Curve Fitting and Parameter Estimation

Glenn Lahodny Jr.

Spring 2015

Least Squares Regression

The first step of the modeling process often consists of simply looking at data graphically
and trying to recognize trends. In this section, we will study the most standard method of
curve fitting and parameter estimation, least squares regression.
Example 1.1. Suppose that eBay hires us to predict its net income for the year 2003, based
on its net income for 2000, 2001, and 2002 (see Table 1.1).
Year
Net Income
2000 $48.3 million
2001 $90.4 million
2002 $249.9 million
Table 1.1: Annual net income for eBay.
We begin by simply plotting this data as a scatterplot of points. In MATLAB, enter the
following commands to obtain Figure 1.1.
>>
>>
>>
>>
>>
>>

Year=[0 1 2];
Income=[48.3 90.4 249.9];
plot(Year,Income,ro)
xlabel(Years Since 2000)
ylabel(Income (in Millions))
axis([-0.5 2.5 25 275])

Our first approach toward predicting eBays future profits might be to simply find a
curve that best fits this data. The most common form of curve fitting is linear least squares
regression.

1.1

Polynomial Regression

In order to develop an idea of what we mean by best fit in this context, we begin by trying
to draw a line through the three points of Example 1.1, in such a way that the distance
between the points and the line is minimized.
1

Net Income (in Millions)

250

200

150

100

50

-0.5

0.5

1.5

2.5

Years Since 2000

Figure 1.1: eBay net income by year

Net Income (in Millions)

250

200

150

100

50

-0.5

0.5

1.5

Years Since 2000

Figure 1.2: Linear least squares curve

2.5

Labeling our three points (x1 , y1 ), (x2 , y2 ), and (x3 , y3 ), we observe that the vertical
distance between the line and the point (xi , yi ) is given by the error Ei = |yi mxi b|. The
idea behind the least squares method is to sum these vertical distances and minimize the total
error. In practice, we square the errors to keep them positive and to avoid possible difficulty
with differentiation (due to the absolute values), which will be required for minimization.
Our total least squares error is
n
X
E(m, b) =
(yk mxk b)2 .
k=1

We note that instead of these vertical distances, we could also use horizontal distances
between the points and the line. Alternatively we could use direct distances (the shortest
distances between the points and the line). While either of these methods can be carried out
for a best fit line, they become considerably more complicated in the case of more general
curves. In the case of a parabola, for example, a point would have two different horizontal
distances from the curve, and while it could only have one shortest distance to the curve,
computing that distance would be a fairly complicated problem.
Returning to our example, our goal is not to find values of m and b that minimize the
error function E(m, b). In order to maximize or minimize a function of multiple variables,
we compute the partial derivatives with respect to each of the variables and set them equal
to zero. Here, we compute
n
X
E
xk (yk mxk b) = 0,
= 2
m
k=1
n

X
E
(yk mxk b) = 0,
= 2
b
k=1
which can be solved as a linear system of two equations for the two unknowns m and b.
Rearranging the terms and dividing by 2 gives
n
n
n
X
X
X
m
x2k + b
xk =
xk y k ,
k=1
n
X

k=1
n
X

xk + b

k=1

k=1
n
X

1 =

k=1

yk .

k=1

Pn

P
Since k=1 1 = n, we multiply the second equation by n1 nk=1 xk and subtract it from the
first to get the relation

!2
! n
!
n
n
n
n
X
X
X
X
X
1
1
xk =
xk
m
x2k
xk y k
yk .
n
n
k=1
k=1
k=1
k=1
k=1
or

n
X

m=

n
X

1
xk y k
n
k=1

!
xk

k=1

k=1

n
X

1
x2k
n
k=1
3

n
X

n
X
k=1

!2
xk

!
yk
.

