Numerical Methods

Some example applications in C++

Numerical methods - applications in C++ 2
Numerical methods apply algorithms that use numerical
approximations to solve mathematical problems.
This is in contrast to applying symbolic analytical solutions,
for example Calculus.
We will look at very basic, but useful numerical algorithms for:
Introduction
1.Differentiation 2. Integration 3. Root finding
Key to the formulation of numerical techniques for differentiation,
integration and root finding is Taylors expansion:
Taylors Expansion
Numerical methods - applications in C++ 3
The value of a function at x + h is given in terms of
the values of derivatives of the function at x
... ) ( ' ' '
! 3
) ( ' '
! 2
) ( '
! 1
) ( ) (
3 2 1
x f
h
x f
h
x f
h
x f h x f
The general idea is to use a small number of terms in
this series to approximate a solution.
In some cases we can improve on the solution by
iterating the procedure ideal task for a computer.
1. Numerical differentiation
Numerical methods - applications in C++ 4
Aim
Given a function f(x), we wish to
calculate the derivative f(x); that is,
the gradient of the function at x.
The Central Difference Approximation,
CDA, provides an approximation to
this gradient:
Numerical methods - applications in C++ 5
The approximation
improves as the size
of h reduces.
Limited precision in
the computer prevents
us from making h very
small!
Proof
Problem
Numerical methods - applications in C++ 6
For the following function, calculate the derivative at
Algorithm
Numerical methods - applications in C++ 7
1. Define the function:
2. Set the parameters:
3 Calculate the CDA:
4 Output the result.
x = 2, h = 0.01
C++ code
Numerical methods - applications in C++ 8
// Central-Difference Approximation (CDA)
// for the derivative of a function f(x).
// Here, f(x)=2*x^3+5*x, h=0.01, x=2.0.
#include <iostream>
using namespace std;
double f(double x) { return 2*x*x*x + 5*x; }
int main() {
double x=2.0, h=0.01;
double cda = (f(x+h)-f(x-h))/(2*h);
cout << "f'(" << x << ") = " << cda << endl;
}
Output
f'(2) = 29.0002
Numerical methods - applications in C++ 9
Verification
The program gives us f'(2) = 29.0002
We can verify that this is what we expect:
The function here is f(x) = 2 x
3
+ 5 x
From calculus we can obtain f(x) = 6 x
2
+ 5
and so the exact solution for f(2) is 6*2
2
+ 5 = 29.0000
We see that the error in the CDA is 29.0002 29.0000 = 0.0002
From analysis of Taylors expansion we predict the error
in the CDA as h
2
f(x)/6
= 0.01
2
.12/6 = 0.0002
Our algorithm is working as predicted.
From Calculus
f(x) = 2 x
3
+ 5 x
f(x) = 6 x
2
+ 5
f(x) = 12 x
f(x) = 12
Numerical methods - applications in C++ 10
A more difficult problem
So far the CDA does not look so useful, we have only solved
a trivial problem. Lets try a more difficult function:
Analytical solution
Evaluate f(4)
Adapt the C++ code for the new calculation
Numerical methods - applications in C++ 11
// Central-Difference Approximation (CDA)
// for the derivative of a function f(x).
#include <iostream>
#include <cmath>
using namespace std;
double f(double x) {
return x*log(pow(x+5,x))/(2*x+pow(3,x));
}
int main() {
double x=4.0, h=0.01;
double cda = (f(x+h)-f(x-h))/(2*h);
cout << "f'(" << x << ") = " << cda << endl;
}
Output
f'(4) = -0.186348
The error is
+0.000002
2. Numerical integration
Numerical methods - applications in C++ 12
Aim
We wish to perform numerically
the following integral:
This is simply the area under the curve f(x) between a and b.
For example,
How can we perform this numerically?
Formulating an algorithm
