Next: Lab 2 - Dispersal Up: Spatio-Temporal Models in Ecology; Previous: Contents

Subsections

Lab 1 - Introducing MATLAB

Goals

The goals of this lab are to introduce students to important MATLAB commands and philosophies, including:
• Basic graphical operations for plotting data in one and two dimensions.

• Defining grids.

• MATLAB's Help facility.

• Special mathematical and logical functions.

• Matrix manipulations (Demos).

• Implementing the Fast Fourier Transform (FFT) in MATLAB (Demos).
MATLAB is built around matrix and vector manipulation, and views everything as vectors or matrices or tensors. It is not a symbolic program, and performs no analytic calculations, but is composed of computationally optimized linear algebra routines for performing operations on and to matrices and vectors.

Introducing the Software

Note: At any time during a MATLAB session you may type help COMMAND for MATLAB's internal information on how COMMAND may be used, including examples. If you do not know what command you are looking for, you may type lookfor SUBJECT to get MATLAB to search for commands which refer to your SUBJECT.

EXERCISE 1: Type

     help linspace

     lookfor space

to see what other relevant commands there might be.

Defining One-Dimensional Functions and Graphing

Let's begin by graphing some familiar one-dimensional functions using MATLAB . The first step is to make a vector corresponding to the real axis, say between -5 and 5:

     x = linspace(-5,5,101);

The above command generates a vector with 101 components, named x', whose elements are evenly spaced numbers beginning at -5 and ending at 5. The semicolon (;) suppresses output - try the command without it and see what happens! Notice also that we get a spacing of = 1/10 using the above command - the 101 allows us to have the endpoints included in the vector of coordinates.

Now define a function, , which we will then plot. The function will be a vector, f, which contains for every component of the coordinate vector, x, a value according to the rule :

     f = sin( x.^2 )./(1 + x.^2);

The notation .' before an operation means to apply the operation element-by-element to an array; thus

x.^2

means the element by element square of x, or a vector of length 101 with values for between -5 and 5 (Look in the Getting Started with MATLAB materials for more information on how basic arithmetic operations are implemented). Now, to plot f issue the MATLAB command
     plot(x,f)

which will plot the ordered pairs given by x,f. Try changing the order of x and f above to see what happens.

Now, to see how to put multiple plots on one graph and change some of the styles, try the following:

     plot(x,f,'r*')
hold on, plot(x,1./(1+x.^2),'g'), hold off
xlabel('x'),ylabel('f(x)'),title('Function and Envelope')

The hold' commands force MATLAB to draw plots without erasing previous plots. One may also plot several functions simultaneously, as in
     figure(2)
plot(x,cos(x),'r',x,exp(-abs(x)),'b',x,sin(x.^2),'g')

Now, just to see what some of the options would be, type
     help plot

which will give you a number of examples, sets of options, and some suggestions for other commands you might find useful. Try some of these just to get the hang of it.

EXERCISE 2: Plot the fundamental solution of the heat equation,

which has the form

for the following values: 1, 2, 4. Plot all on the same set of axes, over an interval from -10 to 10, for a value of .37, and organize the time slices in the colors of the MATLAB spectrum ('r','y','g','b','m'). Label the axes and give the plot a title. The legend command allows you to label the different graphs - use help and check it out!

Scalar Functions of Two Variables

Defining functions of two variables works with the same philosophy in MATLAB , except that the underlying substrate of independent variables becomes a matrix as opposed to a vector. Let's begin by defining two coordinates, x and y and then a two-dimensionally extended version of these coordinates:

     x=.1*[0:100];
y=.05*[-50:50];
[X,Y]=meshgrid(x,y);

(Notice the scalar multiplication of a spacing, either .1 or .05 above, and an array of numbers, [N1:N2], which is all of the integers between N1 and N2, inclusive. Thus, x=.1*[0:100] is equivalent to x=linspace(0,10,101).). Now we may define a function of these two variables analogously to the one-dimensional procedure. For example, to plot the function

as a surface, we could use the sequence of commands
     g = sin(X.^2+pi*Y)./(1+X.^2+Y.^2);
surf(X,Y,g)

