CHAPTER II THE CASE OF MORE THAN

1617122
15 32
5h17
ONE CHOICE VARIABLE
We have discussed optimization in ch g using a single choice variable and exponentialfunctions
in Ch 10 Now we'll discuss how to find extreme values of functions involving two or more choice

variables

First we'll discuss the case of 2 Choicevariables 2 fax y to take advantage of it graphability
Later we'll extend this to the general n variable case Notethat we'll always assume that our
objectivefunction possesses continuous Partial derivatives to any desired order

For functions there are again two kinds of extreme values A absolute
of several variables

or global and 2 relative or local Unless otherwise specified the extrema we'll discuss
andrewstudynotes.com
are relative Conditions for absolute extrema will be discussed in Sec 11.5

71.1 The Differential Version of Optimization Conditions

In Chg the optimization conditions were discussedentirely in me context or derivatives as against

differentials To prepare for problems with two or more choice variables its helpful to know
how these conditions can be eavivalenny expressed interms of differentials

First Order Condition

Given Z fax we can write the differential
dz flexi dx 11.17
and use dz as an approximation to the actual changes AZ induced by the change from
Xo to xo taxi me smaller the Ax the better the approximation

From Fig its clear that f ex 70 dz and dx must take the same algebraic sign illustrate
Il l

by Point A Theopposite case of A where f ex co dz and dx take the opposite sign

andrewstudynotes.com
 Since points like A and A where fix to
cant qualify for stationary points in necessary

that for Z to attain extremum 12 0

More accurately dz o given dx to since

no change in X doesnt make sense

Fig 11.1
A minimum of Z occurs at B and a max at B where 5 ex so in both cases Thus he first
order derivative condition sext o can be translated into First order differential condition de O
Note that while necessary it isnt sufficient because line point C its an inelection point at de o

Second Order Condition

The second order sufficient conditions for extrema a z are in terms of derivatives
5 x so for a max and f x 70 for a min To translate theseinto differentials we need

thenotion of second order differential d del I d z

andrewstudynotes.com

Given dz f x1 dx we can get d't by further differentiating dz In doing so keep in mind

that dx is treated as a constant Thus
dz can only vary with f ext but because fix is
a function of x dz can only vary with x We have
d2z d Cda d f'cx dx by ii I
d f x Jdx
f ex dx
5 x dx 17 2
The result in 11.1 Provides a definite link between dez and f cx As dx to the dx term
is always positive thus d't and 5 ex must take he came sign Similarly he sign f ex

implies a minimax so will die

We can thus state that the second order necessary conditions are
Max Z Six E O MaxZ d2z so
dx to
Min t flex Min 2 diz o

andrewstudynotes.com
Differential Conditions versus Derivative Conditions
We developed the differential conditions as these are stated in forms that can be easily

translated from one to multiple variables cases even though they are the same with me
derivative conditions

We will still work with derivative conditions after we've developed the generalcases win me
differentiation process as they are operationally more convenient to apply

11.2 Extreme Values of a function of Two Variables

With one choice variable an extreme value is a peak a a hill bottom or a valley in 2D space
win a two Choile variables the graph of the function z fix 47 now can be represented in a
3D space but shaped like a dome a bowl

This is illustrated in Fig 11.2 where A in diagram a is a man while B is a min

andrewstudynotes.com

First Order Condition

For the function
Z foxy
the Foc necessary for an extremum is de o Butsince there are two IVs dz is now a
total diceerential thus
dz o for arbitrary values of dx and dy both to 111.37
Therationale is basically me same for me one variable case where an extremum point must

be a stationary point

andrewstudynotes.com
In the present two variable case the total differential is

dz fx dx t fy dy 11.47
To satisfy11.3 Fx and f must be simultaneously zero has the equivalent version of 11 3 is
27 27 2 03 11.57
Fx 57 0 or 2

Graphically this is point A in Fig 11.2 where Tx drawnthrough A and is parallel to me xz plane

holding y constant musthave a zero slope Similarly Ty is parallel to the YZ planecholding x

constant has 57 0 as well

To illustrate that the Foc is a necessary but not surrilient condition consider Fig 11.3 At
C in a Tx T O but this isnt an extremum its a min when viewed against the background
of the Yz plane its a max when viewed against the xz plane This is known as the saddlepoint

Similarly with b whemer viewed against

the xt or 72 Plane heres no
extremum point but theres an inflection
andrewstudynotes.com
Point instead

To develop a sufficient condition we must consider me second order total differential

Second Order Partial Derivatives

There are two first order partial derivatives for the function 2 5 Cx Y

Fx 22 2x and 27 2
fye 1
Since fx is a function of x and y we can measure fx holding y constant by a particular
second order derivative denoted by fxx or 2 dx

Fxx
E Csx or
3 Zx 25
fxx signifies that f has been Partially differentiated with respect to x twice whereas 22212 2
is just the partial version of d't dx We can also write

fxx Ey esy or 31 Ey
andrewstudynotes.com
But as fy is a function of x and fx or y we have two more partial derivatives

txt
I Ee and Sox
3 Ex
These are called cross mix partial derivatives Even though theyre defined separately according to
Young's theorem fxy fxx as long as both are continuous Thus theorder of the derivatives doesnt

matterUsually for specifichnitions that we work with his condition holds but we'll assume the
same for general functions

Fifth y second order partial derivatives or

E X t 5 Xy y
The First Partial derivatives are
f x 3 2 5Y fy 5 x 27
Upon further differentiation
fxx Gx Sexy 5 fry 2 Fxx 5

Istated fxy and fax are identical

andrewstudynotes.com

Éindname
second partial derivatives or Z X é I mine
is lg og so
fx 2X e Y and fy x'et
So we have
fxx 2e Y fxx 2x e Y f xx 2x e 7 fyy x'e Y

Second Order Total Differential

We can derive the second order total diecerential d Z by former differentiation a dz In so
doing remember that dx and de must be treated as constants during dieeerentiation as they
representarbitrary changes in X and Y Also note hat da is a function a x and y

To obtain d z we justapply the definition of a discerential to de itself

andrewstudynotes.com
 J'd 2042 dy
diz d Cdt 2x dx t ii 4
2 2x fx dx t Fydy dx t 724 fxdx f fydy dy
fxx da t Sexy dy dx Cf xx dx t fyydyl dy
Fxx dx t say dyda t fyxdxdy t fyy dy
Fxxdx t 2 fxydydx t fyedy fry fxx 11.61
This shows the magnitude a d't in terms of dx and dy measured from some point Cxo yo
To calculate d't we also need to know he second order partial derivatives fxx Fxx and fyy