Numerical methods - applications in C++ 13
A first approximation can be obtained by forming a trapezoid.
Trapezoid area
= (f(2)+f(4)) (4-2) = (26+148) (2) = 174.
a b
The error in the
result is16%
Numerical methods - applications in C++ 14
An improved approximation can be obtained by forming
two trapezoids.
Trapezoid area
= (f(2)+f(3)) (3-2) + (f(3)+f(4)) (4-3) = 156
a b
The error in the
result is 4%
Numerical methods - applications in C++ 15
Four trapezoids. Trapezoid area
= (f(2.0)+f(2.5)) (2.5-2.0) + (f(2.5)+f(3.0)) (3.0-2.5)
+ (f(3.0)+f(3.5)) (3.5-3.0) + (f(3.5)+f(4.0)) (4.0-3.5)
= 151.5
a b
The error in the
result is 1%
The error 1/n
2
where n is the
number of
trapezoids.
Numerical methods - applications in C++ 16
Formulating an algorithm
Generalising the procedure:
Algorithm
Numerical methods - applications in C++ 17
1. Define the function:
2. Set the limits of the integral, and the number of trapezoids:
3. Set
4. Calculate the ETF as
5. Output the result.
a = 2, b = 4, n = 100
C++ code
Numerical methods - applications in C++ 18
// Numerical integration via the Extended
// Trapezoidal Formula (ETF)
#include <iostream>
using namespace std;
double f(double x) { return 2*x*x*x + 5*x; }
int main() {
double a=2.0, b=4.0;
int n=100;
double h = (b-a)/n;
double etf = (f(a)+f(b))/2;
for (int i=1; i<n; i++) etf = etf + f(a+i*h);
etf = etf * h;
cout << "The integral = " << etf << endl;
}
Output
The integral = 150.002
Error = 0.002
Numerical methods - applications in C++ 19
A more difficult problem
Adapt the previous C++ code
Numerical methods - applications in C++ 20
#include <iostream>
#include <cmath>
using namespace std;
double f(double x) {
return x * pow(0.5+exp(-x)*sin(x*x*x),2); }
int main() {
double a=0.0, b=M_PI;
int n=100;
double h = (b-a)/n;
double etf = (f(a)+f(b))/2;
for (int i=1; i<n; i++) etf = etf + f(a+i*h);
etf = etf * h;
cout << "The integral = " << etf << endl;
}
Output
The integral = 1.46937
The error is
+0.00030
3. Root finding
Numerical methods - applications in C++ 21
Aim
We wish to find the root x
0
of
the function f(x); i.e. f(x
0
) = 0.
How can we perform this numerically?
There are many ways to do this.
We will implement the Newton-Raphson method....
Formulating an algorithm
Numerical methods - applications in C++ 22
Numerical methods - applications in C++ 23
The algorithm so far:
1. define f(x) and d(x)
2. Initialise x
3. Iterate:
e = f(x)/d(x)
x = x e
4. Output x
But how many iterations?
d(x)
Obtaining an
error estimate:
Numerical methods - applications in C++ 24
We have an estimate of the error
Use this to form a termination condition that requires
6 decimal place accuracy:
iterate until < 10
-9

Algorithm
1. define f(x) and d(x)
2. initialise x
3. iterate:
e = f(x)/d(x)
if < 10
-9
terminate
x = x e
4. Output x
Example
C++ code
Numerical methods - applications in C++ 25
// Newton-Raphson method for the root of f(x)
#include <iostream>
#include <iomanip>
#include <cmath>
using namespace std;
double f(double x) { return 2*x*x*x + 5; }
double d(double x) { return 6*x*x; }
int main() {
cout << setprecision(9) << fixed;
double e, x = -1.5;
while (true) {
e = f(x)/d(x);
cout << "x = " << x << endl;
if (fabs(e)<1.0e-6) break;
x = x - e;
}
}
Output
Numerical methods - applications in C++ 26
x = -1.500000000
x = -1.370370370
x = -1.357334812
x = -1.357208820
7 decimal place accuracy
The number of correct digits
doubles on every iteration
(rapid convergence)!
Finally
In this lecture we have looked at Numerical Methods.
More about numerical methods can be found at:
http://en.wikipedia.org/wiki/Numerical_methods