MATLAB Tutorial

ETH Zurich, Department of Biosystems Science and Engineering (D-BSSE)

Contents
1 Introduction 1.1 What is MATLAB ? . 1.2 Getting Started . . . . 1.3 Editor . . . . . . . . . 1.4 MATLAB Help . . . . 1.5 Other Important Info 1.6 Exercise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 3 3 3 5 5 6

2 Basic data types and operations 2.1 Variable names . . . . . . . . . . . 2.2 Numerical variables and operations 2.3 Logical variables and operators . . 2.4 String variables . . . . . . . . . . . 2.5 Vectors and matrices . . . . . . . . 2.6 Exercise . . . . . . . . . . . . . . . 3 Plotting basics 3.1 The plot command . . . . . 3.2 The hist command . . . . . 3.3 Multiple Plots and Subplots 3.4 Exercise . . . . . . . . . . . 4 MATLAB functions 4.1 Simple Built-in Functions 4.2 Exercise . . . . . . . . . . 4.3 MATLAB scripting . . . . 4.4 Dening a function . . . . 4.5 Exercises . . . . . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

. . . . . .

7 . 7 . 7 . 8 . 9 . 9 . 11 12 12 13 14 15 16 16 16 17 17 18 19 19 19 20

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

. . . . .

5 Flow control 5.1 Checking conditions: if, else, ifelse . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2 Iterations: for and while loops . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

6 Ordinary dierential equations in MATLAB 21 6.1 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22 7 Symbolic computations 7.1 Declaring symbols and symbolic expressions . . . . . 7.2 Substituting symbolic variables . . . . . . . . . . . . 7.3 Solving equations of symbolic variables . . . . . . . . 7.4 Dierentiating and integrating symbolic expressions 24 24 24 25 25

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

7.5

Exercises

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26 27 27 28 29

8 Optimization 8.1 Finding roots and minima . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.2 Optimization with constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1
1.1

Introduction
What is MATLAB ?

Programming assignments in this course will almost exclusively be performed in MATLAB , a widelyused environment for technical computing with a focus on matrix operations. The name MATLAB stands for MATrix LABoratory and was originally designed as a tool for doing numerical computations with matrices and vectors. It has since grown into a high-performance language for technical computing. MATLAB integrates computation, visualization, and programming in an easy-to-use environment, and allows easy matrix manipulation, plotting of functions and data, implementation of algorithms, creation of user interfaces, and interfacing with programs in other languages. Typical areas of use include: Math and Computation Modeling and Simulation Data Analysis and Visualization Application Development Graphical User Interface Development

1.2

Getting Started

Window Layout The rst time you start MATLAB , the desktop appears with the default layout, as shown in Figure 1. The MATLAB desktop consists of the following parts: Command Window: Run MATLAB statements. Current Directory: To view, open, search for, and make changes to MATLAB related directories and les. Command History: Displays a log of the functions you have entered in the Command Window. You can copy them, execute them, and more. Workspace: Shows the name of each variable, its value, and the Min and Max entry if the variable is a matrix. In case that the desktop does not appear with the default layout, you can change it from the menu Desktop Desktop Layout Default.

1.3

Editor

The MATLAB editor (Figure 2) can be used to create and edit Mles, in which you can write and save MATLAB programs. A m-le can take the form of a script le or a function. A script le contains a sequence of MATLAB statements; the statements contained in a script le can be run in the specied order, in the MATLAB command window simply by typing the name of the le at the command prompt. Mles are very useful when you use a sequence of commands over and over again, in many dierent MATLAB sessions and you do not want to manually type these commands at the command prompt every time you want to use them.

Use Tab to go to Workspace Browser

Get Help

View or change current Directory

View or Execute previously run commands from the Command History Window Click Start button for quick access to tools and more

Enter Matlab Functions at command line prompt

Figure 1: MATLAB Desktop (default layout)

You can run a script, or a function that does not require an input argument, directly from the Editor/Debugger either by pressing F 5 or selecting Save File and Run from the Debug menu. If you only want to run a part of a script, you can use the mouse to highlight the corrsponding lines in the m-le and press F 9. The results are shown in Command Window.

Figure 2: MATLAB Editor

1.4

MATLAB Help

MATLAB has an extensive built-in help system, which contains detailed documentation for all of the commands and functions of MATLAB . There are dierent ways to ask for help when using MATLAB : Command Line HELP: HELP FUN displays a description of and syntax for the function FUN in the Command Window (e.g., help plot). DOC: DOC FUN displays the help browser for the MATLAB function FUN (e.g. doc help). You can invoke the MATLAB help browser by typing helpbrowser at the MATLAB command prompt, clicking on the help button, or by selecting Start M AT LAB Help from the MATLAB desktop.