Example 3
Find de and d't given z x3t5xy y Substituting various derivatives round in 11.4 and 11.6
dz 3 2 577 dx t 5 147de
and
d't Gxdx t lo dx dy 2 dy
If we Put specific points Say lil

dz 13 dxtdy d z Gdx t to dxdy 2dy

andrewstudynotes.com
Second Order Condition

Similarly with the one variable case once the Foc for a stationary point is established the SoC
Suceicient for a max or 2 f ex yl is

d z co for arbitrary dx and dy both to 111.71

On theother hand for a min a t fax Y
d 270 for arbitrary dx and dy both to 11 8
Because its possible for a't to true a zero value second order necessary conditions must bestated

with weak inequalities

Max Z d't so for arbitrary dx and dy bon yo 11 g
Min Z day o

We can also translate second order differential conditions into equivalent conditions on second order

derivatives This would depend on the signs a me second order partial derivatives We'll discuss
how in Sec 11.3 because we need to know Quadratic forms but the main results are

at co ite fxx so i txt so and fxx fxx 7 fin

70 if fxx o fy y yo and Fxxfyy 7F'XY
andrewstudynotes.com
The sign a d't both fxx and fry and the cross partial derivative fxy This result
hinges on

together with the Foc 11.5 enables us to construct Table 11.1 It should beunderstood that all the
second partialderivatives must be evaluated at the stationary point fx fy o

FxxFxx Fxxfxx
Fxxfxx I FxxFxx Fxx

It should also be stressed that the second order surcilient condition isnt necessary for an

extremom If a stationary value fxx fry fix it violates the condition but it might be an extremum
In the case a fxxfyyaffy we can identity the point as a saddle point

Éh xtreme valuers a z 8 3 taxy 3x t y't l lets rind on the first and second Partials
andrewstudynotes.com
fx 24 2 t Y Gx Fy 2x try
Fxx 48 6 fyy 2 fry 2
The foe calls for the satisfaction for the simultaneous eduations fx o and ty o
2
24 74 Gx 0

24 24 0

The second equation implies that Y X substituting his into the first equation 24 2
oxo
x O

xt b
Applying theSoc we note that when
x ext O
Fxx 6 and fry 2 So hat fxxSry is negative and a f xx failing theSoC The fact that
fxx and fx have opposite signs mean met thesurface will coreupwardin one direction but downward
in another thereby giving rise to a saddle point

For the other solution X we have fxx to and fry fxy 2 meets all the requirements
a the second order surrilient condition for a min Bysetting x t and 7 3 we have the

andrewstudynotes.com
min Zt 27 So here we only have one relative extremum represented by
23
Cx y 2 1 43 13 27

ÉingmmTextreme
valves or zextzey e e

Fx l e f y Ze Ze
2
fxx ex fry 4 fxy o
To satisfy the Foc
1 ex o

Ze teh O
There are solutions x 42 We men evaluate for the Soc at x o and 7 111
only two o and 7

and find hot fxx 1 fyy Ge fxy o Since fxx and fxx co and Syfy 7 47 we can

include that the Z value in Question

2 of e e e it

Imax valve i.e Xt y 2 1 Co 12 17

andrewstudynotes.com

11.3 Quadratic Forms An Excursion

There exist established iriteria for determining he signs a a Quadratic form such as the last line
of the expression d z in 11.6 Since the SoC relies on mesign a d2t those criteria are very view

To begin win we define a form as a polynomial expression in which each component term has a
uniformdegree We encountered Polynomials that were a single variable go taix t tan x

When more variables are involved each term maycontain multiple variables each raised to a certain
Positive integer
e.g 3 4 43 242 In the specialcase where each term has a uniform degree i.e where

his sum of exponents in each term is uniform the polynomial is called a form

For example Ux gy tz is a linear form in 3 variables because each a its term is of the first
2
degree Meanwhile 4 xy t342 in whim each term is a the second degree constitutes a
Quadratic Form in two variables In thethree variable cases its x'taxy y wttw or even n variables

andrewstudynotes.com
Second Order Total Differential as a Quadratic Form
If we let
UI dx dy
v
11.10
a Fxx be fyy he fxy fxx
then the selond order differential
d z Fxx dx t 2 fxx dydx t fyy dy
can be written as a Quadratic form 9 in the two variables u and u
9 an't 2 hurt but 111.6
Note hat in Quadratic form dx and dy are treated as variables while theyretreated as constants

when we're differentiating dz to get dit

Thereason for his reversal is that the SoC requires diz to be a t1 regardless of dy anddx
Thus dx and dymust be treated as variables le we've considering second partial derivatives we're

considering specific valves at points considered to be extremum and thus may be regarded as constant

The major Question is hoc what restrictions should be placed on a b and h while u and u can
andrewstudynotes.com
take any value to ensure a definite sign for a

Positive and Negative Definiteness

For terminology a is said to be
Positive definite Positive 707
Positive semidefinite
ie g is invariably nonnegative 207
Negative definite nonpositive CEO
Negative semidefinite negative C o
k q changes signswhen he variables assume diererent values q is said to be indefinite we
have a saddle pointThecases or positive and negative definiteness are related to the Soc
sureicient where the semidefiniteness are related to the SoC necessary

Determinants Test For Sign Definiteness

We can do thisby examining signs a certain determinants But this is more easily applicable to
definiteness and his SoC sufficient so we'll confine our discussion
only to sufficient conditions

andrewstudynotes.com
Forthe two variable case determining me sign a 9 is relatively easy First he signs of the first
and third term are always positive because they arrear in squaresThus we can ensure the
definiteness of the signs by restricting the signs of a and b

But the issue is in the middle term We can easily overcome his ie we convert the entire
Polynomial into an expression such that u and u appearonly in some squares

To dothis thetrick is to use completing the square By adding h'v2 a to and subtracting
the same Quantity from the rhs of 11.6 we can rewrite the Quadratic form as

a an't chart v t bul hatv2

a cult wut ve t Cb E u
a cut fu t abat cue
Now hat u and U only appear in squares we can predict the sign of 9 if