Can you tell which is the and which the direction? Now let's change the location of the grid. Use the up arrow, , to access previous commands; to access the previous command starting with x=' type x=' on the command line and then hit . Now you can use the back and right to move through the previous command. Typing a character will insert it; backspace removes it. Call up the previous command defining the grid for x and modify it so that x is now an array corresponding to coordinates beginning at -5, extending to +5, and with a grid spacing of .1. Recreate the meshgrid for X, Y, redefine the matrix g whose values are given by the rule above, and plot the result using
     surf(X,Y,g),shading flat

The shading flat option removes the grid lines and shades each portion of the graph as a flat polygon. You can change the color mappings by trying things like colormap hot, and on a computer screen things are more striking if you issue the command whitebg('k') (which makes the background color black, or 'k'). To see a particular slice of the graph, that is a section in either the x or y direction, you can plot a particular row or column of g. For example, to plot a cross section in the direction, with , type
     figure(2),plot(x,g(1,:))

The colon (:) in the second argment of g says all available indices,' and in this case g(1,:)' means From the matrix g take the first row and all columns,' which is, of course the vector containing samples taken from g in the x direction at the smallest y value (which has index 1). Use hold on, hold off to plot some other cross sections of the function until you feel comfortable with which indices correspond to which variables.

Now let's plot contours and densities in another window. Try

     figure(3), pcolor(X,Y,g), shading flat, colormap hot

This will create a new figure, put a density plot of g in it, and color-code the densities in such a way that lower values are dark (cool) and higher densities are white (hot). It is possible to super-pose contours on this graph:
     hold on, contour(X,Y,g,'b'), hold off

This will place 20 aesthetically chosen contours in blue on the color density plot.

EXERCISE 3: The function

is a solution to the wave equation, , which models, among other things, waves on the surface of a canal, in this case moving at speed 2 in either direction. Define a meshgrid in and which will allow you to visualize this solution for times between 0 and 4. Use surface plots, density plots, and contours to visualize the behavior of this solution. Also, in a third figure plot a waterfall' diagram, that is several slices in constant time, varying x, which will illustrate the evolution of the function. Use times 0,1,2,3,4 and color-code the slices in some way so that you know which slice is which.

Basic Logical Functions

One of the nice features of MATLAB is the existence of logical matrices, which are just arrays with elements that are either zero or one depending on whether a statement is true or false. Let's return to x and f given in the first section above. Re-input these into MATLAB (remember !) and now let's manipulate using logical matrices. For example, suppose that we would like to plot only the positive portions of f. Define


fplus=( f>=0 );

The array fplus is exactly as long as f (that is 101 elements) but contains zeros where f0 and ones where f0. To plot f in yellow and only its postive part in little red circles,
     plot(x, f, 'y', x, fplus.*f, 'ro')

By multiplying f and fplus element by element we have created a new entity which is the positive part of f where f is positive and zero where f is negative. If we wanted to show the negative part of f in little blue triangles on the same graph, we could
     hold on, plot(x, (1-fplus).*f,'b^'),hold off

These logical matrices can also be used to create piece-wise functions, which we will see next.

Suppose you wanted to make a function which was Gaussian for two standard deviations (), and then uniform (with value ) for two standard deviations, and then zero, and then that you wanted to re-normalize it so that the resulting function still represented a probability density function (pdf). Pick a standard deviation of .5; a normal pdf with standard deviation .5 is given by

First let's define the space and normal vectors:
     x = 6/100*[-50:50];
N = sqrt(2/pi)*exp(-2*x.^2);

Now we need logical matrices for the different portions of the pdf we are building. These we will call norm' and uni,'
     norm=(abs(x)<=1);
uni=(abs(x)>1 & abs(x)<=2);

