Numerical methods

1 Root finding 1.4 Secant method

1.1 Method of bisection Given an equation in one variable f (x) = 0.

Given an equation in one variable f (x) = 0, • Select two guess values x0 and x1 .

• Calculate f (x0 ) and f (x1 ).

• Select two initial guess values a0 and b0 ,
where a0 < b0 . • Determine the next guess value using the for-
mula
a0 + b0
• Calculate c0 = .
2 f (xn−1 )
xn = xn−1 −
f (xn−1 ) − f (xn−2 )
• Find f (a0 ), f (b0 ) and f (c0 ). xn−1 − xn−2
• Select the new guess interval based on the method of false position.
above values. Call this new interval [a1 , b1 ].
• Repeat step 3 until the guess value converges
– If f (a0 )· f (c0 ) < 0, then a root is some- to a certain value.
where within the interval [a0 , c0 ]. Example. Find the positive solution of the
– If f (b0 )· f (c0 ) < 0, then a root is some- equation 2 sin x = x.
where within the interval [b0 , c0 ].
– If f (a0 ) · f (c0 ) = 0 or f (b0 ) · f (c0 ) = 0,
2 Interpolation
then c0 is a root.
2.1 Lagrange interpolation
• Repeat the previous steps until an and bn are
Given n + 1 points (xi , yi ), it is required to find a
very close to each other.
polynomial function P(x) of degree n such that
P(xi ) = yi , where i ∈ {0, 1, 2, · · · , n}. Then
1.2 Method of false position !
n
x − xm
Same as method of bisection, except use P(x) = ∑ y j ∏
j=0 0⩽m⩽n ∩ m̸= j x j − xm
an f (bn ) − bn f (an )
cn = . 2.2 Newton’s divided difference method
f (bn ) − f (an )
Given n + 1 points (xi , yi ), it is required to find a
1.3 Newton-Rhapson method polynomial function P(x) of degree n such that
P(xi ) = yi , where i ∈ {0, 1, 2, · · · , n}. Then
Given an equation in one variable f (x) = 0. !
n j−1

• Select a guess value x0 . P(x) = ∑ a j ∏ (x − xi )

j=0 i=0

• Calculate f (x0 ) and f ′ (x0 ). where a j is computed using the following table.

x y 1st 2nd ···

• Determine the next guess value using the for-
mula x0 y0 [y0 , y1 ] [y0 , y1 , y2 ] · · ·
f (xn−1 ) x1 y1 [y1 , y2 ] [y1 , y2 , y3 ] · · ·
xn = xn−1 − ′ . x2 y2 [y2 , y3 ] [y2 , y3 , y4 ]
f (xn−1 )
x3 y3 [y3 , y4 ]
• Repeat step 3 until the guess value converges x4 y4
.. ..
to a certain value. . .

Engr. Jaydee N. Lucero 1

 Numerical methods

where 4.2 Trapezoidal rule

[yk , yk+1 , · · · , yk+n ] Assume that f (x) is continuous on [a, b]. Let n be
[yk+1 , yk+2 , · · · , yk+n ] − [yk , yk+1 , · · · , yk+(n−1) ] b−a
= a positive integer and ∆x = . If [a, b] is di-
xk+n − xk n
vided into n subintervals, each of length ∆x, with
Example. If P is a cubic function such that endpoints at P = {x0 , x1 , x2 , · · · , xn }, then
P(2) = 5, P(3) = 7, P(4) = 12 and P(5) = 19,
Z b
find P(6). ∆x  
f (x) dx ≈ 1 2 2 ··· 2 1
a 2

3 Differentiation 4.3 Simpson’s one-third rule

3.1 First derivative Assume that f (x) is continuous on [a, b]. Let n be
b−a
Forward difference a positive integer and ∆x = . If [a, b] is di-
n
f (x0 + h) − f (x0 ) vided into n subintervals, each of length ∆x, with
f ′ (x0 ) ≈ endpoints at P = {x0 , x1 , x2 , · · · , xn }, then
h
Backward difference Z b
∆x  
′ f (x0 ) − f (x0 − h) f (x) dx ≈ 1 4 2 4 2 ··· 2 4 1
f (x0 ) ≈ a 3
h
Central difference In the Simpson’s rule, n must be an even integer.
f (x0 + h) − f (x0 − h) If n is odd, the last subinterval must be approxi-
f ′ (x0 ) ≈ mated using trapezoidal rule.
2h

3.2 Second derivative Example. In a standard normal distribution,

find the probability that a z-score lies between
Central difference 1 and 2. In other words, calculate the definite
f (x0 + h) − 2 f (x0 ) + f (x0 − h) integral
f ′′ (x0 ) ≈ Z 2
h2 1 2
√ e−z /2 dz.
Example. A particle moves along a straight 1 2π
line with +x-axis to the right. Its position x(t)
in meters at any time t in seconds is given by
the function 5 Systems of linear equations
x(t) = sin(cos(tan x)) 5.1 Jacobi iteration
Given a system of n linear equations in n variables
Find the velocity and acceleration of the parti- {x1 , x2 , · · · , xn }.
cle at t = 2 seconds.
• Assume guess values for x1 , x2 · · · , xn .

4 Integration • Using these guess values, calculate new val-

ues of x1 , x2 , · · · , xn at the same time.
4.1 Midpoint rule
• Repeat step 2 until the values converge.
Assume that f (x) is continuous on [a, b]. Let n
b−a
be a positive integer and ∆x = . If [a, b] is 5.2 Gauss-Seidel iteration
n
divided into n subintervals, each of length ∆x, and
mi is the midpoint of the ith subinterval, then Given a system of n linear equations in n variables
Z b n {x1 , x2 , · · · , xn }.
f (x) dx ≈ ∑ f (mi )∆x
a i=1
• Assume guess values for x1 , x2 · · · , xn .

Engr. Jaydee N. Lucero 2

 Numerical methods
 
• Using these guess values, calculate new val- h k2
• k3 = h f xn + , yn +
ues of x1 , x2 , · · · , xn one at a time. Once the 2 2
new value of a variable is calculate, immedi-
• k4 = h f (xn + h, yn + k3 )
ately replace the previous value and use it in
subsequent calculations. k1 + 2k2 + 2k3 + k4
• yn+1 = yn +
6
• Repeat step 2 until the values converge.
Example. Solve the first-order differential
Example. Solve the system of equations equation
dy √ √
= x+ y

2x + 3y − 5z = −4
 dx
3x − 5y + 6z = 7 . with initial condition x = 0, y = 0.

6x − y − 4z = 9


6 Differential equations
6.1 Euler method
Given the differential equation dy/dx = f (x, y)
with initial condition f (x0 ) = y0 ,

• Pick a marching step h.

• k1 = h f (xn , yn )

• yn+1 ≈ yn + k1

6.2 Midpoint method

Given the differential equation dy/dx = f (x, y)
with initial condition f (x0 ) = y0 ,

• Pick a marching step h.

• k1 = h f (xn , yn )
 
h k1
• k2 = h f xn + , yn +
2 2
• yn+1 = yn + k2

6.3 Runge-Kutta method

Given the differential equation dy/dx = f (x, y)
with initial condition f (x0 ) = y0 ,

• Pick a marching step h.

• k1 = h f (xn , yn )
 
h k1
• k2 = h f xn + , yn +
2 2

Engr. Jaydee N. Lucero 3