Vous êtes sur la page 1sur 4

Computational Economics WS 14/15

Prof. Neumann

Exam Preparation 1:
Basic Concepts and One-Dimensional Minimization

1. What is Computational Economics?


- using computational tools for both economic modelling and for the computational solution
of analytically formulated economic problems.

2. Can you name some of the benefits of the computational analysis?


- Identification of what is important and what is not
 Conventional analyses is of qualitative nature creating intuition and often
ignores quantitative importance of the results
 Theory can identify qualitative features of an equilibrium but cannot(easily)
identify the significance of any feature
- Relaxation of simplifying assumptions
 Theoretical analyses often make simplifying assumptions that lead to non-
robust conclusions
 Conventional theory is very good at
 Telling us what is possible
 Exploring the qualitative nature of economic phenomena

Computational analysis is needed to determine the actual importance of the analysis

3. What is the binary equivalent of the decimal number 156?


- 156/2 = 78 remainder is 0
- 78/2 = 39 remainder is 0
- 39/2 = 19 remainder is 1
- 19/2 = 9 remainder is 1
- 9/2 = 4 remainder is 1
- 4/2 = 2 remainder is 0
- 2/2 = 1 remainder is 0
- 1/2 = 0 remainder is 1

Binary number is 10011100

4. How many methods can you name for economizing on computational costs of evaluating the
following : (polynomial evaluations costs)
n
k 0 ak xk n
k 0 ak xk ∑𝑛 𝑎 𝑥 𝑘
𝑘=0 𝑘
- Direct Method 1
- Direct Method 2
- Horner’s Method
5. Show how Horner's Method can be applied to transform the octal number 13418 into
decimal. Write down the required steps and equations. What is the result?
- 3+8*(5+8*(2+8*1))=683

6. What is the most effective way to compute derivatives?


- Finite differences

7. Which forms of finite differences are commonly considered? Please name all of them and
write down their formulas.
- Forward
- Backward
- Central
o For exact formulae, see Professor’s slide No. 49!

2
8. Approximate the derivative (using forward differences) of f(x) = −2𝑥 3 + 7𝑥 2 − at point
𝑥
x=2 with initial distance h=0.022 and step-size ϵ=0.005. In your calculations, you will get the
required h for each evaluation by applying the formula: hi=hj-1-ϵ for i=2, ..., 5. Provide just
two calculation steps.
"i" "x" hi "f(x + "hi ")" "f(x)" "f"′fi "(x)"
- 1 2 0.022 11.0964 11 4.3818
2 2 0.017 11.075 11 4.4118

9. What can be the sources of most errors?


- Rounding
 a computer must round off the results of many arithmetic results
 whether a rational number has a terminating expansion depends on the
base
- Mathematical truncation
 Mathematical objects are often defined as infinite sums
 At some point of time the process needs to be ended
- Turning a continuous function into a discrete function
- Underflow, overflow

10. Calculate eigenvalues and eigenvectors for the following matrix:


1 −1
( )
−1 1
- 𝜆1 = 2, 𝜆2 = 0 (eigenvalues)
- So, the eigenvectors (−1,1) correspond to eigen value 𝜆1 .
- the eigenvectors (1,1) correspond to eigenvalue 𝜆2 .

11. What is the Spectral radii? Calculate spectral radii for the above matrix.

- p(A) =max{λ}
where λ refers to eigenvalues of A

- p(A) =max{1,1} =2
12. Please deliver the first two terms (n=2) of the Taylor Approximation for the function cos(x)
centered at point 𝑥0 = 0
- cos(x)= cos(𝑥0 )- sin(𝑥0 )𝑥 − 0.5 cos(𝑥0 ) 𝑥 2
- cos(x) = 1-1/2𝑥 2

13. Deliver the Hessian matrix for the function f(x) = sin(x) * cos(x) and analyse the nature of the
stationary point using Hessian characteristics.
𝜋
(𝑥, 𝑦) = ( , 0)
2

𝜋 −Sin[𝑥]Cos[𝑦] −Cos[𝑥]Sin[𝑦] −1 0
- (𝑥, 𝑦) = ( , 0) , 𝐻 = ( )=( )
2 −Cos[𝑥]Sin[𝑦] −Sin[𝑥]Cos[𝑦] 0 −1
- H1 = −1 ≺ 0, H2 = 1 ≻ 0
𝜋
- Hessian matrix is negative definite, ( 2 , 0) is a maximum point

14. What is the other definiteness of Hessian and what are the corresponding nature of
stationary point?

See slide 177.

