Sie sind auf Seite 1von 3

Nonlinear Optimization

Tutorial 1
School of Mathematics
The University of Edinburgh
Year 2016/2017

1. Find the local minima and maxima of the function f (x) = (ax1 − x2 )2 + (1 − ax1 )2 ,
where a 6= 0 is a real number.
Solution:
The gradient of f is ∇f (x) = (4a2 x1 − 2ax2 − 2a, 2x2 − 2ax1 ).
Since a 6= 0, the only stationary point is x∗ = (1/a, 1). Finally, the Hessian matrix of f is
constant:  
2 4a2 −2a
∇ (x) =
−2a 2
Since the minors are ∆1 = ∆2 = 4a2 > 0, the matrix is globally positive definite, which
means not only that x∗ is a local minimum but also that f is convex, and, therefore, x∗ is
a strict global minimum.
2. A manufacturer is designing a biodegradable carton with the shape of a rectangular
prism made from recycled materials. The cost of the material is based on the surface area
of the rectangular container that will house the product. The cost per unit area is £1.5
per square meter. The container must hold a volume of exactly 0.032 m3 . The perimeter
of the base must be less than or equal to 1.5 m. Its sides are scaled geometrically to hold
information labels. The width should not exceed three times the length. Its height must
be less than two thirds the width. Its length and width are less than 0.5 m.

(a) Write a model to determine the measures of the container at minimum cost.
(b) Find the stationary points for the unconstrained problem. What can you say about
them and the solution(s) of the constrained problem?

Solution:

(a) Let x1 be the width, x2 the length, and x3 the height of the container.
The area of the front/rear side is x1 x3 , the area of the left/right side is x2 x3 , and the
area of the top/bottom side is x1 x2 . If we multiply the sum of these areas by 2 (because
there two of each) and by 1.5, the objective function is

f (x1 , x2 , x3 ) = 3(x1 x2 + x1 x3 + x2 x3 ).

Now, we have the following constraints:


• Volume: x1 x2 x3 = 0.032.
• Maximum perimeter of the base: 2(x1 + x2 ) ≤ 1.5.
Nonlinear Optimization - Tutorial 1 2

• Ratio width/length: x1 ≤ 3x2 .


• Ratio height/width: x3 ≤ 2x31 .
• Maximum length and width: x1 , x2 ≤ 0.5.
Finally, the measures are nonnegative: x1 , x2 , x3 ≥ 0.
(b) The gradient is ∇f (x) = 3(x2 + x3 , x1 + x3 , x1 + x2 ). The only stationary point
is x∗ = (0, 0, 0), but this point is not feasible for the constraint problem. Therefore,
constrained problems must be solved with different techniques.

3. Let f (x1 , x2 ) = −2x21 + x1 x22 + 4x41 .

(a) Find the local minima and maxima.


(b) Are any of them global?

Solution:

(a) The gradient is ∇f (x1 , x2 ) = (−4x1 + x22 + 16x31 , 2x1 x2 ).


In order to find the stationary points we need to solve

−4x1 + x22 + 16x31 = 0,


2x1 x2 = 0.

Using the second equation, we have that either x1 = 0 or x2 = 0.


If x1 = 0, then x2 = 0. If x2 = 0, then −4x1 + 16x31 = 0. So, x1 (1 − 4x21 ) = 0, which
1 1
means that x1 = 0, x1 =
1 1
 − 2 , or x1 = 2 . Therefore, we have three stationary points:
(0, 0), − 2 , 0 , and 2 , 0 .
Now, the Hessian matrix of f is
 
2 −4 + 48x21 2x2
∇ f (x1 , x2 ) = .
2x2 2x1

Let us evaluate it at the stationary points.


 
2 −4 0
• ∇ f (0, 0) = is negative semidefinite. This could indicate either a local
0 0
maximum or a saddle point. However, f (0, 0) = 0 and, if α > 0, then f (α3 , α) = −2α6 +
α5 + 4α12 = α5 (−2α + 1 + 4α7 ) > 0 for all α sufficiently small. Thus, (0, 0) cannot be
a local maximum,which meeans that it is a saddle point.
8 0
• ∇2 f − 21 , 0 = is indefinite. Thus, − 21 , 0 is a saddle point.
 
0 −1
 
1 8 0
is positive definite. Thus, 12 , 0 is a strict local minimum.
2
 
• ∇ f 2, 0 =
0 1
(b) Since lim f (−1, x2 ) = lim 2 − x22 = −∞, then f is not bounded below and
1
 x2 →+∞ x2 →+∞

2
, 0 is not a global minimum.
Nonlinear Optimization - Tutorial 1 3

0.8

0.6 3

0.4
2

0.2
1
0

−0.2 0

−0.4
−1
1
−0.6
0.5 1
−0.8 0 0.5
0
−0.5 −0.5
−1
−1 −0.5 0 0.5 1 −1 −1

Figure 1: f (x1 , x2 ) = −2x21 + x1 x22 + 4x41 .

Das könnte Ihnen auch gefallen