1.5

Other Important Info

MATLAB as well as a large number of associated toolboxes are available for students on the ETH campus and are installed on the machines in the IFW student computer pools (Windows and Linux). The campus license runs under Windows, Linux, and Mac OS X (cf. IDES [?] catalog for system

requirements). There is an excellent, but quite extensive documentation [?] available on the product web page as well as a number of built-in demos [?]. Numerous free tutorials exist on the web (see here for one with ETH avor [?]).

1.6

Exercise

OK, now its the time to play around a bit with MATLAB : 1. Start MATLAB . 2. Feel free to click around dierent segments in the MATLAB window, try resizing or closing some of them. 3. Now recover the desktop default layout, so that your MATLAB window contains the main features shown in Figure 1 again. 4. Change the working directory to Bioinformatics2010/your name. (Make the directories if needed!) 5. Try out the command window, the easiest way to use command window is to think of it simply as a normal calculator. 6. Done!

2
2.1

Basic data types and operations

There are some specic rules for how you can name your variables, so you have to be careful. Only use primary alphabetic characters (i.e., A-Z), numbers, and the underscore character in your variable names. You cannot have any spaces in your variable names, so, for example, using this is a variable as a variable name is not allowed (in general, you can use the underscore character to replace space in your variable name). MATLAB is case sensitive. What this means for variables is that the same text, with dierent combinations of capital and small case letters, will not be interpreted the same in MATLAB . For example, VaRIAbLe, variable, VARIABLE and variablE would all be considered distinct variables in MATLAB .

2.2

Numerical variables and operations

Variables are dened in MATLAB by usage, i.e. theres no explicit declaration section needed. Numerical variables are by default dened as double. The most basic operations with MATLAB are arithmetic operations. The basic arithmetic operators are +, , /, and (power). These operators can be used together with brackets (). As with all programming languages, special care should be given to how a mathematical expression is written. For example, 5 + 10/2 3 = 5 + (10/2) 3 = 20, which is not equal to 5 + 10/(2 3) = 5 + 5 . In general, MATLAB follows the order: 3 1. Parentheses ( ) 2. Exponentiation 3. Multiplication and division, , /, from left to right 4. Plus and minus, +, , from left to right For example: 3 + 5/2 4 2 3 + (5 2) = = = = = 3 + 5/2 4 2 3 + 10 3 + 5/2 4 8 + 10 3 + 2.5 4 8 + 10 3 + 10 8 + 10 15

MATLAB always stores the result of the latest computation in a variable named ans. To store variables for a longer time we can specify our own names: >> x = 5+2^2 x = 9 and then we can use these variables in computations: 7

>> y = 2*x y = 18 These are examples of assignment statements: values are assigned to variables. Each variable must be assigned a value before it may be used on the right of an assignment statement. If you do not want to see the result of a statement in the Command Window, which is typically the case for intermediate statements, we can terminate the line with a semicolon: >> x = 5+2^2; >> y = 2*x; >> z = x^2+y^2 z = 405 Notes: Other types of numerical variables can be dened explicitly if needed as:

char, single, int8, int16, int64, uint8, uint16, uint64. MATLAB is accurate but does not do exact arithmetic ! >> 3*(5/3 - 4/3) - 1 ans = 4.440892098500626e-16 MATLAB identies i, j, pi also as numbers and to some extent Inf and N aN : >> pi ans = 3.1416 >> (-1)^(0.5) ans = 0.0000 + 1.0000i >> log(0) ans = -Inf >> 0/0 ans = NaN

2.3

Logical variables and operators

A logical variable can be true or false, or one and zero in binary system. >>S = 2>4 S = 0 8

Boolean algebra can be done in MATLAB using the logical operators AND (&) , OR (|) and NOT (). For example given an arbitrary boolean variable S: >> S | ~S ans = 1 Note: MATLAB implicitly casts data types to avoid syntax errors: Logical values are converted to 1 and 0 when used as numbers. >> (2 < 3) * 3 ans = 3 >> true * 3 ans = 3 All numbers hold true boolean value when used in logical expressions, except for 0 itself which is false. >> false | -4.03 ans = 1

2.4

String variables

You can also assign strings of text to variables, not just numbers, with single quotes (not double quotes) around the text. For example: >> w = Goodmorning; >> w w = Goodmorning

