NumericalExamples.

nb 1

Illuminating
Numerical Analysis
using Mathematica
Jennifer Voitle and Edward Lumsdaine

Abstract
This paper discusses the experiences of the authors teaching courses in Numerical Analysis to engineering
students at two universities, both pre- and post-introduction of Mathematica. Use of Mathematica in
such courses really enhances and accelerates student learning and comprehension, providing a foundation for
success in subsequent courses. With Mathematica, more realistic and complex applications can be
assigned, since Mathematica has many of the required intrinsic functions to solve real-life problems. Tips
are given to increase student acceptance of Mathematica, as some students will resist it for various reasons.

This paper also discusses the use of Mathematica as a programming language, including transition from
and communication with other languages. Much of the course can be taught using the intrinsic functions
available in Mathematica, but for certain applications extension of the built-in capabilities is required. For
example, in earthquake and vibration engineering, systems of differential equations must be solved whose
forcing functions are given as tabular data. This paper shows how to read in these data from a file and manipu-
late them so that NDSolve can work with them.

One of the most powerful ways to use Mathematica as a teaching tool is via Mathematica Note-
books. The course can be taught electronically - as a "live" textbook. Students can obtain the needed informa-
tion from these Notebooks, which include text, algorithms, animations and problem sets. The paper discusses
such Notebooks to teach Numerical Analysis, including guidelines for effective design. Where desired,
assignments can be given in the "traditional" way of writing computer programs by following algorithms, but
using Mathematica as the programming language. Other assignments can be given which involve the use
or modification of existing Mathematica functions. Freed from the tedious program writing, testing,
debugging and compiling of programs, this method encourages students to explore "what-if" questions by
altering parameters and methods.
NumericalExamples.nb 2

Mathematica as a Programming Language

Mathematica is a rule-based language. Transition from other languages is easy, as Mathematica
provides capabilities for all of the standard constructs, eg. looping, conditional testing, etc. The development
time and size of a Mathematica program will generally be substantially less than in traditional languages
such as FORTRAN, BASIC, C and so forth.

Summing Integers
As an example, consider the problem of summing integers from 1 to some ending value. That is, we seek
100
i
i=1

The FORTRAN code to accomplish this objective is

SUM = 0
WRITEH,L ' ENTER ENDING VALUE TO SUM TO : '
READH,L N
DO 10 I = 1, N
SUM = SUM + I
10 CONTINUE
WRITEH,L SUM
END

Similar code would be written in BASIC:

sum = 0
INPUT "enter ending value to sum to:", n
FOR i = 1 TO n
sum = sum + i
NEXT
PRINT sum
NumericalExamples.nb 3

or in C:
sumint.c program to sum from 1 to n
 include < stdio.h >

int main HvoidL

9
long int i, n, sum;
sum = 0;
printf I"Enter ending value to sum to:\n"M;
scanf H"%ld", & nL;
for H i = 1; i <= n; i ++L
sum += i;
printf H"The sum is %ld.", sumL;
return 0;
=

Mathematica code
In Mathematica, this program can be written in a single line,since there is a Sum function built in. The upper
index n is passed as a function argument. Suppose that we choose n=100:

100
i
i=1

5050

One unfortunate fact regarding other programming languages is that they only have finite precision. Mathe-
matica has infinite precision, which means that one could ask for the sum of i from one to one billion and
the correct answer would be obtained, an impossible feat in other languages. However, the way that Mathe-
matica does the sum makes the result, while correct, prohibitively slow. To speed things up, change the way
the sum is evaluated. Mathematica has a Range function, which outputs a list ranging over the limits
provided by the user. For example, Range[10] produces the following:

Range@10D

81, 2, 3, 4, 5, 6, 7, 8, 9, 10<

Mathematica has another function, Plus, which takes a set of numbers as its input and returns their sum.
NumericalExamples.nb 4

For example:

1 + 2 + 3 + 4 + 5 + 6 + 7 + 8 + 9 + 10

55

To avoid having to type the arguments to Plus, combine it with Range using the command Apply as
follows:

Plus @@ Range@10D

55

Apply works by taking the output form of Range, which was a list of numbers, and replacing this by Plus.
So the output is the sum of the given set of numbers. To appreciate this, compute the sum of integers from 1 to
10000 using both Sum and Apply with Range. Wrap the function Timing around these for comparison. I
am running this on a Macintosh SE/30 running an unenhanced copy of Mathematica. Your times will vary.

$Version

4.0 for Microsoft Windows HJuly 26, 1999L

10 000
TimingB iF
i=1

80.05 Second, 50 005 000<

Timing@Plus @@ Range@10 000DD

80. Second, 50 005 000<

Fixed Point Iteration

Solve x = g(x) given an initial guess p0, a tolerance tol, and a maximum number of iterations itmax.
NumericalExamples.nb 5

BASIC:

'*************** user's function goes here ***************

DEF fng(x) = (x+1)^(1/3)
'***************************************************
tol = 0.0001
itmax = 100
p0 = 1.5 ' initial guess
i=1
status$= "iterating"
PRINT" Solution of x = g(x) by Fixed Point Iteration"
PRINT"iteration x g(x) "
WHILE (i<=itmax AND status$ <>"Converged")
pnew=fng(p0)
PRINT i,p0,pnew
IF ABS(pnew-p0)<=tol THEN
status$ = "Converged"
END IF
p0 = pnew
i = i +1
WEND

IF status$ = "Converged" THEN

PRINT"Method converged in ",i -1
PRINT" iterations to root ",pnew
ELSE
PRINT"Method did not converge in ",itmax," iterations."
END IF
NumericalExamples.nb 6

C:

fixedpoint.c program to perform fixed point iteration  include <

stdio.h >  include < math.h >  define TOL 0.0001
 define ITMAX 100
 define CONVERGED 1
 define ITERATING 0
double g Hdouble xL; function prototype

int main HvoidL

9
double p0, pnew;
int i = 1;
int status = ITERATING;
printf I"Fixed Point Iteration\n"M;
printf I"Please enter an initial guess:\n"M;
scanf H"%lf", & p0L;
printf I"\nIteration\tx\t\tgHxL\n"M;
while Hi ITMAX && status CONVERGEDL 9pnew = g Hp0L;
printf I"%d\t\t%6.8lf\t\t %6.8lf\n", i, p0, pnewM;
if HHfabs Hpnew p0LL TOLL status = CONVERGED;
p0 = pnew;
i ++;= end of while block if Hstatus CONVERGEDL
9printf I"\nMethod converged in %d iterations.", iM;
printf I"\nRoot is %6.8lf", pnewM;= else printf
I"\nMethod did not converge after %d iterations.\n", ITMAXM;
return 0;
=

double g Hdouble xL
8
return pow HHx + 1L, H1. 3LL; users function definition
<
NumericalExamples.nb 7

Mathematica as a C (or BASIC) interpreter:

Clear@gD
g@x_D = Hx + 1L ;
13

tol = 0.0001;
itmax = 100;
p0 = 1.5;
i = 1;
status = "iterating"
Print@" Solution of x = ", g@xD, " by Fixed Point Iteration"D;
Print@"iteration x gHxL "D;
WhileAi itmax && status "Converged",
pnew = g@p0D; PrintAi, "\t\t\t", p0, "\t", pnewE;
If@Abs@pnew p0D tol, status = "Converged"D; p0 = pnew; i += 1E
If@status === "Converged",
Print@"Method converged in ", i = 1, " iterations to root ", pnewD,
Print@"Method did not converge in ", itmax, " iterations."DD

the Mathematica programming language

Of course, Mathematica provides the functions NestList and FixedPoint:

Clear@gD
g@x_D = Hx + 1L13 ;
itmax = 5;
p0 = 1.5;
NestList@g, p0, itmaxD

81.5, 1.35721, 1.33086, 1.32588, 1.32494, 1.32476<

NumericalExamples.nb 8

Rootfinding
Example: Heat Transfer through Concentric
Cylinders
The analysis of heat transfer through a series of concentric cylinders requires the solution of the following
equation. Determine the thickness Dr = r2 - r1 of the inner cylinder given that


r2 0.2640
ln = 2.2185 ln
r1 r ln r2

2 r
1
by all root-finding methods. Note that r2 and r1 are in meters. Take r1 = 0.1575 m. Summarize results. Did
Steffensen's method result in any improvement?

