Definition 1.

A function z = f(P) is said to be strictly globally concave (strictly globally

convex) in his domain D, if for any two distinct point P, Q D and for any number (0,1)
we have
f(P) + (1- ) f(Q) < f(P + (1-)Q)
(f(P) + (1- ) f(Q) > f(P + (1-)Q), respectively)

In Figure 8 is presented the graphic of the strictly globally concave function and in
Figure 9 the graphic of the strictly globally convex function. It is known, that a strictly
globally concave (strictly globally convex) function has not more than one local maximum
(local minimum).

Theorem 7. If a function z = f(P) is strictly globally concave (strictly globally convex) and A
is an extreme point of f, then this extreme point A is also the global maximum (the global
minimum) point of f.

Theorem 8. If (P) > 0 and z 'xx' ( P ) < 0 (z 'xx' ( P ) > 0 ) for each P D, then a two time
differentiable function z = f(P) is a strictly globally concave (strictly globally convex) in D. A
two time differentiable function y = f(x) is a strictly globally concave (strictly globally
convex) in D, if f ( x) < 0 ( f ( x) > 0 ) for each x D.

y
x

Figure 8: the graphic of the strictly globally concave function z = f (P ).

Example 28. To find the extreme points of the function z = ln(xy) (x + y) 2

Solution. First we find the partial derivatives
'
z x = y / (xy) 2(x + y) = 1/x 2(x + y)
z 'y = 1/y 2(x + y)

Equalizing these partial derivatives to zero, we get the system

46
 z

y
x
Figure 9: the graphic of the strictly globally convex function z = f (P ).

1
x 2( x + y ) = 0
1
2( x + y ) = 0
y
As now
1
x = 2( x + y )
1
= 2( x + y )
y
then
1/x = 1/y y = x

Replacing y = x into the equation 1/x = 2(x + y), we find

1/x 4x = 0 (1 4x 2 ) / x = 0 x 1 = 1/2, x 2 = 1/2 y 1 = 1/2, y 2 = 1/2
Consequently we have two stationary points A = (1/2, 1/2) and B = (1/2, 1/2). As
z 'xx' = 1/x 2 2 < 0; z 'yy' = 1/y 2 2; z 'xy' = z 'yx' = 2
and
= (1/x 2 + 2) (1/y 2 + 2) 4 > 0

for each P D, then A and B are the global maximum points of the given function by
Theorems 7-8 and z(A) = z(B) = ln (1/4) 1.

13. Some multivariable economical models

I. Let
p the unit price of the given product,
A the advertisement cost,
Q the consumers demand of this product,
- the profit of the firm.

47
We assume that Q is a function of p and A, it means Q = Q(p,A) and suppose that this function
has the form
Q = (a p)A b ; a, b are positive parameters.
We see that Q is linear and decreasing in price and nonlinear and increasing in advertaising.
We also assume that the firm operates with a constant marginal cost of c dollars per unit of
production. Then the profit function is
= pQ cQ A or = (p c) (a p) A b A
Of course, this model has the sense only if p > c and p < a.
Now we find the maximal profit. For this purpose we at first find the stationary points.
We find the partial derivatives
'p = (a p) A b (p c) A b = (a 2p + c) A b

'A = b (p c) (a p) A b 1 1

and equalize these partial derivatives to zero:

(a 2 p + c) A b = 0
b 1
b( p c)(a p ) A 1 = 0
From the first equation of this system we have
p* = (a + c)/2 (18)
Hence
p* c = (a c)/2, a p* = (a c)/2

and from the second equation of the system we have

b [(a c)/2] 2 A b 1 = 1 Ab 1 = b 1 [ (a c) / 2]
2

A* = {b [(a c)/2] 2 } 1b (19)

Thus (p*, A*) is the stationary point. Now we find the second partial derivatives:

'p' 2 = 2A b , 'A' 2 = b (b 1) (p c) (a p) A b 2 , 'pA' = 'Ap' = (a 2p + c) b A b 1

As
b
b 1b
(p*, A*) = = 2 (a c) 2 < 0
''
p2
4
1
b2
b b 1
'A' 2 (p*, A*) = b (b 1) [(a c)/2] 2 {b [(a c)/2] 2 } 1 b
= (b 1) (a c) 2
4
'pA' (p*, A*) = 'Ap' (p*, A*) = 0
and
1
b
((p*, A*)) = 2(b 1) (a c) 2 > 0
4

for 0 < b < 1, then (p*,A*) is the local maximum of by Theorem 6.

