1
The two variable Newton method for solving equations
Let a(x,y) and b(x,y) be two differentiable functions of x and y. In calculus we
sometimes need to solve the equations
(*)
a(x,y) = 0 and b(x,y) = 0
simultaneously. Here are two examples.
Example 1 Find the critical points of a function f(x,y).
f
f
We need to find simultaneous solutions to the equations x(x,y) = 0 and y(x,y) =
f
f
0. This is (*), where a(x,y) = x(x,y) and b(x,y) = y(x,y).
Example 2 Find the maximum and minimum values of a function f(x,y) on a curve of the
form g(x,y) = c. The method of Lagrange multipliers says that we must solve the equation
-f(x,y) = -g(x,y), which becomes the two equations
f
g
f
g
(1)
=
and
(2)
=
x
x
y
y
g
g
f g f g
Multiplying (1) by y and multiplying (2) by x we obtain x y = y x. Hence we must
f g f g
solve the equations (*), where a(x,y) = g(x,y) c and b(x,y) = x y y x.
Newton's method of solution
In general there is no method for getting solutions to (*) that you can write down.
However the Newton method gives a numerical procedure that solves the equation (*) to
any desired degree of accuracy. The idea behind the Newton method is very simple. We
describe it first and then illustrate the method with two examples.
Step 1 By some method we find an approximate solution (xo,yo) to (*).
One way to find an approximate solution to (*) would be to graph the curves a(x,y)
= 0 and b(x,y) = 0, and then look to see where these two curves intersect. A computer
program with a zoom finder can be useful here.
Step 2 We replace the linear equations (*) by the approximate equations
(*)o ao(x,y) = 0 and bo(x,y) = 0
where ao(x,y) and bo(x,y) are the linear approximations of a(x,y) and b(x,y) at (xo,yo). The
equation (*)o is a linear equation in two unknowns x and y, and it can be easily solved by
2
hand. The solution (x1,y1) to (*)o is not the true solution to the original equation (*), but
in general it is a better approximation than (xo,yo).
Recall that the linear approximation of a function f(x,y) at a point (xo,yo) is given by
fo(x,y) = A + B(x xo) + C(y yo)
f
f
where A = f(xo,yo) , B = x(xo,yo) and C = y(xo,yo), all real numbers.
Step 3 We replace the linear equations (*) by the approximate equations
(*)1 a1(x,y) = 0 and b1(x,y) = 0
where a1(x,y) and b1(x,y) are the linear approximations of a(x,y) and b(x,y) at (x1,y1). Let
(x2,y2) be the solution to (*)1.
Step 4 Continue as described above. In general we get a sequence of approximate
solutions (x1,y1), (x2,y2), ... , (xk,yk) that converge rapidly to the true solution if the
beginning approximate solution (xo,yo) is reasonably accurate. To obtain the next
approximate solution (xk y ) we solve the equations
+1 k+1
(*)k ak(x,y) = 0 and bk(x,y) = 0
where ak(x,y) and bk(x,y) are the linear approximations of a(x,y) and b(x,y) at (xk,yk).
A matrix formula for the approximate solutions (xk,yk)
In order to solve the approximate equations (*)k above rapidly with a computer it is
useful to have a formula that one can program into the computer. First we define the
a a
x y
Jacobian matrix J(x,y) = b b. If we write (x ,y ) and (x ,y ) as column vectors
k k
k+1 k+1
x y
x
x
k and k+1 then the equation (*) can be written in matrix form as
y
yk+1
k
k
xx a(x )
a(xk)
xk
0
x
k = k
+ J(x ,y )
=
J(x
,y
)
+
J(x
,y
)
b(y )
y
0
yy b(y )
y
k
k
k
k
k
k
k
k
k
k
[]
+1
The solution yk to (*)k in matrix form now becomes
k+1
x
x
1 a(x )
(*)k yk+1 = yk J(x ,y ) b(yk,)
k k
k+1 k
k
1
where J(xk,yk) denotes the inverse of the matrix J(xk,yk).
[]
3
Example 1 Find an intersection point of the circle x2 + y2 = 1 and the parabola y = x2.
(*)
We need to solve the equations
a(x,y) = 0
where a(x,y) = y x2
b(x,y) = 0
where b(x,y) = x2 + y2 1
We pick (xo,yo) = (1,1) as an approximate solution. If ao(x,y) and bo(x,y) denote the linear
approximations at (xo,yo), then we obtain ao(x,y) = 2x + y + 1 and bo(x,y) = 2x + 2y
3. Hence the equations ao(x,y) = 0 and bo(x,y) = 0 have the solution (x1, y1) =
(5 / 6, 2 / 3).
If a1(x,y) and b1(x,y) denote the linear approximations at (x1,y1), then we obtain
a1(x,y) = (5/3)x + y + (25/36) and b1(x,y) = (5/3)x + (4/3)y (77/36). Hence the
equations a1(x,y) = 0 and b1(x,y) = 0 have the solution (x2, y2) = (331 / 420, 13 / 21).
If we continue this process we obtain the following approximate solutions
(x1, y1) = (5 / 6, 2 / 3) (.8333, .6667)
(x2, y2) = (331 / 420, 13 / 21) (.7881, .6190)
(x3, y3) (.7861, .6180). In two more steps we obtain 15 digit accuracy
(x5, y5) (.786151377757423, .618033988749894)
Example 2 Find a critical point of the function f(x,y) = 2x2y2 + x2y 2x y2
f
We calculate a(x,y) = x(x,y) = 4xy2 + 2xy 2
f
b(x,y) = y(x,y) = 4x2y + x2 2y
The critical points of f(x,y) are the solutions to the equations a(x,y) = 0 and b(x,y) = 0.
We start with the approximate solution (xo,yo) = (1,1) and we obtain the equations
0 = ao(x,y) = 0 + 2(x1) 6(y+1) = 2x 6y 8
0 = bo(x,y) = 1 6(x1) + 2(y+1) = 6x + 2y + 7
The solution to these equations is the next approximate solution
(x1, y1) = (13/16 , 17/16) = ( .8125, 1.0625) Continuing we obtain
(x2, y2) ( .8069, 1.0759) In two more steps we obtain 15 digit accuracy
(x4, y4) (.807068419897086, 1.075848752008690)
