ME 2016 - Midterm Exam 1 (PRACTICE)

Instructor: Prof. Emily D. Sanders

Georgia Institute of Technology

Name:

Instructions:

1. This examination contains 15 pages, including this page.

3. You may use one (1) double-sided 8.5” × 11” page with notes that you have prepared.
4. You may use a calculator (that is not on your phone).
5. Please read and sign the below Honor Agreement.

Having read the Georgia Institute of Technology Academic Honor code, I understand and accept my respon-
sibility as a member of the Georgia Tech Community to uphold the Academic Honor Code at all times. In
addition, I understand my options for reporting honor violations as detailed in the code.

Signature:

1
Problem number Available points Your points

1 16
2 6
3 10
4 10
5 18
6 10
7 12
8 18

Total 100

2
Problem 1: Numerical methods and mathematical models [16 pts]
(a) [6 pts] Define numerical methods.

(b) [6 pts] List one reason numerical methods are useful in engineering.

3
(c) [5 pts] A mathematical model for a freefalling parachutist of mass, m, before releasing his parachute,

dv c
= g − v, (1)
dt m
can be solved to determine his velocity, v, as a function of time, t, where c is the drag coefficient and
g is the gravitational constant. Identify the independent and dependent variables.

4
Problem 2: Precision and accuracy [6 pts]
You design a part with an expected mass of 25 g. You print the part 10 times on two different 3D printers
and record the mass of each print. With Printer A, the mass is consistently 3 g less than expected. With
Printer B, the mass is consistently within +/- 3 g of the expected value. Which printer is more accurate?
Which printer is more precise?

5
Problem 3: Roundoff vs. truncation error [10 pts]
(a) [5 pts] Explain the difference between roundoff error and truncation error.

6
(b) [5 pts] If I were to approximate the integral,
Z 2
−x2 − 4 dx,


I=
−2

as I ≈ 2 × (1 × 3) + (1 × 4), where the approximation comes from summing the rectangles as shown
below, would the error in my computation be due to roundoff error or trunction error? Explain
your reasoning.

7
Problem 4: Floating point representation [10 pts]
Consider the single-precision IEEE floating point system in which a number, x, is represented by,

f l (x) = (−1)s (1.f )(2(b−bias) ),

and 1 bit is used to store the sign, s, 8 bits are used to store the b portion of the exponent (recall exponent,
e = b − bias), and 23 bits are used to store the fraction, f , in the mantissa.
Discuss the how the allocation of bits to the exponent versus the fraction influences the range of numbers
that can be represented and the precision of those numbers (i.e., what if I took some of the bits from the
fraction and gave them to the exponent and vice versa?).

8
Problem 5: Root-finding [18 pts]
In Simple Fixed Point Iteration, we iterate according to,

xi+1 = g (xi ) , i = 0, 1, 2, . . .

Consider g (x) = −0.25x2 + 1 shown in the figure below.

(a) [6 pts] Given initial guess, x0 = 1.5, draw the sequence of root estimates based on Simple Fixed Point
Iteration directly on the plot. Based on your drawing, does Simple Fixed Point Iteration converge or
diverge?

9
(b) [6 pts] Given initial guess, x0 = 1.5, quantify the values of the next three root estimates, x1 , x2 , x3 , and
show that the absolute approximate percent relative error, |εa |, associated with each of these iterations,
agrees with your answer to (a).

10
(c) [6 pts] Use arguments of the Fixed Point Theorem to further justify your answer to (a).

11
Problem 6: Root-finding [10 pts]
(a) [5 pts] What is the difference between bracketing (e.g., bisection, false-position) and open methods
(e.g., Fixed-point, Newton-Raphson, Secant method) of root-finding? Which ones are (typically) more
robust? Which ones are (typically) more efficient?

(b) [5 pts] In root-finding, describe a circumstance in which Newton-Raphson might be preferred over
Secant method. Describe a circumstance in which Secant method might be preferred over Newton-
Raphson. Explain your reasoning.

12
Problem 7: Convergence [12 pts]
Sketch directly on the plot below how the (upper bound) ratio of true errors at consecutive iterations,
|Et,i+1 |/|Et,i |, changes over the iterations for linear convergence and superlinear convergence. Explain
how you arrived at your sketch.

13
Problem 8: Regression [18 pts]
In linear least-squares regression, we seek to solve the optimization problem,

n
X 2
min Sr = (yi,meas − yi,model ) . (2)
a0 ,a1 ,...,am
i=1

Given a set of measured data shown in the plot below, the optimization problem in Eq. (2) was solved to
find the polynomial fit indicated on the plot.

(a) [6 pts] Based on the plot above, fill in the table below. Note: order does not matter.

i xi yi,meas yi,model
1
2
3
4
5
6
7

14
(b) [6 pts] Write the Vandermonde matrix, [Z], that maps between the polynomial coefficients of our model
and the y1,model , . . . , yn,model associated with x1 , . . . , xn . Please plug values into the Vandermonde
matrix.

(c) [6 pts] The coefficient of determination,

St − Sr
r2 = ,
St

is r2 = 0.58 for this problem. Explain what the value of r2 tells us in general, making use of the
definitions of St and Sr . What does r2 = 1 mean? What does r2 = 0 mean?

15