UNIT 9 MULTISTEP AND PREDICTOR -

- CORRECTOR METHODS FOR

SOLVING INITIAL VALUE
PROBLEMS
Structure Page No.
9.1 Introduction ' 29
Objectives
9.2 Multistep Methods 29
Interpolation Approach
Method of Undetermined Parameters
9.3 Predictor-Corrector Methods
9.4 Stability of Multistep Methods
9.5 Summary
9.6 Solutions/Answers
Practical Assignments

9.1 INTRODUCTION
- -

In Unit 8 you have studied singlestep (Taylor series and Runge-Kutta) methods for
solving the initial value problem (IVP)
y l = f(x,y), y(xo) =yo, (1)
These methods require only the immediate previous value yi at the nodal point x i to
calculate y,,, at the nodal point x,,, . The order of the highest derivative used in
Taylor series methods and the number of function evaluations per step in the Runge-
Kutta method increases with the order of the method. Thus very high order singlestep
methods are not used for solving the initial value problems.
There is another class of methods for solving IVP's which require the k solution
values yi-,,,at previous k nodal points xi-,, , m = O,l,. .. ,(k - 1) to obtain the value of
yi+, at the nodal point xi+,. Such methods are called multistep or k-step methods.
These methods are classified as explicit or implicit according to whether they do not
use or use yi+, in addition to the previously calculated k values. These methods
required only one extra function evaluation per step irrespective of the order of the
method. In this unit we shall discuss both explicit and implicit multistep methods.
In Sec.9.2 we have used two approaches to discuss the derivation and implementation
of various explicit and implicit multistep methods. The explicit and implicit methods
are combined to define a new class of methods called the predictor-corrector
methods which we shall discuss in Sec.9.3. Finally, in Sec.9.4 we have discussed the
stability of the multistep methods.
Objectives
After reading this unit you should be able to
explain the multistep methods;
explain the predictor-corrector methods;
derive the explicit and implicit multistep methods;
obtain the solution of initial value problems using multistep and predictor-
corrector methods;
obtain the interval of absolute stability of multistep methods.

9.2 MULTISTEP METHODS

A numerical method for solving the NP(1) viz.,
y1= f(x,y), y(x0) = Y o
Numerical Solutions is called a multistep or a k-step method if it uses the values of y(x) and f(x, y) at the
of ODES
k nodal points xi-, , m = 0, I,. .. ,(k - 1) to determine the solution value yl+, at the
nodal point x,+,. Such a method is called an explicit or a predictor method. If the
method also uses the values of y(x) and f(x, y) at the nodal point xl+,, then the
method is called an implicit or a corrector method. The value yi is the initial value
and the values y,-, ,y,-, ,...,y,-,+, are called the starting values. The starting values are
obtained using some singlestep method before using the multistep method. Generally,
Taylor series methods and Runge-Kutta methods are used for starting a multistep
method. Thus multistep methods are not self-starting methods. A predictor method is
used to predict a value of y,+, and a corrector method is used to improve upon this
value. In this section we shall use two approaches for the derivation of multistep
methods.
9.2.1 Interpolation Approach
Integrating y' = f (x, y) in the interval [xi, xi+,1, we obtain

Xi

To derive multistep methods, we replace f(x, y) in the integral in Eqn.(2) by some

suitable interpolating polynomial. We shall now discuss some of these methods.
Adams-Bashforth methods
These are explicit methods. Through the k data values
(x, , f i), (xi-,, fi-]), .. .% ( x , - ~,+fi-k+l),
, we fit the Newton's backward difference
interpolating polynomial of degree k - 1as
(x-xi) (x-xi) (x-xi-,)
Pk-,(x) = f i + - Vfi + v 2 f i +...
h 2! h 2

+ ( X - X ~ - ] ) . . . ( X - X ~ -v~k-'+ ~f i).
(k - l)! h k-l
The error of interpolation is given by
TE = ( x - x i ) ( X - X ~ - , ) . . . ( X - X ~ f- 'k'~ +(5)
~ ),
(4)
k!
where 6 lies in some interval containing the points x i , xi-,,... ,x i-k+I and x.
Replacing f (x, y) by Pk-,(x) in Eqn.(2), we get

Set x - x i = h s . Then x - x , = x - x i + x i - x , = ( s + i - m ) h . Hence,weget

yi+,= yi + 1
0
[f, + sVf + -1 s (s + l ) v 2 f i +
2

Error term can be obtained by integrating Eqn.(4) in the interval [xi, xi+,]or, directly
by using Taylor series. For different values of k, we obtain different methods.
k = 1: we get a singlestep method y,,, = yi + h f i (6)
which is the Euler method. The error term is given by
T E = ~ ( x i + l ) - ~ i=+Y! ( x ~ + I ) - [ +Yh~f i I
= y ( x i + l ) - ~ i-'Y;
 Multistep and Predictor-
where f, = y { . Corrector Methods for
Expanding each term in Taylor series about the point x i , we obtain Solving 1 3 ~ s

This method is of order 1.

The error term is given by

TE = Y(xi+l)- Y i + l
=y(xi +h)-[yi +h(3y: -y:-,)]
S
=-h3ym(5), < 5 < xi+[
12
This method is of order 2.

The error term is given by

This method is of order 3.