48
 In the economical analysis are used short-time and long-run analysis. Duration of the
short-time and the long-time period is relative and depend on the problem. In the short-time
analysis we can consider some variables as constants, in the long-run analysis all variables are
changing.
The short-time analysis. We consider two cases:
1) Let p = p 0 be fixed for a short time. Then
= (p 0 c) (a p 0 ) A b A
is a one-variable function = (A) and
d
= b (p 0 c) (a p 0 ) A b 1 1 = 0
dA
1
1 b
A* = [b (p 0 c) (a p 0 )] (20)
As
d 2
= b (b 1) (p 0 c) (a p 0 ) A b 2 < 0
dA2
for each A, 0 < b < 1, then = (A) is the globally concave function by Theorem 8 and hence
A* is the global maximum point.
2) Let A = A 0 be fixed for a short time.. Then
= (p c) (a p) A0b A 0
is a one-variable function = (p) and
d
= (a 2p + c) A0b = 0 p* = (a + c)/2
dp
i.e.we have got the point, determined by (18). As d 2 / dp 2 = 2 A0b < 0 for each p, then p* is
the global maximum point of = (p).
The long-time analysis. Then p and A both are changing. First we note that in the short-
time analysis the value of p* does not depend on A. Therefore the global maximum of the
two-variable function = (p, A) can be only in the point, where p = p* = (a + c)/2. As for
fixed p, i.e. for p = p 0 , the maximum of lies in such a point (p 0 , A), where A = A* is
determined by (20), then taking p 0 = p* in (20) we get that the value of A* is determined by
(19), as before. Thus the long-run global max also lies in (p*, A*), determined by (18) and
(19).

Remark. It is possible to show, that the strict global concavity is not usable in this case,
because ((p,A)) is not positive for all p and A.

II. Cobb-Douglas technology.

One way of writing a profit function is the difference between the total revenue (TR) and the
total costs (TC):
= TR TC = pQ wL rK,
where p is unit price, Q output, L labour, K kapital, w and r are the unit prices of L and
K. In addition to it, the production or output Q is a function of L and K, it means Q = F(L,K).
The last function called the production function. A function
Q = cL a K b ,

49
where a, b, c are positive parameters, is called the Cobb-Douglas production function. For
simplicity we consider only the special case where c = 1 and a = b, it means Q = L a K a .
Thus we have now the profit function
= p L a K a wL rK
We consider here L and K as the variables and p, w and r as parameters. First we find the
stationary points. For this purpose we find the partial derivatives of and equalize these
derivatives to zero:
'L = apL a 1 K a - w = 0 (21)
'
K = apL a K a 1 - r = 0 (22)
From (21) we have
1

K a = [w/(pa)] L 1 a K = {[(w/(pa)]L 1 a } a

Substituting the last equality into (22), we get

1 1 a 1 a a 1 (1 a )( a 1)
1 a a 1
a
apL {{[(w/(pa)]L a
} } = r pp a
aa a
w a
L L a a
=r

1 1 a 1 2 a 1 2 a 1 1 1 1 a

paaa w a
L a
=rL a
=p a
a a
w a
r
1

L* = (apw a 1 r a ) 1 2 a
(23)

As the Cobb-Douglas production function is symmetrical of variables L and K, then

1
a 1 a 1 2 a
K* = (apr w ) . (24)
Now we find the second partial derivatives:

''
LL = a(a 1)pL a 2 K a , ''
KK = a(a 1)pL a K a 2 , ''
LK = a 2 pL a 1 K a 1

= p 2 a 2 (a 1) 2 L 2 a 2 K 2 a 2 - p 2 a 4 L 2 a 2 K 2 a 2 = p 2 a 2 L 2 a 2 K 2 a 2 (1-2a)

As > 0 and 'LL

'
< 0 for all K and L, if a 1 < 0 and 1 2a > 0 or 0 < a < ., then has the
global max imum in (L*,K*) by Theorems 7-8.