Schematic

Analysis
The given equation must be solved for r2. r1 is given as 0.1575 m. It would be reasonable to expect that r2 >
r1 but r2< 1, say. The solution will be carried out by the Bisection Method, Fixed Point iteration, Steffensen's
Method, Newton's Method and the Secant Method. The solution will be carried out to a tolerance of 0.001 and
all results compared.
NumericalExamples.nb 9

Solution

Function Definition
Define r1 and and f[r2]:

r1 = 0.1575;
r2 .264
f@r2_D = LogB F 2.2185 LogB F;
r1 r2 LogA r2
r1
E

Graphical Estimation of Root

Use the Plot command to plot the function for a graphical estimation of the root. We could plot the left hand
side against the right hand side and look for an intersection of the two curves, or set the right hand side to zero
and look for roots of the function. It is the latter approach that we use here:

Plot@f@r2D, 8r2, r1, 1<, AxesLabel 8"r2", "f@r2D"<D

f@r2D

r2
0.2 0.4 0.6 0.8 1

-10

-20

-30

The root is at approximately r2 = 0.3. For a better estimate on the

Macintosh, click anywhere on the graph. A bounding box will occur and the cursor will change to a compass.
Move the cursor along the curve (don't press down) while holding down the Command (Apple) key. The point
coordinates will be displayed on the lower left of the screen. By this method, the root is at about .302.

Solution by Solve
Mathematica has the Solve function, which works really well only for polynomials and simpler non-
polynomial equations. It will not work on our particular function, which is why numerical methods are
necessary.
NumericalExamples.nb 10

N@Solve@f@r2D == 0, r2DD

Solve::tdep : The equations appear to involve

the variables to be solved for in an essentially nonalgebraic way.

Solve::tdep : The equations appear to involve

the variables to be solved for in an essentially nonalgebraic way.

0.264
[email protected] r2D 2.2185 LogB F == 0., r2F
r2 [email protected] r2D

Bisection Method
This example uses our package, bisection.m, described in the appendix. For now, just use the function
definition:

Bisection@f_, var_, 8LeftEndPoint_, RightEndPoint_<, tol_, itmax_D :=

ModuleA8iteration, func<, a = N@LeftEndPointD; b = N@RightEndPointD;
func = Function@var, fD; PrintA"i\ta\tb\tp\tf@aD\tf@bD\tf@pD"E;
ForAiteration = 1, iteration <= itmax, iteration ++,
IWhile@func@aD func@bD > 0, HMessage@Bisection::nosignchangeD;
8a, b< = Input@"Enter new values 8a,b<:"DLD;
MidPoint = N@Hb + aL 2D;
PrintAiteration, "\t", N@a, 4D, "\t", N@b, 4D, "\t", N@MidPoint, 6D,
"\t", N@func@aD, 4D, "\t", N@func@bD, 4D, "\t", N@func@MidPointD, 4DE;
If@func@aD func@MidPointD > 0, a = MidPoint, b = MidPointD;
If@Abs@func@MidPointDD <= tol, Break@DD;ME;
If@iteration <= itmax, Message@Bisection::converged, iteration 1D,
Message@Bisection::unconverged, itmaxDD;
PrintA"\nThe root is found approximately as \n", N@MidPointDE;
Print@" where the function value is ", func@MidPointD, "."D;E
NumericalExamples.nb 11

Bisection@f@r2D, r2, 8.2, .4<, .001, 10D

i a b p f@aD f@bD f@pD

1 0.2 0.4 0.3 3.55336 1.69772 0.0470806

2 0.3 0.4 0.35 0.0470806 1.69772 0.924903

3 0.3 0.35 0.325 0.0470806 0.924903 0.470302

4 0.3 0.325 0.3125 0.0470806 0.470302 0.220581

5 0.3 0.3125 0.30625 0.0470806 0.220581 0.0891619

6 0.3 0.30625 0.303125 0.0470806 0.0891619 0.0216669

7 0.3 0.303125 0.301563

0.0470806 0.0216669 0.0125472

8 0.301563 0.303125 0.302344

0.0125472 0.0216669 0.00459939

9 0.301563 0.302344 0.301953

0.0125472 0.00459939 0.00396397

10 0.301953 0.302344 0.302148

0.00396397 0.00459939 0.000320186

Bisection::converged :
The Bisection Method converged to the root of the function in 9 iterations.

The root is found approximately as

0.302148

where the function value is 0.000320186.

NumericalExamples.nb 12

Solution by Fixed Point Iteration

The given equation has to be reformulated in the form r2 = g[r2].
This may most easily be done by adding r2 to both sides of the equation:

Trial 1

Clear@gD
g@r2_D = f@r2D + r2

0.264
r2 + [email protected] r2D 2.2185 LogB F
r2 [email protected] r2D

Plot to see if we expect convergence:

Plot@8r2, g@r2D<, 8r2, r1, 1<D

0.2 0.4 0.6 0.8 1

-5

-10

-15

It is clear from the plot that this choice of g will cause the method to diverge. A few iterations will verify this:

NestList@g, .4, 5D

80.4, 2.09772, 11.3956, 27.2563, 46.335, 67.3389<

This is a case of monotone divergence. Different formulations of g can be attempted. In this case, conver-
gence will be attained if we solve for the innermost r2 in the nested logs. We find:
NumericalExamples.nb 13

0 . 2640
1


r2 2 . 2185

r1
r2 = r1 Exp
r2




Clear@gD
.264
r2 12.2185
K O r2
g@r2_D = r1  r1

0.114756

0.1575  r2 1.45076

Twenty five iterations of the fixed point method and a plot of the path are generated by the code below:

ShowFixedPath@g_Symbol, initguess_, numiterations_, 8a_, b_<D :=

Module@8graph1, lines, x<,
iterates = NestList@g, initguess, numiterationsD; lines = Partition@
Flatten@Table@8iteratesPiT, iteratesPiT<, 8i, Length@iteratesD<DD,
2, 1D; lines = lines . linesP1T 8linesP1, 1T, 0<;
graph1 = Plot@8x, g@xD<, 8x, a, b<, DisplayFunction IdentityD;
FixedPointPath = Graphics@Line@linesDD; Show@FixedPointPath,
graph1, DisplayFunction $DisplayFunctionD; iteratesD
NumericalExamples.nb 14

ShowFixedPath@g, .4, 25, 8.2, .4<D

80.4, 0.242991, 0.384908, 0.249113, 0.372905, 0.254543, 0.363136,

0.259386, 0.355046, 0.263721, 0.348254, 0.267613, 0.342492,
0.271111, 0.337561, 0.27426, 0.333314, 0.277095, 0.329634, 0.279648,
0.326432, 0.281947, 0.323634, 0.284018, 0.321183, 0.285881<

This method appears to be exhibiting spiral convergence. It will take a large amount of iterations to converge
since the slope of g[r2] approaches unity (the upper asymptote for convergence) near the root.

