MA1023-Methods of Mathematics-15S2

Tutorial 1
There will be a spot test on something similar to this during the Tutorial hour on
the week starting from 19/09/2016.

Q1. Consider the function =

1. Write down the nth degree Taylor Polynomial near > 0.

2. Show that the remainder satisfies

if > >0
| , |<

if > >0

3. Show that the value of 1000 accurate to 3 decimal places is 1.995.

4. Find the value of 1025 accurate to 10 decimal places.

Q2. Consider the biquadratic equation #

− − 1 = 0.

1. Find real intervals that contains real roots.

2. Propose a method to narrow down the intervals that contains roots.

Arrange the equation as = %

3. Repeat the process in 2 and find the roots.

For each , perform iterations according to & = % & .

4. in 4 different ways.

Which % gives convergence?

5.
6.
7. Explain the above behavior by a diagram.
8. Can we use the same method to find complex roots, give it a try!
 MA1023-Methods of Mathematics-15S2
Tutorial 3
There will be a spot test on something similar to this during the Tutorial hour on
the week starting from 17/10/2016.

Q1. Consider finding roots of = − − 1 = 0.

1. Write down the method to be solved by the Newton’s method.

2. Analyze the convergence treating it as an iterative method i.e. taking
= −
3. Find the no of iterations needed to find the real root in [1,2] to an accuracy
of 0.001 and find the root as an iterative method described in (2).
4. Analyze the convergence and find the no of iterations needed to find a real
root in [1,2] to an accuracy of 0.001 treating the process as the original
Newton’s method.
5. Suggest and implement, possibly a faster method by taking more terms in
the Taylor series expansion of .
6. What modifications/generalizations are needed in the Banach Fixed point
theorem to find the complex roots by the iterative method?
7. Predict convergence and try to find complex roots using ideas in (6).
8. Can we use the Newton’s method or the method described in (5) to find
complex roots?
 MA1023-Methods of Mathematics-15S2
Tutorial 4
There will be a spot test on something similar to this during the Tutorial hour on
the week starting from 07/11/2016.

Numerical solutions to non-linear equations of one variable.

Q1. Find the height of the circular sector with arc length 2 and chord length .

Q2. Your CASIO calculator can integrate. is found by , , .

Let ~ 0,1 which has a PDF = . Find the value of such that
√
< = 0.8 using the Newton’s method.

Q3. Consider finding roots of = tan starting from $ > 0. Find the value
of = $ such that the Newton’s method enters into a cycle. (If $ < the
method will converge and if $ > the method will diverge).

Numerical Integration.

Q4. Derive Numerical Integration Formulas that uses the left end point/right end
point/middle point of each partition interval to approximate .

Q5. Derive error formulas for the same.

Q6. Find the number of intervals required to get the values of $ √

and

$
sin accurate to 0.001 and find those integrals to that accuracy.
 MA1023-Methods of Mathematics-15S2
Tutorial 5
There will be a spot test on something similar to this during the Tutorial hour on
the week starting from 14/11/2016.

Numerical Integration.

Q1. Use the error formula for the Trapezoidal rule − to find the number
of divisions needed to find the value of sin accurate to 0.001 and find
the integral to that accuracy.

Q2. Use Taylor series to derive an error formula for the Trapezoidal rule.

Q3. Use the error formula for the Simpson’s rule − to find the number
of divisions needed to find the value of sin accurate to 0.001 and find
the integral to that accuracy.

Q4. Use Taylor series to derive an error formula for the Simpson’s rule.

Q5. Directly show that the Simpson’s formula is exact(zero error) for all cubic
polynomials.
 MA1023-Methods of Mathematics-15S2

Tutorial 6

There will be a spot test on something similar to this during the Tutorial hour on the week
starting from 28/11/2016.

Numerical Integration

Q1. Use the error formula in the Lagrange Interpolation to prove the error formula for the
Trapezoidal rule.

Q2. One method of doing numerical integration is Gaussian quadrature. Note that both the
Trapezoidal and the Simpson’s rules looks like ≈∑ and we knew and
found . In this method we find both and so that the integral and the sum are equal for a
given degree polynomial , by forcing both sides equal for each power of for =
0,1,2, … , . For = 3 how many and can we find?. For , = −1,1 find them and use it
! !
to approximate !
and thereby "
.