Finally, substituting m into the one of the previous equations yields

! n
!
n
n
X
X
1 X
!
xk
yk
xk yk
n
n
n k=1
X
1X
k=1
k=1
b =
yk
xk
!2
n
n
n k=1
X
X
1
k=1
x2k
xk
n
k=1
! n
!
! n k=1!
n
n
X
X
X
X
yk
x2k
xk
xk yk
=

k=1

k=1

n
X

k=1

x2k

n
X

k=1

k=1

!2

xk

k=1

By continuity, there must be at least one local minimum for E. Since m and b are uniquely
determined, these values yield the minimum error.
Observe that we can proceed similarly for any polynomial. For second degree polynomials
of the form y = a0 + a1 x + a2 x2 , the error becomes
E(a0 , a1 , a2 ) =

n
X
(yk a0 a1 xk a2 x2k )2 .
k=1

In this case, we must compute a partial derivative of E with respect to each of the three
parameters, and consequently solve three linear equations for the three unknowns.
The MATLAB command for polynomial fitting is polyfit(x,y,n), where x and y are vectors
representing the data and n is the degree of the polynomial. For the eBay data, we have
>> polyfit(Year,Income,1)
ans =
100.8000 28.7333
Note that, left to right, MATLAB returns the coefficient of the highest power of x first,
the second highest power of x second, etc., continuing until the y-intercept is given last.
Alternatively, for polynomial fitting up to degree 10, MATLAB has the option of choosing
it directly from the graphics menu. In the case of our eBay data, while Figure 1 is displayed
in MATLAB, we choose Tools, Basic Fitting. A new window opens and offers a number
of fitting options. We can easily experiment by choosing the linear option and then the
quadratic option, and comparing. (Since we only have three data points in this example,
the quadratic fit necessarily passes through all three points. This, of course, does not mean
that the quadratic fit is best, only that we need more data.) For this small a set of data, the
linear fit is safest, so select that one and click on the black arrow at the bottom right corner
of the menu. Checking that the fit given in the new window is linear, select the option Save
to workspace. MATLAB saves the polynomial fit as a structure, which is a MATLAB array
variable that can hold data of varying types. The elements of a structure can be accessed
through the notation structurename.structureelement. Here, the default structure name is fit,
and the first element is type. The element fit.type contains a string describing the structure.
4

The second element of the structure is fit.coeff, which contains the polynomial coefficients of
our fit. Finally, we can make a prediction with the MATLAB command polyval(P,X). This
command returns the value of a polynomial P evaluated at X. In particular, P is a vector
of length n + 1 whose elements are the coefficients of the polynomial in descending powers.
For the eBay example, we have
>> polyval(fit.coeff,3)
ans =
331.1333
We obtain the prediction 331.1333 or approximately $331.13 million in net income. Finally,
we mention that MATLAB refers to the error for its fit as the norm of the residuals, which
is precisely the square root of E as weve defined it.
Example 1.2. (Crime and Unemployment) Suppose we are asked to model the connection
between unemployment and crime in the United States during the period 19942001. We
might suspect that in some general way increased unemployment leads to increased crime,
but our first step is to collect data. First we contact the Federal Bureau of Investigation
and study their Uniform Crime Reports (UCRs), which document, among other things,
the United States crime rate per 100,000 citizens. Next, we contact the U.S. Bureau of
Labor and obtain unemployment percentages for each year in our time period. The data is
summarized in Table 1.2.
Year Crime Rate
1994
5,373.5
1995
5,277.6
1996
5,086.6
1997
4,922.7
1998
4,619.3
1999
4,266.8
2000
4,124.8
2001
4,160.5

Percent Unemployment
6.1%
5.6%
5.4%
4.9%
4.5%
4.2%
4.0%
4.8%

Table 1.2: United States crime rate and unemployment data.