2.5

Vectors and matrices

Matrices are not a type of data but they are n-dimensional arrays of basic MATLAB data-types. MATLAB treats all variables equally as matrices. Traditional matrices and vectors are two- and onedimensional cases of these structures, respectively, and scalar numbers are simply 1-by-1 matrices. A matrix is dened inside a pair of square brackets ([]). The punctuation marks comma (,), and semicolon (;) are used as a row separator and column separator, respectively. The examples below illustrate how vectors and matrices can be created in MATLAB . >> A = [1, 2, 3] A = 1 2 >> A = [1; 2; 3] A = 1 2 3 >> A = [1 2 3; 4 A = 1 2 4 5

5 6] 3 6 9

Simple matrices can also be created using functions such as: ones, zeros, >> A = eye(3) A = 1 0 0 0 1 0 0 0 1 >> A = diag([1, 2, 3]) A = 1 0 0 0 2 0 0 0 3

eye, diag. For example:

Concatenating matrices are straightforward MATLAB as long as their dimensions are consistent. For example: >> A = [1, >> C = [A, C = 1 >> C = [A; C = 1 4 2, 3];B = [4, 5, 6]; B] 2 B] 2 5 3 4 5 6

3 6

Colon operator The colon operator allows you to create vectors with a sequence of values from the start value to the stop value with a specied increment value. The increment value can also be negative or non-integer. For example: >> A = [-2: 2] A = -2 -1 0 >> A = [2: -1: -2] A = 2 1 0 >> A = [0.5: 1.5: 6] A = 0.5000 2.0000

-1

-2

3.5000

5.0000

Vector and matrix indexing Once a vector or a matrix is created you might needed to access only a subset of the data. This can be done with indexing. Each element of a matrix is indexed with the row and column of the element. The entry in the ith row and jth column is denoted mathematically by Ai,j and in MATLAB by A(i, j). Consider the matrix: >> C = [1 2 3;4 5 6;7 8 9;10 11 12] C = 1 2 3 4 5 6 10

7 10

8 11

9 12

the element on the 3rd row of the 2nd column can be accessed like: >> C(3, 2) ans = 8 The colon operator can also be used to access the whole or a set consecutive elements within a dimension of a matrix. Given the data matrix C from above we have: >> C(:, 3) ans = 3 6 9 12 >> C(2:3, 2:3) ans = 5 6 8 9

Matrix operations Basic matrix operations are straightforward in MATLAB : Addition: >> C = A+B; Subtraction: >> C = A-B; Transposing: >> C = A; Matrix multiplication: >> C = A*B; Element-wise multiplication: >> C = A .*B; Exponentiation: >> C = A^p; (where p is a scalar.) Element-wise exponentiation: >> C = A.^p; (where p is a scalar.) Note: Matrices must have compatible dimensions!

2.6

Exercise

1. Create a row vector v with values (1, 2, 3, 5, 11, 7, 13). 2. Change the value of the 5th and the 6th element of the v to 7 and 11 respectively. Try doing this with only one command as well. 3. Create a vector Five5 out of all multiples of 5 that fall between 34 and 637. By the way, how many are they? (Tip: look up the command length in help.) 4. Double click on the variable Five5 in the Workspace to inspect your results in the Variable Editor. 5. Can you use the colon operator to extract the numbers in the vector Five5 that are multiples of 5 only but not 10 ?

11

3
3.1

Plotting basics
The plot command

The most basic plotting command in MATLAB is plot. The plot function has dierent forms, depending on the input arguments. If y is a vector, plot(y) produces a piecewise linear graph of the elements of y versus the index of the elements of y. If you specify two vectors as arguments, plot(x,y) produces a graph of y versus x, i.e. the points ((x1 , y1 ), (x2 , y2 ), (x3 , y3 ), . . . ), are connected with lines.
1 0.8 0.6 0.4 0.2 0 0.2 0.4 0.6 0.8 1

Figure 3: Plot of the sine function

For example, these statements use the colon operator to create a vector of x values ranging from 0 to 2 , compute the sine of these values, and plot the result: >> x = 0:pi/100:2*pi; >> y = sin(x); >> plot(x,y) The result is show in gure 3. However, it is not always desirable to have the consecutive points to be connected to each other. In these cases the plot command can be used to draw scatter plots by other symbols instead of lines, for example with dots: plot(X, Y, .). In general it is possible to specify color, line styles, and markers (such as plus signs or circles) when you plot your data using the plot command: plot(x,y,color_style_marker) where color style marker is a string containing from one to four characters (enclosed in single quotation marks) constructed from a color, a line style, and a marker type. The strings are composed of combinations of the elements shown in gure 4. You can also edit color, line style, and markers interactively. See MATLAB help of Editing Plots for more information.