In general a k-step method of the form (5) gives a method of order k. We require one
function evaluation per step for the application of the multistep method.
Example 1: Find the approximate value of y(1 .O) using the multistep method
(Adams-Bashforth second order method)

with h = 0.2 for the initial value problem

1
y' = -2x y', y(0) = 1
!
Calculate the starting value using the Runge-Kutta second order method
1
Yi+[ = Y I +-(Kl + K , )
2
Kl = h f ( x , , Y , )
K, = h f ( x i + h , yi + K l )
with the same step size h.
1 Solution : For the given multistep method

we get for i = 0
Numerical Solutions Now yo is the given initial value and y, is the starting value. We calculate the
of ODES
starting value using the given second order Runge-Kutta method. We obtain from
f(x y)=-2x y 2 , x o = 0 , yo =1, h = 0 . 2
K , = h f ( x 0 , y o ) = h f(0, 1 ) = 0
K, = h f ( x , + h , yo + K , ) = h f(0.2, I)=-0.08

After the starting value has beendewmined, we use the multistep method (9). We
obtain for
i=0: x, =O,yo =1, fo =-2xoyo2 =O;
x , =0.2,y, =0.96, f, =-2x, y: =-0.36864

h
y5 y(1.0) = y, + -(3 f, - f,) = 0.482963.
2
***
Example 2: Solve the initial value problem
y l = x + y 2 , y(O)=l
in the interval [0,1] with step length h = 0.2 using the multistep method

(Adams-Bashforth 'third order method).

Compute the starting values using third order Taylor series method with the same step
size h.
Solution: From the given multistep method

we obtain for i = 0

Now yo is the given initial value, whereas y, and y, are the starting values. We
compute the starting values using the third order Taylor series method

We have
Y ' = X + ~ ~y a, = 1 + 2 y y ' , ym=2yy"+2(y')2
We obtain from Eqn.(l 1) for h = 0.2 and
 Multistep and Predlctor-
Corrector Methods for
Solving lVPs

= 1.270667.
j=l: x , =0.2,y, =1.270667,~',=1.814595,
y; = 5.61 1492, yy= 20.846185 and

= 1.773611.
After determining the starting values, we use the multistep method (10). We have
h =0.2 and
2
xo = 0, yo = 1, fo = xo + yo = 1,

x, = 0 . 4 , =
~ 1.773611,f2
~ = x, + y: = 3.545696
We obtain from Eqn.( 10) for

Adams-Moulton methods
These are implicit methods. Through the k + 1data values
fi+,),(xi,fi),.. , fi-k+l),
we fit the Newton's backward difference
interpolating polynomial of degree k as

Note that we are using the current data point (xi+,, fi+l)also.
I
The error of interpolation is given by
Numerical Solutions
of ODES

where 6 lies in some interval containing the points xi+,, x i , . ..,xi-,+, and x. Replacing
f (x, y) by P, (x) in Eqn.(2), we get

Set x - x i = h s . Then x-x, = ( s + i - m ) h . Hence, weobtain

The error term can be obtained by integrating (13) in the interval [ x i ,xi+,] or, directly
by using the Taylor series. For different values of k, we obtain different methods.
k = 0 : We get a singlestep method. Retaining one term inside the brackets in
Eqn.(l4), we get the method
Yi+l=Yi +hfi+~ (15)
which is the backward Euler method. The error term 1s given by
TE =y1(xi+l)=yi+1= Y ( x +h)-[yi
~ +h~i+,]
where fi+,= yl+,. Expanding each term in Taylor series about xi , we get
h
TE=--yR(&), xi <C<xi+]
2
This method is of order 1.
k = 1 :We again get a singlestep method. Retaining two terms inside the brackets in
Eqn.(l4), we get the method

which is called the trapezoidal method.

The error term is given by
TE = (xi+!) - Y i + ~

This method is of order 2.

k = 2 :Retaining three terms inside the brackets in Eqn.(l4), we get the method

h
+-[5fi+, + Sf, - fi-,]
= yi
12
The error term is obtained as
TE = y(xi+~
) - Yi+l
This method is of order 3.
In general, the k-step method of the form (14) is of order (k + 1)
Since the methods are implicit, we obtain a non-linear equation in y for non-linear
,+,

initial value problem. The solution can be obtained by using Newton-Raphson method
or, any other iterative method. Generally we take the initial approximation yi+l= yi .
For a linear initial value problem, we can obtain yi+, directly without iteration.
Milne-Simpson methods
These are implicit methods. If we integrate y' = f(x, y) in the interval xi+,],we
get
Xl+l

J
y ( x i + , ) = ~ ( x i - I ) + f(x7 y)dx (18)
Xi-I

Now we replace f(x, y) in Eqn.(l8) by the interpolating polynomial Pk(x) given by

Eqn.(l2) based on the data values (xi+l,fi+l), ( x i , fl ), ... ,fi-k+l) . Using the
substitution x - x i = h s , we obtain as in the case of Adams-Moulton methods

By choosing different values of k, we get different methods.

k = 0 : Retaining one term inside the brackets in Eqn.(l9), we get the method
Y i+l = Y i-I + 2h f i + ~ (20)
The error term is given by
TE = Y(X,+I ) - Y ,+I
= xi + h ) - [ Yi-1 + 2h ~ l + l I
=-2h2y"(5), xi-l<5<~i+,
This method is of order 1.
k = 1 : Retaining two terms inside the brackets in Eqn.(l9), we get the method
I
Y i + l = Y i-1 + h[ ' f i + l - 'Vfi+I
= Y i-1 + h[ 2 f , + l - 2 - fi ) I
= yi-, + 2h f i
This is an explicit method. The error term is given by
TE = ~ ( x i + - l )Y i + l
= y(xi + h) - [ yi-l + 2h yi]

This method is of order 2.