Now we study more precisely relations (21) and (22). From (21) and (22). we get
w = apL a 1 K a (25)
r = apL a K a 1 (26)

Definition 2. Let two economical variables x and y are connected by the function y = f(x).
Then the derivative y = f(x) is said be the the marginal of y with respect to x and denoted by
My, i.e. My = y.

The marginal My shows the change of y per infinitely small changes of x (that we
measure by the change of y per unit change of x. We can also say, that the marginal shows the
approximate change of y per unit change of x, if this unit is sufficiently small for given value
of x.

50
 As the labour costs CL = wL, then the marginal costs of labour
M(CL) = CL= w
As the capital costs CK = rK, then the marginal costs of the capital
M(CK) = CK= r
As the revenue R = p L a K a , then the marginal revenue with respect to the labour
MR(L) = R 'L = p L a K a
and the marginal revenue with respect to thecapital
MR(K) = R 'K = apL a K a 1 .
Thus the relation (25) we can interpretate economically as follows: the additional revenue,
generated from hiring an extra unit of the labour, must equal to the wage rate paid for extra
unit of labour. Similarly the relation (26) we can interpretate as the additional revenue,
generated from using an extra unit of the capital, must equal to the price, paid for extra unit of
the capital.

14. Comparative statics

In the Cobb-Douglas model L and K are variables and p, w, r and a are parameters.
Also L and K are called endogeneous variables, it means they are variables, which value can
be determined by agents (for example by firms) decisions (they are determined from inside
the model). Parameters p, w, r and a also are called exogeneous variables, it means they are
variables, which are beyond the control of the agents in the model (they are determined from
outside the model).
We have solved the Cobb-Douglas model for the endogeneous variables, we have
found L* and K* via exogeneous variables p, w, r and a. Now it is important to predict how
the changes in the environment (it means changes in p, w, r and a) will affect these values L*
and K*. This type of analysis is called the comparative statics. Mathematically the
comparative statics involves taking the derivatives or partial derivatives of our solution
functions with respect to the parameters.
In the Section 13 we have found (see (23) and (24)):
1 1

L* = (apw a 1 r a ) 1 2 a K* = (apr a 1 w a ) 1 2 a .

If we wish to know, how the changes in p affect to the values of L* and K*, we take the
partial derivatives:
1 1
1
' a 1 a 1 2 a 1 2 a
L* = (aw
p r ) (1/(1-2a))p >0
1 1
1
K* 'p = (ar a 1 w a ) 1 2 a
(1/(1-2a))p 1 2 a
>0 (1 2a > 0).

It means that increasing p also increase values of L* and K*. However for impact of r we find
1 a
1
' a 1 1 2 a 1 2 a
L* = (apw
r ) (-a/(1-2a))r < 0.

Thus increasing the price of the capital r the optimal value of L* decreases.

51
 Problems

20. To find local (and if possible) global extreme points

z = x 3 y 3 + 3 xy + 8
21. To find local (and if possible) global extreme points and comparative statics results
b) z = ax x 2 + by y 2 ( x + y ) 2
22. The amount of theatre-goers depend on the price of ticket x (Estonian Crones - EEK) for
adults and on the price of ticket y (EEK) for children. The amount of children is given by the
function z1 = 2000 2x 15y and the amount of adults by z2 = 1450 5x 2y. For which
prices x and y the theatre performance gives maximal revenue?
23. A firm sells goods with the price 10 EU per unit and pays for used capital K and labour L
respectively 12 EU and 8 EU per unit. To find the values of K and L ,which guarantee the
maximal profit for the firm and to find corresponding production, if production function is
Q= 4 K3 L.
24. A firm produces products X and Y with quantities x and y respectively. To find the profit
maximazing values for x and y, if the cost function of firm is
1 2 1
C(x, y) = x + axy + y 2
4 4
where a is a positive parameter and demands for X and Y are represented by p X = 8 x and
pY = 16 2 y. To find comparative statics results.
25. A company produces 2 products, whose demand functions are
p1 ( x, y ) = 40 + 2 x 5 y and p2 ( x, y ) = 80 5 x + 16 y
where x and y are respectively the produced quantities of first and second products. The costs
of company are represented by function
C ( x, y ) = 3 x 2 9 xy + 17 y 2 + 34 x + 71 y 24
How much both products must the company produce for getting maximal profit? Find the
maximal profit and the prices of both products in this case.
26. A company sells produced goods with the price p and pay for applied capital K and labour
L respectively 6 and 3 money unit. To find the values of K and L ,which guarantee the
maximal profit for the firm and find corresponding production, if the production function is Q
= a K 3 L , where a is a positive parameter. To find comparative statics results.
27. A firm produces goods X and Y. The demand functions for X and Y are p X = 10 2 x y
and pY = 10 x 2 y. To find the production quantities, quaranteeing maximal profit, if firms
cost function is C = ax + by, where a and b are positive parameters. To find comparative
statics results.