Positive definite in ab h
9 is and so cil.ie
negative definite

Now we have
andrewstudynotes.com
1 ab he should be positive in both cases
4 ab must be positive since it must 7h2 thisimplies that a and b musthave the
same sign
We can state this with the usage or det First rearrange 17.6 into
a a cu th luv
th Cvn t b cu
with the squared terms on the principal diagonal and the 2 hav term splitsinto 2 and placed
at the diagonal Thecoerts now form a symmetric matrix the Quadratic form is men to be
a scalar resulting from the following matrix multiplication

a Cavs Ib Y
The det co the 2 2 Coele matrix If I I referred to as he discriminant of 9 which we'll
denote as IDI supplies the clue to me criterion in 11.11 for the latter can be expressed as
Positive definite
9 is
negative definite lid and
195 so 11.11

andrewstudynotes.com
lat a is simply the first leading principal minor co Ibl while the det of an bl is the second

leading principal minor of IDI In this case there are only two leading Principal minors and these

will determine the sign of 9

If we rewrite the shorthandsinto hair original form

Positive definite fxx o f
d z is ire and 7 fxx fry f2xy 70
negative definite g fyx fyy

In general he disiriminant as a Quadratil form

9 an't 2h nut but

is he symmetric determinant I 951 In the particular case of
d2z fxx dx f 2 fxy dxdy t fry dye
the discriminant is a det with the second order part der as its elements Such det is called
a Hessian In the two variable rose in

my stil andrewstudynotes.com

ÉImi q sunt smut 202 the discriminant is if tI win leading principal minors
5
570 and I 7.7570

Iis a definite positive

ÉÉ 2 fry I and say l at a certain point on a function 2 toxin does d t have

a definite sign at thatpoint regardless or ax and dy The discriminant or a is Y I win
i o and TI 170

Idt is a negative derinite

andrewstudynotes.com
Three Variable Quadratic Forms
A Quadrat corm withthree variables hi U2 and us may be generally represented as
a cut uz Us die Guilt diz Cu un t dis er us
dz chew td 22 Cui t d23 Urus 11.12
tds cus u t der casual tdrs Cui
III dijuin

This square array is always considered to be a symmetric one even though we write he
Cortes diz das as it the pair is different

This three variable Quadratic is expressible as a product or three matrices

arm must Cummins In In Du ii in

As with me two variable case the first row and the third column matrices list me variables
while the middle one D is a symmetric coers matrix This time here are 3 leading principal
minors from in discriminant

land 10.1 lost

andrewstudynotes.com

Where Di denotes me in leading principal minor or he discriminant DI The definiteness can

also be stated in terms a certain sign restrictions on these principal minors

The Quadratic form in 11.2 can be converted into an expression in which the three variables appear
only as iomponent a some squares Specifically as a12 az we have

This sum of squares will havethe same sign for any valves or ten ten and us with the Cortes
or the threesquared expressions But he 3 cores can be expressed in terms or he 3 leading

andrewstudynotes.com
Principal minors as follows

in

Hence for positive definiteness

i
he sufficient and necessary condition is threefold
Dil 70
Dal 0 given that Dil so already
103170 given mat 1172170 already

All the leading principal minors must be positive For a negative definiteness

IDil so
102170 given hat ID Ico already
1031co given hat Del so already
The leading principal minors must alternate in the specified manner

I in q uit6uit3up 2u.mz 4azus The discriminant a 9 is

Heil andrewstudynotes.com

with the following leading principal minors

ID l l Dal
I 1,81 570 Dsl 291 1170

Ihauadratic is a positive definite

n Variable Quadratic Forms

We'll state without proof hat for the n variable case for me avadrati corm
a cnn.mn mn É E dij Will
the necessary and sureicient condition for positive definiteness is that ID
Died Del 911 tent
It
ad

all be positive While for negative definiteness he sign must alternate

Dil co 102170 1031 co Ceta

so that all odds are negative and all evens are positive

andrewstudynotes.com
Characteristic Root Test For Sign Definiteness
Theres also an alternative to me det test of a Quadratic u Du whith uses the concept of

characteristic roots a the matrix D

This concept crises from whether given an nxn matrix D can we find a scalar r and

an nxt vector x to such that

D r x
Cnn hi Cnxi
11 13

is satisfied If so r is referred to as a characteristic root of matrix D and X as a

characteristic vector of that matrix

1717172 1917122
Dx rx can be rewritten as DX r IX O or 1741 08 43
9h48
D rt x 0 where O is nel 11.13
Which represents a system of n homogeneous linear equations Since we want a nontrivial solution

For x the core matrix CD rt calledthe characteristic matrix a D is required

to be singular i.e its determinant must be made to vanish
andrewstudynotes.com

11.14

11.4 is called the characteristic equation of matrix D Since he det ID rel will yield upon a
Laplace expansion an n th degree polynomial in the variable r Thus there will be n roots each of
which Qualifies as a characteristic root If D is symmetric suit as in the present Quadratic form context
the characteristic roots willalways be real numbers but they can be either sign or even zero

When we substitute
any valve a r into 11.13 will produce a corresponding vector X Irri More
accurately the system being homogeneous it will yield an infinite number of vectors corresponding
to the root ri

We'll apply a process called normalisation explained in Example 5 and select a particular member

of that infinite set as the characteristic vector corresponding to vis his vector will bedenoted by vi

andrewstudynotes.com
Éingitcharacteristic
root and vectors or the matrix Bysubstituting megivenmatrix
for D in 11.14 we get the equation

2 r 2 r
f 6
r 0
2 i r

with roots ri 3 and re 2 When thefirst root is used the matrix equation in 11.13 kneeme form

I C3 a 63
Because the corer matrix being linearly dependant as expected in 11.14 which means theres an

infinite number of solutions which can be expressed by x 2 2 To force out a unique solution

we normalise the solution by imposing the restriition kit x2 l Then since

txt 2 27 txt 5 5 1
We can obtain x2 155 and also x 2 2 255 Thus the first characteristic vector is
vi
HE
Similarly by using the second root re 2 in 11.13 we get
andrewstudynotes.com
Z c 27 2

i n
1 11 c
Upon normalisation we have
which has the solution x t x2
txt C E x2 t XE E xi L
which yields x2 2155 and Xi 1155 Thus the second characteristicvector is
ve YE
The set a characteristic vectors obtained in this manner possesses two important Properties
First the scalar product Vi Vi i 1,2 n most equal I sine

Vivi Cx xz xn
II II xi 1 by normalisation

Second the scalar Product vi

Vj where it can always be then to be zero Thus we may write
Vivi I and Viv O 11.15

These properties
will prove useful later Example 6 As a matter of terminology when two vectors

yield a hero valued scalar product the vectors are said to be orthogonal to eachother
andrewstudynotes.com
Henieeach pair a characteristic vectors mustbe orthogonal The other property vivid is indicative of
normalisation Together housetwo properties account for the fact that the characteristic vectors Cui vn

are said to be a set of orthonormal vectors

Now we're ready to explainhow characteristic vectors and characteristic roots a matrix D can be
used to determine thesign definiteness of the Quadratic form a Da