-
k = 2 : Retaining three terms inside the brackets in Eqn.(l9), we get the method
7

The error term is given by

TE = ~ ( x i + l-) Yi+l
 h'
= 90
- - Xi-,<~<Xi+,
Thus the method is of order 4. The method (22) is called the Mllne-Slmpson method
of order 4.
For-k = 3, we obtain the same method as (22). Hence, for k = 2 and k = 3 , we obtain
the same method.
Example 3: Find the approximate value of y(0.5) for the initial value problem
y' = x + y, y(0) = 1,
using the multistep method

(Milne-Simpson fourth order method) with h = 0.1

Compute the starting value using classical Runge-Kutta fourth order method with the
same step .---L
iengcn Ln.
Solution: The given multistep method is an implicit method. Since the given initial
value problem is linear we do not require any iteration. We can compute yi+, directly.
From the given multistep method

h
yi+, = yi + -(fi+, + 4fi+1+ f i )
3
we obtain for i = 0
h
Y2 = Y, +-(f, + 4f, + f*)
3
Now yois the given initial value, wherezls y, is the starting value. We compute y,
using classical fourth order Runge-Kutta method.
Wehave f ( x , y ) = x + y , x o = O , y o = l a n d h = O . l . Weobtain
K,= h f(x,, y o ) = h f(O,1)=0.1

After determining the starting value, we use the multistep method (23). From
Eqn.(23), we obtain

Using xi+, = xi + 2h, xi+,= x i + h ,we get

 Usingx, = 0, yo = 1, x , = 0.1, y, = 1.1 10342, h = 0.1, we obtain from Eqn.(24) for Multistep and Predietor-
Corrector Methods for
Solvlng lVPs

Example 4: Find an approximate value of y(l.2) for the initial value problem
y l = x 2 + y 2 , y(l)= 2
using the Adam-Moulton third order method

with h = 0.1. Calculate the starting value using third order Taylor series method with
h = 0.1: ,

Solution: The Adam-Moulton third order method can be written as

h
yi+, = yi+,+-[ sfi+, + 8fi+,- fi], i =0,1,... (25)
12
For i=O,weget

Now, yo is the initial value and y, is the starting value. We calculate y, using Taylor
series third order method

r
with h = 0.1. We have
yo = 2, y; = x,2 + yo2 = 5 y; = 2xo + 2y, y; = 22,
y: = 2 + 2y0 y; + 2 ( ~ b =) ~140
i Therefore, we obtain from Eqn.(26)

=2.633333 .
After determining the starting value, we use the multistep method (25). From
Eqn.(25), we obtain for i = 0.

Using yo = 2, y, = 2.633333, x, = 1, x , = 1.1, x, = 1+ 2h = 1.2 and simplifying, we

Be*
Numerical Solutions
of ODES
,
y =- 0.04 1667 y: + 3.194629 ,which is a nonlinear equation in y,
We use Newton-Raphson method to determine y, .
Let F(y,) =0.041667~: - y 2 + 3.194629
We.get, F1(y,) = 0 . 0 8 3 3 3 4 ~-~1
The Newton-Raphson method is gven by

Starting with y?) = y, = 2.633333, we get

since ~ ( ~ 7= 0.000000
) ) 1, we take y(1.2) = y2 = 3.794588

Note that we have to use iteration at every step.

And now some exercises for you.

E l ) Find the truncation error and the order of the method

3h
~ i + =3 ~ +-[
i
8
Y;+~I
YI + ~ Y I + I + ~ Y ; + Z+
for solving the initial value problem y' = f(x, y), y(x,) = yo.
E2) Find an approximate value of y(0.6) for the initial value problem
y l = x - y 2 , y(O)=l
using the multistep method
h
yi+l=myi + - (3f, -
2
with the step length h = 0.2. Compute the starting value using Taylor series
second order method with the same steplength h.
E3) Find an approximate value of y(1.5) for the initial value problem
y l = x2 + y 2 , y(l)=O
using the mult'istep method

with h = 0.1 . Calculate the starting values using third order Taylor series
method with the same steplength h.
E4) Find an approximate value of y(2) for the initial value problem
Y 1 = y + y 2 ,y(l)=l
using the multistep method
h
Yi+1 = yi-, +- [ 7 f i - 2fi-, + f i 4 ]
3
with h = 0.2. Compute the starting values using the Heun's method
with the same steplength h.
E5) 'Solve the initial value problem
y' = 2x + 3y, y(1) = 2
 on the interval [l, 1.41using the multistep method Multistep and Predict
Corrector Methods for
h Solving IVPs
, (y:+l + 4 Y: + Y:-I
Y,+l = Y,-1 +

with h = 0.1 . Calculate the starting value using classical fourth order
Runge-Kutta method, with the same steplength h.
- - - - - - -- - - - - - - - - - - - --

Let us now discuss the second approach for the derivation of multistep methods.
9.2.2 Method of Undetermined Parameters
The general multistep or k-step method for the solution of the initial value problem (1)
can be written as

or, Y ~ +=
I C a r n ~ i - ~ i++hCbmy:m+I
I
m=l m=O I
Changing i to i + k - 1 in Eqn.(24), we can also write the method as

In method (27), yi is the initial value and yi-, ,...,yi-,+, are the starting values.
If b, = 0, the method (27) defines an explicit method and if bo # 0, the method (27)
defines an implicit method.
In method (27) a',,s and b',s are unknowns which are to be determined such that the
method is of a particular order. We write the error term as
TE=y(xi+,>-Yi+l

Expanding each term in (29) in Taylor series about the point x i and collecting the
terms of various powers of h, we can write
T E = C O y ( x i ) + C l h y ' ( x i ) + . . . +C , ~ ~ ~ ( ~ ' ( X ~ )
P+l (,+I)
(30)
+ C,+Ih Y (xi) +;.
where C;,s are independent of h and depend on a',s and b',s . The multistep method
(27) is said to be of order p, if
C, = C, = ... = C, = 0 and C,+, # 0
The truncation error in this case is given by

Solving the system of equations obtained from

co=c,=...=cp
=o
we determine the constants a',,s and b',,s and hence the method. The first non-zero
term in Eqn.(30) gives the error term and the order of the method.
A method is said to be consistent ifp 2 1.
Note that we have (2k + 1) arbitrary unknowns in the method (27) and we need
2k + 1equations to determine these unknowns. Hence, a method of the form (27) can
be of maximum order 2k. However, the stability conditions restrict the order of the
method to k for an explicit method. For an implicit method, the order is restricted to
k+l, when k is odd; and k+2, when k is even. We shall discuss the stability of these
methods in Sec.9.4.
Adarns-Bashforth method
The form of a few important class of methods is
70
Numerical Solutions
of oms
m=l
This is an explicit method and is of order k.
Nystrom method

This is an explicit method and is of order k.