Abs@g @r2DD . r2 .305

0.915321

In fact, fixed point iteration will take 208 iterations to converge to the root of 0.302134. Either NestList or
FixedPoint can be used to illustrate this:

8FixedPoint@g, .4, 207D, FixedPoint@g, .4, 208D<

80.302133, 0.302134<

Steffensen's Algorithm
To increase the speed of convergence of the fixed point method,
compute p1 = g[p0], p2 = g[p1]. So far, this is identical to the fixed point method. But now compute
( p1 p 0 )2
p = p0
p 2 2 p1 + p 0
where p is assumed to be a better estimate of the root than p0, p1 or p2. Convergence is attained when |p - p0|
<= tol. If not, p0 is set equal to p and the process is continued.
This algorithm can be easily programmed in Mathematica as:
NumericalExamples.nb 15

HxP2T xP1TL2
Steffensen@x_ListD := xP1T
xP3T 2 xP2T + xP1T

Start by generating the required estimates p1, p2 by NestList. These will be used by Steffensen to get the next
estimate p:

Iteration 1

x = NestList@g, .4, 2D

80.4, 0.242991, 0.384908<

p = Steffensen@xD

0.317532

Iteration 2

Now use this as p0 to generate p1 and p2:

x = NestList@g, p, 2D

80.317532, 0.288745, 0.315811<

The improved estimate p is

p = Steffensen@xD

0.302695

Iteration 3

x = NestList@g, p, 2D

80.302695, 0.301605, 0.302635<

p = Steffensen@xD

0.302135
NumericalExamples.nb 16

Iteration 4

x = NestList@g, p, 2D

80.302135, 0.302133, 0.302135<

p = Steffensen@xD

0.302134

This is the solution. It took four iterations compared to the two hundred and eight required by fixed point,
dramatically demonstrating the acceleration of convergence.
A simple Steffensen program can be written as follows:

x = NestList@g, .4, 2D;

SteffensenTable = 8<;
Do@p = Steffensen@xD; x = NestList@g, p, 2D;
SteffensenTable = Append@SteffensenTable, pD, 8k, 5<D
SteffensenTable

80.317532, 0.302695, 0.302135, 0.302134, 0.302134<

Solution by Newton-Raphson Method

The Newton-Raphson method uses the scheme
f (x i )
x i +1 = x i
f ( x i )
to approximate the solution of the equation f(x) = 0.
Mathematica has the built-in function FindRoot to approximate the solution of nonlinear equations and
systems of equations by the Newton-Raphson and the Secant methods. The syntax is:
FindRoot[lhs == rhs, {var,guess}] performs rootfinding by Newton-Raph-
son iteration to approximate the solution of the
equation lhs = rhs using the initial estimate guess for the unknown
var.

FindRoot[lhs == rhs, {var,{guess1,guess2}}] performs rootfinding by

Secant iteration using two initial estimates guess1 and guess2 for
the unknown var.

FindRoot[{lhs1==rhs1,lhs2==rhs2,...,lhsn==rhsn}, {var1,guess1},
NumericalExamples.nb 17

{var2,guess2}, ..., {varn,guessn}] returns the numerical approximation

of the system of equations with given initial estimates guess1, guess2,
... for the unknowns.

FindRoot@f@r2D == 0, 8r2, .4<D

8r2 0.302134<

To see the intermediate calculations, NestList can be used. Newton's method is a special case of fixed
point iteration, with g defined as follows:

Clear@gD
f@r2D
g@r2_D = r2
f @r2D
Plot@8r2, g@r2D<, 8r2, 0.2`, 0.4`<, PlotLabel "Newton Form"D

0.264
[email protected] r2D 2.2185 LogB r2 [email protected] r2D
F
r2
1.
r2
8.40341 r2 K 0.264
0.264
O [email protected] r2D
r22 [email protected] r2D2 r22 [email protected] r2D

Newton Form

0.32

0.25 0.3 0.35 0.4

0.28

0.26

It is clear that this scheme will converge. NestList will show the calculations:

NestList@g, .4, 10D

80.4, 0.278707, 0.300405, 0.302125, 0.302134,

0.302134, 0.302134, 0.302134, 0.302134, 0.302134, 0.302134<

Convergence attained in five iterations with 0.4 as the initial guess.

NumericalExamples.nb 18

Solution by the Secant Method

The Secant method is a variation of Newton's method which replaces the analytical derivative f'[x] by a numeri-
cal approximation

f (x + h ) f (x )
f ( x )
h
Thus, the secant scheme is

f ( x i +1 )( x i +1 x i )
x i + 2 = x i +1 , i = 0,1,2,..., done
f ( x i +1 ) f ( x i )
Note that two initial guesses x0, x1 are required. The Secant Method is carried out in Mathematica by
calling FindRoot with two initial guesses:

FindRoot@f@r2D == 0, 8r2, 80.2, 0.4<<D

8r2 0.302134<
NumericalExamples.nb 19

Function Interpolation and Approximation

Example: Force on a Weightlifter's Back

Polynomial Interpolation
The problem may be solved by fitting an interpolating polynomial through the (angle,force) data. This approxi-
mates the function force(angle) which is then solved for the angle giving a force of 1250 pounds. Alterna-
tively, inverse interpolation may be performed. In this case, an inverse function angle(force) is generated
which may then be evaluated at 1250. Mathematica has the function InterpolatingPolynomial to
perform these tasks. The output is identical to the expanded Lagrange or Newton interpolating polynomial. A
short program, NewtonDividedDifference, is presented to show the intermediate steps of the calcula-
tions.

Data Definition and Plot

force@60D = 2103; force@70D = 1410;

force@80D = 1092; force@90D = 497;
WeightLifterData = Table@8i, force@iD<, 8i, 60, 90, 10<D

8860, 2103<, 870, 1410<, 880, 1092<, 890, 497<<

NumericalExamples.nb 20

DataPlot = ListPlot@WeightLifterData,
PlotStyle [email protected]`D, AxesLabel 8"angle", "force"<D

force

2000

1750

1500

1250

1000

750

angle
65 70 75 80 85 90

Lagrange Form of Interpolating Polynomial

P3@angle_D = Expand@InterpolatingPolynomial@WeightLifterData, angleDD

22 795 angle 4939 angle2 163 angle3

50 648 +
12 200 1500

Plot@P3@angleD, 8angle, 60, 90<, PlotLabel "Force on back doing squats"D

Force on back doing squats

2000

1750

1500

1250

1000

750

65 70 75 80 85 90

The above graph shows that an angle of approximately 74 degrees will result in a force of 1250
pounds on the back. We will verify this with FindRoot:
NumericalExamples.nb 21

FindRoot@P3@xD == 1250, 8x, 74<D

880.302134, 0.302134, 0.302134< 74.8241<

Print@"The force exerted on the back while squatting at an angle of ",

%P1, 2T, " degrees is ", P3@xD . %, " pounds."D

The force exerted on the back while squatting at an angle of

74.8241 degrees is 850 076.3, 50 076.3, 50 076.3< pounds.

Interpolation by Newton's Divided Differences

Description of Procedure NewtonDD

Procedure NewtonDD takes a list of n+1 {x,y} pairs as its input and returns a interpolating polynomial of
degree n in the variable var.

Clear@NewtonDD, xD

NewtonDD@data_List, var_SymbolD :=
Module@8i, n, poly, xx, f<,
xx = var;
Do@xx@i 1D = dataPi, 1T;
f@i 1D = dataPi, 2T, 8i, Length@dataD<D;
f@i_, 0D := f@iD;
Hf@n_, i_D := Hf@n, i 1D f@n 1, i 1DL Hxx@nD xx@n iDL ; i > 0L;
Sum@f@i, iD Product@var xx@j 1D, 8j, i<D, 8i, 0, Length@dataD 1<D D

NewtonDD@WeightLifterData, xD

693 15 163 H80 + xL H70 + xL H60 + xL

2103 H60 + xL + H70 + xL H60 + xL
10 8 1500

In this form, the coefficients a0, a1, a2, ... can easily be identified for checking hand calculations. Expand this
to verify that this form is identical to the Lagrange polynomial:

Expand@%D === P3@xD

True
NumericalExamples.nb 22

Solution by Inverse Interpolation

Here we construct a function angle(force) by interchanging the (angle,force) data. This function is simply
evaluated at force = 1250 for the desired angle.

angle@force_D = N@Expand@
InterpolatingPolynomial@Table@8force@iD, i<, 8i, 60, 90, 10<D, forceDDD

73.9901 + 0.0656302 force 0.0000774083 force2 + 2.04647 108 force3

angle@1250D

75.0472

This compares reasonably well to the previous answer.