Interpolation

Q1. Consider the data set { 1,1 , 2,1 , 3,2 , 5,5 }

1. Find the Lagrange polynomial directly by matrix inversion.
2. Find the Lagrange polynomial by the above formula.
3. Use it to find the values at 0, 4 and 6.
4. Use it to find the derivative at 0, 4 and 6.
5. Use it to find the integral from 0 to 5.
6. Assume that the above data are obtained from the Fibonacci sequence with
!
& = )* − + * where ) > 0 and + < 0 are the roots of . / − . − 1 = 0. If the continuous
√(
!
version is = Re & = ) 3 − −+ 3
cos 7 find an upper bound for the error.
√(

Q2. One method of finding the Lagrange polynomial is to use the Newton’s divided differences. For
8 39 8 3: 8 39 ,3 8 3: ,39
{ ", !, /} we define ", ! = 39 3:
and ", !, / = 3 3:
and so on and the
Lagrange polynomial is given by ; = " + ", ! − " + ", !, / − " − ! .
See why the formula is working and use it to find the Lagrange polynomial for
{ 1,1 , 2,1 , 3,2 , 5,5 }.

Q3. Yet another way of finding the Lagrange polynomial is to define it as the iterative process
; = =! + >! − " and >! = =/ + >/ − ! and so on. Use this to find the find
the Lagrange polynomial for { 1,1 , 2,1 , 3,2 , 5,5 }.

Q4. Consider the data set { 1,1 , 2,1 , 3,2 }. Find a polynomial (Hermite polynomial) that goes
though the above points and satisfying ;′ 1 = 0, ;′ 2 = 1, ;′ 3 = 0.

Q5. Consider the data set { 1,1 , 2,1 , 3,2 }. Find a piecewise (defined on each sub interval),twice
differentiable cubic polynomial (Cubic Spline) that goes though the above points and
satisfying;′′ 1 = 0, ;′′ 3 = 0.
 MA1023-Methods of Mathematics-15S2-ucjaya@uom.lk-2016/12/09

Tutorial 7-Numerical Questions

Note: You are supposed to know the answers to the bolded questions only which are also given with
regular Tutorial 7. The rest is given to introduce you to the other related topic in Numerical Analysis that
can be done with your S2 knowledge. However those topics will not be tested at the final exam and you
are encouraged to solve them as a challenge after the exam. You are also highly encouraged to take the
other two specialized courses MA3023-Numerical Methods, MA4053-Numerical Methods for Scientific
Computing in your future semesters, where a detailed discussion of these topics will be done.

Q1. Use a suitable numerical method to solve ( ) = 0 and find the global maximum of

( ) = |( − 2)( − 3)( − 4)( − 6)| on [2,6].

with an interval [ , ] containing the unique maximum. Then we find [ , ] ⊂ ( , ). If ( ) < ( ),

Q2. One numerical method of finding a local maximum is called the Golden Section Search. Here we start

maximum is on [ , ] and otherwise the maximum is in [ , ]. So we rename [ , ] accordingly and repeat

the process. We can select [ , ] = − , + for small . But it is advantageous to select [ , ]

= !" = % where & =

!" #" $ $ √(
"
such that is the golden ratio (hence the name of the method) which
satisfies & − & − 1 = 0, why? Note that need not be differentiable to employ this method.

Q3. Use the above method to find the maximum in Q1.

Q4. Suppose we are going to fit ( ) = 4sin - $ 0 by a 1 th degree interpolating polynomial 2( ) on [0,6]
./

with equispaced points such that the error is smaller than 0.001 in magnitude. Find 1.

Q5. Find the Natural Cubic Splines for the data 3(1,1), (2,1), (3,2)4 and 3(1,1), (2,1), (3,2), (5,5)4.

+ + that minimizes the error

6 ( , , ) = ∑8;<( 8 + 8 + − 98 ) for the data set 3( 8 , 98 )4, = = 1, … , 1 is given by the formula
Q6. The least square parabola
:

∑ 8@ ∑ 8A ∑ 8 ∑ 8 98
?∑ 8 ∑ 8 ∑ 8 B C D = ?∑ 8 98 B .
A

∑ 8 ∑ 8 ∑1 ∑ 98
Find a condition for 8 such that , , can be found (ie. matrix is invertible).