Adnms-Moulton method

This is an implicit method and is of order k+l.

MUne-Simpson method

m=O
This is an implicit method and is of order k + 1,when k is odd and of order k + 2 ,
when k is even.
We illustrate the derivation of multistep methods by taking some examples.
Example 5: Obtain the constants a , , b, and b2in'the explicit multistep method
i h i b i ~ +I ~ , Y I - , ]
Yi+l = a i ~ +
Determine the truncation error and the order of the method.
Solution: We write
TE = Y(xi+l),- Yi+l
=Y(xi+I)-{ a,Y(xi) + h{biyr(xi)+ b , ~ ~ ( x)I] i-~
=y(xi + h ) - a l y ( x i ) - h [ b1yf(xi)+b2yV(xi
-h)]
Expanding each term in Taylor series about the point x i and collecting the coeff~cients
of various power of h ,we get
TE=(~ - a I ) y ( x i ) + (1-(b, + b 2 ) ) h yl(xi)

We have three unknowns and we need three equations to determine these unknowns.
Setting the constant term and the coefficients of h, h 2 in Eqn.(3 1) to zero, we obtain
the system of equations
1-a, =0, or, a l = l
1-(b, +b,)=O, or, b, + b , = I
1 1
- + b , =0, or, b, = --
2 2
3 I
whose solution is a , = I , b, =-, b, =-- . Thus we obtain the method as
2 2

and fiom Eqn.(31) , we'obtain

Therefore, the method is of order 2.

 Multistep and Predictor-
The method (32) is the Adams-Bashforth method of order 2. Corrector Methods for
*** Solving IVPs

Example 6: Derive the constants in the explicit method

Yi+l = a,yi + h[b,Yl + bzyl-1 + b,Y:-,l
Determine alsothe truncation error and the order of the method.
Solution: We write
TE=y(xi+l)-~i+l
=Y(xi+l)-[a,~(x,)+-h{b,y'(xi) + b , ~ ' ( x i - ~ ) + b ~ y ' ( x i ~ ) ) ]
=y(xi + h ) - a , y ( x i ) - h [ biy'(xi)+b2y1(xi-h)+b,yl(xi -2h)]
Expanding each term in Taylor series about the point x i and collecting the
coefficients of various powers of h, we get
~ ~ = ( l - a , ) ~ ( x ~1-(b, ) + {+ b, + b 3 ) ) h y ' ( x i )

We have four unknowns and we need four equations to determine these unknowns.
Equating the constant terms and the coefficients of h, h Z and h3 in Eqn.(33) to zero,
we obtain the system of equations.
1-a, = O or, a , =1
1-(b,+b,+b,)=O or, b , + b , + b 3 = 1
1 1
-2 + (b, + 2b,) = 0 or, b, + 2b3 = --
2
1 1 1
I
- - -(b, + 4b3)= O or, b, + 4b3 =-
6 2 3
Solving this system of equations, we obtain

and the method is given by

which is Adams-Bashforth third order method. From Eqn.(33), we get

Example 7: Derive the method

= a , y i +azy1-2 +h(b, yl+, + b , Y: +b,yl-,)
Yi+l

Determine the truncation error and order of the method.

Solution: We write
*E = Y(xi+I) - Yi+l
=Y(xi+\)-[aI y(xi)+a, y(xi-Z)+h(bo yt(xi+l)+b1~'(xi)+b2y'(xi-1)))
= y(xi + h) -[ a,' y(x,) + a , y(xi - 2h) + h(b, y'(x, + h)
+ b l y l ( x i ) +b, yt(xi - h)]
Expanding each term in Taylor series about the point xi and collecting the
coefficients of various power of h ,we get
Numerical Solutions
of ODES

+ ...
(35)
We have five unknowns, and we need five equations to determine these unknowns.
Setting the constant term and the coefficients of h, h 2 , h3 and h4in Eqn.(35) to zero,
we obtain the system of equations

Solving these equations, we obtain

and the method is given by

1 h
yIt1=-(9yi - Y ~ - ~ ) + - [ ~ Y+ I~, Y~ I-3~;-11
8 8
We obtain from (35)

Hence, the method (36) is implicit and is of order 4.

***
And now some exercise\s for you.
v
E6) Derive the method
Yi+l =a,Y,-l +h[ b,yl +b,yl-, +b,yf-,I
for solving the initial value problem y' = f(x, y), y(xo) = yo. Find the
truncation error and the order of the method.
E7) Derive the method
Yi+l =alyi +a,yi-l + h [ b l ~ +b2yl-1
l +b,yl-2]
for solving the initial value problem y' = f(x, y), y(x0) = y o . Find the
truncation error and the order of the method.
E8) Derive the method
Yi+l = a ~ y +
i h[ boy:+, + b l ~ +b,yl-II
l
for solving the initial value problem y' = f(x, y), y(xo) = y o . Find the
truncation error and order of the method.
E9) Derive the method
Y I t l = a,yi +a,y:-, +hbOy:+l
for solving the initial value problem y' = f(x, y), y(x,) = yo. Find the
truncation error and the order of the method.
 So far we have developed a number of explicit and implicit multistep methods to solve Multistep and Predictor-
Corrector Methods for
an initial value problem y' = f(x, y), y(xo) = y o . Based on the stability properties Solving IVPs
of the method which we shall be discussing- in Sec.9.4, it is seen that all the exvlicit
methods converge only if the step size h is sufficiently small. Hence, if the solution of
the initial value problem is required in a large interval, then we may require a large
number of integration steps, which besides increasing the round-off error, is not
economical. On the other hand, most implicit methods have strong stability
properties, that is one can choose sufficiently large value of h for integration.
However, we need to solve a non-linear equation by iteration at each step, which is
also expensive. Hence, we combine the explicit and implicit methods to define a new
class of methods called the predictor-corrector methods which we shall discuss
now.