12

Figure 4: The plot command: Line styles and colours.

3.2

The hist command

Histogram plots are one of the other basic types of plots in MATLAB and can be produced using the hist function. The easiest form of using this function is hist(x), in which x denotes a vector. The command divides the range of the values in x to ten bins and plots the distribution of the element counts in each bin using bar plots. An additional scalar parameter can be added afterwards in order to tune the number of bins, for example the following commands produce 10, 000 normally distributed random numbers, plot their

13

histogram using 100 bins: >> x = randn(10000,1); >> hist(x, 100) The result is shown in Figure 5.
350

300

250

200

150

100

50

Figure 5: Histogram plot of 10,000 normally distributed data.

3.3

Multiple Plots and Subplots

You may also want to create more than one plot in the same gure. This can be done using the hold command. Normally when you use the plot command any previous plot in the gure is erased, and replaced by the new plot. However, if you rst type hold on in the command prompt, all new plots will be superimposed in the same gure. The command hold o tells MATLAB to return to the default mode (i.e., each new plot will replace the previous plot in the gure). For example: >> >> >> >> >> >> >> >> X = [1 3 4 6 8 12 18]; Y1 = 3*X; Y2 = 4*X+5; Y3 = 2*X-3; plot(X,Y1); hold on plot(X,Y2); plot(X,Y3);

You can also create multiple plots in the same gure, but with each of them in a separate subgure (i.e., each with their own axes). You can do this with the subplot command. If you type subplot(M,N,P) at the command prompt, MATLAB will divide the window into a bunch of subgures. There will be M rows and N columns of subgures and MATLAB will place the result of the next plot command in the Pth subgure (where the rst subgure is in the upper left). For example: 14

>> >> >> >> >> >>

subplot(3,1,1) plot(X,Y1) subplot(3,1,2) plot(X,Y2) subplot(3,1,3) plot(X,Y3)

3.4

Exercise

1. Initialize the random seed with command: >> S= RandStream(mrg32k3a); 2. Create a random vector u of 1, 000 uniformly distributed numbers. (Tip: use the rand command.) 3. Create a random vector n of 1, 000 normally distributed numbers. (Tip: use the randn command.) 4. Create a new plot window using the command Figure. 5. Plot the distributions and scatter plot of the elements in u and n in the gure window. (Three subplots in total.) 6. Make yourself familiar with the interactive plot editor, accessible from View Property Editor in the gure window menu bar. Give fancy labels to your axes and gures, so that they look scientic enough! 7. Save your gures with .eps extension in your folder, and close the gure. 8. Save your workspace to a le called My Workspace in your folder using save command. 9. Close the MATLAB , done!

15

MATLAB functions

MATLAB has an extremely wide range of built-in functions. Additionally, it is also simple to dene your own functions in MATLAB as any other programming languages.

4.1

Simple Built-in Functions

Here are some basic instances of the ready MATLAB functions: Trigonometric functions: sin, cos, tan, asin, acos, atan (their arguments should be in radians). Exponential: exp, log, log2, log10 Random number generator: rand, randn, randperm, mvnrnd Data analysis: min, max, mean, median, std, var, cov Linear algebra: norm, det, rank, inv, eig, svd, null, pinv Strings: strcmp, strcat, strfind, sprintf, sscanf, eval Files: save, load, csvwrite, csvread, fopen, fprintf, fscanf Other: abs, sign, sum, prod, sqrt, floor, ceil, round, sort, find Master key: help, doc, and the external function Google!

4.2

Exercise

1. Open MATLAB , and load the data le My Workspace from the previous section into MATLAB . (Tip: you can use the load button in the Workspace or the load command.) 2. Can you nd how many elements in n are larger than 0.5 ? 3. What is the index of the smallest element in in the vector n ? (Tip: check out the min function.) 4. Try calculating the value of x from the following expression: x=
1000 i=1 ui 1000 j=1 uj

The correct answer is obviously 1. Can you reach it? (Tip: try using prod function, and separately exp, log, sum functions. ) 5. Compute meanU as the average of the vector u. (Tip: use mean function.) 6. Find the location of the elements in u that are larger that 0.9 using the command >> I = find(u>0.9). 7. Set all of the elements in u with larger than 0.9 to zero. How much does the mean of u shrink ?

