the maximum eigenvalue of the system. In general, if a). SR is about an order of magnitude (10,20, etc), the system is NOT stiff. b). SR is about three orders of magnitude (1000, 2000, etc) the system is STIFF c). SR is about six orders of magnitude (106, etc), the system is VERY STIFF. The eigenvalues are obtained from the matrix of the coefficients of the set of ODEs.
Prof. A.A. Adesina, ChSE, UNSW 23

Comments on Runge-Kutta method selection


In principle, a stiff ODE is one whose general solution contains an exponential term, e.g. ex for some constant . When is large and negative, the ODE is especially troublesome because it permits the solution to decay quickly to zero. For instance, the solution to:
dy = f ( x, y ) (7.26) dx with y( x0 ) = y0 , the eigenvalue, , may be obtained as

f y xi

(7.27)

where can vary in magnitude at each step of the integration. f is NEGATIVE at some point in the range of integration. Eqn (7.26) will be stiff if y Thus, the solution will be unstable.
Prof. A.A. Adesina, ChSE, UNSW 24

Real stability regions for selected solution methods


Method Explicit Euler Implicit Euler Euler Predictor-Corrector 2nd order Runge-Kutta 3rd order Runge-Kutta 4th order Runge-Kutta 5th order Runge-Kutta Adams-Moulton Stability boundary

-2 h 0 0 < h < , for < 0; -2 h 0, for > 0 -1.077 h 0 -2 h 0 -2.5 h 0 -2.785 h 0 -5.7 h 0 -1.285 h 0

Prof. A.A. Adesina, ChSE, UNSW

25

For a set of n simultaneous ODEs, we can obtain the eigenvalues from the Jacobian matrix of the functions, fi written;
f1 f1 f1 ....... y y2 yn 1 f 2 f 2 f 2 ........ y y yn 2 1 J = . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . f n f f n ........ n y1 y2 yn
26

Prof. A.A. Adesina, ChSE, UNSW

Comments on Runge-Kutta method selection


It is apparent from the foregoing considerations that implicit methods are recommended for handling the solution to stiff ODEs. This is why Predictor-Corrector methods are somewhat popular among chemical engineers since we often deal with nonlinearity due to the Arrhenius or vant Hoff terms arising from modeling of nonisothermal tubular reactors/columns or dynamics of reactive lumped systems with nonlinear kinetics.

Prof. A.A. Adesina, ChSE, UNSW

27

Example problem
Solve

d 2y = x 2 dx

+ xy + y

for y at x = 2.0 if y = 4.0 at x = 1.0 and

dy = 0.5 at x =1.0 dx

Prof. A.A. Adesina, ChSE, UNSW

28

Solution Let

dy = f1 ( x , y , z ) dx d2y dz = = x 2 + xy + y 2 = f2 ( x, y, z ) 2 dx dx z=
Assume h = 0.5. From the initial conditions x 0 = 1 .0 , y 0 = 4 .0 , z 0 = 0 .5 1st R-K constants k 1 1 = h f 1 ( x 0 , y 0 , z 0 ) = 0 . 5 x 0 . 5 = 0 .2 5
k 1 1 = h f 2 ( x 0 , y 0 , z 0 ) = 0 . 5 (1 + 4 + 1 6 ) = 1 0 .5 0

2nd R-K constants h k k k 21 = h f1 ( x 0 + , y 0 + 11 , z 0 + 12 ) 2 2 2 h 0 .5 x0 + =1+ = 1 .2 5 2 2 k 0 .2 5 y 0 + 11 = 4 + = 4 .1 2 5 2 2 k 1 0 .5 = 5 .7 5 z 0 + 1 2 = 0 .5 + 2 2 k 2 1 = 0 . 5 x 5 .7 5 = 2 .8 7 5


k 22 = h f 2 ( x 0 + = 1 1 .8 6 7 2 h k k , y 0 + 11 , z 0 + 12 ) 2 2 2 2 = 0 . 5 [ (1 .2 5 ) + 4 . 1 2 5 x 1 . 2 5 + ( 4 .1 2 5 ) 2 ]

Prof. A.A. Adesina, ChSE, UNSW

29

3rd R-K constants h k 21 k 22 , y 0 + , z0 + ) k 31 = h f1 ( x 0 + 2 2 2 0 .5 h = 1+ = 1 .2 5 x0 + 2 2 2 .8 7 5 k 21 = 4 + = 5 .4 3 7 5 y 0 + 2 2 1 1 .8 6 7 2 k 22 = 0 .5 + = 6 .4 3 3 6 z0 + 2 2 k 31 = 0 .5 6 .4 3 3 6 = 3 .2 1 6 8


k
32

h k 21 k 22 , y 0 + , z0 + ) 2 2 2 = 0 . 5 [ (1 . 2 5 ) 2 + 5 . 4 3 7 5 x 1 . 2 5 + ( 5 . 4 3 7 5 ) 2 ] = 1 8 .9 6 2 9 = hf2 ( x
0