9.3 PREDICTOR-CORRECTOR METHODS

The explicit methods are called the predictors which predict or give an initial
I
approximation to y(x,+,). This initial approximation is denoted by yi,P,).The implicit
methods are called the correctors which use yi,P,) to obtain an improved value of Y , + ~ .
This corrected value is denoted by y!::. The corrector method is used till convergence
5 is obtained. We define predictor-corrector methods (P-C sets) as follows:
P : Predict an approximate value of y(x i+l ) . This approximate value is denoted
by y{,P,' or y:!/ .
C : Correct this value using a corrector method to obtain yip{.
We take y(xi+,) = y!p:.
When the predictor and corrector methods are of the same order, we need only one or
two corrector iterations per step. However, if predictor is of lower order, we may
require more number of corrector iterations for convergence. The predictor-corrector
methods (P-C methods) are used as sets. Some examples of P-C sets are
1. P: Y::,) = y i + h f ( x i , Y,) (Euler method)
(S+l)
C: ~ i + l =~i+hf(xi+l,~;s),) (backward Euler method)
s=o,1, ...
The order of both the methods is one.
2. P: yI:/ = Yi + hf(xi, Y,) (Euler method)
t C: Y!s+l)
I+I = yi + [f (x i, y, ) + f (x y!:\ )] (Trapezoidal method)
s =0,1, ...
I
The predictor is of first order and the corrector is of second order.

s =0,1, ...
Both the methods are of fourth order. This P-C set is also called Milne's P-C set.
The starting values are obtained using some singlestep explicit method like Taylor
series or Runge-Kutta method. The order of the singlestep method may be same or
less than that of the corrector method.
Example 8: Obtain the approximate value of y(0.3) for the initial value problem

using the method:

Numerical Solutions P = predictor:
of ODES
yl,P; = yi + h fi
h
C = corrector: y:: = y, + -[fi + f(x,+,,y,+,)] (37)
2
with the step length h = 0.1. Perform two corrector iterations per step.
Solution: Since both the predictor and corrector methods are singlestep methods, we
do not need any starting value.. We write the methods (37) as
P: yj:I = y i + h f ( x i , y i ) = y i + h [ x : + # ]

since f(x,y) = x 2 + y2.

Using the given predictor method for h = 0.1, we ge;
i =0: x0 =0, yo = 1,
P : yio' = y , + h ( x i + y i ) = l . l

Therefore, after two corrector iterations, we get

y, = y(O.1) = 1.1 12216
i=1: x, =0.1, y, =1.112216
P : y): = y , +h(x; +y:)=1.236918

Therefore, after two corrector iterations, we obtain

y 2 = y(0.2) = 1.255076
i=2: x, = 0.2, y, = 1.255076

Therefore, after two corrector iterations, we obtain

y3 = y(0.3) = 1.444114
***
Example 9: Solve the initial value problem
l, y ' = x 2 +y3, y(l)=O
on the interval [l, 1.61using the predictor-corrector method
 Muhistep and Predictor-
Carrector lWetkods for

with the step 1ength.h = 0.2. Perform three corrector iterations per step. Compute the
starting value using Taylor series second order method with the same step length h.
Solution: We have f(x, y) = x 2 + y3. We write the method (38) as

where s=0,1,2 and i =0,1, ...

"For i = 0 , we get from Eqns.(39) and (40)

Now yo is the given initial value, whereas y, is the starting value. We compute the
starting value using the Taylor series second order method

L
with h = 0.2. We have
y ' = x-7 + y 3 , ya =2x+3y2y'
Therefore, for x, = 1, yo = 0, we obtain yb = 1, Y: = 2

,
and y = y(1.2) = 0 + 0.2(1) + -
("'2)2 (2) = 0.24
2
After determining the starting value, we use the predictorcorrector method given by
Eqns.(39) and (40). We obtain for

1 Now x ? =1.4, f(x,, y y ) ) = x ; + ( Y ? ) ) =2.151250

~

! and f(x,, y f ) ) = x i + ( ~ f ' =) 2.172185

~

I and f(x,,'yY)) = x i + ( Y ? ) ) =~ 2.174053

I Therefore, after three corrector iterations

y, = y(1.4) = 0.598348
Numerical Solutions
of ODES

p: yy) = y2 +-[h
3f2 -f.,]=1.105232
2
Now x3 =1.6, f ( ~ , , ~ r ) ) = x+ :(yy))3=3.910083

and f(x3, y y ) ) = x i + (yj'))3 = 4.365709

Hence, after three corrector iterations

y3 c y(1.6) = 1.227823
***

El 0) Solve the initial value problem

yr = y + sin y, y(1) = 1
in the interval [ 1, 1.61 using the predictor-corrector method

h
C : yi+,= yi + -(fi+,+fi)
2
with h = 0.2 . Calculate the starting value using second order Taylor
series method with the same steplength. Perform two corrector iterations per
step.
El 1) Obtain the approximate value of y(1 .O) for the initial value problem
y'= x 2 + y 2 , y(0) = 1
using the predictor-corrector method

with h = 0.2. Calculate the starting values using theEuler method with the
same stepleqgth. Perform two corrector iterations per step.

Let us now discuss the stability of multistep methods.