15. Method of Lagrange for optimization with equality constraints

Let us consider the opt. problem with the constraints:
max {f(x); F(x) 0, x 0} (27)
where x = (x 1 ,,x n ), F(x) = (F 1 (x),, F m (x)) and

52
 F 1 (x) = f 1 (x) - g 1 ; . , F m (x) = f n (x) - g m ,

where g 1 ; . , g m are constants. For solving this problem we define

G(x, ) = f(x) - 1 F 1 - - m F m ,
called the function of Lagrange.

Definition 3. Point (x 1 *,,x n *, 1 *, m *) is called the saddle point of G in domain D, if

(x 1 *,,x n *, 1 *, m *) D and for each (x, ) holds the condition
G(x*,) G(x*,*) G(x, *)
Thus the function G has in saddle point max by x and min by .
Theorem 9. If (x*, *) is the saddle point of G for x, 0, then x* is an optimal solution of
the problem (27).

Using this result for finding the optimal solution is difficult, and every function does
not have saddle point. Therefore we apply the other possibilities. First we consider the case, if
instead of inequalities there are equalities, it means we have the problem
max {f(x); F(x) = 0, x 0}.
We form for solving this problem the system
G x' i = 0
, i = 1,,n; j = 1,,m (28)
G' j = 0

Solving this system we get all the critical points of our problem. It is known, that the local
extremum of our problem can be only in such a point, for which equalities (2) are satisfied.
The global extremum can be also in the border of domain, where some x i = 0. For checking
the extremum in the critical points we can use the next results:
Theorem 10. If all the constraints are linear and f is strictly globally concave, then the local
maximum of the Lagange function G is also a global maximum of this function.

Theorem 11. If the Lagange function G for * has the local (global) extremum in the point
x*, then f also has the local (global) extremum in the point x*.

Now we explain the sense of . For simplicity we consider only the case of two-
variable fn z = f(x, y) with one constraint h(x, y) = g (g is a constant): to find max f(x, y), if
h(x, y) = g. Let x= (x*, y*) be the optimal solution of our problem. Clearly x depends on g, it
means x= (x*, y*) = (x(g), y(g)). We denote f*(g) = f(x(g)) and differentiate f* by g:
f * f x f y
= + (29)
g x * g y * g
The Lagrange function has now the form
G(x, y, ) = f(x, y) - (h(x, y) g)
and the system (28) takes the form

53
 G f h
= =0
x x x
G f h
= =0
y y y
G
= g h ( x, y ) = 0

Consequently
f h f h
= =
x x x * x *
f h f h
= =
y y y * y *

Therefore we have
f * h x h y h x h y
= + = + (30)
g x * g y * g x * g y * g

by (29). Further we differentiate the equality h(x*,y*) = g by g and get

h x h y
+ =1
x * g y * g
g
because = 1 . Therefore from (30) we have
g
f *
=
g
Hence we may assert, that show approximately how much increases the maximal value of f,
if g increases by unit.
For general case we can say, that j show approximately how much increases the
value of f, if g j increases by the unit.