In essence theidea is to transform u Du into a form mat contains only squared terms Thus its

similar to completing the square method but now we have the additionalfeature mat the
transformation Possesses the additional feature hat each squared term has as its coefer one of
the characteristic roots so that he signs of the n roots will provide sufficient into on the sign
of the Quadratic form

The transformation is as follows Let the characteristic vectors vi un constitute the columns

or a matrix T
I Cui we Un
andrewstudynotes.com
Thenapply the transformation u T y to the Quadratic form u Du
u Du Ty DCF
Y'T TOTH 4.11
Y Ry where R e T DT
Now we've turned the original Quadratic in the form of hi to me form a Yi Since ni and
i taus the same range of values the transformation doesnt arrest the sign definiteness of
the Quadratic form

Thus we
might as well consider the sign of the Quadratic Y Ry instead which will turn out to
be a diagonal one with the roots ri rn of matrix D displayed along its diagonal with
O everywhere else so that we have

11.16

andrewstudynotes.com
which is an expression involving squared terms only Thus RIT DT Provides us a procedure

for diagonalising me symmetric matrix D into me special diagonal matrix R

Fefighff given in Example 5 can be diagonalised into Iri On the basis

of the characteristic vectors in Example 5 the transformation matrix should be
2155 1155
T C V2
1155 2155
Thus we may write

f IIe X it C
whips the diagonalisation Process

To prove the diagonalisation process lets partially write out the matrix R as follows

andrewstudynotes.com
ReTDT
Y D CU Uz Un

We ran easily verify that DEV Uz Un can be rewritten as Due Dua Dun
By 11.13 we can rewrite his as Cru run run Hence we see hat

Which is what we're trying to show

According to 11.16 we may summarise the characteristic root test for sign definiteness or a

Quadratic form as follows

andrewstudynotes.com
 I g u Du is
positive definite ice every inaracteristic root is positive and vice versa
7 9 W Dm is positive semidefinite ire all characteristic roots a D are nonnegative and

vice versa
3 q n Dn is indefinite ie some characteristic roots are positive and ie some are negative
Note that only need the characteristic
we roots i he vectors arent required unless we wish

to find the transformation matrix T We are also able to check the SoC necessary Part21
simultaneously with me sufficient Part D

However the drawback is that the computation is tedious when higher degrees are involved

so he def test might be preferable

11.4 Objective Function or More Than Two Variables

First Order Condition For Extremum

Lets consider

Z f Cx X2 V37
andrewstudynotes.com
with the first Part devs Fi fi and 3 and second Part ders Fi 212 2x 2x win
i i 1 2,3 By virtue of Young'stheorem fiji fii

To have a stationary pointsimilarly we must have dz o for arbitrary valves of da din and
dx3 not all zero Since dz is now
dz fidx t fed xu t fade 11 17
Since dx dxz and do are arbitrary changes in the IVs not all zero the only way to
guarantee hero dz is to have fi fi 3 0 Thus again the condition For extremum is hat
all the first order part der to be zero

Second Order Condition

Afterwe've established a stationary point and we findthat d't is a positive definite this would
sureile to establish z as a min pointsimilarly ie its negative it would be a max point The
issue is then to express d't when there are 3 variables and hair sign definiteness

We can differentiate dz to get dez in such a process in 11.6 we should treat me derivatives
andrewstudynotes.com
fi as variables and the differentials dxi as constants we men have

11.18

which is a Quadratic form similar to 11.12 consequently the criteria for positive and negative
definiteness will also apply here

In determining the positive negative definiteness of dez we must again in ii o regard dxi as
variables that can take any values chough not all hero while considering the derivatives ti as
andrewstudynotes.com
confers upon whim to impose certain restrictions

The corers in 11.18 give rise to the symmetrin Hessian det

its 1
While the leading principal minors as
I hit fin Hel fi I'll 1431 141
Thus based on the det criteria for positive and negative definiteness we
may establish the
SoC for an extremum Of Z as follows

2 is a is tail so i 142170 I Holco Cd't is a neg doe

I Hit o 142170 431 0 dez is a pos der
s

In using his condition we must evaluate all the leading principals at the stationary point
f fi 53 0

We can also apply the characteristin root test and associate the positive negative definitene

andrewstudynotes.com
or d't with the positivity negativity of all the characteristinroots of the Hessian matrix

fi Fil 13
fy tie fr In
fact we can just say that the Hessian matrix is negative positive instead of
si 532832 the SoC total differential d't

Éigifextreme
value is a

Z 2x taxi taxi txixstxst2

The Foc requires satisfying the following 3 equations
fi Ux f Xz t Xs 0
fi X t 82 t o

1537 Xi 0
Because this is a homogeneous linear equation system in whith all three Ivs are independent he

def wont vanish theres only one solution Xt Xt Xt o Theonly stationary value is 2

The Hessian det is

andrewstudynotes.com

t 1 1 1
whose leading Principalminors are all positive
1411 4 1421 31 1431 54

Incan conclude that 2 2 is a min

Érimmfextreme
value's a

Z xp 3 1 3 t2 2 XE 3 32
The first part ders are
81 3 12 3 3 52 2 2 2 83 3 1 6 3

Bysetting all Si equal to nero we get three simultaneousequations one nonlinear and two linear

3 7 3 3 p
2x L

3 1 643 0
Since the second equation gives x I and thethird implies x 2 3 Substituting her into the
andrewstudynotes.com
First eavation yields two solutions

exit xt xt 10.1.0 implying 27 1

Chility implying Zt
The second order Partders give us
6 1 0 3
141 o r o

3 o o

In which the first element reduces to hero under the first solution x o and 3 under the

second exit Il Thefirst solution doesnt satisfy the SoCsufficient as IHit o We

may though
use the characteristic root test for Armer into

Hence for the first solution the characteristic equation is

r o 3

f o
3
c r
o
o
o r
0

which upon expansion gives me equation

3 tore Jr 18 0 andrewstudynotes.com

We have the integer root 2 Thus the cubit action should be divisible
by Crts and we can
rewrite it as
ertz Cr Gr g 0
So we have the first roof 2 from rtn while after applying the Quadratic formula to the other
term we have re 3 1572 and re J 1572 Because r and ro are
negative but re is
positive he Quadratic dez is indefinite thusviolating the Sol for a min max Thus 25 1 isnt on extremum

for thesecond solution we have

Hit 3 1421 6 and 1431 18

fignsalternating the det test suggests that 2 716 is a max

n Variable Case
When there are n choice variables the objective function may be expressed as
Z fCxi x2 Xn

andrewstudynotes.com
The total differential then will be
dz fide t fi da t f fndxn
so that the necessary condition for an extremum de o for all arbitrary dxi not all nero

is that all the n first order part devs are reauired to bezero

The Soc ion be round similarly using the Hessian

with leading principal minors Hit Hel Hnl The SoCsufficient is hus all positive for a
min and alternating for c max the first one being a negative