4th R-K constants k 41 = h f1 ( x 0 + h , y 0 + k


x y z k
0 0 0

31

, z

+ k

32

+ h = 1 .5 + k + k
41 31 32

= 7 .2 1 6 8 = 1 9 .4 6 2 9
0

k
42

= 0 .5 x 1 9 .4 6 2 9 = 9 .7 3 1 4 5 + h, y
0 2

= hf2 (x

+ k

31

, z

+ k

32

= 0 .5 [ ( 7 .2 1 6 8 ) = 3 2 .5 7 8 7 y =

+ 1 . 5 x 7 . 2 1 6 8 + (1 . 5 ) 2 ]

1 [ k 11 + 2 k 21 + 2 k 31 + k 41 ] = 3 .6 9 4 1 8 6 1 z = [ k 12 + 2 k 22 + 2 k 32 + k 42 ] = 1 7 .4 5 6 5 6 y (1 ) = y 0 + y = 7 . 6 9 4 1 8 z
(1 ) (1 )

= z

0 0

+ z = 1 7 .9 5 6 5

x = x + h = 1 .5 We will now use x(1), y(1), and z(1) as the starting point for the computation of x ( 2 ) , y ( 2 ) , a n d z ( 2 ) .

Prof. A.A. Adesina, ChSE, UNSW

30

2nd integration step


1st R-K constants k11 = hf1 ( x (1) , y (1) , z (1) ) = 0.5 x 17.9565 = 8.97824
k11 = hf 2 ( x (1) , y (1) , z (1) ) = 0.5[(7.69418) 2 + (7.69418 x 1.5) + (1.5) 2 ] = 36.4958

2nd R-K constants k k h k 21 = hf1 ( x (1) + , y (1) + 11 , z (1) + 12 ) 2 2 2 h x (1) + = 1.75 2 k y (1) + 11 = 12.1833 2 k z (1) + 12 = 36.2044 2 k 21 = 0.5 x 36.2044 = 18.1022

k 22 = hf 2 ( x (1) +

h (1) k11 (1) k12 ,y + ,z + ) 2 2 2 = 0.5[(1.75) 2 + 1.75 x 12.1833 + (12.1833) 2 ] = 92.4997


Prof. A.A. Adesina, ChSE, UNSW 31

3rd R-K constants h k k k 3 1 = h f 1 ( x (1 ) + , y (1 ) + 2 1 , z (1 ) + 2 2 ) 2 2 2 h = 1 .7 5 x (1 ) + 2 k y (1 ) + 2 1 = 1 6 . 7 4 5 3 2 k z (1 ) + 2 2 = 6 4 . 2 0 6 3 2 k 31 = 0 .5 x 6 4 .2 0 6 3 = 3 2 .1 0 3 1 5

k 3 2 = h f 2 ( x (1 ) + = 1 5 6 .3 8 6

h k k , y (1 ) + 2 1 , z (1 ) + 2 2 ) 2 2 2 2 = 0 . 5 [ (1 . 7 5 ) + 1 . 7 5 x 1 6 . 7 4 5 3 + (1 6 . 7 4 5 3 ) 2 ]

4th R-K constants k 4 1 = h f 1 ( x (1 ) + h , y

(1 )

+ k 3 1 , z (1 ) + k 3 2 )

x (1 ) + h = 2 . 0 y
(1 )

+ k 31 = 3 9 .7 9 7 3

z (1 ) + k 3 2 = 1 7 4 . 3 4 2 k 41 = 0 .5 x 1 7 4 .3 4 2 = 8 7 .1 7 1 k 4 2 = h f 2 ( x (1 ) + h , y
(1 )

+ k 3 1 , z (1 ) + k 3 2 )

= 0 .5 [( 2 ) 2 + 2 x 3 9 .7 9 7 3 + (3 9 .7 9 7 3 ) 2 ] = 1 6 6 7 .4 2 y = 1 [ k 11 + 2 k 21 + 2 k 31 + k 41 ] = 3 2 .7 6 6 1 [ k12 + 2 k 22 + 2 k 32 + k 42 ] = 3 6 6 .9 4 8 z = 6 y ( 2 ) = y (1 ) + y = 4 0 . 4 5 4 2 z ( 2 ) = z (1 ) + z = 3 8 4 . 9 0 4 x ( 2 ) = x (1 ) + h = 2 . 0
at x = 2.0, y = 40.4542

Prof. A.A. Adesina, ChSE, UNSW

32

Das könnte Ihnen auch gefallen