Definiteness H Nature x Eigenvalues


Positive d. Minimum >0
Positive semi-d. Valley >= 0
Indefinite Saddle point =/ 0
Negative semi-d. Ridge =< 0
Negative d. Maximum <0

15. What is the difference and similarities between Bracketing and Sectioning methods?
- The goal of bracketing is to find an interval [a0,b0] that contains the minimum of the
function, while with the sectioning method we aim at iteratively reducing the size of
this interval until a certain precision is reached.
- Both work better with unimodal functions.

16. How to use the Bracketing method if the function is multimodal?


- Break up the function into parts, and treat each of them as unimodal and apply the
method
𝑥4 2𝑥 3 𝑥2
17. 𝑏[x]: = 40 + 5
+ 4
− 12𝑥

Using Δ=2 and γ=1.2, identify the sub-intervals of b in (-12,5) that are provided by the Bracketing
approach.

Let x1=-12. Just calculate the Bracketing distances. For now, only calculate x2 and x3 .

x2 −10
-
𝑥3 −7.6
18. Still referring to function b(x) above, Apply the Sectioning using Dichotomous Search using a
resolution of δ=0.01 to identify the minimum in the given interval [-12,-7] within 5% of its
exact value. Please, perform the initial step (n=0) and n=1.

"n" ln rn Ln x"n"",""1" x"n"",""2" "f("x"n"",""1" ")" "f("x"n"",""2" ")" "center"


0 −12 −7 5 −9.505 −9.495 −2.7899 −2.7318 −9.5
1 −12 −9.495 2.505 −10.7525 −10.7425 −5.1535 −5.1825 −10.7475

Log(2)
19. Perform the Golden Sectioning method on the function 𝑓(𝑥) = 𝑥 ∗ Log(𝑥) on the interval
[1.5,7.5]. Please, perform the initial step n=0 and n=1.

"n" "ln" "rn" "Ln" X "1"",""n" X "2"",""n" "f("x"1"",""n" ")" "f("x"2"",""n" ")"
- 0 1.5 7.5 6 3.79 5.21 1.97 2.19
1 1.5 5.21 3.71 2.92 3.79 1.89 1.97

𝑥 3 −4𝑥 2
20. Apply the Fibonacci search to the function g(x)=Cos[√𝑥]* 1000
on the interval [50,134].
Before you start your calculations, determine which Fibonacci number you have to start the
sectioning with, such that the length of the final interval is not more than 10% of the initial
interval's length. Please, perform initial step n=0 and n=1.
"n" "ln" "rn" "Ln" F N X "1"",""n" X "2"",""n" "g("x"1"",""n" ")" "g("x"2"",""n" ")"
0 50 134 84 21 82 102 −489.09 −796.18
1 82 134 52 13 102 114 −796.18 −447.65

21. Given the function f(x)=1/2 x3+(10-x)2+7x-75 and initial interval L0=[-2,5] present the first
step of the Bisection Minimization Method.
"n" an bn cn "f′["an "]" "f′["bn "]" "f′["cn "]"
-
0 −2 5 1.5 −11. 34.5 −6.625
22. Why Secant method does not always converge?
- since the property of the unimodality of the function does not need to hold in order
to apply this method. The convergence is, therefore, dependent on the initial guess
of x0 and x1
23. Given the function f(x)=(ln(x))4+0.2x3+1/x2 and initial points x0=0.5 and x1=4 perform just
ONE iteration of Secant method (calculate x2) to find the minimum of f(x).
𝑥 −𝑥 4−0.5
- 𝑥2 = 𝑥1 − 𝑓′(𝑥1 ) · 𝑓′(𝑥 1)−𝑓′(𝑥
0
)
= 4 − 𝑓′(4) · 𝑓′(4)−𝑓′(0.5) = 2.6075
1 0

24. Minimize f(x)=30.5x+x/(5+x2)-0.25x3 using Newton's method given initial point x0=5. Show the
calculation of first iteration(that is x1).
𝑓′(𝑥 ) 𝑓′(5)
- 𝑥1 = 𝑥0 − 𝑓''(𝑥0 ) = 5 − 𝑓''(5) = 1.3442
0
25. What is the disadvantage of the Newton’s Method?
- Jumping from one point to another one
- It may diverge
26. What is the difference of Quasi-Newton from Newton’s method?
- If the function f(x) is not available in closed form or it is difficult to differentiate (not
possible to use Newton's method), then the finite differences are used to approximate
the derivatives, that is Quasi-Newton method is applied.

Vous aimerez peut-être aussi