16

4.3

MATLAB scripting

To avoid writing the same code over an over again you can create so called m-les (they have the le extension .m). These can be either scripts or functions. The main dierences are: Scripts work like an extended command line statement. All variables are shared between the workspace and the script. Scripts have no input or output arguments. Functions have their own workspace. They are usually called with one or several input arguments and return one or several output arguments. They are declared by the keyword function. Creating MATLAB scripts is easy. Just call the MATLAB editor: >> edit example_script.m Now write your lines of code and save the le somewhere in the MATLAB search path (e.g. the current working directory) and call it from the command line by: >> example_script The script will be executed line-by-line.

4.4

Dening a function

You can easily extend the functionality of MATLAB by creating new functions. The general form of a function declaration is: function [return_args] = function_name(input_args) function body ... Thus, a MATLAB function has a name and optionally input and return arguments. The return arguments must be assigned somewhere in the function body. Example: the following function computes the product of two numbers a and b. function p = product(a,b) p = a*b; Saved under the le name product.m it can be called from the command line. >> x = product(3,4) x = 12 Example: this function sets to zero all values in the input vector which are below a given threshold. It returns the altered vector and also the indices of the replaced values. function [v,idx] = findBelowThresh(v,thresh) idx = find(v<thresh); v(idx) = 0; As a matter of convention, you should always call the m-le and the function by the same name. However, make sure that it does not conict with a name that is already taken by another function. Otherwise the original function is overwritten, i.e., cannot be called any more.

17

4.5

Exercises
n k

1. Create a function which returns the average of given input vector. 2. Create a function which computes the binomial coecient =
n! k!(nk)! .

3. Create a function which plots sin(x) for a given input frequency in the range 0 x 2. 4. Create a function which plots the surface dened by sin(1 x) cos(2 y) in the range 0 x, y 2. You can use the meshgrid function to create the x and y matrices.

18

5
5.1

Flow control
Checking conditions: if, else, ifelse

Conditional statements check a given expression and based on the outcome execute certain parts of the code. You can also check several conditions in a sequential order with elseif statements. if expression statements elseif expression statements elseif expression statements ... else statements end Example: the following function computes the factorial n! = n k by recursive function calls. k=1 function f = fac(n) if n == 0 f = 1; else f = n*fac(n-1); end

5.2

Iterations: for and while loops

Iterations are dened by for and while loops. The dierence between both is that for-loops have a xed number of cycles. While-loops stop until a certain condition is matched. The for loop has the general syntax: for variable = vector statements end Example: the following function computes the Fibonacci series. function f = fibonacci(n) f = zeros(n,1); f(1:2) = [1 1]; for l=3:n f(l) = f(l-2) + f(l-1); end The while loop format is while condition statements end Example: the following function sorts the elements of a vector in ascending order.

19

function v = gsort(v) pos = 2; while pos <= length(v) if v(pos)>v(pos-1) % move to the next position pos = pos + 1; else % swap the values foo = v(pos-1); v(pos-1) = v(pos); v(pos) = foo; if pos > 2 % decrease position pos = pos - 1; end end end For and while loops can be interrupted with the break statement. Generally, for and while loops should be avoided in MATLAB as they are very slow. Instead you should try to use built-in MATLAB functions or vectorize the code with the dot-operator.

5.3

Exercises

1. Extend the factorial function dened above, such that it checks for a positive input argument. 2. Write a function that calculates a random walk trajectory based on the iteration xi+1 = xi + , where is a uniformly distributed random variable between -1 and 1. Plot the resulting trajectory. 3. Change the sorting algorithm given above such that it sorts the elements in descending order. Compare the speed of your function with the built-in MATLAB function sort. 4. Approximation of . The area of a circle is given by A = r2 where r is the radius. Assume a circle with r = 0.5 embedded in a unit square. A point in the unit square is dened by p = (x, y) where 0 x, y 1. If we draw ntot times two uniformly distributed random numbers (x, y) and count how often the corresponding points falls into the circle (ncirc ), we get an approximation of the circle area by ncirc and by that of the number . Write a function which approximates ntot by the described method. How many draws do you need to get the rst three digits right?