Cubic Spline Interpolation

Solution by Natural Cubic Splines

Mathematica has a built-in spline function, but it does not give the spline equations for evaluation. Here,
we use our own spline package to construct S[0,x] on the interval [60,70]; S[1,x] on [70,80], and S[2,x] on
[80,90]. To locate the angle at which the force is equal to 1250 pounds, note that f[80] = 1092 and f[70] =
1410. Since
1092 < 1250 < 1410, the desired angle is between 70 and 80 degrees. Thus we will perform rootfinding on
S[1,x] = 1250.
The natural cubic spline uses the boundary conditions S''[0,x0]=0 and S''[n-1,xn] = 0.

Here, we import our spline interpolation package, which we have stored in the standard package library in the
folder NumericalMath. ( Of course, Mathematica now has its own spline routines.)

Needs@"NumericalMath`SplineInterpolation`"D

Information@"SplineInterpolation", LongForm FalseD

The package SplineInterpolation contains code for construction of

natural, clamped, periodic and B cubic splines. A plotting utility,
PiecewiseCubicPlot, and integration procedure, IntegrateSpline,
are also provided. For information on usage of these procedures,
type ?NaturalCubicSpline, ?ClampedCubicSpline, ?PeriodicSpline,
?BSplineInterpolation, ?PiecewiseCubicPlot or ?IntegrateSpline.
NumericalExamples.nb 23

Information@"NaturalCubicSpline", LongForm FalseD

NaturalCubicSpline@88x0,y0<,8x1,y1<,...8xn,yn<<D constructs and prints

a natural cubic spline interpolating the given data points. The
natural spline boundary conditions S''@x0D = S''@xnD=0 are used.

NaturalCubicSpline@WeightLifterDataD

The spline is constructed as follows:

S@0,xD = 18 617. + 1198.29 x 21.324 x2 + 0.118467 x3 for 60. < x < 70.

S@1,xD = 96 562.4 3737.97 x + 49.194 x2 0.217333 x3 for 70. < x < 80.

S@2,xD = 65 332. + 2333.07 x 26.694 x2 + 0.0988667 x3 for 80. < x < 90.

PiecewiseCubicPlot@PlotLabel "Natural Cubic Spline"D;

Natural Cubic Spline

2000

1750

1500

1250

1000

750

65 70 75 80 85 90

S@1, x_D = 96 562.4 3737.97 x + 49.194 x2 0.217333 x3 ;

N@Solve@S@1, xD == 1250, xDD

88x 74.7558<, 8x 75.7987 11.0028 <, 8x 75.7987 + 11.0028 <<

NumericalExamples.nb 24

Solution by Clamped Cubic Splines

The package SplineInterpolation.m also has a utility for construction of clamped cubic splines, which match the
function derivatives at the endpoints x0, xn. Then S'[0,x0] = f'[x0], S'[n-1,xn] = f'[xn]. For this example
involving tabular data, we shall approximate the derivatives numerically using three-point forward and back-
ward difference formulas. These formulas are:
3 f (xi ) + 4 f ( xi +1 ) f (xi + 2 )
f (xi ) =
2h
forward difference O h 2 ( )
3 f (xi ) 4 f ( xi 1 ) + f ( xi 2 )
f (xi ) =
2h
backward difference O h 2 ( )
The Mathematica code to perform these derivatives is:

Clear@FD3Pt, BD3PtD
3 fP1T + 4 fP2T fP3T
FD3Pt@f_List, h_D :=
2h
3 Last@fD 4 fPLength@fD 1T + fPLength@fD 2T
BD3Pt@f_List, h_D :=
2h

General::spell1 :
Possible spelling error: new symbol name "BD3Pt" is similar to existing symbol "FD3Pt".

These equations can be used to approximate our endpoint derivatives as follows:

Derivs = N@8FD3Pt@82103, 1410, 1092<, 10D, BD3Pt@81410, 1092, 497<, 10D<D

888.05, 73.35<

The procedure ClampedCubicSpline requires the given data as input and these derivatives. The usage is:

Information@"ClampedCubicSpline", LongForm FalseD

ClampedCubicSpline@88x0,y0<,8x1,y1<,...8xn,yn<,8y'0,y'n<<D
constructs and prints a clamped cubic spline interpolating
the given data points. The clamped spline boundary
conditions S'@x0D = f'@x0D,S'@xnD = f'@xnD are used.

Modify the weight lifter data to include the derivatives using Append:

Append@WeightLifterData, DerivsD

8860, 2103<, 870, 1410<, 880, 1092<, 890, 497<, 888.05, 73.35<<
NumericalExamples.nb 25

ClampedCubicSpline@%D

The spline is constructed as follows:

S@0,xD = 2294.4 + 469.35 x 10.513 x2 + 0.0652 x3 for 60. < x < 70.

S@1,xD = 87 160. 3364.41 x + 44.255 x2 0.1956 x3 for 70. < x < 80.

S@2,xD = 46 369.6 + 1642.95 x 18.337 x2 + 0.0652 x3 for 80. < x < 90.

PiecewiseCubicPlot@D;

2000

1750

1500

1250

1000

750

65 70 75 80 85 90

Show@DataPlot, %D

force

2000

1750

1500

1250

1000

750

angle
65 70 75 80 85 90
NumericalExamples.nb 26

NASolveA87 160. 3364.41 x + 44.255 x2 0.1956 x3 == 1250, xEE

88x 74.8099<, 8x 75.7213 11.7188 <, 8x 75.7213 + 11.7188 <<

Function Approximation by Regression

Regression can be performed in Mathematica by either the Fit function or by loading the Statistics`LinearRe-
gression. This package computes the best fit polynomial as does Fit, but also prints a complete report of the
results including an ANOVA table.

The Regression Fits and Roots

DoAfunc = FitAWeightLifterData, TableAanglei , 8i, 0, k<E, angleE;

result = Solve@func == 1250, angleDP1T;
PrintAk, "\t", func, "\t", resultE, 8k, 1, 3<E

1 5127.5 51.36 angle 8angle 75.4965<

2 6475. 88.11 angle + 0.245 angle2 8angle 74.9003<

3 50 648. 1899.58 angle + 24.695 angle2 0.108667 angle3

8angle 74.8241<

N@P3@angleDD

50 648. 1899.58 angle + 24.695 angle2 0.108667 angle3

Note that the cubic regression polynomial is identical to the cubic generated by InterpolatingPolynomial.
(WHY?)
What general conclusions can you draw from this?
NumericalExamples.nb 27

Numerical Solution of
Differential Equations
Application
(Dr. Naser Mostaghel's problem)
In earthquake engineering, the following system of ODE's requires solution:

dz
+ 2 z (t ) + 2 2 u (t ) = u g (t )
dt
du
= z (t )
dt
z (t 0 ) = y (t 0 ), y (t 0 ) = y 0 , t 0 t t max
where ug(t) represents the displacement of the ground with time and u(t) is the resulting displacement of the
structure under consideration. Although Mathematica's intrinsic functions DSolve and NDSolve should
be able to solve this system, the difficulty in solving this particular problem is that ug(t) is usually discrete data
points which must be read in from a data file. Currently, DSolve and NDSolve require continuous inputs,
so some manipulation of the data must be performed before these functions see it.