As in Example 1.1, we begin by looking at a scatterplot of the data (see Figure 1.3).
Certainly the first thing that we observe about the scatterplot is that their does seem
to be a distinct connection: as unemployment increases, crime rate increases. In fact, aside
from the point for 2001, the trend appears fairly steady. In general, we would study this
point at the year 2001 very carefully and try to deterine whether this is an anomaly or a
genuine shift in the paradigm. For the purposes of this example, however, were going to
treat it as an outlyer a point that for one reason or another does not follow an otherwise
genuinely predictive model. The important point to keep in mind is that discarding outlyers
when fitting data is a valid approach, as long as you continue to check and reappraise your
model as future data becomes available. Discarding the outlyer, we try both a linear and
quadratic fit, each shown in Figure 1.5.
5

5800

Crime Rate per 100,000 Individuals

5600
5400
5200
5000
4800
4600
4400
4200
4000
4

4.5

5.5

6.5

Unemployment Percentage

Figure 1.3: Scatterplot of crime rate versus unemployment percentage in the U.S.

5800
Data
Linear
Quadratic

Crime Rate per 100,000 Individuals

5600
5400
5200
5000
4800
4600
4400
4200
4000
4

4.5

5.5

6.5

Unemployment Percentage

Figure 1.4: Best fit curves for crime and unemployment data.

Clearly the quadratic fit is better, and though we have not yet quantitatively developed
the socioeconomic reasons for this particular relation, we do have a genuinely predictive
model that we can test against future data. For example, the unemployment percentage in
2002 was 5.8%, which our quadratic model would predict should be associated with a crime
rate of 5,306.2 crimes per 100,000 individuals. The actual crime rate for 2002 was 4118.8.
At this point, we are led to conclude that our model is not sufficiently accurate. In this
case, the problem is most likely a lag effect. While a short-term rise in the unemployment
rate does not seem to have much effect on crime, perhaps a sustained rise in unemployment
would.

1.2

Alternative Methods of Curve Fitting

Although least squares regression is the most popular form of basic curve fitting, it is not
the only method that can be applied. We could also think, for example, of defining our error
as the maximum of all the vertical distances between the points and our curve. That is, for
n data points, we could define
E = max |yk f (yk )|.
k=1,...,n

This is called the Chebyshev method. We could also sum the unsquared absolute values,
E=

n
X

|yk f (xk )|,

k=1

or take any other reasonable measure.

1.3

Regression with More General Functions

Example 1.3. Yearly temperature fluctuations are often modeled by trigonometric expressions, which lead to more difficult regression analyses. Here, well consider the case of
monthly average maximum temperatures in Big Bend National Park. In Table 1.3 the first
column lists raw data of average maximum temperatures each month. In order to model
this data with a simple trigonometric model, well subtract the mean (which gives Column
3) and divide by the maximum absolute value (which gives Column 4) to arrive at a column
of dependent variables that vary like sine and cosine between -1 and 1.
A reasonable model for this data is T (m) = sin(m a), where T represents scaled temperatures and m represents a scaled index of months (m [0, 2]), where by our reductions
weve limited our analysis to a single parameter, a. Proceeding as above, we consider the
regression error
n
X
E(a) =
(Tk sin(mk a))2 .
k=1

To minimize the error, we set the derivative E 0 (a) equal to zero which gives
n
X
dE
=2
(Tk sin(mk a)) cos(mk a) = 0,
da
k=1

Month Average Max Temp. Minus Mean

Jan
60.9
-18.0
Feb
66.2
-12.7
Mar
77.4
-1.5
Apr
80.7
1.8
May
88.0
9.1
Jun
94.2
15.3
Jul
92.9
14.0
Aug
91.1
12.2
Sep
86.4
7.5
Oct
78.8
-0.1
Nov
68.5
-10.4
Dec
62.2
-16.7