9.4 STABILITY OF MULTISTEP METHODS

If we apply the multistep method
yi+,= a , y i + a,yi-, i...+ ~ i - k + ~

+h[ boy;+,+ b , yl bk ~ f - ~ , , ]
to the test equation
Y' = LY, y(xo)= Yo
we get
Yi+l =a1 yi yi-l + ' ' ' + a k Y i - k + l
+ hh[ + b l yi + ' ' ' + b k ~i-k+l]
 Multistep and Predictor-
Or, yi+l- a l yi - a 2 yi-l - -..- Yi-,+I Corrector Methods for
- hh [ b, yi+l+ b, yi + + b k Yi-k+l]= 0 (42) Solving lVPs
The exact solution of Eqn.(42) is given by
(x-~0)).
y(x) = y(x0 e
Therefore, we get
y(Xi) = Y(x,) e ( X ~ - X ~ ) A = y(x,) eAhi
and y(xi+,)= y(xo) e ( x i + l - x o ) A = y(xo) eAh(i+lf
Hence, we obtain
Y(X. 1
-=eAh or, y ( ~ ~ + ~ ) = ye A ( xh ~ )
xi)
We find that the exact solution of Eqn.(41) grows (A > 0) or, decays (h < 0) by the
factor e A h .
Substituting yi = y(xo) e A h iin Eqn.(42), we get
(e'h(i+l) -a l - ...- a k e Ah(i-k+l)) Y(Xo)
-Ah( boefi(i+l)+ bIe"i + ...+ bk e ~ h ( i - k + l ) ) Y(XO)
=0
i ~ ~ , we get
~ i ~ ibyd e"h(i-k+l)
[(eAhk-a, e ~ h ( k - 1 ) - a,) +
- hh(boeAhk b,e"h(k-') + ...+ b,)] = 0 (43)
Substituting eAh= €, in Eqn.(43), we get

where,
p(€,)=€,k- a l ck -...- ak
a(€,) = b, + b l ck + . - . + b k
We note that a(€,) is a polynomial of degree k for an implicit method (b, # 0) and a
polynomial of degree k - 1 for an explicit method (b, = 0) .
The polynomial equation p(€,)- h h a (5) = 0 is called the characteristic equation
associated with the multistep method (41).
The polynomial p(€,)= 0 is called the reduced.characteristic equation.
LetE,1,E,2,...,tkbe therootsof p(€,)=Oand €,lh,E,2h ,...,Sul be therootsof
~ ( € , ) - h ha(€,)=O.
We note that for a consistent method, ,€ = 1 is always a root of p (5)= 0.
The multistep method (41) is said to be
stable : if 1 Si 1 < 1, i f 1 and 5, = 1
unstable: if ( Si 1 > 1 for some i or, if there is a multiple root of p (5)= 0 of
modulus unity.
conditionally stable (weakly stable): if €,f s are simple and if more than one of these
roots have modulus unity.
For a stable method, we further define
absolutely stable :if h < 0 and 1 Ci I < 1 for all i
relatively stable :if h > 0 and 1 Si I < e" for all i
interval of absolute stability :The set of values of Ah for which the method is
absolutely stable, that is the interval (Ah, 0), h < 0 and I Si I < 1 for all i
Similarly, we define the interval of relative stability.
We generally use the Routh-Humitz criterion to obtain the interval of absolute
stability. To use this criterion, we substitute
Numerical Solutions in the characteristic equation p (5) - Ah o (5) = 0 and obtain the transformed
of ODES
equation.
P ( z ) = v o z k+ v , zk-I +..-+Vk-, Z+Vk = O
where vl s depend on Ah.
Let
v I v3 vs .-- Vzk-,
vo v2 v4 ..- V2k-2
0 Vl V3 V2k-3
D=
0 V0 V2 VZk4

0 0 0 ... v,
then the method is absolutely stable if
(i) vi > 0 for all i and
(ii) leading principal minors of D are positive.
-- We obtain the condition of absolute stability as
k = l : v,>O, v, >O
k = 2 : vo >O, v, >O, v, >O
k = 3 : v, >O,. v, >O, v, >O and v, v, -vo v, > 0
Note that if any of vi s is zero and other vi s are positive, then it indicates that a root
lies on the unit circle ( 5 1 = 1 . If any of vI s is negative, then there is at least one root
for which 1 ti 1 > 1.
In both the cases, the method is unstable. --
Without giving proof we state the following result.
Theorem 1 The order of a stable explicit k-step method of the form (24) cannot
exceed k and the order of a stable implicit k-step method of the form (24) cannot
exceed k + 1 if k is odd and k + 2 if k is even.
We now take up a few examples
Example 10: Find the interval of absolute stability of the Adams-Bashforth methods
h
(1) YI+, = YI + ~ O Y ;- Y;-, )

(ill Y l + , = Y , + ~(23 yI -16 y:-, +Sy:-,)

Solution: We apply the method to the test equation y' = Ay .

(1) we get the characteristic equation (k = 2) as

Ah
c2 -5--(35-1)=0
2
3hh Ah
8F 4'-(l+-)5+-=o
2 2
l+z
Substituting 5 = -
1-2
,we get

(E)~ 3Ah 1+ z
[L)+y=o
Ah

Ah
or, (I+~)~-(I+?] (1-z2)+- 2 (1-zl2 = O

or, ( 2 + 2 h h) z2 +(2-Ah) z+[-Ah]=O

or, v0 z2 + v , z + v 2 = o .
where, vo = 2 + 2 h . h , v, = 2 - A h , v, =-Ah.
 Multk&p md Prallctor-
We require that Corrector Metbeds for
v , > O , v, >O v 2 > 0 when Ah<O Solving IVPs
We find that v, and v, are always positive.
Now v2 >0=2(l+Ah)>O or, Ah>-1
Hence, interval of absolute stability is ] - 1, 0[
(ii) we obtain the cliaracteristjc equation (k = 3) as

23H 16H 5H
or, 5) E,' +-
12
5---
12
-0
where, H = Ah.
1+ z
Substituting 5 = -,we get
1-z

or, (l+3z+3z2 +z3)- 1+-H

:: ) (l+z-z2-z3)

~omparingwithv 0 z 3 + v , z 2 + v 2 z + v 3=0,weget
1 2
vo =-(6+1 lH), v, =-(6-H), v, = 2(1 -H), v3 = -H . Since Ah < 0 , we have
3 3
H < 0 . We find that v,,v,,v, are always positive. From v, > 0 , we get the
condition H > - 6 / 1 1 . Wealsofindthat v , v , - v o v 3 = 5 -(22/3)H+8,whichis
~ ~
positive for all H < 0 . Hence, the interval of absolute stability is ] - 6 / 11, 0[ .