In summary men We have the criteria listed in Table 11.2

andrewstudynotes.com

11.5 Second Order Conditions in Relation to Concavity and Convexity

SoC whether stated in terms of Principalminors or the characteristic roots of the Hessian are

always concerned whether a stationary point is he recu of the hill or bottom of the valley
how a curve hypersurface or surface bends itsele around a stationary point

Afunction that has a rise to hill valley over the entiredomain is said to be a concave convex
a

Function For now we'll take the domain tobe the entire Rn where n is the choice variable We

can boil down concavity and convexity into strict concavity and strict convexity

In the nonstrict case the hill valley is allowed to contain one or more flat portions such as

andrewstudynotes.com
line segments on a curve or Plane segments on a curve while these are ruled out in the Strait ice

The two surfaces in fig 11.2 illustrate the strict cases

goncave ponvex

Fig 6.5 meanwhile shows he nonstrist convex rose as it contains line segments A strictly
concave convex function must be concave convex but he reverse isnt true

In relation of the association a concavity and strict concavity win a global hill configuration
an extramum of a concave function must be a max and also an absolute max as the hill covers
andrewstudynotes.com
the entire domain

However such Max

may not beunique may occur is the peak of the
because multiple maxima

hill contains a fact horizontal too But his may be dismissed if we have a strictly concave anation
as here would only be a point and the absolute max must be unique A unique absolute point
is also referred to as a strong absolute max while nonunique is referred to as wean

Similar idea also applies to a min point and convexity

If an absolutepoint is only valid for a Portion of the curve surface then the associated minimax
are relative local to the subset for the domain since we dont know me situationoutside the subset s

In our earlier discussion on sign definiteness we evaluated the leading principalminors or the
Hessian dat
only at the stationary point But because we'reonly limiting me evaluation to a small
area around the stationarypoint we could only discuss relative minimax It may be that d z he
c definite sign everywhere in which the point covers the entire domain and thus is absolute in naw

andrewstudynotes.com
More specifically if dez is everywhere negative positive semidefinite men the auction is concave

convex.ie its negative Positive definite Men its swishy concave convex

The discussion is summarised in Fig 11.5 for twice continuously differentiable unction This is
for a concave action but its the same with convex as well

andrewstudynotes.com

The most important message from his fig is me two extended Passing through the two
diamonds The one on the left States mat given a concave objective anction any stationary point can
beidentified as absolute max when its strictly concave its a unique absolute max

In either case once the foe has been satisfied concavity and strict concavity effectively replaces
the SoC sufficient for absolute max This is Powerful when we're considering mat d z can happen

to bezero at a peace causing the Sol surrilient to fail

andrewstudynotes.com
Its for his reason why economists just outright assume concavity convexity when formulating
a general objective function because then only the foe needs to be checked Note that whenever a
special objective function is used the convexity concavity cant be assumed and has to be checked

Checking Concavity and Convexity

There are several ways to define check strict nonstriit concavity convexity We'll First

introduce a geometric definition of a concavity and convexity for a two variable unction
2 5Lxi xel similar to the one variable version in Sec 9.3
The function 2 5Cx x2 is concave convex ice for any pair of distinct
Points Mand N on in graph a surface line segment MNlies on either
below abovethe surfaceThe function is strictly concave convex in line
segment MNlies entirely below above the surface except M and N

The lace or strictly concave function is illustrated in Fig 17.6 where M and N two arbitrary
points on the surface arejoinedby an are and a broken line segment with me former consisting
of points on thesurface that lie directly above the linesegment
andrewstudynotes.com

Since strict concavity requires the line

segment MN to lie entirely below me are

MN for any points M and N the source

must be typically domeshaped while its

a bowl for a strictly convex anition

Fornonstrist concave convex unctions

since the segment MN is allowed to lie

on thesurface itself some portion of the
surface may be a plane rather man curved

To generalise to me nongraphable n dimension case we need to havethe equivalent algebraic

version Returning to fig 11.6 let Me Cut me and u cvisual be any two distin it orderedpairs
G vestorst in the domain of Z f x X2

andrewstudynotes.com
Then the E values height of thesoreceel corresponding to these will be feat ten mil
and tcu f Vi Val We've assumed that thevariables can true all real values so ie u and
v are in the domain men all points on the line segment au are also in the domain

Each point on the line segment is me weighted avg of u and u Thus we can denote his
line segment by Out Cl O v where 0 is a variable scalar win the range of values
0 E GE l By the same token line segment MN representing me set of all weighted augs
of 5cal and fcu can beexpressed
by Of Cal t C1 07 foul with O varying from o l

With regards to the arc since it shows the values of f evaluated at the various points at
the line segment nu it can be written as f Out C1 o u Win an hese we may state

A action t is Intel ire for any pair or points u and u in the domain
and for OC OC 1

E Cmos
t andrewstudynotes.com
Note that he restrition O e Oct is to exclude the two end points M and N From the height
lomparison

We can transform his into a strict convexity concavity by replacing E win c Theadvantage
of his is that u and u in his definition can be interpreted as n vectors From 11.20 we have
three theorems which are generalisable to any number of functions

Theorem I Clinearanction k fax is a linear function men in a concave function as well

as a convex function but not strictly so

Theorem I negative or a function II fax is a concave men fix is a convex and

vice versa Similarly ie theyre strict

TheoremIesumannchon tt gal and fix are both concave convex then fix t
gox is also a concave convex action In the strict case is either or both are strict men text t

got is also strict

andrewstudynotes.com
Theorem I follows from the fact that a linear plots as a straight line Plane or hyperplane

so that line segment MN will always coincide with the arc MN Consequently the two
weak inequalities are always satisfied although not the strict ones making it a nonstrict

Underlying Theorem It is mat me definitions of concavity and convexity dieter in the ineauality