In the following example, we shall first assume that ug(t) is well-represented by a sine function and will apply
DSolve directly to the problem. We then read in the actual data from the file, manipulate it into a continuous
form and compare the results from DSolveing this to the preceding results.
NumericalExamples.nb 28

Parameter Definitions

Clear@u, z, omega, alpha, beta, y0, y0prime, u, z, t, xiD

xi = 0.05; omega = 2. ; alpha = 1.5; beta = 1.0;
t0 = 0.48843; yprime0 = 18.1134; y0 = 1.0;
tend = 6; a = 240; delta = 2.5 omega;
ug@t_D = a Sin@delta tD;

Solution with ug[t] = a Sin[d t]:

Mathematica cannot solve this! This is unfortunately what often happens with highly complex real problems...

Clear@u, zD
solution =
FlattenADSolveA9z @tD + 2 alpha beta xi z@tD + alpha2 omega2 u@tD == ug@tD,
u @tD == z@tD, z@t0D == yprime0, u@t0D == y0=, 8z@tD, u@tD<, tEE

FactorSquareFree::lrgexp : Exponent is out of bounds for function FactorSquareFree.

PolynomialGCD::lrgexp : Exponent is out of bounds for function PolynomialGCD.

FactorSquareFree::lrgexp : Exponent is out of bounds for function FactorSquareFree.

FactorSquareFree::lrgexp : Exponent is out of bounds for function FactorSquareFree.

General::stop :
Further output of FactorSquareFree::lrgexp will be suppressed during this calculation.

PolynomialGCD::lrgexp : Exponent is out of bounds for function PolynomialGCD.

PolynomialGCD::lrgexp : Exponent is out of bounds for function PolynomialGCD.

General::stop :
Further output of PolynomialGCD::lrgexp will be suppressed during this calculation.

Had Mathematica had been able to solve this, we would have plotted using the following command:

plot1 = Plot@Evaluate@u@tD . solutionD,

8t, t0, tend<, PlotLabel "displacement u@tD"D
NumericalExamples.nb 29

displacement u[t]

2 4 6 8 10
-1

-2

Solution with ug[t] read in from a file

The data file, collected from data acquisition software, is shown below. The data are stored in the file
"quakedata".

0.48843 236.435
3.20602 22.9008
5.92361 -223.666
8.6412 -144.339
10. 0.0304909

All that needs to be done is to read in the data in the proper form for Mathematica manipulation, that is, in
the form
{{pair1},{pair2}, ... ,{pairn}}. This can be accomplished with the OpenRead command (here,
it is assumed that the data are contained in the Mathematica folder. If not, the required information would
be included.)

OpenRead@"quakedata.txt"D
GroundData = ReadList@"quakedata.txt", 8Number, Number<D

[email protected], 17D

880.48843, 236.435<, 83.20602, 22.9008<,

85.92361, 223.666<, 88.6412, 144.339<, 810., 0.0304909<<

inputdataplot = ListPlot@GroundData, PlotStyle [email protected]`DD

NumericalExamples.nb 30

200

100

4 6 8 10

-100

-200

A curve-fit of the data (regression, interpolating polynomial, spline, etc) could be attempted. Please refer back
to the previous section if this is desired.
(For example, we could type Fit[GroundData,{Sin[N[delta] t]},t] to obtain the result
240.1597 Sin[15.7079 t]).
Here, we will construct a piecewise-continuous function using Which. First, put the data into the proper form
using the Table command to automate things. This creates a list. Then Which is mapped onto the list, which
is named ug[t].

Clear@ugD
ug@t_D := Which @@ Flatten@Append@
Table@8t GroundDataPi, 1T && t < GroundDataPi + 1, 1T, GroundDataPi, 2T<,
8i, 1, Length@GroundDataD 1<D, 8t GroundDataPLength@GroundDataD, 1T,
GroundDataPLength@GroundDataD, 2T<DD

Plot@ug@tD, 8t, GroundDataP1, 1T, GroundDataPLength@GroundDataD, 1T<,

PlotLabel "ug@tD from data file"D

ug@tD from data file

200

100

2 4 6 8 10

-100

-200
NumericalExamples.nb 31

Comparison of actual data and Which approximation

Show@%, inputdataplotD

ug@tD from data file

200

100

2 4 6 8 10

-100

-200

Note that if this approximation is unsatisfactory, (perhaps one desires the points to be at the midpoint of the
interval rather than at the left endpoints), this is easily accomplished by modification of the code used to
construct ug[t].

Numerical Approximation of Solution of the system of differential equations

We may not have been able to obtain an analytic solution, but we can solve numerically using NDSolve:

Clear@u, zD
solution2 =
FlattenANDSolveA9z @tD + 2 alpha beta xi z@tD + alpha2 omega2 u@tD == ug@tD,
u @tD == z@tD, z@t0D == yprime0, u@t0D == y0=, 8z@tD, u@tD<, 8t, t0, 6<EE

8z@tD [email protected], 6.<<, <>D@tD,

u@tD [email protected], 6.<<, <>D@tD<

plot2 = Plot@Evaluate@u@tD . solution2D,

8t, t0, tend<, PlotLabel "displacement u@tD"D
NumericalExamples.nb 32

displacement u@tD

1 2 3 4 5 6
-1

-2

Comparison of ug[t] from function and from input data

Show@plot1, plot2, PlotLabel "Displacement from function and data file"D

What If
The input data for ug[t] did not give an accurate result because there were too few points to adequately
describe the forcing function. Try solving the problem again with the file quakedata2, which contains the
following points:

0.48843 236.047
0.964009 128.571
1.43959 -139.806
1.91517 -233.222
2.39074 -34.7715
2.86632 207.194
3.3419 189.866
3.81748 -65.0713
4.29306 -238.575
4.76864 -113.512
5.24421 153.605
5.71979 228.493
6.19537 17.4318
6.67095 -215.444
7.14653 -178.702
7.62211 81.6783
8.09769 239.841
8.57326 97.8538
9.04884 -166.594
9.52442 -222.557
-15
10. -1.35308 10

A regression approximation of these data in the form a Sin[d t] yields 240.00 Sin[15.7079 t]. This is a
NumericalExamples.nb 33

good approximation which will yield a good result.

Solution of 2-D Laplace

Equation
Introduction
The two-dimensional Laplace Equation

2 2 1
2
+ 2
+ g ( x, y ) = 0
x y k
describes two-dimensional, steady state conduction heat transfer. The solution (x,y) gives the temperature
T(x,y)- Tinf on a rectangular plate, where Tinf is the temperature of the surroundings. The physical domain is
given as 0 <= x <= PlateLength, 0 <= y <= PlateHeight.

An energy source term g(x,y) is allowed. Admissable boundary conditions are Type 1. The user specifies the
material thermal conductivity k; the boundary functions; the values of the plate dimensions PlateLength and
PlateHeight; and the desired numbers of subintervals in each direction. The program then constructs the finite
difference equation at each interior node and solves for the desired nodal temperature excesses via the Solve
function. The results are presented in the form of contour and surface plots.
The program may be modified to handle three-dimensional geometry, time-dependence and boundary condi-
tions of Types 2 and 3. Also note that Solve is used only for illustrative purposes on this small toy problem. In
other documents, we have used more sophisticated matrix methods. Also note that a variant of the heat equa-
tion is the fundamental formula underlying many financial derivatives, and in other papers these techniques are
applied to problems of financial analysis.
NumericalExamples.nb 34

Example
Determine the temperatures on a rectangular plate. The plate is internally heated by an energy source term
g(x,y) = 1000 Exp[-4 Pi^2 x y]. The plate is 1.0 m by 1.0 m in dimension with thermal conductivity k = 275
W/m-K. The boundary conditions are:
(1) T[x,y] = 100 Sin[Pi y/PlateHeight], x = 0, 0 <=y<=PlateHeight;
(2) T[x,y] = 400 Sin[Pi x/(2 PlateLength)], y = PlateHeight, 0 <=x<=PlateLength;
(3) T[x,y] = 400 y/ PlateHeight, x = PlateLength, 0 <=y<=PlateHeight;
(4) T[x,y] = 0, y = 0, 0 <=x<=PlateLength.