Scaled
-1
-0.71
-0.08
0.10
0.51
0.85
0.78
0.68
0.42
-0.01
-0.58
-0.93

Table 1.3: Average maximum temperatures for Big Bend National Park.
a nonlinear equation for the parameter a. Though nonlinear algebraic equations are typically
difficult to solve analytically, they can be solved numerically.
In this case, we will use MATLABs fzero(FUN,X0) function which tries to find a zero
of the function FUN near X0, if X0 is a scalar. First, we write a script M-file that contains
the function were setting to zero, listed below as bigbend.m.
% Glenn Lahodny Jr.
% Math 442 - Mathematical Modeling
% This function defines the derivative of the error function needed to
% minimize error for fitting average maximum temperatures for Big Bend
% National Park.
function E = bigbend(a)
% Scaled Temperatures
T=[-1 -0.71 -0.08 0.1 0.51 0.85 0.78 0.68 0.42 -0.01 -0.58 -0.93];
% Initial Value
E=0;
% Loop defining the derivative of the error function.
for k=1:12
m=2*pi*k/12;
E=E+(T(k)-sin(m-a))*cos(m-a);
end

Finally, we solve for a and compare our model with the data, arriving at Figure 1.5.
% Glenn Lahodny Jr.
% Math 442 - Mathematical Modeling
clear
% Data
Months=1:12;
Temps=[60.9 66.2 77.4 80.7 88 94.2 92.9 91.1 86.4 78.8 68.5 62.2];
% Computes the zeros of the derivative of the error function defined by
% bigbend.m needed to obtain the best fit.
a=fzero(bigbend,1);
modeltemps=18*sin(2*pi*Months/12-a)+mean(Temps);
% Plot Information
plot(Months,Temps,ro,Months,modeltemps,b-,linewidth,1.5)
xlabel(Months)
ylabel(Average Max Temperature)
legend(Data,Model)

100
Data
Model

95

Average Max Temperature

90
85
80
75
70
65
60
0

10

12

Months

Figure 1.5: Trigonometric model for average monthly temperatures in Big Bend.

For models with multiple parameters, we must solve a system of nonlinear equations,
which can be quite difficult in general, even computationally. The MATLAB function
lsqcurvefit can be used to solve nonlinear least squares problems.
Example 1.4. Let us consider a model for population growth in Bryan, TX from 19002010
(see Table 1.4).
Year Population
1900
3,589
1910
4,132
1920
6,307
1930
7,814
1940
11,842
1950
18,072

Year Population
1960
27,542
1970
33,141
1980
44,337
1990
55,002
2000
65,660
2010
76,201

Table 1.4: Population of Bryan, TX from 19002010 according to the U.S. Census.
Let N (t) denote the population size (in thousands) at time t 0. The Verhulst (logistic)
growth equation is given by


N
dN
= rN 1
, N (0) = N0
dt
K
where N0 > 0 is the initial population, r > 0 is a growth parameter for the population, and
K > 0 is the carrying capacity. This differential equation can be solved using separation of
variables and partial fractions. The solution is given by
N (t) =

N0 K
.
N0 + (K N0 )ert

Although the time t = 0 corresponds to the year 1900, we assume that the estimate that
year was fairly crude and obtain a value of N0 by fitting the entirety of the data. In this
way, we have three parameters to solve for, and carrying out the full regression analysis can
be tedious.
The first step in finding values for the parameters with MATLAB consists of defining the
model, N (t), as a function M-file, with the three parameters N0 , r, and K stored as a vector
p = (p(1), p(2), p(3)) = (N0 , r, K).
% Glenn Lahodny Jr.
% Math 442 - Mathematical Modeling
% This function computes the logistic model N=N0*K/(N0+(K-N0)*exp(-r*t)).
function N = logistic(ArK,t)
N=p(1)*p(3)./(p(1)+(p(3)-p(1))*exp(-p(2).*t));
10