Sign Suppose we have a concave

fault C1 07 foul E f Out C1 O u
Multiplying through by 1 and reversing the inequality sign
C seal Cl 07 C 5 432 f out CI OI V
but his is a definition for fax to be convex The geomeni representation is very simple
as the mirror image of a hill is a valley

For Theorem I i suppose that text and g ex are both concave Then the following hold
feel C1 Q f Ch E f Out Cl O v 111.211 1917172 2017177
18 32 05 08
7h50
gcu Cl a govEg out 4 01 v 11.72 75h08

Adding these we obtain a new inequality

andrewstudynotes.com
0 faulty cul t CI O Scu g cul E f Out CI G u t
g out CI o
U 1.73
Butthis is also the condition Jor text t gex to be concave thus the theorem horde

The idea is similar with thesecond part of the strint case let fax becomes
strictly concave
then 11.21 becomes a strict inequality
O fault l 01 foul a f Out I 07 u 11.217
Thus the sum a the Ihs will always be strictly less than the rhs regardless of the c or sign
meaning that 11.23 is now strictly concave similarly with me convex anition

This theorem is especially useful when there are multiple terms involved that contain additive
terms to the additive terms are individually concave convex the sum a me function would
be surriliently concave convex

ÉÉÉÉ it xi for concavity convexity To apply ii no let u Cui mil and v Cu ve be

any two distinctpoints in the domain Then we have
andrewstudynotes.com
 5cal four Mil Mit mi
f cu FCU Unl Vit Ui
scout Cl 07v3
flaggy
and vi
outfit
On Cl a u t Court CI Ove
Substituting these into 11.20 subtracting the rhs from the Ihs and collecting the terms
Ci O Car for t O CI O Vit Vi 20 C1 G Cu v Fuzuz
C1 o Chi ul't car v27
Since O is a positive traction OCI G must be positive Since we me and vi ve are distinct
Points so mi tu and ur ve the bracketed expressions must also be positive Thus the strict

inequality holds and 2 exit Xi is strictly convex

Alternatively we can check the xi and xi terms separately since each a mem is striity convex the
sum must be the case as well We can easilydedule that the unique absolute min is Z o at

Xt x o and hats unique because any ordered pair x x2 co 07 yields a 2 value greater

Manned andrewstudynotes.com

i is the negative version a Example 2 Thus byTheorem I its strictly concave

IÉÉÉ city for concavity convexity We can still use n n me and u cu ve

to denote the two distinct points in the domain Then we have

fsu flee Mi fait Mi

foul F Cui Val Vitor
and f Out Cl Olu COM th O v t Omit Cl O uz
Osu tail t CI G Vitor
Subtituting subtracting and simplifying into 11.20
11 07 I wit un 20 1 0 I wit na CutV27 t 011 O Cut v27
D CI O cult un Coital
Similarly the bucketed expression is nonnegative as new cant be ruled out and thus the weak
ineauality holds City is a nonstriit convex

andrewstudynotes.com
Diceerentiable Functions
We dont need derivatives to deeine concavity and convexity although its possible to do so win
first derivatives ie the hnition is diceerentiable In the one variable case the definition is
A dieverentiable auction fix is Intel ire for any given point u

and another point u in the domain

foul E ten t s cu cu n 41.247

Forthe strict version just replace the weak inequalities with strict ones Geometrically his
depicts a ioniavellonvex curve mat lies on or belowlabour all its tangent linerTo Quality as
a stri it concave convex curve the curve must lie belowabove all the tangent lines except at the point

In Fig 11.7 let

A be any point on the curve with
the height fan and tangent line AB Let x increase
From u Then a strictly concave curve in order to form
hill must turn away from AB so that point C with
a

height foul has to lie below point B Thus theslope

andrewstudynotes.com
a ACC AB

In the nonstrict case it may contain a line segment

so that for instance are AC may turn into a line
segment and be coinident win line segment AB as
a linear portion or the curve in this case the slope AC AB Together these two imply that

slope a line segment Ac

If SWISH E slope or ABA flu

When multiplied by the Positive Quantity cu al this inequality yields the result in 11.29 for me concave

anition The same result can be obtained in we consider instead x values less man u

Withtwo or more tu we need to modify it slightly

A diererentiable anction Sext fax xn is 988 ice for
any
given point u sun an l and any other point ve Cui sun in thedomain

f cu E fault I fi su Uj Uj 2
where tical If 2x is evaluated at u Cu un

andrewstudynotes.com
This definition requires the graph or a concave convex unction fax to lie on or belowabove

all its tangent Planes and hyperplanes For stri it concavity and convexity replace me weak to
strict inequality and for the graph to strictly lie belowabove the tangents except at the point

Finally consider 2 fax xn which is twice continuously dieoerentiable Second part ders
exist and d't is defined Concavity and convexity can then be checked with mesign a d't
Atwine continuously dicoerentiable function 2 taxi xn is 8 94
in d't is everywhere 85k semidefinite Its strictly 81813
iz d't is everywhere 889k decinite 11.25

ÉÉÉÉx4 for concavity convexity by thederivative conditions First

apply 11.24 The Ihs
For the ineduality is v4 and he rhs v4 4mscu ut Subtracting the latter cromthe former
we can find their diererence to be
v4 tu4 4 us cu u u u f t 4ns

Cu ul C cu tu ut vertus t 4ns
andrewstudynotes.com
We can divide the bracketed expression by v m which means that we can obtain a squared term
u ul to facilitate the evaluation a sign
v ul Curt zum t sur cu un cut ul tzu
Given that ut u the sign of his expression must be negative Wim c holding Z X is strictly

concave meaning it has a unique absolute max 2 0 attained at X o

Because he function is twice continuously differentiable we may also apply 11.25

5 X1 dx 12x dx
We know that dx is positive as only nonhero changes are being considered but 12 2 can be
either negative hero Thus we can only conclude that d't is everywhere negative semidefinite and
thus Ze x nonstrictly concave

This is weaker than the previous method and his is because d't can take a hero value at a

stationary point causing it to rail This is why of course that he sign a dez is presented only

Isurricient but not necessary condition for strict concavity convexity

andrewstudynotes.com
ÉIdÉÉ itxi using the derivative conditions For his we need to use 11.24 with a can we

and v Vi Val as
any two points in the domain
left side Vitve
Right side Mit ai t 2h CU a t Zar Cve real
Subtracting theletter from the former and simplifying
Ui Zu n t hit the 2V2me the Cv mil t cuz my
Given that Cui uz cu me this difference is always positive 7 holds and its strictly convex