Five subintervals will be taken in each direction.

Solution Technique
The finite-difference methods will be used to solve for the steady temperatures on a square plate subject to
Type 1 boundary conditions T(x,0) = T(0,y) = 100; T(x,H) = T(L,y) = 0. The plate is subdivided into four
equally-spaced subdivisions in each direction. The plate thermal conductivity is uniform and there is no
internal energy generation. The governing equation is
D[T[x,y],[x,2}] + D[T[x,y],{y,2}] == 0 which is written in finite difference form as Ti,j+1 + Ti,j-1 + Ti+1,j +
Ti-1,j - 4Ti,j = 0,
i = 1, ... , 4, j = 1, ... , 4.

Solution
The finite-difference equation is defined as eq[i,j]:
NumericalExamples.nb 35

Clear@u, g, eq, TD
PlateLength = 1.0; H m L
PlateHeight = 1.0; H m L
NumberofXSubintervals = 5;
NumberofYSubintervals = 5;
DeltaX = PlateLength NumberofXSubintervals;
DeltaY = PlateHeight NumberofYSubintervals;
k = 275; H WmK L
g@i_, j_D = 1000 Exp@N@4 Pi ^ 2 i DeltaX j DeltaYDD;
equation@i_, j_D := HHu@i + 1, jD 2 u@i, jD + u@i 1, jDL DeltaX ^ 2 +
Hu@i, j + 1D 2 u@i, jD + u@i, j 1DL DeltaY ^ 2 == 0L

General::spell1 :
Possible spelling error: new symbol name "NumberofYSubintervals" is similar
to existing symbol "NumberofXSubintervals".

General::spell1 :
Possible spelling error: new symbol name "DeltaY" is similar to existing symbol "DeltaX".

Boundary Conditions

H Along the side x = 0, 0 y PlateHeight L

u@0, j_D := N@100 Sin@Pi j NumberofYSubintervalsD D

H Along the side y=PlateHeight,0<=x<=PlateLength L

u@i_, NumberofYSubintervalsD := 400 Sin@Pi i H2 NumberofXSubintervalsLD

H Along the side x=PlateLength,0<=y<=PlateHeight L

u@NumberofXSubintervals, j_D := 400 j NumberofYSubintervals
H along the side y=0,0<=x<=PlateLength L
u@i_, 0D := 0

Now construct and print out the finite difference equations to be solved (one equation is written for each node
at which the temperature is unknown.:
Mathematica constructs the list of equations, one at each point at which the temperature is unknown, with the
following command:
NumericalExamples.nb 36

Equations = Flatten@
Table@
Simplify @ equation@i, jD D ,
8i, NumberofXSubintervals 1<, 8j, NumberofYSubintervals 1<D
D

81469.46 + 25. u@1, 2D + 25. u@2, 1D == 100. u@1, 1D,

2377.64 + 25. u@1, 1D + 25. u@1, 3D + 25. u@2, 2D == 100. u@1, 2D,
2377.64 + 25. u@1, 2D + 25. u@1, 4D + 25. u@2, 3D == 100. u@1, 3D,
4559.63 + 25. u@1, 3D + 25. u@2, 4D == 100. u@1, 4D,
25. Hu@1, 1D + u@2, 2D + u@3, 1DL == 100. u@2, 1D,
25. Hu@1, 2D + u@2, 1D + u@2, 3D + u@3, 2DL == 100. u@2, 2D,
25. Hu@1, 3D + u@2, 2D + u@2, 4D + u@3, 3DL == 100. u@2, 3D,
5877.85 + 25. u@1, 4D + 25. u@2, 3D + 25. u@3, 4D == 100. u@2, 4D,
25. Hu@2, 1D + u@3, 2D + u@4, 1DL == 100. u@3, 1D,
25. Hu@2, 2D + u@3, 1D + u@3, 3D + u@4, 2DL == 100. u@3, 2D,
25. Hu@2, 3D + u@3, 2D + u@3, 4D + u@4, 3DL == 100. u@3, 3D,
8090.17 + 25. u@2, 4D + 25. u@3, 3D + 25. u@4, 4D == 100. u@3, 4D,
2000. + 25. u@3, 1D + 25. u@4, 2D == 100. u@4, 1D,
4000. + 25. u@3, 2D + 25. u@4, 1D + 25. u@4, 3D == 100. u@4, 2D,
6000. + 25. u@3, 3D + 25. u@4, 2D + 25. u@4, 4D == 100. u@4, 3D,
17 510.6 + 25. u@3, 4D + 25. u@4, 3D == 100. u@4, 4D<

Now write the unknowns

Unknowns = Flatten@ Table@u@i, jD,

8i, NumberofXSubintervals 1<, 8j, NumberofYSubintervals 1<DD

8u@1, 1D, u@1, 2D, u@1, 3D, u@1, 4D, u@2, 1D, u@2, 2D, u@2, 3D, u@2, 4D,
u@3, 1D, u@3, 2D, u@3, 3D, u@3, 4D, u@4, 1D, u@4, 2D, u@4, 3D, u@4, 4D<

Show a schematic of the plate nodal temperatures:

MatrixForm@ Temperatures = Table@u@i, jD,

8i, 0, NumberofXSubintervals<, 8j, 0, NumberofYSubintervals< D D

0. 58.7785 95.1057 95.1057 58.7785 0.

0 u@1, 1D u@1, 2D u@1, 3D u@1, 4D 100 J1 + 5N

0 u@2, 1D u@2, 2D u@2, 3D u@2, 4D 100 2 J5 5N

0 u@3, 1D u@3, 2D u@3, 3D u@3, 4D 100 J1 + 5N

0 u@4, 1D u@4, 2D u@4, 3D u@4, 4D 100 2 J5 + 5N

0 80 160 240 320 400

NumericalExamples.nb 37

Now solve the system of linear equations for the unknowns:

result = Solve@Equations, UnknownsD

88u@1, 1D 50.8159, u@1, 2D 90.5593, u@1, 3D 113.124, u@1, 4D 120.593,

u@2, 1D 53.9256, u@2, 2D 103.192, u@2, 3D 146.236, u@2, 4D 186.863,
u@3, 1D 61.6947, u@3, 2D 122.046, u@3, 3D 181.767, u@3, 4D 245.51,
u@4, 1D 70.8067, u@4, 2D 141.532, u@4, 3D 213.275, u@4, 4D 289.802<<

Put the temperatures in a convenient form for plotting:

Temperatures = Temperatures . Flatten@resultD

:80., 58.7785, 95.1057, 95.1057, 58.7785, 0.<,

:0, 50.8159, 90.5593, 113.124, 120.593, 100 J1 + 5 N>,

:0, 53.9256, 103.192, 146.236, 186.863, 100 2 J5 5 N >,

:0, 61.6947, 122.046, 181.767, 245.51, 100 J1 + 5 N>,

:0, 70.8067, 141.532, 213.275, 289.802, 100 2 J5 + 5 N >,

80, 80, 160, 240, 320, 400<>

Plots

ListPlot3D@Temperatures, PlotLabel "Steady Plate Temperature"D

NumericalExamples.nb 38

Steady Plate Temperature

400
300 6
200
5
100
0 4
1
2 3
3
4 2
5
61

ListContourPlot@Temperatures,
PlotLabel "Steady Plate Temperature", ColorFunction HueD
NumericalExamples.nb 39