Next, we write a MATLAB script M-file which uses the function lsqcurvefit to obtain
least squares parameter estimates for N0 , r, and K.
% Glenn Lahodny Jr.
% Math 442 - Mathematical Modeling
% This script uses the function lsqcurvefit to fit the solution of the
% logistic growth model to the Bryan, TX population data for 1790-2000.
clear
close all
% Data
Time=0:10:110;
Pops=[3.589 4.132 6.307 7.814 11.842 18.072 27.542 33.141 44.337...
55.002 65.660 76.201];
Year=Time+1900;
% Initial Guess for Parameters
N0=3.589;
r=0.01;
K=2*max(Pops);
p0=[N0 r K];
% Nonlinear Least Squares Best Fit
[p,Error] = lsqcurvefit(logistic,p0,Time,Pops)
N = logistic(p,Time);
% Plot Information
plot(Year,Pops,ro,Year,N,b-,linewidth,1.5)
xlabel(Year)
ylabel(Population (in Thousands))
legend(Data,Logistic Model,location,NorthWest)
After defining the data, we entered our initial guess as to what the parameter values should
be, the vector p0 . (It should be noted that MATLAB uses a routine similar to fzero, and can
typically only find roots reasonably near our guess.) In this case, we guessed a small value
of r, a carrying capacity of twice the maximum population in the dataset, and an initial
population equal to the population in 1900. Finally, we use lsqcurvefit, entering our function
file, our initial parameter guesses, and our data, respectively.
The function lsqcurvefit provides two outputs, our parameters and a sum of squared
errors, which we call Error. Although the error seems large, keep in mind that this is a
sum of all errors squared:
X
Error =
(Pops(Time) N (Time))2 .
Time

11

The output is displayed in the MATLAB Command Window:

p =
2.7583
0.0410
Error =
9.0175
>> sqrt(Error)
ans =
3.0029

107.8958

A more reasonable measure of error is the square root of this value, from which we see that
over 12 decades our model is only off by around 3,000 people. The data and logistic fit are
illustrated in Figure 1.6.
80
70

Data
Logistic Model

Population (in Thousands)

60
50
40
30
20
10
0
1900

1920

1940

1960

1980

2000

2020

Year

Figure 1.6: Population data for Bryan, TX and the logistic model approximation.

1.4

Transformations to Linear Form

In the previous examples, we found that nonlinear regression can be used to estimate the
parameters for arbitrary functions. So it is not required that our functions be in linear form.
However, lines are certainly the easiest fits to recognize visually, and so it is often worthwhile
to at least look for a linear relationship for your data.
Each of the following nonlinear relationships can be put into a linear form:
y = axb
y = a bx
1
y =
.
ax + b
12

This is not an exhaustive list, but these are some very useful examples. For example, if we
suspect our data has the form y = axb , we might take the natural logarithm of both sides
to write
ln y = ln a + b ln x.
If we now plot ln y versus ln x, the slope of the plot will be b and the y-intercept will be ln a,
from which we can obtain a value for a by exponentiation.

1.5

Parameter Estimation Directly from Differential Equations

In modeling a new phenomenon, we often find that we can write down a differential equation
that models the phenomenon, but we cannot solve the differential equation analytically. We
would like to solve it numerically, but all of our techniques thus far for parameter estimation
assume we know the exact form of the function.
1.5.1

Derivative Approximation Method

Example 1.5. In population models, we often want to study the interaction between various
populations. One of the simplest interaction models is the Lotka-Volterra predator-prey
model
dx
= (a1 b1 y)x,
dt
dy
= (a2 + b2 x)y,
dt
where x(t) and y(t) denote the number of prey and predators at time t 0, respectively.
Observe that the interaction terms, b1 xy and b2 xy correspond to death of prey in the
presence of predators and proliferation of predators in the presence of prey, respectively.
The Lotka-Volterra predator-prey model is a classic population model. However, it is
often too simple to capture the complex dynamics of interacting species. One classic example
in which the model describes the interaction fairly well is the interaction between lynx and
hares, as measured by the number of pelts collected by the Hudson Bay Company from
19001920. The data from the Hudson Bay Company is given in Table 1.5.
Year Lynx Hare
1900 4.0
30.0
1901 6.1
47.2
1902 9.8
70.2
1903 35.2 77.4
1904 59.4 36.3
1905 41.7 20.6
1906 19.0 18.1

