in Eqs. (3.11) and (3.12) state that the maximum values of the minimum values are uni
ique and global
Te hare nitimum. Not al functions have a unique and global maximum ofa minimum valu.
‘We have shown that we can find without drawing the graph whether the function flx) will have a maximum value o
minimum value. We hi
ave shown thar the graph of a quadratic Function has a peak or a trough depending on the sign of he
Coefficient of x. We urge the reader to ionen graphs (see Problems section at the end of the chapter) to verify and ine.
nalite this behaviour of quadratic Functions. We shall sas more about the quadratic function inthe nex seton
Ta the fllowing ls, we summarise the important results and introduce concave and convex functions
1. We define a quadratic function as fx) = ax? + bx-+ e, where a #0.
2 Ifa <0 then the function fts) = ax? + bx +c has a maximum forall real x, when «--(2) and the maximum valu i
or}
AO
4a
3. If a@>0 then the function flx) = ax? + x + ¢has a minimum value when x=
4a}
4 In Mathematics the function then the function fls) = ax + bx + efor a <0 is called a concave function.
5+ Similarly, fle) = a + bx+ for a> 0 is called a convex function. Concave and convex functions have important applice-
‘tons in Statistical Decision ‘Theory and therefore we will discuss these functions in the next subsection.
(2) and the minimum value is
2a,
3.4.1 Concave and Convex Functions
Definition 3.1 (Convex Set): Let X be a set such that if x and
J ate two members of the set and if the point Ax+ (I-2ly
belongs to the set X, where 0 < A-< 1, then the set is defined as a
‘convex set.
Examples of Convex Set: The set of real numbers isa convex
also form a convex set. However, the set of positive integers does not form a convex set.
‘Throughout this book, we will be concerned only with real numbers. It is obvious that the ‘set of real numbers also forms
‘convex set. Therefore, we can say chat any function that we have defined on a set of real numbers is also a function defined
set. All the feasible solutions of a Linear Programming problem
Definition 3.2 (Concave and Convex Functions): In Mathematics concave and convex functions are defined as follow
1. A fanction fis a concave function if (Ax, +(1~ A)(ay)) 2 Aflx,)+(-A)f(e,), IF serce inequality holds, chen the
function f is called a strictly concave function,
2. A function f is a convex ea if fAx +0-ANlx))S AF(%) +A) flx,). LE strice inequality holds then the func:
tion is called a strictly convex function,
Bclow we will show, with the help of Example 1 and Example 2, the idea of concave and conves Function, respectively
[Example 1
Can we prove that a fanction fis 2 concave function sisting he allowing ineuaiy fds « cy )
Fd fix) for 0
fax +0-ax, (las
Consider line (, passing through the point P, and pale
to the xaxis. Now the coordinates of the point P, ae
(x, +0 -A) x3) AFC) +A) fx)
Therefore, the equation of fis given by
I=AF(H)+(L-A) fle)
‘The line 4, will intersect the graph of the function fat the
point P, whose coordinates are
CF )+O-AYF(x,), fA F(a) +
G-Afle)
Now the coordinates of the point ?, on the graph of fate
Ux, +0-a)x,, flax, +0 Ae.
Further, we have
AF )+O-A) fle) > flax, +(1-A)x,}
Since fis an increasing function of its argument, we have
SUF) +0-A)Fle)}
> flax, +0-a)x,} auln)
Again because of Eq,
ing function we have
(Al1.26)
(AI1.27) and because fis an increas
Af) +
This proves the result.
‘We will now examine -
when dre —e the special case where A= I/2and
ction fis an increasing funcion-
Seen Fis an increasing Funct
AF lx:)> Ax, +(1—a)x, (AIL28)
we have
1
ee
2 M287 fled feu) (all) igure 3.14 shows the result that we have obtained is
| Eq, (AIL29) diagrammatically, Its also impor, we
that the graph shown in Figure 3.14 is an increasing fune-
| ion ofits argument.