20

Ordinary dierential equations in MATLAB

Dierential equations are the most common tool to model time and state-continuous dynamical processes. Many of these ODEs have to be analyzed numerically. MATLAB provides a powerful suite for the numerical integration of ODEs. We will focus on two types of ODE solvers: ode45 and ode15s. ode45 is the rst method of choice, but whenever the ODE is sti, i.e., the dynamics happen on very dierent time scales, you should use ode15s for integration. The general call of the two ODE solvers is [t,x] = ode45(@rhs_handle,time_span,x0,options,args) [t,x] = ode15s(@rhs_handle,time_span,x0,options,args) where rhs handle is a handle to the right hand side dening the ODE (see below). The vector dening the lower and upper integration boundaries is time span, the vector x0 denes the initial conditions. Options like relative and absolute error tolerances can be specied within the options structure, which is generated with the odeset function. However, they can also be omitted (with []) to use standard options. All additional parameters (args) are passed to the rhs handle le. As a rst example consider the logistic dierential equation, which is a model for the growth of populations. x(t) = rx(t)(K x(t))/K (1)

The time-dependent population density is x(t), parameter r is the growth rate and K is called the carrying capacity (Can you explain why?). In order to implement this equation in MATLAB we have to dene the right-hand-side of equation 1 as a MATLAB function. function dxdt = logistic(t,x,r,K) dxdt = r*x*(K-x)/K; We create this function using the MATLAB editor and save it in the current working directory under the le name logistic.m. Lets assume the following parameter values: r = 2, K = 10 and as initial conditions x0 = 0.1. We can now integrate the ODE for the time interval 0 t 10 and plot the results with the call [t,x] = ode45(@logistic,[0 10],0.1,[],2,10); plot(t,x) Note that we have left the options structure empty as we want to use the standard integrator options. If we want to get the solution x(t) for specic time points, e.g., at each 1/10 unit of t, we can set tspan = [0:0.1:10]. However, the internal step size of the integrator is chosen automatically and cannot be controlled with the tspan argument. Lets assume, we want to integrate the logistic equation for two dierent parameter sets K1 = 10, K2 = 5. We can do this of course sequentially by calling twice ode45 with two dierent parameter sets. But we can do this also in parallel by adding a second ODE to the right hand side function. function dxdt = logistic(t,x,r,K1,K2) dxdt = zeros(2,1); % column vector dxdt(1) = r*x(1)*(K1-x(1))/K1; dxdt(2) = r*x(2)*(K2-x(2))/K2; Thus, x and dxdt are now two-dimensional. Note, that we have to explicitly dene dxdt as a column vector because MATLAB expects this format for the return argument. Lets set both initial conditions two 21

12 10 8 x(t) 6 4 2 0

4 Time t

10

Figure 6: Solution of the logistic dierential equation with r = 2, K = 10 and x0 = 0.1.

x0 = [0.1 0.1]; We integrate the ODEs and plot both solutions with [t,x] = ode45(@logistic,[0 10],x0,[],2,10,5); plot(t,x(:,1),-k,t,x(:,2),-r)

12 K1 10 8 x(t) 6 4 2 0 K
2

4 t

10

Figure 7: Two solutions of the logistic dierential equation with K1 = 10 and K2 = 5.

6.1

Exercises
x(t) = r(sin(t) x(t)) 22

1. Implement the following non-automomous ODE in MATLAB .

Simulate the ODE with r = 0.01 and r = 100 on the interval 0 t 100. Plot your solution x(t) together with sin(t). 2. The Lotka-Volterra model describes the dynamics of two interacting populations consisting of predator (x) and pray (y). x(t) y(t) = ax(t)y(t) bx(t) = cy(t) dx(t)y(t)

Implement the Lotka-Volterra model in MATLAB and simulate it with ode15s using as parameter values a = 1, b = 1, c = 1.5, d = 1 and as initial conditions x0 = 2, y0 = 5. Plot x(t) and y(t) as well as the y(x). Hint: you can plot the phase space x(t) vs. y(t).

23

Symbolic computations

Symbolic computations allow you to calculate mathematical operations analytically without assigning any numerical values.

7.1

Declaring symbols and symbolic expressions

In order to distinguish regular variables that can take on numerical values from symbolic variables, you have to declare them as symbols before hand. >> syms x y; >> x+x+y ans = 2*x + y Note, that you can perform symbolic calculation also with natural numbers if you declare them as symbols. >> sym(2/5) + sym(1/3) ans = 11/15 which is in contrast to: >> 2/5 + 1/3 ans = 0.7333 You can create symbolic expressions by combining previously declared symbols within mathematical expressions. For example to dene the quadratic function f (x) = a x2 + b x + c just type: >> syms a b c x; >> f = a*x^2 + b*x + c f = a*x^2 + b*x + c