The MATLAB command abs(x) returns the absolute value of x; the statement (abs(x)<=1)' tests whether or not the absolute value of x is smaller or equal to 1, returning the value 1' if it is, and 0' if it is not. The arrays norm and uni will thus contain only ones where the pdf will be normal and then uniform, respectively. Now we can create a vector, pdf, which will have the proper shape but will not have unit integral:
     pdf=N.*norm + sqrt(2/pi)*exp(-2)*uni;

You should plot this to see that it is correct. Finally we must normalize. We have defined a grid with spacing 6/100=.06, so a quick approximation of the actual integral would be
     mass=.06*sum(pdf)

The command sum adds up all the elements of a vector input, so that the above is equivalent to a simple Reimann-sum approximation to the integral. A better approximation would come from the trapeziod rule
     mass=.06*trapz(pdf)

which sums the average value in each cell. Now we can normalize,
     pdf=pdf/mass;

Now plot this. You can use up-arrows to just find the command you used last time, or you can compare it to the previous (unnormalized) function by using hold on and hold off.

EXERCISE 4: A function which we will see again in dispersal is the steady-state solution to the diffusion-loss model with a localized source at the origin,

the steady-state ( ) solution of which is

Unfortunately this predicts a small, but perceptible presence of a diffusing particle infinitely far away from the origin. One way to address this would be to chop off the function at some realistic point and then normalize the resulting function. Let and and truncate the function at 50, 25, and 10 percent of its peak value, then re-normalize so that the resulting function is a pdf. On a single graph, compare these three pdfs to a (normalized) version of the original function.

Learning about Matrix Manipulation and Fourier Transforms using the Demos

Now that you have a MATLAB basis, you might find it useful to tour through some of MATLAB 's cababilities using demos. The demos are a set of GUI slide shows which illustrate various MATLAB functions. So, on the command line type

     demos

and a window should pop up on screen with a variety of MATLAB subjects. To get comfortable with the demos and see some of how MATLAB handles linear algebra click on Matrices in the left hand window, followed by Basic matrix operations in the right hand window, and then Run Basic matrix... on the lower-right hand corner button. A new window will appear (the Slideshow Player) which you can use then to page through slides which illustrate various MATLAB matrix operations, which are useful to know. You can actually type the commands appearing in the slide show in your command window to see how they work, which is probably the best way to learn. Pay particular attention to how to set up matrices and find eigenvalues.

EXERCISE 5: The Nicholson-Bailey model for host-parasite interactions is

The steady state corresponding to coexistence of host and parasite in this model is given by

The Jacobian matrix (found by hand, not MATLAB !) is

Show that when (corresponding to the interaction between a greenhouse whitefly and its chalcid parasitoid) that the steady state is unstable. At the very least you will need to take the eigenvalues of a 2x2 matrix and to see if any of these eigenvalues have magnitude greater than one. Can you think of a way to plot a stability diagram in which will illustrate that these equilibrium populations are never stable?

In the rest of the course we will be using the Fast Fourier Transform (FFT) to evaluate the consequences of dispersal. You can use the demos facility to start learning about FFTs and their implementation in MATLAB . To begin with, click Numerics in the left-hand window and then Fast Fourier Transform in the right hand window, followed by clicking on Run Fast Fourier... on the button in the lower right hand corner. You can now page through slides which illustrate how one can use the FFT to analyze sunspot data. I suggest typing

     help fft

first to get a idea of what the FFT is all about, if you have never used Fourier transforms for data analysis. We will be using FFTs a lot, so don't despair if these seem a little cryptic just now. Remember, if you want to see what any of the component commands in the demo do, you can always type help SUBJECT in the command window and it will almost certainly give you either no information or more information than you really need.

Congratulations! You have `jumped in the deep end' with MATLAB , but you didn't sink!

Next: Lab 2 - Dispersal Up: Spatio-Temporal Models in Ecology; Previous: Contents   Contents
James Powell
2001-04-06