Similarly, we can show that for an increasing concave
function f
a
2
|
|
Afls)+-Afle,) fos) (Alli)
Explanatory notes on Figure 3.14: In Figure 3.14 we have
shown the line whose equation is given in Eq, (AIL.19),
‘The line P,P, is the chord joining the points P, and P
AIL6 Concave Function
In Section AUIL3 we have defined g as a concave function
if g=f, where fis a convex function, Therefore, if gis a
convex function then from the above definition we have
gldx, +(1-A)}x, 2 Agls,)+(1— Adg(s,) —(AIL32)
| forall 0 < A< 1. Putting A= 1/2, we get
1 1
Byaln)+ 7 sls) (ANL33)
____s REPRESENTATION OF CONSEQUENCES: SOME SPECIAL FUNCTIONS | 65
By following the same approach that we followed in
Eqs. (AII.14)-(AII.28), and assuming that g is an increas-
ing concave function ofits arguments, we can prove that
tet 1
Sh tox, 2 gly) to glx, (AII.34)
itm east Tele)
2
Figure 3.15 shows the result that we have obtained in
Eq, (AIL.34) diagrammatically. Here again, we must note
that the graph is an increasing function of its argument.
AIL7 Relation between Concave Functions and Risk
Aversion
Let us consider 2 50:50 gamble (lottery). By a 50:50
gamble, we mean 2 lottery that pays an amount x, with
probability 0.5 and an amount x, with a probability 0.5.
‘Without limiting generality, we can assume that x, > x,. Let
f be the preference function of the decision maker. Now let
q be the certainty equivalent of the gamble. The certainty
equivalent (cash equivalent)" is defined as an amount such
that the decision maker is indifferent between choosing the
amount and choosing the 50:50 gamble. IF q is the cer-
tainty equivalent and fis the decision makers preference,
then we can define the certainty equivalent q of the lorery
(gamble) as
F)=FFls)+F fl) (AlL35)
eee
20%) CE
Figure 3.14 Showing a convex preference function.
rs
use the expression cash equivalent when the lottery payoffs are in monetary units. For a discussion of these ideas ee Chapter 12 66 1 OPERATIONS RESEARCH
ox)
Blot) 9143)
CE
Figure 3.15 Showing a concave preference function.
John Pract, Howard Raiffa, and Robert Schlaifer have used
the symbol 1 to define the decision maker's preference func-
tion. For maintaining consistency in notation, we have used
the symbol fto define the preference function ofthe decision
maker. In Chapter 4 of the book “Introduction to Statistical
Decision Theory” (p. 4.2.2)", the authors have defined the
certainty equivalent (cash equivalent) q by Eq, (AIL.32)
‘Again, the authors define a decision maker to be risk
averse when
1
SOD= 5 flo )+4 fos) S thy (AlL36)
Now we have shown in Section 3.2 that Eq. (AII.36) isthe
condition for fto be a concave function. Similarly, we have
shown in Section 3.2, that when
1d gl 1
pete Sg FO +5 fey)
fis a convex function
john Prat, Howard Raff, and Robert Schafer (1965),
Company, p. 4.2.5, Eas. 3-4)
(3.16).
Fae
“Introduction t0 Statistical Decision
4
%
Again, John Pratt, Howard Raiffa, and Rober
Schlaifer'’ have shown that the person whose preferenct
function is a concave function is a risk avers, the pest
whose preference function is a convex function is a tit
taker, and the person whose preference function is lint
is a risk-neutral person. Therefore, in this section we hit
shown the relation between concave functions, come
function, risk aversion and risk seeking. Below we si
marise the conclusion
1. A person whose preference function is concave 1
Person is risk averse.
A person whose preference function is convex ©
person is a risk seeker \
3. A person whose preference function is linear @
Person is a risk neutral,
2
Theory’, New York: McGrew
"1 Satie! Decision Theory”, New York: McGaw
)