Year Lynx Hare

1907 13.0 21.4
1908 8.3
22.0
1909 9.1
25.4
1910 7.4
27.1
1911 8.0
40.3
1912 12.3 57.0
1913 19.5 76.6

Year Lynx Hare

1914 45.7 52.3
1915 51.1 19.5
1916 29.7 11.2
1917 15.8
7.6
1918 9.7
14.6
1919 10.1 16.2
1920 8.6
24.7

Table 1.5: Pelts collected by the Hudson Bay Company (in thousands).

13

Our goal is to estimate values of a1 , b1 , a2 , and b2 without finding an exact solution

of the differential equations. Assuming the predator population is nonzero, the differential
equation for y can be written as
1 dy
= b 2 x a2 .
y dt
By treating the expression y1 dy
as a single variable, we see that b2 and a2 are the slope and
dt
intercept of a line, respectively. Thus, we would like to plot the values of y1 dy
versus x and
dt
fit a line through the data. Since we have the values of x and y, the only difficulty is in
. In order to do this, recall the definition of a derivative
finding the values of dy
dt
dy
y(t + h) y(t)
= lim
.
dt h0
h
Following the idea behind Eulers method for numerically solving differential equations, we
conclude that for h sufficiently small,
dy
y(t + h) y(t)

,
dt
h
which is called the forward difference approximation.
How good is this approximation? To answer this question, recall that the Taylor series
for a function f (x) centered at x = a is given by
f (x) =

X
f (n) (a)
n=0

n!

(x a)n = f (a) + f 0 (a)(x a) +

f 00 (a)
f 000 (a)
(x a)2 +
(x a)3 + .
2!
3!

Setting x = t + h and a = t, we have the expansion

f 00 (t) 2 f 000 (t) 3
h +
h + .
2
3!
Finally, subtracting f (t) from both sides and dividing by h gives the approximation
f (t + h) = f (t) + f 0 (t)h +

f (t + h) f (t)
f 00 (t)
f 000 (t) 2
0
= f (t) +
h+
h + .
h
2
3!
It follows that the error in our approximation is proportional to h. Thus, the forward
difference approximation is an order one approximation. This can be written as
f 0 (t) =

f (t + h) f (t)
+ O(h),
h

where g(h) = O(h) means that | g(h)

| remains bounded as h 0.
h
Can we do better than this approximation? Consider the central difference approximation
dy
y(t + h) y(t h)

.
dt
2h
It is not difficult to show that this is a second order approximation. That is,
f (t + h) f (t h)
+ O(h2 ).
2h
Therefore, we will use the central difference approximation for the data.
f 0 (t) =

14

Returning to the data, we find that h = 1 for our approximation. This value for h is not
particularly small. However, our goal is to estimate the parameters, and we can check the
validity of our estimates by checking the model against our data. Since we cannot compute
a central difference approximation for the first year of data, we begin in 1901 and compute


1 dy
1 y(t + h) y(t h)