As for 11.75 since fi 2x and 52 2 2 we have

fi 270 and I I I 881 470

regardless of where the second order part devs are evaluated thus in positive definite everywhere

andtussatiseies the sufficient condition for strict convexity

Convex Functions versus Convex Sets

lets establish the difference between convex sets and actions as they are relatedbut distinct concepts
andrewstudynotes.com

First he geometric Charaiterisation co a convex set Let S be a set co points in a 2 space or

3 space le For any two points in the sets the line segment connecting these two segments lie
entirely on S then S is said to be a convex set Asingle point and a null set is also a convexset

For homer examples consider Fig 11.8 The solid circle is a convex set as any linejoining any
twopoints in the disk lies entirely in the dish exemplicied by ab and Cd

because he presence a holes A convex

set must not have any hole and its
boundary not indented anywhere

andrewstudynotes.com
This geometri interpretation also applies to points in a 3 space where a solid cube is a convex

set while a hollow blinder isnt But with higherspaces we need to turn to the algebraic definition

To do this we need to introduce the concept of convex combination of vectors which is a special

type of linear combination A linear combination of two vectors u and u can be written as
k ut ke U
where he and he are scalars when these two lie in the closed interval o 1 and add up to unity
the linear combination is said to be a convex combination and expressed as
Ou t CI G V CO E OEI 11.267
For example he combination I 8 Y is a convex combination as the two sealers
are positive fractions adding up to 1 such a convex combination may beinterpreted as a weighted

average a the two vectors

The unique characteristic that for every acceptable value 0 the resulting sum veitor
of 11.26 is

lies on the line segment connecting the points u and u In Fig II g where we have plotted
two veitors a Mt and V Y as two points Mi Mi and vi Val
andrewstudynotes.com

If we plot another vector 9 such that Oguu

Forms a parallelogram then we have

a at u or g u v

It Follows that the convex combination of vectors

u and u denoted as w can be expressed in

terms of 9 because
we Out Cl OI V Out v Ov OCU v to
09 tu
To plot when we can simply add Og and u by the familiar parallelogram method le G is
a positive fraction then 09 will be a shortened version of vector q thus Gg must lie on the line
segment og

Thus
by adding 09 and u we must find vector w lying on the line segment nu for the new
smaller parallelogram is just the original one with the gu side shirteddownward The exact

andrewstudynotes.com
location or w will depend on the scalar 0 by varying 0 from o l thelocation co w will shirt
from u to u

Thus he set co
all points on theline segment uv including u and u themselves correspond to
the set or all convex combinations of vector u and u

We can redefine a convex set as follows A convex sets ire anyfor two points UES andUES

and for every siclar O E CO I in true that we Out CI O V E S

Because this is stated algebrainally its applinable regardless of the dimension in which he

vectors u and u are located

Theres a big difference between the usage of convex in functions and sets Wim convex
specifies how a curve surface bends itself to form a valley In describing a set the word

spelities how he points are packed together They must not have any holes and the boundary
must not be indented
andrewstudynotes.com

But they are related for one in defining a convex action we need a convex set in me doma
This is because thedefinition in 11.70 redviresmat for any two points u and u in the domain
all the convex combinations of u and u specifically out C1 07 u of 0 El must also be
in the domain i.e he domain must be a convex set

To satisfy his requirement we earlier stated hat me domain consists co me entire

Étipice
which is indeed a convex set However with me knowledge of the convex set we can relax this
assumption by simply needing a convex subset of Rn rather man Rn i sere

Theres another relationship between the two to fix is a convex function then for
any constant
K it can give rise to a convex set
s x1 sext Ek tax convex 1.277

This is illustrated in fig 11.10 a for me one variable case The set S consists or all he
valves associated with the segment fax lying on below the broken horizontal line Its me
andrewstudynotes.com
line segment marked with heavy dots which is a convex set

We can go one stepfurther by observing that a concave anition is related to a convex set
in similar ways First the definition of a concave nation in 11.20 is determined by thedomain
thats a convex set A concave unction say gal can generate an associated convex set

given some constant k i e

S xigex 2k girl concave 11.28
in whish he sign 2 appears instead of E

11.6 Economia Applications

Problem or a Multiprodust Firm

ÉÉ f consider a two product firm under pure competition In a pure competition Price is

exogenously determined so we have Pio and Pao thus we have

R PioQ t Pro Qr
andrewstudynotes.com
We have
2QitQiQzt2Qi
Note that the port ders A each MC also depends on the other Product i.e the two
commodities are seen to be technically related in production

The profit function is men

Tu R C Pio Q t RoGe Zai Qian Zai
We are thus trying to find the levels of Q that maximise to to do this we rind by first
Partial derivatives a the prosit anation
Tu Mda Pio 40 Q2
2T l 29
Tu z CI day Pro gaz Q
Setting these to zero we get
49 t Q2 Pio
Qi 492 Pro
whithyield
Qt 4119ft and Kat 49th
andrewstudynotes.com
Thus it Pio 12 and Pro 18 we have Qi 2 and 92 4 implying Tut ad We can check

in this is a max using the Hessian by partially differentiating 11.2g

inte I L it
Because Hiko andMalco we have a negative definite and the solution is a max prosit
Since the signs of the leading principal minors are everywhere negative the action is smithy

concoueardthe max profit is a unique absolute max

IIIIFonsider a monopolistic market where me revenue will depend on me demandauction

Q 40 2 PitP2
11 30
Q2 15 tp p
These imply that the two goods are related in consumption i e theyre substitute as each other
as meincrease a price in one will raise his demand for my omer Its more convenient for us
to have Pi and Pz expressed in terms of Qi and Qe i.e the AR anction
2 Pit P2 Q1 40
andrewstudynotes.com
Pi Pz Q2 15

We may apply Cramer's rule to solve for Pi and Pr as Qi and 92 are parameters

Pi 55 Ai a
11.30
Ri 70 a 292
Consequently he firm's TR ion be written or

R P Q t P2Q2
55 a Q2 Q t 70 Q1 292762
5501 7092 29 Qi Qi Zai
le we have
C Qft Q art QE
Then
TV R C 559 7002 39102 2Qi 392 11.317
Once we find Qi and Qi piandPat are easy enough to find from 11.30 Theobjective
function yields the first and second Part ders
Tu 55 392 40 2 70 391 692

TV11 4 Tu z q Tu 22 G

andrewstudynotes.com
 To satisfy the Foc we have Tur Tur O
4 Q t 302 55
3Q1 692 70
we have
cat at 8.73
Substituting into 11.30 and 11.31
PF 395 P 46 and TV 4085
The Hession is