Steady Plate Temperature

6

1
1 2 3 4 5 6
NumericalExamples.nb 40

Appendix
Code for CubicSplineInterpolation.m
BeginPackage@"NumericalMath`SplineInterpolation`"D
SplineInterpolation::"usage" =
"The package SplineInterpolation contains code for construction of
natural, clamped, periodic and B cubic splines. A plotting utility,
PiecewiseCubicPlot, and integration procedure, IntegrateSpline,
are also provided. For information on usage of these procedures,
type ?NaturalCubicSpline, ?ClampedCubicSpline, ?PeriodicSpline,
?BSplineInterpolation, ?PiecewiseCubicPlot or ?IntegrateSpline. "
NaturalCubicSpline::"usage" =
"NaturalCubicSpline@88x0,y0<,8x1,y1<,...8xn,yn<<D
constructs and prints a natural cubic spline
interpolating the given data points. The natural spline
boundary conditions S''@x0D = S''@xnD=0 are used."
ClampedCubicSpline::"usage" =
"ClampedCubicSpline@88x0,y0<,8x1,y1<,...8xn,yn<,8y'0,y'n<<D
constructs and prints a clamped cubic spline interpolating
the given data points. The clamped spline boundary
conditions S'@x0D = f'@x0D,S'@xnD = f'@xnD are used."
BSplineInterpolation::"usage" =
"BSplineInterpolation@88x0,y0<,8x1,y1<,...8xn,yn<,8y'0,y'n<<D
constructs and prints a Bspline interpolating the given
data points. Both natural and clamped spline boundary
conditions S'@x0D = f'@x0D; S'@xnD = f'@xnD; S''@x0D =
S''@xnD = 0 are used. The BSpline does not interpolate the
given data at the points 88x1,f@x1D<,8xn1,f@xn1D<<.
NOTE: n must be >2 for BSplineInterpolation."
PeriodicSpline::"usage" = "PeriodicSpline@88x0,y0<,8x1,y1<,...8xn,yn<<D
constructs and plots a periodic spline. This procedure is used for
plotting smooth, closed curves. The periodic boundary conditions
S@x0D = S@xnD, S'@x0D = S'@xnD and S''@x0D = S''@xnD are used."
IntegrateSpline::"usage" = "IntegrateSpline computes the
integral of the spline over the interval @x@0D,x@nDD.
It is not intended for use with PeriodicSpline."
CubicSpline::"smallnerr" = "Number of data points, n =`1` is insufficient
for spline interpolation: you must provide at least `2` data points."
CubicSpline::"nonfunctionerr" = "The x coordinates provided
`1` do not describe a function. The x's must be unique."
PiecewiseCubicPlot::"usage" = "PiecewiseCubicPlot@D constructs a plot
of the constructed spline. The option PlotPoints>num may be
included. Otherwise, the default value of PlotPoints is used.
NumericalExamples.nb 41

Note that setting num=1 yields a piecewise linear plot, identical

to the result obtained via ListPlot, with PlotJoined>True.
NOTE: Either NaturalCubicSpline or ClampedCubicSpline
must be executed prior to calling PiecewiseCubicPlot."
Begin@"`Private`"D
Unprotect@ClampedCubicSpline, NaturalCubicSpline, IntegrateSpline,
PiecewiseCubicPlot, BSplineInterpolation, S, PeriodicSplineD
S@i_, x_D := a@iD + b@iD Hx t@iDL + c@iD Hx t@iDL2 + d@iD Hx t@iDL3
WriteEquations := BlockA8i<, EquationList = 8<;
EquationList = Append@EquationList, S@n 1, t@nDD a@nDD;
DoAEquationList = Append@EquationList, S@i, t@i + 1DD S@i + 1, t@i + 1DDD;
EquationList = Append@EquationList,
Expand@x S@i, xD . x t@i + 1DD Expand@x S@i + 1, xD . x t@i + 1DDD;
EquationList = AppendAEquationList, ExpandA8x,2< S@i, xD . x t@i + 1DE
ExpandA8x,2< S@i + 1, xD . x t@i + 1DEE, 8i, 0, n 2<EE
MakeUnknowns := Block@8i<, Unknowns = 8<;
Do@Unknowns = Append@Unknowns, b@iDD, 8i, 0, n 1<D;
Do@Unknowns = Append@Unknowns, c@iDD, 8i, 0, n 1<D;
Do@Unknowns = Append@Unknowns, d@iDD, 8i, 0, n 1<DD
SolvetheSystem := BlockB8<, Result = N@Solve@EquationList, UnknownsDD;

DoBb@i 1D = ResultP1, i, 2T; c@i 1D = ResultP1, i + n, 2T;

3n
d@i 1D = ResultP1, i + 2 n, 2T, :i, 1,>F;
3
Print@"The spline is constructed as follows: "D;
Do@Print@"S@", i, ",xD = ", Expand@S@i, xDD,
" for ", t@iD, " < x < ", t@i + 1DD, 8i, 0, n 1<DF
SolvetheBSplineSystem := Block@8<, Result = N@Solve@EquationList, UnknownsDD;
a@1D = ResultP1, 1, 2T; a@n 1D = ResultP1, 2, 2T;
Do@b@i 3D = ResultP1, i, 2T; c@i 3D = ResultP1, i + n, 2T;
d@i 3D = ResultP1, i + 2 n, 2T, 8i, 3, 3 + n 1<D;
Print@"The B spline is constructed as follows: "D;
Do@Print@"S@", i, ",xD = ", Expand@S@i, xDD,
" for ", t@iD, " < x < ", t@i + 1DD, 8i, 0, n 1<DD
SolvePeriodicSystem := Block@8<, LHS = Table@Row@iD, 8i, 0, 3 n 1<D;
RHS = Table@constant@iD, 8i, 0, 3 n 1<D;
Result = LinearSolve@LHS, RHSD; Do@b@i 1D = ResultPiT;
c@i 1D = ResultPi + nT; d@i 1D = ResultPi + 2 nT, 8i, 1, n<D;D
WriteandSolvePeriodicEquations := BlockA8i<, Clear@Row, Equation, b, c, dD;
DoAEquation@iD = HS@i, t@i + 1DD a@iDL HS@i + 1, t@i + 1DD a@i + 1DL;
constant@iD = a@iD + a@i + 1D; Equation@i + n 1D =
Expand@x S@i, xD . x t@i + 1DD Expand@x S@i + 1, xD . x t@i + 1DD;
constant@i + n 1D = 0; Equation@i + 2 n 2D =
ExpandA8x,2< S@i, xD . x t@i + 1DE ExpandA8x,2< S@i + 1, xD . x t@i + 1DE;
E
NumericalExamples.nb 42

constant@i + 2 n 2D = 0, 8i, 0, n 2<E;

