You are on page 1of 6

CHAPTER FOUR

SYSTEMS OF NONLINEAR EQUATIONS

We have already seen that systems of simultaneous equations are those which could be
given as

f 1 ( x 1 , x 2 , x 3 , … … … , x n ) =b1
f 2 ( x 1 , x 2 , x 3 , … … … , x n ) =b2
.
.
.
f n ( x 1 , x 2 , x 3 , … … … , xn ) =b n

The system is linear if each of the function is expressed as summation of terms that are
composed of only one of the independent variables.

i.e. f k ( x1 , x2 , x3 , … … … , x n) =a1 x 1+ a2 x 2 +… a n x n−b 1=0 1≤k≤n

The nonlinear equations are those that do not fit to this format. e.g.

f 1 ( x , y ) =x2 + xy −10=0 …………………………………………………….4.1(a)

f 2 ( x , y ) = y +3 x y 2−57=0…………………………………………………………4.1(b)

This solutions would be the values of x and y that make the functions f1 and f2 equal to
zero. The open methods for solving single equations could easily be extended to solve
non liner equations.

4.1 Iterative methods

We will first apply fixed point iteration method to a simultaneous equation consisting of
two equations which are depicted above. Eqn 4.1 (a) and (b) could be solved for

10−x 2i
x i+1= and y i+1 =57−3 x i y 2i ………………4.2
yi

Note that the correct pair of roots is x=2 and y=3

Let us initiate the computation with guesses x=1.5 and y=3.5 and eqn 4.2 would give us
new values of x as x=2.21429 and using this value of x in the second eqn we have

1
y=-24.37516 .

Thus the approach seems to be diverging. The second iteration is even worse giving

x=-0.20910 and y= 429.709

The next thing we are left to do with is to change the format of the eqn 4.2

57− y
x i+1= √10−xy and y i+1 =
√ 3x
…………..4.3

Now the results will be very satisfactory giving the following values after the second
iteration

X=1.94053 and y=3.04955.

A final deduction we can take from the ongoing method is the fact that the convergence
depends on the manner in which we formulated our equation. The other unforeseen
shortcoming is the divergence that will result due to initial guesses that are insufficiently
close to the true solution.

It could easily be stated that fixed point iteration converges if in the region of interest
the slope of the function (for instance g(x)) in absolute value is less than the slope of the
line f(x)=1.

In our case we will define a sufficient condition for the convergence of two equation
system.

∂f1 ∂f2 ∂f ∂f
| || | | || |
∂x
+
∂x
<1∧ 1 + 2 <1
∂y ∂y

4.2 Newton Raphson

This method employs the derivative (slope) of a function to estimate its intercept with
the axis of the independent variable i.e. the root. This estimate was based on first order
Taylor series.

f ( x i +1) =f ( x i ) + ( x i+ 1−xi ) f ' (xi )

2
x iis the the initial guess at the root and x i+1 is the point at which the slope intercepts the
x axis. By definition at x i+1 , f ( x i +1) =0 and the above equation becomes

f ( xi )
x i+1=x i−
f '( xi )

This is the single equation form of Newton Raphson method.

For a two variable case i.e. x and y with two functions u(x,y) and v(x,y) we have a Taylor
series for u as follows

∂u i ∂ ui
ui +1=ui +(x ¿ ¿ i+1−x i) +( y ¿ ¿i +1− y i) ¿¿
∂x ∂y

The same equation holds for v.

Rearranging the above equation for the situation ui +1=0 would led us to

∂u i ∂u i ∂u i ∂ ui
x i+1+ y i+1=−ui+ xi + yi ………………4.4(a)
∂x ∂y ∂x ∂y

∂ vi ∂ vi ∂ vi ∂ vi
x i+1+ y i+1 =−v i + x i + yi ……………….4.4(b)
∂x ∂y ∂x ∂y

Solving for this set of simultaneous equation we have

∂ vi ∂u
ui −v i i
∂y ∂y
x i+1=x i− …………………..4.5(a)
∂u i ∂ v i ∂ ui ∂ v i

∂x ∂ y ∂ y ∂x

∂ ui ∂v
vi −ui i
∂y ∂y
y i+1 = y i− …………………….4.5(b)
∂ ui ∂ v i ∂ ui ∂ v i

∂x ∂ y ∂ y ∂ x

Note: the determinant is referred to as the determinant of the Jacobean of the system.

