Armstrong State University

Engineering Studies
MATLAB Marina – Integration Primer

Prerequisites
The Integration Primer assumes knowledge of the MATLAB IDE, MATLAB help, arithmetic
operations, built in functions, scripts, variables, arrays, logic expressions, conditional structures,
iteration, functions, debugging, characters and strings, cell arrays, structures, file input and
output, 2D plotting, 3D plotting, interpolation, curve fitting, and numerical differentiation.
Material on these topics is covered in the MATLAB Marina Introduction to MATLAB module,
MATLAB Marina Variables module, MATLAB Marina Arrays module, MATLAB Marina Logic
Expressions module, MATLAB Marina Conditional Structures module, MATLAB Marina Iteration
module, MATLAB Marina Functions module, MATLAB Marina debugging module, MATLAB
Marina Character and Strings module, MATLAB Marina Cell Arrays module, MATLAB Marina
Structures module, MATLAB Marina File Input and Output module, MATLAB Marina Plotting
module, MATLAB Marina Interpolation Module, MATLAB Marina Curve Fitting Module, and
MATLAB Marina Differentiation Module.

Learning Objectives
1. Be able to use MATLAB to compute numerical integrals.
2. Understand the limitations of numerical integration.

Terms
numerical integration, complete (definite) integral, continuous integral, noisy data

MATLAB Functions, Keywords, and Operators

sum, trapz, cumsum, cumtrapz, quad, quad2, integral, integral2
Note: quad, quad2, integral, integral2 are not used in ENGR 1371 but are used in ENGR 2010
and are not covered in this primer.

Numerical Integration
b
The definite integral of a function f ( x ) can be interpreted as an area R = ∫ f ( x ) dx and the
a

indefinite integral of a function can be interpreted as the anti-derivative of a function

df ( x )
f ( x) = ∫ dx . Similar to numerical differentiation, numerical integration is typically used
dx
when a formula for the function to integrate is not available or the integral would be difficult to
determine analytically. Numerical integration can also be used to compute estimates of
integrals that may be solved analytically later. Numerical integration can yield accurate results if
the function is piecewise continuous, relatively smooth, and the increment between function
points is relatively small. Numerical integration typically yields better results than numerical
differentiation as it is less sensitive to inaccuracies in the data (noisy data).

1
The Trapezoidal rule and Simpson's rule are two common methods of performing numerical
integration for definite integrals. For the Trapezoidal rule, the area between the function and
the x-axis is approximated by a trapezoid.
b
f ( a ) + f (b)
∫a f ( x ) dx ≈ ( b − a )
2
For numerical integration using the composite Trapezoidal rule, the interval of integration is
split into N small uniform subintervals, each approximated by a trapezoid, and the area of all
the trapezoids is summed.
b−a
b

∫a f ( x ) dx ≈ 2 N ( f ( x0 ) + 2 f ( x1 ) + 2 f ( x2 ) +  + 2 f ( xN −1 ) + f ( xN ) )
Where the xk are the end points of the trapezoids
b−a
xk= a + k , k = 0,1, , N
N

For numerical integration using Simpson's rule, the interval is approximated by a quadratic
function, and the integral is approximated by the area between the quadratic functions and the
x-axis.
b
(b − a )   a+b 
∫a f ( x ) dx ≈ 6  f ( a ) + 4 f  2  + f ( b ) 
Breaking the interval up into 2N subintervals and approximating each interval by a quadratic
function, and summing the areas between the quadratic functions and the x-axis, the integral is
approximated by the composite Simpson's rule
b

∫a f ( x ) dx ≈ 3 ( f ( x0 ) + 4 f ( x1 ) + 2 f ( x2 ) + 4 f ( x3 ) + 2 f ( x2 N −2 ) + 4 f ( x2 N −1 ) + f ( x2 N ) )
h

Where the xk are the end points of the trapezoids

b−a
xk= a + k , k = 0,1, , 2 N
2N
and
b−a
h=
2N

Cumulative numerical integration is used when we want to determine an approximation of the

anti-derivative of a function.
t

∫ f (t ) dt
−∞

Numerical Integration using MATLAB

The MATLAB function trapz is used to approximate a definite integral of a function using the
Trapezoidal rule. The trapz function takes two arguments x and y and returns the integral of y

2
with respect to x. The vectors x and y must be the same length and the interval for the
approximate definite integral is specified via the x vector values.

The MATLAB functions cumtrapz and cumsum are used to approximate the anti-derivative of
a function (cumulative numerical integral). The cumtrapz function takes two arguments x and
y and returns the cumulative integral of y with respect to x using the trapezoidal method. The
vectors x and y must be the same length, the vector x should have uniform spacing, and the
result is a vector the same size as x and y. If the spacing in x is one (distance between x values is
one), then cumtrapz can be used with a single argument y.

The MATLAB program of Figure 1a computes the approximate definite integral of

5 x

∫ (x )
+ 2 x + 1 dx and cumulative integral (approximate anti-derivative) of ∫ (u )
+ 2u + 1 du .
2 2

−5 −5

Figure 1b shows the function and its cumulative integral.