Example 29. Let Q = L a K a be the Cobb-Douglas production function, where Q is output, L

labour, K kapital and a is a positive parameter. Find minimal costs for producing Q 0 units
of the given production, if w and r are the unit prices of L and K.
Solution. We need to find the minimum of the cost functionn C = wL + rK for producing
Q 0 = L a K a units of the given production. The Lagrange function is now
G = wL + rK - (L a K a - Q 0 )
and the system (28) has the form:
G 'L = w - aL a 1 K a = 0
G 'K = r - aL a K a 1 = 0 (31)
' a a
G = L K - Q0= 0
From the first two equation we have
= w/(aL a 1 K a ) = r/(aL a K a 1 ) w/K = r/L K = Lw/r
Replacing the last equality into third equation of the system (31), we have

54
 1

L (Lw/r) - Q 0 = 0 L
a a 2a
= Q 0 (r/w) L= [Q 0 (r/w) ]
a a 2a

Consequently
1 1 1 1
L* = (Q 0 ) 2 a (r/w) 2 K* = (Q 0 ) 2 a (w/r) 2
Now
G 'LL
'
= -a(a 1)L a 2 K a , G 'KK
'
= -a(a 1)L a K a 2 , G ''
LK = -a 2 L a 1 K a 1
and
= 2 a 2 (a 1) 2 L 2 a 2 K 2 a 2 - 2 a 4 L 2 a 2 K 2 a 2 = 2 a 2 L 2 a 2 K 2 a 2 (1-2a)

Now > 0 and G 'LL'

> 0 for all K and L, if a 1 < 0 and 1 2a > 0 or 0 < a < . Thus the
global conditional minimum of G is. in the point (L*,K*) by Theorems 7-8 thus the minimal
costs are wL* + rK*.

Problems

28. To find minimal and maximal value of function z = x 3 y + 4 x 2 y x 2 y 2 in triangle,

bordered by lines x = 0, y = 0 and y = 6 x.
29. With the help of method of Lagrange to find conditional extremes:
z = x y; x 2 + y 2 = 1
30. Student Eugen can spend for cigarettes and bear not more than 100 EEK in week. To find,
how much he must spend for both goods in week in order to get maximal utility, if Eugens
utility function for consuming these goods is
x2 y
U ( x, y ) = xy
4 2
where x is money, spended for bear and y is money, spended for cigarettes (in hundreds
crones).
31. A firm sells product X with price 50 EU and product Y with price 45 EU. Besides it is not
allowed to produce these products not more than 100 units. Which quantities of X and Y
maximize the profit, if cost function is
C(x, y) = 2 x 2 + 0,5 xy + y 2 ,
where x is a produced quantity of X and y is a produced quantity of Y, if
a) that firm must produce at least 5 unit of product X
b) that firm must produce at least 15 unit of product X
32. To solve the previous problem for the case that these products not allowed to produce not
more than 25 units.
33. The production function of a company is Q = 3 4 LK ,
unit prices of labour L and capital K are 3 and 5 money unit respectively and resource for
production is 1200 money unit. To find values of L and K, which maximize the profit and
corresponding profit and production quantity, if the production is selled with price 40 money
unit and all 1200 money unit are used thoroughly for production. Is it possible to increase the
profit, if the part of money resources to save? If possible, that values of K and L will
maximize the profit?
34. A student Helmiine works also in restourant. For it she gets 8EU per hour. The utility
function of Helmiine is u = 4 r 4 s 3 , where r denotes solary in euros and s denotes the time in

55
hours, spended for learning. In week Hermiine spends for working and learning alltogether 90
hours. How she must divide time for working and learning for getting maximal utility?
35. To find the minimal and the maximal value of the function z = 2 x 3 + 4 x 2 + y 2 2 xy in
domain, which is determined by inequalities x 2 y 4 .
36. With the help of method of Lagrange to find conditional extremes::
z = x2 + y 2; x + y = 2
37. Companys production function is Q = 3 3 LK and this production are selled with price 20
EU. To find values of L and K, which maximize the profit, if unit prices of labour L and
capital K are 2 and 4 money unit respectively and for production that company can use
a) 3000 eurot, b) 1000 eurot

16. Optimization with inequality constraints. Kuhn-Tucker conditions

Let us consider the one-variable optimisation problem without constraints:

max y = f ( x)
(32)
x 0
where y = f(x) is a differentiable function. It is known, that x* is a local maximum, if f (x*)
= 0 and f (x*) < 0. If there exists only one such x* and x* > 0, then x* is also a global
maximum and we say that problem (32) has the interior solution (see Figure 10). If x* < 0,
then the global maximum is in the point x = 0, and this solution is called the boundary
solution.

Figure 10: the interior solution x* > 0 (x* =2).

Thus the local maximum is in the point x, if

f (x) = 0 and x > 0 (33)

or
f (x) 0 and x = 0 (34)

56