Let us apply this method to the system given in equation 4.1 with f 1=u∧f 2=v

The correct value of the roots is x=2 and y=3

Initiate the computation with guesses of x=1.5 and y=3.5

3
∂u 0 6.5∧∂ v 0 2
=2 x + y=2 ( 1.5 ) +3.5= =3 y =36.75
∂y ∂x

∂u 0 1.5∧∂ v 0
=x= =1+6 xy =32.5
∂y ∂y

Thus the determinant of the jacobian in this case is 6.5(32.5)+1.5(36.75)=156.125

Additionally u0 =x2 + xy =−2.5 v 0= y +3 x y 2=1.625

Inserting all these values into eqn 4.5 (a) and (b) will finally result in

x=2.03603∧ y=2.84388

a much closer value to the actual solution after a single iteration.

Similar to the case for fixed point iteration the Newton Raphson method will diverge if
the initial guesses are not sufficiently close to the true errors.

The approach for solving two nonlinear equations could be further extended to the
general case where there are n simultaneous non linear equations. i.e

f 1 ( x 1 , x 2 , x 3 , … … … , x n ) −b 1=0

f 2 ( x 1 , x 2 , x 3 , … … … , x n ) −b2=0

f n ( x 1 , x 2 , x 3 , … … … , xn ) −b n=0

The Newton Raphson method applied in this case is referred to as the multidimensional
version. At first a Taylor series expansion is written for each equation like the k th
equation given below.

∂ f k ,i ∂ f k ,i ∂ f k ,i
f k ,i +1=f k ,i +(x ¿ ¿ 1 ,i+1−x 1 ,i ) +( x ¿ ¿ 2 ,i+1−x 2 ,i ) + …( x ¿ ¿ n ,i+1−x n , i) ¿ ¿ ¿ …...4.
∂ x1 ∂ x2 ∂ xn
6

4
Here k represents the equation or unknown and the second subscript denote whether
the value is taken at the present value (i) or at the next value (i+1).

Each f k ,i +1 term are set to zero at the root for each equation. Hence we have,

∂ f k ,i ∂ f k, i ∂ f k,i ∂ f k ,i ∂ f k,i ∂ f k, i
−f k ,i + x 1 ,i + x2 , i +… x n ,i =x 1 ,i +1 + x 2 ,i +1 +… x n ,i +1 …….4.7
∂ x1 ∂ x2 ∂ xn ∂ x1 ∂ x2 ∂ xn

With the unknown terms ( x 1 ,i+1 , x2 , i+1 … ….. x n ,i +1) collected on the right side.

Such equations are generated for each equation with k=1, 2, ……..n constituting a set of
linear simultaneous equations that are solved by techniques we have discussed in
chapter three. We ended up converting a nonlinear set of equations to linear set of
equations. Equation 4.7 could be fully expressed including all the remaining equations
by using four separate matrix notations.

∂ f 1 ,i ∂ f 1, i

[ ]

∂ x1 ∂ xn
[ Z ]= ⋮ ⋱ ⋮
∂ f n ,i ∂ f n, i

∂ x1 ∂ xn

T
[ X i ] =[ x 1 ,i x 2 ,i … … … . x n ,i ] the initial values in vector format, transposed
T
[ X i +1 ] =[ x 1 ,i +1 x 2 ,i+1 … … … . xn , i+1 ]the final values in vector format, transposed
And finally,
T
[ F i ] =[ f 1 ,i f 2 ,i … … … . f n ,i ]
The resulting finalized representation is then,

[ Z ] [ X i+1 ] =−[ F i ] + [ Z ] [ X i ]………………………………………………………………………4.8

This equation could be solved by applying techniques such as gauss elimination and
could be repeated iteratively to obtain a refined estimate in similar manner to the two
equation system case.

There are two major shortcomings in the foregoing approach. First, [ Z ] is inconvenient
to evaluate. The most straight forward method is to use finite difference approximations

5
for the partial derivatives in [ Z ] . . Second, we are expected to give a precise initial guess
to be ensured of convergence.

You might also like