Example 11: Find the interval of absolute stability of the methods.

h
(ii) yf+,= yi +-[ 5yi+, + 8y; - Y:-,]
12
Solution:
(i) We have k = 1 . The characteristic equation is obtain as

Solving for 5 , we get

1+Ah/2
'= I-Ah12
For Ah < 0, 1 E, 1 < 1for all Ah . Hence, the method is absolutely stable for all h. The
interval of absolute stability is ] - a,0[ and therefore the method is A-stable.
(ii) We have k = 2 . The characteristic equation is given by
Numerical Solutions
of ODES

where H = Ah .
1+ z
Substituting 6 =- , we get
1-2

Comparing with v, z 2 + V , z + v 2 = 0 , we get

Since h < 0, we find that H < 0 . Now v, an v 2 are both positive for H < 0 .
v, > 0 if H > -6 . Hence, the interval of absolute stability is ] - 6, O[.
***
Example 12: Show that the Milne-Simpson method
h
Y n + l =Yn-l + - ( Y ~ + +
I 4 ~ +; Y ~ - I )
3
is not absolutely stable for any h.
Solution: Here k = 2 . The characteristic equations is obtained as
<' -1-$(52 +4,+1)=0

c2 +
Note that the roots of p(<) = - 1 = 0 are 1 . Since two roots of p(5) = 0 are
simple and of modulus 1, the method is conditionally stable.
Alternatively
l+z .
Substitute 5 =- in the characteristic equation and obtain
1-2

Comparing with v, z 2 + V , z + v2 = 0 , we get

Since h < 0, H < 0 . For H < 0, v, and v, are always positive. However, v, is not
positive for any H < 0 . Hence, the method is not absolutely stable for any H < 0 .
***
And now some exercises for you.
E12) Determine the interval of absolute stability for the method

when applied to the test equation y' = h y, h < 0 .

 Multistep and Predictor-
E13) Show that the method Corrector Methods for
h Solving IVPs
yi+,= 8y, + 9yi_,+-(17yl + 14~:-1- Y;-2)
3
is not stable when applied to the test equation y' = hy, h c 0.
E14) Determine the interval of absolute stability for the method

when applied to the test equation y' = hy, h < 0 .

E15) Show that the method

is A-stable when applied to the test equation y' = hy, h < 0 .

w e n o w end the unit by giving a summary of what we have covered in it. -

9.5 SUMMARY
In this unit, we have learnt the following points:
1. In multistep methods for solving the IVP(1)
y1=f(x9y), y(x,)=y,
the solution value y +,at the nodal point x ;+,is obtained by using previously,
calculated k values y , - , at the nodal points xi-,, ,m = 0, I,. ..,k - 1. Such a
method is called an explicit multistep method. If the method also uses the
,+,
solution value y , then the method is called implicit multistep method.
2. The multistep methods are not self starting. yi is called the initial value and
yi-, , yi-?,. . ., Y , - ~ +are
, called the starting values. The starting values are to be
calculated using some singlestep method before the multistep method can be
used.
3. .A general multistep method or a k-step multistep method can be written in the
form
Yi+l =alyi + a 2 Y i - 1 + " ' + a k Y i - k + l

+ h(boy:+l + b i y f + ...+ b , ~ ; - ~ + , )
The method is explicit if b, = 0 and implicit if b, t 0 . The unknowns a , 's
and bi 's are obtained using (i) interpolation methods (ii) method of
undetermined parameters.
We have the following form of a few important class of methods
(a) Adams-Bashforth method: a , = I,a, = a , = - . . a k = 0, b, = O . These
methods are explicit methods and are of order k .
(b) Nystrommethod: a, = L a , = a , = . . . = a , =O,b, = O . Thesemethods
are explicit methods and are of order k .
(c) Adams-moulton methods: a , = 1,a, = a , = .;. = a , =O,b, # 0 . These
methods are implicit methods and are of order k + 1.
(d) Milne-Simpson methods: a, = 1,a, = a , = . . - = a k = O,b, ;t 0 . These
methods are Implicit methods and are of order- k + 1 if k is odd and of
order k + 2 ~f k is even.
An explicit method is called a predictor and an implicit method is called a
~- ~

4.
corrector. We use the explicit method to predict the value y$ for yi+, and
use the Corrector, which uses ye/ iteratively to obtain y,,, ,the improved value
yi+,; After the required number of corrector iterations, we take y,+,= y;:/ .
Numerical Solutions 5. If we apply the multistep method (44) to the test equation y' = hy ,we obtain the
of ODES
characteristic equation
~ ( 5 -) Ah ~ ( 5=) 0
where

,<,
Let t1,<,,. .. be the roots of the reduced characteristic equation
p(5) = 0 and el,,ezh,.
..,<& be the roots of the characteristic equation. Then,
the method is
(i) stabie:iflk,1<1, i # l and €,,=I.
(ii) unstable: if I > 1for some i , or if there is a multiple root of p(<) = 0 of
modulus one.
(iii) conditionally stable (weakly stable): if t i ' s are simple and if more
than one of these roots have modulus one.
The method is called
absolutely stable: if h < 0 and 1 < 1for all i . The set of values
of hh which satisfy this condition give the interval of absolute stability.
If this interval is ] -.a,0 [ ,the method is called A-stable.
I 1
relatively stable: if h > 0 and ti,,< e U for all i . The set of values of
hh which satisfy this condition give the interval of relative stability.
The order of a stable k-step explicit method cannot exceed k and the order of
a stable k-step impl~citmethod cannot exceed k + 1if k is odd and k + 2 if k is
even.