Equation@3 n 3D = S@n 1, t@nDD a@n 1D S@0, t@0DD;
constant@3 n 3D = a@0D a@n 1D; Equation@3 n 2D =
Expand@x S@0, xD . x t@0DD Expand@x S@n 1, xD . x t@nDD;
constant@3 n 2D = 0; Equation@3 n 1D =
ExpandA8x,2< S@0, xD . x t@0DE ExpandA8x,2< S@n 1, xD . x t@nDE;
constant@3 n 1D = 0; Do@Row@jD = 8<; Row@jD = Append@Row@jD,
Table@Coefficient@Equation@jD, b@iDD, 8i, 0, n 1<DD; Row@jD = Append@
Row@jD, Table@Coefficient@Equation@jD, c@iDD, 8i, 0, n 1<DD; Row@jD =
Append@Row@jD, Table@Coefficient@Equation@jD, d@iDD, 8i, 0, n 1<DD;
Row@jD = Flatten@Row@jDD, 8j, 0, 3 n 1<D;
MakeUnknowns; SolvePeriodicSystemE
NaturalCubicSpline@slist_ListD :=
BlockA8i<, Clear@a, b, c, dD; SplineData = N@slistD;
MinNumberofPoints = 3; If@Length@SplineDataD < MinNumberofPoints,
Message@CubicSpline::"smallnerr", Length@SplineDataD,
MinNumberofPointsD; Return@Hold@NaturalCubicSpline@dataDDDD;
If@SplineData Sort@SplineDataD, SplineData = Sort@SplineDataDD;
xlist = Table@SplineDataPi, 1T, 8i, Length@SplineDataD<D;
If@xlist Union@xlistD, Message@CubicSpline::"nonfunctionerr", xlistD;
Return@DD; n = Length@SplineDataD 1; Do@t@iD = SplineDataPi + 1, 1T;
a@iD = SplineDataPi + 1, 2T, 8i, 0, n<D; WriteEquations;
EquationList = AppendAEquationList, ExpandA8x,2< S@0, xD . x t@0DE 0E;
EquationList = AppendAEquationList, ExpandA8x,2< S@n 1, xD . x t@nDE 0E;
MakeUnknowns; SolvetheSystemE
ClampedCubicSpline@slist_ListD :=
Block@8i<, Clear@a, b, c, dD; SplineData = N@slistD;
MinNumberofPoints = 4; If@Length@SplineDataD < MinNumberofPoints,
Message@CubicSpline::"smallnerr", Length@SplineDataD,
MinNumberofPointsD; Return@Hold@ClampedCubicSpline@dataDDDD;
LeftDeriv = SplineDataPLength@SplineDataD, 1T;
RightDeriv = SplineDataPLength@SplineDataD, 2T;
SplineData = Drop@SplineData, 1D;
If@SplineData Sort@SplineDataD, SplineData = Sort@SplineDataDD;
xlist = Table@SplineDataPi, 1T, 8i, Length@SplineDataD<D;
If@xlist Union@xlistD, Message@CubicSpline::"nonfunctionerr", xlistD;
Return@DD; n = Length@SplineDataD 1;
Do@t@iD = SplineDataPi + 1, 1T; a@iD = SplineDataPi + 1, 2T, 8i, 0, n<D;
WriteEquations; EquationList = Append@EquationList,
Expand@x S@0, xD . x t@0DD LeftDerivD; EquationList =
Append@EquationList, Expand@x S@n 1, xD . x t@nDD RightDerivD;
MakeUnknowns; SolvetheSystemD
BSplineInterpolation@slist_ListD :=
BlockA8i<, Clear@a, b, c, dD; SplineData = N@slistD;
MinNumberofPoints = 5; If@Length@SplineDataD < MinNumberofPoints,
NumericalExamples.nb 43

Message@CubicSpline::"smallnerr", Length@SplineDataD,
MinNumberofPointsD; Return@Hold@BSplineInterpolation@dataDDDD;
LeftDeriv = SplineDataPLength@SplineDataD, 1T;
RightDeriv = SplineDataPLength@SplineDataD, 2T;
SplineData = Drop@SplineData, 1D;
If@SplineData Sort@SplineDataD, SplineData = Sort@SplineDataDD;
xlist = Table@SplineDataPi, 1T, 8i, Length@SplineDataD<D;
If@xlist Union@xlistD,
Message@CubicSpline::"nonfunctionerr", xlistD; Return@DD;
n = Length@SplineDataD 1; Do@t@iD = SplineDataPi + 1, 1T, 8i, 0, n<D;
a@0D = SplineDataP1, 2T; a@nD = SplineDataPn + 1, 2T;
Do@a@iD = SplineDataPi + 1, 2T, 8i, 2, n 2<D; WriteEquations;
EquationList = AppendAEquationList, ExpandA8x,2< S@0, xD . x t@0DE 0E;
EquationList = AppendAEquationList, ExpandA8x,2< S@n 1, xD . x t@nDE 0E;
EquationList = Append@EquationList,
Expand@x S@0, xD . x t@0DD LeftDerivD; EquationList =
Append@EquationList, Expand@x S@n 1, xD . x t@nDD RightDerivD;
MakeUnknowns; Unknowns = Prepend@Unknowns, a@n 1DD;
Unknowns = Prepend@Unknowns, a@1DD; SolvetheBSplineSystemE
PeriodicSpline@slist_List, opts___RuleD :=
BlockB8i, plotpoints, X, Y<, Clear@a, b, c, dD;
plotpoints = PlotPoints . 8opts< . Options@PlotD; SplineData = slist;
MinNumberofPoints = 3; If@Length@SplineDataD < MinNumberofPoints,
Message@CubicSpline::"smallnerr", Length@SplineDataD,
MinNumberofPointsD; Return@Hold@NaturalCubicSpline@dataDDDD;
n = Length@SplineDataD; Do@t@iD = i + 1; a@iD = SplineDataPi + 1, 1T;
aa@iD = SplineDataPi + 1, 2T, 8i, 0, n 1<D; t@nD = n + 1; a@nD = a@0D;
WriteandSolvePeriodicEquations; title = "PeriodicCubicSpline";
t@i + 1D t@iD
DoBX@iD = DropBTableBN@S@i, xDD, :x, t@iD, t@i + 1D, >F, 1F,
plotpoints
8i, 0, n 2<F;
t@nD t@n 1D
X@n 1D = TableBN@S@n 1, xDD, :x, t@n 1D, t@nD, >F;
plotpoints
AbscissaPoints = Flatten@Table@X@iD, 8i, 0, n 1<D, 1D; Clear@a, b, c, dD;
Do@a@iD = aa@iD, 8i, 0, n 1<D; a@nD = a@0D; WriteandSolvePeriodicEquations;
t@i + 1D t@iD
DoBY@iD = DropBTableBN@S@i, xDD, :x, t@iD, t@i + 1D, >F, 1F,
plotpoints
8i, 0, n 2<F;
t@nD t@n 1D
Y@n 1D = TableBN@S@n 1, xDD, :x, t@n 1D, t@nD, >F;
plotpoints
OrdinatePoints = Flatten@Table@Y@iD, 8i, 0, n 1<D, 1D; SplineData = Table@
8AbscissaPointsPiT, OrdinatePointsPiT<, 8i, Length@OrdinatePointsD<D;
ListPlot@SplineData, Joined True, PlotLabel titleDF

B
NumericalExamples.nb 44

PiecewiseCubicPlot@opts___RuleD := BlockB8i, plotpoints<,

plotpoints = PlotPoints . 8opts< . Options@PlotD;
title = PlotLabel . 8opts< . Options@PlotD;
t@i + 1D t@iD
DoBplot@iD = TableBN@8x, S@i, xD<D, :x, t@iD, t@i + 1D, >F,
plotpoints
8i, 0, n 1<F; SplinePlot = ListPlot@Flatten@

Table@plot@iD, 8i, 0, n 1<D, 1D, Joined True, PlotLabel titleDF

IntegrateSpline := BlockB8i<, SplineIntegral =

n1
1
a@iD Ht@i + 1D t@iDL + b@iD Ht@i + 1D t@iDL2 +
i=0
2
1 1
c@iD Ht@i + 1D t@iDL3 + d@iD Ht@i + 1D t@iDL4 ;
3 4
Print@"The integral of the cubic spline is ", SplineIntegral, "."DF
End@D
Protect@NaturalCubicSpline, ClampedCubicSpline, BSplineInterpolation,
PiecewiseCubicPlot, PeriodicSpline, IntegrateSplineD
EndPackage@D