Final Exam Semester 1, 2023

The University of Sydney

Faculties of Arts, Economics, Education,
Engineering and Science

MATH5410: Special Topics in Applied Mathematics

Lecturer: Nalini Joshi and Ayesha Sohail

Time allowed: 2 hours

This booklet contains 6 pages.

• This is a 2-hour exam.

• There are two questions; each is worth half the total mark for the exam.
Attempt all questions.

SOLUTIONS
Final Exam Semester 1, 2023 page 2 of 6

1. Assume that a weight function w defined for x ∈ R, x ≥ 0, and t ∈ R, is given by

w(x; t) = xρ exp(−x2 + t x), ρ > −1,

and that we have a family of monic polynomials {Pn }∞ n=0 , which satisfy
Z ∞
⟨Pn , Pm ⟩ = Pn (x)Pm (x)w(x; t)dx = hn δnm . (1.1)
0

Define the moments µn (t), n ∈ N by

Z ∞
µn (t) = xn w(x; t)dx.
0

Note 1: Each of the functions and coefficients considered in this question may depend
on t, even when we do not state its dependence on t explicitly.
Note 2: Cramer’s rule: Given a system of equations of the form A v = b, where A
is an n × n matrix and v and b are n-column vectors, the solution is given by vi =
det(Ai )/ det(A), where vi is the i-th component of v and Ai is the matrix formed by
replacing the i-th column of A by the column vector b.
(a) Show that for all integers k ≥ 0, we have

dk µ0
= µk .
dtk
R 
xm w(x; t)dx = xm wt (x; t)dx. The result follows from
R
Solution: Note that
t
the fact that wt (x; t) = xw(x; t). Mark 2

(b) Assume N ≥ 1 is an integer and write

N
X −1
PN (x) = xN + cj xj .
j=0

(i) Show that, for 0 ≤ m ≤ N , we have

Z ∞ N
X −1
m
x PN (x)w(x; t)dx = µN +m + cj µj+m . (1.2)
0 j=0

xm PN (x)w(x; t)dx = µN +m +
R
Solution: The definition of moments gives
PN −1
j=0 cj µj+m . Mark 1

(ii) Use equation (1.2) to write down a system of linear equations of the form
A v = b, where A is an (N + 1) × (N + 1) matrix and v and b are N + 1-
component column vectors involving cj , 0 ≤ j ≤ N − 1, hN , and µN +k ,
0 ≤ k ≤ N.
Solution: By the orthogonality condition for m < N , the integral in Equa-
PN −1
tion (1.2) must be zero, which gives a system of linear equations j=0 cj µj+m =
−µN +m , for 0 ≤ m ≤ N − 1. The case m = N gives a similar equation but

SOLUTIONS turn to page 3

Final Exam Semester 1, 2023 page 3 of 6

the integral is now equal to hN . The resulting system of equations has the Mark 2
form A v = b, where
     
µ0 . . . µN −1 0 c0 µN
 .. .. .. ..   ..   .. 
 . . . .  . 
A= , v =  , b = −  . .
  
µN −1 . . . µ2N −2 0 cN −1  µ2N −1 
µN . . . µ2N −1 1 −hN µ2N
Mark 3

(iii) Use Cramer’s rule (see Note 2 above) to show that

△N +1 (t)
hN = ,
△N (t)

where, for n ≥ 1, △n (t) are defined to be the determinants

 
 µ0 µ1 . . . µn−1 

 µ1 µ2 ... µn 
△n (t) =  . .

. .. ..
 . . ... . 
 
 µn−1 µn . . . µ2n−2 

Solution: The result follows from Cramer’s rule applied to the last com-
ponent of the vector v. Mark 2

(c) Show that the polynomials {Pn (x)}∞n=0 , satisfy a 3-term recurrence relation of the
form
x Pn (x) = Pn+1 (x) + an (t) Pn (x) + bn (t) Pn−1 (x), (1.3)
where we assume the convention P−1 = 0, P0 = 1. Find the explicit relations giving
an , bn in terms of hn .
Solution: Completeness allows us to expand xPn (x) as a linear combination of
{Pk (x)}n+1
k=0 . But orthogonality shows that ⟨xPn , Pk ⟩ = 0 for k < n−2. So Equation
(1.3) holds. The coefficients an , bn can be expressed as an = ⟨xPn , Pn ⟩/⟨Pn , Pn ⟩,
bn = ⟨xPn , Pn−1 ⟩/⟨Pn−1 , Pn−1 ⟩. But, reusing Equation (1.3) shows that we have Mark 2
bn = hn /hn−1 . To find an , note that because xw = ∂t w, we get ⟨xPn , Pn ⟩ = ∂t (hn ),
which implies an = ∂t (hn )/hn .

(n)
(d) Show that the following equation holds and find the coefficients αj explicitly
n
∂Pn X (n)
= αj Pj . (1.4)
∂t
j=0

Solution: Since the left side is a polynomial, we can expand it in terms of {Pj }
and since Pn is monic, the first term of the expansion must have degree n − 1, i.e.,
(n)
αn = 0. Applying orthogonality to find ⟨∂t Pn , Pm ⟩ for 0 ≤ m ≤ n − 1, shows that
(n) (n)
αj = 0 for 0 ≤ j ≤ n − 2, while αn−1 = −hn . Mark 2

SOLUTIONS turn to page 4

Final Exam Semester 1, 2023 page 4 of 6

(e) Use Equations (1.3) and (1.4) to show that an and bn satisfy

dan
= hn+1 − hn
dt
dbn

= hn an − an−1 .
dt

Solution: Differentiate Equation (1.3) with respect to t and use orthogonality and
Equation (1.3) again to get an identity involving {Pn , Pn−1 , Pn−2 }. Since these are
orthogonal polynomials, their coefficients vanish identically, leading to the desired
result. Mark 4

2. (a) Show that the Chebyshev polynomials of first kind defined as

Tn (x) = cos nθ

with
x = cos θ
are orthogonal but not orthonormal with respect to the following inner product:
Z 1
Ti (x)Tj (x)
⟨Ti , Tj ⟩ = √ dx (2.1)
−1 1 − x2

Solution:

(b) Define an orthonormal set of Chebyshev polynomials of the first kind by the method
of scaling.
Solution:

SOLUTIONS turn to page 5

Final Exam Semester 1, 2023 page 5 of 6

(c) Find the four Chebyshev base points for interpolation on the interval [0, π/2], and
find an upper bound for the Chebyshev interpolation error for f (x) = sin x on the
interval.
Solution:

(d) To verify that the discrete orthogonality holds “exactly” for the Chebyshev poly-
nomials of first kind, first verify the equality:

n+1
X sin ((n + 1)θ)
cos ((2k − 1) θ/2) = , (2.2)
2 sin(θ/2)
k=1

and then show that aij = 0 ∀i ̸= j, where aij is defined as:

n+1
X
aij = Ti (xk )Tj (xk ) (2.3)
k=1

Solution:

SOLUTIONS turn to page 6

Final Exam Semester 1, 2023 page 6 of 6

SOLUTIONS This is the end of the examination paper