3(1,1), (2,1), (3,2)4 and 3(1,1), (2,1), (3,2), (5,5)4.

Q7. Fit the least square line/parabola/cubic (equations are identical to the above) for each of the data

Q8. Fit a least square function of the form 2( ) = E / + sin + cos for each of the above data. The
method is identical to finding the least square parabola which also has 3 coefficients.

Q9. Let the data set 3( 8 , 98 )4, = = 1, … , 1 has a least square line given by + . Show that the least
∑ /J ∑ /J
square line passes through ( ̅ , 9I) where ̅ = :
, 9I = :
are the means of 8 , 98 respectively.

= K where N/O = : ∑( − ̅ )(98 − 9I) is the

KLM
+ , show that
$
Q10. For the least square line 8
LL

and N// = ∑( − ̅ ) is the variance of

KLM
8 , 98 Show that P given by P =
$
: 8 8.
QKLL KMM
covariance of is
P = ±1 when all data are on a straight line. P is the correlation coefficient and is a measure of the degree
to which a given set of points falling on a straight line.
Q11. One method to find the maximum of a multivariate function ( , 9) is called the Steepest Descent
Method. Here we start at a given point ( < , 9< ) and select the direction of the maximum slope at ( < , 9< ).

variable function, say at ( $ , 9$ ) and we repeat the process. Show that the maximum directions at
Then we follow that maximum slope direction till we get the maximum along that direction as a one

( $ , 9$ ) and ( < , 9< ) are perpendicular.

Q12. Explicitly write the error functions 6 for each of the data sets in Q7 and use the steepest descent
method to minimize it and find the coefficients of the least square polynomials.

Q13.One application of Taylor series it to solve the ODE !/ = ( , 9), 9( < ) = 9< . Show that by starting
!O

from 9< = 9( < ) followed by 98 $ = 98 + ℎ ( 8 , 98 ) will give you a good approximation to 9( ) at

8 $ = 8 + ℎ if ℎ is suitably small. The above method is called the Euler’s method.
Use it to solve !/ = + 9, 9(0) = 1 and find the value 9(1). Directly solve the ODE and compare the
!O

results.

Q14.Use the Taylor series for the Euler’s method to show that the local truncation error
9( + ℎ) − 9$ = 9′′(W) and express error in terms of .
TU
<

Q15. We saw in both the natural cubic spline interpolation and in Gaussian quadrature that we will be
forced to solve large systems of linear equations. There are numerical methods for those too. One of them
is called the Jacobi Method. Suppose we want to solve

$ + $9 + $X = $ 8 $ = ( $− $ 98 − $ X8 )/ $
+ 9+ X = which can be re written as 98 $ = ( − 8 − X8 )/
A + A9 + AX = A 8 $ = ( A− A 8− A 98 )/ A

and start at ( < , 9< , X< ) and do iterations. Use this method to solve the systems of linear equations you
obtain in Q5. Will the method always converge?

Q16. We see that we already have new values for some variables after the 1st and 2nd steps which can be
used in the next step immediately. Use this idea to write a better method (Gauss-Seidel Method) to solve
the systems in Q5. Will the method always converge?

Q17. Suppose we have an inconsistent (more variables than equations) system of equations. Use the ideas
of least square curve fitting to come up with a method to find a least square solution.

Q18. Richardson Extrapolation is a technique to make a numerical solution Z(ℎ) which depends on a step
length ℎ > 0 for an actual solution Z, more accurate.
^ _- 0"_(T)
Assume Z − Z(ℎ) = ℎ\ + ℎ: + ⋯where < 1 . Show that Z − Ù
^ "$ is having a lesser error.
^ _ - 0"_ (T)
So if Z8 (ℎ) is the = th iteration for Z a better one is Z8 $ (ℎ) =
J Ù J
^ "$ .

Q19. Assume that we are going to approximate Z = ( ) by Z< (ℎ) =

a(/ T)"a(/)
T
approximation Z$ (ℎ).
. Find a better

Q20. Use the above method to improve the answer (called Romberg Integration) obtained from the
LU
Trapezoidal method for b< E " U
$
.