7.2

Substituting symbolic variables

The subs command can be used in order to substitute symbolic variables either by other variables or by numerical values. E.g., the following command substitutes x by 2.3 in the above expression for f: >> subs(f,x,2.3) ans = (529*a)/100 + (23*b)/10 + c 24

You can also substitute variables by expressions, as in the following example. >> g = y/pi; >> subs(f,x,g) ans = (a*y^2)/pi^2 + (b*y)/pi + c

7.3

Solving equations of symbolic variables

The command solve can be used to solve a symbolic expression S for a given variable for which S is zero. For example, if you apply the solve command to the previous symbolic expression f you get the ! solutions of the quadratic equation f (x) = a x2 + b x + c = 0. >> solve(f,x) ans = -(b + (b^2 - 4*a*c)^(1/2))/(2*a) -(b - (b^2 - 4*a*c)^(1/2))/(2*a) The solve command can also be used to solve multiple expression simultaneously. >> syms x y u v >> [x y] = solve(x + 2*y - u, 4*x + 5*y - v) x = (2*v)/3 - (5*u)/3

y = (4*u)/3 - v/3

7.4

Dierentiating and integrating symbolic expressions

The diff and int commands perform symbolic dierentiation and integration, respectively. >> syms x y; >> f = sin(x)^2 + cos(y)^2; >> diff(f, y) ans = (-2)*cos(y)*sin(y) >> int(f,y) ans = sin(2*y)/4 + y*(sin(x)^2 + 1/2) 25

Partial derivatives and mixed derivatives can be generated in a similar manner. For example >> diff(diff(f, y),y) ans = 2*sin(y)^2 - 2*cos(y)^2 while
2f yx

2f y 2

is:

is:

>> diff(diff(f, y),x) ans = 0 In the example above, we calculated the indenite integral of f with respect to y. If you want to calculate the denite integral, i.e., the integral between two given integration boundaries, you have to 2 pass these as additional arguments. For example, the denite integral 0 f (x, y)dy can be calculated as >> int(f,y,0,2*pi) ans = pi*(2*sin(x)^2 + 1) MATLAB provides many more tools for symbolic computation, consult the documentation! However, symbolic computation is not the primary aim of MATLAB and many people prefer to use MATHEMATICA or MAPLE instead. We will encounter symbolic computation in the context of sensitivity analysis and parameter estimation.

7.5

Exercises
f (x, y) = a + b exp((x mx )(y my ))

1. Find the maximum of the following function by symbolic computation.

2. Write a function which calculates the n-th derivative of an input function with respect to a certain argument of that function. Both, the input function and the function argument should be passed as strings (instead of symbols or symbolic expressions). Note: you can use the findsym function to search for all symbols within the input strings. 3. The MATLAB inline function allows you to create a function from a string. For example, >> fun = inline(exp(t)*sin(w*x)) fun = Inline function: fun(t,w,x) = exp(t)*sin(w*x) creates the inline function fun(t,w,x) with three input arguments. Extend your symbolic derivative function such that it also returns the derivative as an inline function.

26

Optimization

Optimization plays a central role in parameter estimation. Whenever you want to nd a parameter set that leads to an optimal t of your model to some measurements, you will have to employ some kind of optimization procedure.

8.1

Finding roots and minima

Assume you want to nd values for the arguments of a function for which it is zero. These are called the roots of a function. The fzero function can be used to nd the root of a continuous function of one variable. You can either specify your own function x = fzero(@myfun,x0); where myfun is an M-le function that you dened before hand, or you specify an inline function (in MATLAB this is called an anonymous function) such as: x = fzero(@(x)sin(x*x),x0); Beside an input function, fzero requires a starting value, from where the search for the root starts. For example, the nearest roots of sin(x2 ) starting from x0 = 2 and x0 = 3 are: >> x = fzero(@(x)sin(x*x),2) x = 1.7725 >> x = fzero(@(x)sin(x*x),3) x = 3.0700 Note, that fzero only nds a root if your function is continuous and crosses the x-axis at the root (i.e. changes sign). The minimum of a continuous function of one variable is computed by the fminbnd function: x = fminbnd(fun,x1,x2) where x1 and x2 are the boundaries within which the minimum is attained and fun is a function handle to an M-le function or a build in MATLAB function. For example, >> x = fminbnd(@cos,3,4) x = 3.1416 nd a numerical approximation to . Similarly the minimum of an multi-variable, scalar function is computed by the fminsearch function. x = fminsearch(fun,x0)