% function f(x) = x^2 + 2x + 1

coef = [1, 2, 1];
x = -5 : 0.1 : 5;
f = polyval(coef, x);

% approximate definite integral of f(x) from -5 to 5

defintf = trapz(x,f);
% approximate anti-derivative of f(x)
cumintf = cumtrapz(x,f);

figure(1)
plot(x,f,'b-',x,cumintf,'g-')
xlabel('x'), legend('f(x)','approximate anti-derivative of f(x)')

Figure 1a, MATLAB Program Performing Numerical Integration

If the spacing in x is not uniform, then cumtrapz(y) can be used to compute the cumulative
integral with unit spacing. This result must then be element by element multiplied with the
spacings. The cumsum function can be used similar to cumtrapz to perform cumulative
numerical integration except it only takes one argument and must either be multiplied by the
spacing increment or element by element multiplied with the spacings for non-uniform spacing.

Remember that the MATLAB function trapz is for computing approximate definite integrals
and the result will be a scalar and that cumtrapz is for computing cumulative (indefinite)
integrals and the result will be a vector.

3
 100
f(x)
90 approximate anti-derivative of f(x)

Continuous
80 Integration

70

60

50

40

30

20

10

0
-5 -4 -3 -2 -1 0 1 2 3 4 5
x

Figure 1b, Resulting Plot of Cumulative Integration Example

Numerical Integration and Differentiation of Noisy Data

One must be careful when performing numerical operations on noisy data (data with
uncertainties or data with measurement error). The noise or error associated with measured or
uncertain data is often small in magnitude but rapidly changing.

Integration tends to reduce the effect of error or noise whereas differentiation tends to
magnify the effect of error or noise. Integration is similar to averaging and tends to reduce the
effects of things that are small but vary rapidly. Differentiation which gives the rate of change
of something, magnifies things that vary rapidly even if the magnitude of the variation is small.
The derivative of the error or noise in data may be larger in magnitude than the derivative of
the underlying function of the data.

Figures 2a shows a MATLAB program that generates a noisy signal and determines the
approximate integral and approximate derivative. The underlying function is a unit amplitude
cosine of 250Hz. The additive noise is a cosine of 60Hz with amplitude uniformly distributed
over the range -0.05 to 0.05.

Figures 2b -2d show plots of the noisy data, the approximate integral of the noisy data, and the
approximate derivative of the noisy data. Notice in the plot of Figure 2b, that the noise in the
data is barely discernible except at a few of the peaks. The signal looks like a unit amplitude
cosine of 250Hz (period 4msec).

4
 % noisy data, 250Hz cosine with additive noise
% modeled as a 60Hz cosine with varying amplitude
t = 0:1/250/20:0.02;
% unit amplitude cosine of 250Hz
g = 1.0*cos(2*pi*250*t);
% noise, random amplitude 60Hz cosine
noise = (-0.05 + (0.05 - -
0.05)*rand(1,length(t))).*cos(120*pi*t);
% data plus noise
gNoisy = g + noise;

% approximate integral and derivative of the noisy data

intgNoisy = cumtrapz(t,gNoisy);
dgNoisy = diff(gNoisy)./diff(t);

Figure 2a, Numerical Integral and Approximate Derivative of Noisy Data

Noisy Data
1.5

0.5
g(t)

-0.5

-1

-1.5
0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02
t (sec)

Figure 2b, Plot of Noisy Data

The integral of the noisy data, shown in Figure 2c, is a sin of approximately 250Hz. The
amplitude is much smaller than one might expect, but recall that ∫ cos = (ωt )dt ω1 sin (ωt ) + c , so
the amplitude the sine wave should be scaled by 1
ω times the amplitude of the noisy signal.

5
 -4
x 10 Numerical Integral of Noisy Data
8

2
integral g(t)

-2

-4

-6

-8
0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02
t (sec)

Figure 2c, Approximate Integral of Noisy Data

The derivative of the noisy data, Figure 2d, should be a negated sin scaled by ω. The plot is
similar to this but the initial value and frequency are off. There are five periods of the original
signal in 20msec but only around 4.5 periods of the signal’s derivative. Notice that in several
areas the derivative of noise is easily apparent (all the negative peaks, second through fifth
positive peaks, down slope of second peak, etc.).

Approximate Derivative of Noisy Data

2000

1500

1000

500
d g(t)/dt

-500

-1000

-1500

-2000
0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02
t (sec)

Figure 2d, Approximate Derivative of Noisy Data

In this example the magnitude of the noise was 5% or less of the original signal and was
multiplied by a cosine of 60Hz which for most points further reduced the magnitude of the
noise. The plots of Figures 3a and 3b shows the effect of purely additive noise. Changing the
state in the program Figure 2a the computes the noise to

6
noise = (-0.05 + (0.05 - -0.05)*rand(1,length(t)));
The effect of the noise is barely noticeable on the approximate integral but more pronounced
on the approximate derivative.
-4
x 10 Numerical Integral of Noisy Data
8

2
integral g(t)

-2

-4

-6

-8
0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02
t (sec)

Figure 3a, Approximate Integral of Noisy Data

Approximate Derivative of Noisy Data

2000

1500

1000

500
d g(t)/dt

-500

-1000

-1500

-2000
0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02
t (sec)

Figure 3b, Approximate Derivative of Noisy Data

Last modified Thursday, November 13, 2014

This work by Thomas Murphy is licensed under a Creative Commons Attribution-NonCommercial-NoDerivs 3.0
Unported License.