y dt
y
2h


1 9.8 4.0
=
6.1
2
= b2 (47.2) a2 .
Repeating this for each year, up to 1919, we obtain a system of equations which can be
solved by regression. This can be done by writing a script M-file in MATLAB.
% Glenn Lahodny Jr.
% Math 442 - Mathematical Modeling
% Data
Hare=[30 47.2 70.2 77.4 36.3 20.6 18.1 21.4 22 25.4 27.1 40.3 57 76.6...
52.3 19.5 11.2 7.6 14.6 16.2 24.7];
Lynx=[4 6.1 9.8 35.2 59.4 41.7 19 13 8.3 9.1 7.4 8 12.3 19.5...
45.7 51.1 29.7 15.8 9.7 10.1 8.6];
% Central Difference Approximation
for k=1:19
y(k)=(1/Lynx(k+1))*(Lynx(k+2)-Lynx(k))/2;
x(k)=Hare(k+1);
end
% Linear Fit
P=polyfit(x,y,1);
L=P(1)*x+P(2);
% Plot Information
plot(x,y,ro,x,L,b-,linewidth,1.5)
title(Linear Regression for Predators)
xlabel(x)
ylabel((1/y)(dy/dt))
legend(Data,Linear Fit,location,Northwest)
The data from the central difference approximation is plotted in Figure 1.7. The best fit
line for this data is y = 0.0234x 0.7646. Therefore, we have estimates of b2 = 0.0234 and
a2 = 0.7646.
Similarly, we approximate the prey equation using a central difference approximation to
obtain the data in Figure 1.8. The best fit Line for this data is y = 0.0240x + 0.4732.
Therefore, we find estimates of a1 = 0.4732 and b1 = 0.024.
15

Linear Regression for Predators

1.5

Data
Linear Fit

(1/y)(dy/dt)

0.5

-0.5

-1
0

10

20

30

40

50

60

70

80

Figure 1.7: Linear fit for predator equation.

Linear Regression for Prey

0.6

Data
Linear Fit

0.4
0.2

(1/x)(dx/dt)

0
-0.2
-0.4
-0.6
-0.8
-1
-1.2
0

10

20

30

40

50

Figure 1.8: Linear fit for prey equation.

16

60

Finally, we would like to check our parameter estimates for the model by plotting time
series of the numerical solution of the model against the Hudson Bay Company data. To do
this, we first define the Lotka-Volterra predator-prey equations with a function M-file.
% Glenn Lahodny Jr.
% Math 442 - Mathematical Modeling
% This function defines the Lotka-Volterra predator-prey model.
function dx = LotVolt1(t,x)
% Parameters
a1=0.4732;
a2=0.7646;
b1=0.024;
b2=0.0234;
% ODEs (Here, x1=x and x2=y)
dx=zeros(2,1);
dx(1)=(a1-b1*x(2))*x(1);
dx(2)=(-a2+b2*x(1))*x(2);
Then we solve the model using the built-in MATLAB ODE solver ode45 with the following
script M-file.
% Glenn Lahodny Jr.
% Math 442 - Mathematical Modeling
% This program solves the Lotka-Volterra predator-prey model defined by
% the function LotVolt1.m using the ODE solver ode45 and plots the
% solutions and data over time.
% Time Vector
Year=1900+0:1:20;
% Maximum Time
tmax=20;
% Time Span
tspan=[0 tmax];
% Initial Conditions
x0=30;
y0=4;
Init=[x0,y0];

17

% Solves ODEs
[t,x]=ode45(LotVolt1,tspan,Init);
% Plot Information
subplot(2,1,1)
plot(Year,Hare,ro,1900+t,x(:,1),b-,linewidth,1.5)
title(Prey (Hare))
xlabel(Year)
ylabel(Population (in Thousands))
legend(Data,Model)
subplot(2,1,2)
plot(Year,Lynx,ro,1900+t,x(:,1),b-,linewidth,1.5)
title(Predator (Lynx))
xlabel(Year)
ylabel(Population (in Thousands))
legend(Data,Model)

Population (in Thousands)

Population (in Thousands)

The data, along with the numerical solution, are illustrated in Figure 1.9
Prey (Hare)

80

Data
Model

60
40
20
0
1900

1905

1910

1915

1920

1915

1920

Year
Predator (Lynx)

80

Data
Model

60
40
20
0
1900

1905

1910

Year

Figure 1.9: Data and model for lynx-hare system.

References
[1] P. Howard, Modeling Basics, Lecture Notes for Math 442 - Mathematical Modeling,
2009, http://www.math.tamu.edu~/phoward/M442.html.
18