3
El) TE= --hSyV(xi)+ 0(h6); order:4
80
E2) xo = 0, yo = 1, yb = -1, y", 3; starting value: y, = y(0.2) = 0.86;
We obtain from the multistep method
f o = yb = -1, f, = yi = -0.5396, y, c ~ ( 0 . 4 =0.79812;
)
f, = y; = -0.236996, y, = y(0.6) = 0.780981. ,
E3) x, = 1, yo = 0, yb = 1, y t = 2, yy = 4; Starting value:
yl c y(l.1) = 0.110667
x, =1.1, y, =0.110667, y', =1.222247, yl=2.861296,
y r = 5.621078; starting value y, su y(1.2) = 0.248135.
From the multistep method we get
f o = 1, f, = 1.222247, f, =1.501571; y, = y(1.3) =0.414637;
,
f, = 1.861924; y w y(1.4) = 0.622223;
f, = 2.347162; y, = y(1.5) =0.886405.
E4) Starting values:
xo =1, yo =I, K,=0.4, K, =0.672, y, =y(1.2)=1.536;
x , = 1.2, y, = 1.536, K, =0.779059, K, = 1.534912,
y, = (1 -4) = 2.692985.
From the multistep method, we get
fo = 2, f, = 3.895296, f, = 9.945153, y, y(1.6) = 5.791032;
f, = 39.327084, y = y(1.8) = 19.979290;
, ,
f = 419.151320, y ~(2.0)= 196.814380.
 Multistep and Predictor-
E5) Starting value: Corrector Methods for
x, =I, yo = 2, K, =0.8, K, =0.93, K, =0.9495, Solving IVPs
K, = 1.10485, y, = y(l.1) = 2.943975.
From the multistep method, we get
h
(1 - h)yi+, = yi + $2xi+, + 4fi+,+ fi), i = 0,1,2
We obtain
x o = 1 , y 0 = 2 , f o = 8 , X, =1.1,
y, =2.943975, f, =11.031925, X, = 1.2
y, x y(1.2)= 4.241767; f2 =15.125301, X, = 1.3,
y, =y(1.3)=6.016755; f, =20.650264, X, =1.4,
y, z y(1.4) = 8.436273.

E 10) Starting value:

xo = 1, yo = 1, yb = 1.841471, y", 2.381773,
y, P y(1.2) = 1.415930.
From the given P-C set, we get
f, = 1.841471, x, = 1.2, f, = 2.403962, X, = 1.4,
yr' = 1.952972; f ( x 2 , y r ' ) = 2.880828
Yy' = 1.944409; f (x, , yy)) = 2.875424
y:2) = y(l.4) = y2 = 1.943869, f2 = 2.875081; x, = 1.6
yiO'= 2.565997; f(x,, yy') = 3.1 10332,
yy' = 2.542410; f(x,, yy)) = 3.106377,
Yi2) y(1.6) = y, = 2.542015.
P

E 11) Starting values

y1=y(0.2)=1.2, y2*y(0.4)=1.496,
y, r*: y(0.6) = 1.9756.
Using the P-C set, we ge't
y$ = 3.4235, y:' = 3.6167, y\2' = 3.7074
Therefore, y, = y(0.8) = 3.7074.
3' NumericalSolutioac
y?) =9.0140, y:' =11.5792, yy' = 15.1009
I of ODES
Therefore, y, % y(1 .O) = 15.1009
E12) Characteristic equation is obtained as

Sbstituting { = (1 + z) 1 (1*- z) ,we obtain

v o z3 + v , z 2 + v 2 z + v 3 = 0
where

Since vo < 0 for all H < 0 , method is not stable for any H.
. .
~i 3) The roots of the reduced characteristic equation
p(5)=k2 +8{-9=0
are 5, = 1, t2= -9.
Since 1 €,1,> 1, the method is unstable.
E14) Characteristic equation is obtained as

Substituting 6 = (1 + z) I (1 - z) , we get
v,, z3 + v , z2 + v 2 z + v 3 = O
where

3
and D = v,v, -vov3 = - [ 9 - 1 3 ~
4
+3 ~ ~ Interval
1 . of absolute stability is
i
El 5) The characteristic equation is obtained as

Substituting E, = (1 + z) I (1 - z) ,we get

2
VoZ +VIZ+VZ= O
where

since v,, v, ,v2 are all positive for all H < 0 ,the method is A-stable.
-X-
 Muttlstep and Predlctor-
PRACTICAL ASSIGNMENTS Corrector Methods for
Solving lVPs
Session 2:

1. Write a program to solve the initial value problem y' = f (x, y) , y(xo) = y o , using
Euler method as predictor and Backward Euler method as corrector. Use the
corrector thrie times. Input the step length h and the number of steps of
~ntegration.

2. Write a program to solve the initial value problem y' = f(x, y), y(xo) = y o , using
Euler method as predictor and Trapezoidal method as corrector. Use the corrector
three times. Input the step length h and the number of steps of integration.

3. Write a program to solve the initial'value problem y' = f(x, y), y(xo) = y o , using
Milne's predictor-corrector method. Required starting values are to be computed
using the Euler method.

4. 'Write a program to solve the initial value problem y' = f (x, y) , y(xo) = y o , using
Milne's predictor-corrector method. Required starting values are to be computed
using the Taylor series method of second order. Test your program on Example 9.