27

As an example consider the function f (x, y) = ax2 + by 2 which has a uniquely dened minimum at (x, y) = (0, 0) for a, b > 0. To solve for the minimum we need to dene the function in a separate M-le. function f = myfun(x,a,b) f = a*x(1)^2 + b*x(2)^2; Now the call >> x = fminsearch(@(x) myfun(x,1,2),[0,1]) x = 1.0e-03 * 0.9298 0.0312

yields a numerical approximation to the true solution (starting at the value (x, y) = (0, 1) and setting a = 1, b = 2). The result might come as a surprise, since we already started a the correct value of x and end up wit a deviation in the order of 103 . If we increase the default accuracy from 103 to 108 the deviation from the true value decreases accordingly. >> opt = optimset(TolX,1e-8); >> x = fminsearch(@(x) myfun(x,1,2),[0,1],opt) x = 1.0e-08 * 0.1105 0.1613

The optimset function allows to adjust many more optimization settings. Consult the documentation for details!

8.2

Optimization with constraints

Often the objective should be attained under a certain set of constraints, e.g., that parameters are positive. The fmincon function allows for equality as well as inequality constraints during the optimization. It is a gradient-based method, that is designed to work on problems where the objective and constraint functions are both continuous and have continuous rst derivatives. Consider the following example: arg max f (x, y)
x,y

= x2 + y 2 1 > 0 > 0

(2) (3) (4) (5)

x+y x y

Note that maximizing f (x, y) is equivalent with minimizing f (x, y) (minimization is the default of fmincon). The inequality constrains can be written in matrix form Av b where 1 1 0 , v = (x, y)T A = 1 b = (1, 0, 0)T . (6) 0 1 28

You can nd one solution to the problem by calling fmincon as: >> res = fmincon(@(x) -sum(x.^2), [0.2 0.7],[1 1; -1 0; 0 -1],[1 0 0]) res = 0 1

8.3

Exercises
x x ) 2 0.02 + x2

1. Assume the following ODE. x = x(2 (7)

Use the fzero function to nd steady state solutions, i.e., roots of x. Compare your results by plotting the function in the range 0 x 5. 2. Find the maximum of the following equation under the given constrain. arg min f (x, y)
x,y

= x + y,

x2 + y 2 = 1

(8)

Note: nonlinear constraints have to be code in a separate function: [ineq, eq] = confun(x) ineq = ... % can be empty if not specified eq = ... Consult the documentation how to call fmincon in this case. 3. Previously, you have encountered the logistic dierential equation. Its analytic solution is: x(t) = K 1 + (K/x0 ) exp(rt)

Assume you have the following data set to estimate the unkown parameters K and r. Time Data 0 0.16 1 0.59 2 1.75 3 4.49 4 7.14 5 7.78 6 8.68 7 8.77 8 8.67

A common way to nd an optimal parameter set is by minimizing is the sum of squared deviations of the solution x(ti ) from the data points Di .
n

arg min S(K, r) =

(x(ti , K, r) Di )2

For simplicity we will assume that x0 = D1 . Implement the function to be minimized. Call the fminsearch function to nd a minimum. Use K = 1, r = 1 as starting conditions. Plot your nal solution together with the data points. How stable is your nal estimate to dierent starting conditions for r and K? 29

Extension No. 1: Now consider the initial condition x0 also as an unkown parameter. Change your code accordingly. Does the estimation improve or worsen if x0 is also unkown (i.e. to be estimated)? Extension No. 2: Each data point is subject to a certain observational error. You can include this information by minimizing the scaled squared deviation:
n

arg min S(K, r) =

2 (x(ti , K, r, x0 ) Di )2 /i

(9)

where i is the standard deviation of the mean Di . Adopt your parameter estimation code to include information on observational errors. Assume the following data: Time Mean STD 0 0.16 0.1 1 0.59 0.12 2 1.75 0.2 3 4.49 0.46 4 7.14 0.72 5 7.78 0.75 6 8.68 0.92 7 8.77 0.89 8 8.67 0.91

Plot the data with the MATLAB function errorbar. Additionally, plot the standard deviation over the mean. What do you observe? Estimate the parameters K, r and x0 from the data. Does the t improve when taking the errors of the measurement into account?

30

References:
For ETH MATLAB licences contact: http://ides.ethz.ch Ocial MATLAB page: http://www.mathworks.com/ User community: http://www.mathworks.com/matlabcentral/ User contributed packages: http://www.mathworks.com/matlabcentral/fileexchange/ Parts of this tutorial were taken from the ocial MATLAB documentation available online at http://www.mathworks.com/help/techdoc/.

31