I HI 4 so and 1421 1570

Thus its a max and because the principal minors are independent co where theyre evaluated

have a negative definite its strictly concave and it has a unique absolute mat

Price Discrimination
Even a one productfirm can have an optimization problem such as when there are two
andrewstudynotes.com
markets and thus must decide the Q supplied to maximise profitTheseveral markers will have
demand conditions and it there are different elasticities profit maximisation entail price distriminatio
Yo 3 oo so
5h56
2 boy
eÉ w have 3 choir variables Marwari and write in general forms Assuming a monopolist we ter

R R CQ Recant R3CQ3
CQ where Q Git Art Qs
Each revenue function would depend on each market There is also a single cost function as here

only a single firm producing We then have

Tu Ri CQilt R2CQ l RsCas C Q
with Part ders Tui ITU 2Qi
Tui Ri can C a 3 Rican C Q
C la fate 1 also the chain
Tun Rican Rican a rule is used
d
Tus Rz 931 Cal 3 Rz 931 C Cal BE TE JE
Setting all these simultaneously to zero will give us
Q Rican Rican p can
andrewstudynotes.com
Thatis
MC MR MRI MR
Thusthe levels of Qi Q2 and as should besuch that Mc MR o each co product To see the
implications of his with regard to prise discrimination let first find out how MRis specifically related
to me poise in that market Since Ri PiQi the MR must be
Mr II Pi III Qi tf
P It 8 8 Pi Iteat
Where Edi is normally negative Consequently the relationship between Mri and Pi can be expressedby

Mri Pi l 11.37 É
IEdit is a function of Pi so when Qi is chosen and Pittis spelieied Edit will assume some
specieis value But Qi must be such that ledit I as if IEdit at the parenthesised expression
will be negative thereby implying negative Mri Similarly it Edit O Mr is also 0 Because Mci
Positive and the Foc requires Mc MR the firm needs a positive Mri

The FOC MR MRL Mrs can now be

Pif g Paf eat B t

andrewstudynotes.com d
Thus the smaller the value of led1 the higher the price charged in that market must be hence

Price discrimination To ensure Profit maximisction lets check the Foc

Tui Rican Q Mda Rican C Cal
Tu z Ri fay C Q Taz Rica Q
24
TV33 RsCQ C CQ 203 R'sCa C ca
2
and Tull Turn His TU31 Tues Tzu C CQ Qi I
so that after shortening the second derivative
12 C c

ar c is
C Rs c

To satisey the Soc we need to have

l Hit Ri c co i e the slope a MR is less than the slope a Mc or the entireoutput
2 Itht CR c Cri c C 70 or R Re Cri't Re 7C so
341,1 Riker's Ri Ri't Rik's Rik's7C co
The last two parts of his condition arent economically easier to interpret man the first Note

andrewstudynotes.com
though that we've assumed that the general Ri Cai actions are all concave and Cal convex
so that CQ is concave then the profit function should be concave and we dont needto check

MI

E.IT
tiateanomeri aexamne
P 63 4Q R P'Q1 63 a 4g
R 105 592 Re P2Q2 10592 595
Ps 75 ga Rz Po93 7503 695
and
6 20 15Q
The HR will be
Ri 63 86 R2 105 1092 Ro 75 1293
When each Mc is set to MR we have

QI G Art g at 5

at E Qi 10
andrewstudynotes.com
Substituting these into the tu function we get to 67g We need to check the soc as his is a
specific model Since the second derivatives are
R 8 122 10 Ra 12 C O
We can check the ARanctions to see that the firms should Pri le discriminate Pi 3gPetGo and

PI45 TN ledl is lowest in the second market mus it has thehighest price charged

Input Decisions or a Firm

Éonsidette
competitive arm win
Tu R C PQ WL rk 11.34
Since the firm operates in a competitive market the exogenous are P w and r while he endogenou

are kit and Q However Q is a nation a K and L via

Q Q Chill
We shall assume this is a Cobb Douglas function discussed in Sei 12.61
a LdKB
where a and 13 are positive parameters Assuming decreasing returns to scale at Bal for

andrewstudynotes.com
simplicity we'll assume a p Ct
Q Ckd 11 357
Substituting into 11.34
TUCK c P Lt K WL rk
The Foc for profit maximisation
2712L pact K W o
11 36
272k palter t
r o
We would then have the optimal L and K for profit max lets verify the Hessian
Paca blank Palatka
14
1 fit PALE KI PL ca 174kt 2

The suecisient requires Hilco and 14170

Z
IH Paca 1 La H co
2 2 2 24 za 2
1H P'd ca 11 224 k p224 2k
p221 222 2 K K Z CI 22770
Therefore for act the Soc sufficient is satisfied

andrewstudynotes.com
We ran now return to the Foc to solve for k and Lt Rewriting 11.36 to isolate k
part Ka a
w
a
2
k
Substituting into the second equation or 11.36
L a a
Parka t r Pace I 24 Did
r so
Ca 1712
play t
w r

Rearranging to solve for r

ICI 221
Pa wa tr 2

Because or the symmetry of the model we also have

part d i ex
k w
Substituting It and letinto the production function
at Lt Kt
a C d ICI la
Pawa tr a para lw a
ga
419 24
II 11.37

andrewstudynotes.com
 ÉmmIw have me following circumstances
41 Two inputs a and b are used in the production of Q
4 The Prite a bom inputs and output Prine P are exogenous denoted as Pao Pbo Po
3 The Production Process tones to years to complete and thus the revenue from the sales must
be discounted before it can be compared win me cost at present timeThe rate of
discount on a continuous basis is assumed to be ro
Upon assumption I we can write a general Production action Q Qcarb with MPP Qa and Qb
Assumption 2 enables us to write

Pao at Pbob
and
R Po QCarb
To write the Profit action we must discount the revenue by multiplying it by e roto We'll
rt Thus
simplify this by writing e
rt
Tu PoQ cab e Pao a Pbo b

In which a and b are the onlychoice variables

andrewstudynotes.com

To maximise profit its necessary that

Tua CE JI PoQue
re
Pao 111 30
tubCE II Poab e
rt
pyo
both be zero meaning
rt r
Po Que Pao and Po Qb e Pbo 11.39
Since PoGa Price times marginal product a input a represents my valve or marginal Product

of input a CVMPal the first equation is saying that UMPa must be equated to price input a

Similarly with me second equation

To satisfy 11.3g both me MPP Qa and Qb must be Positive because Po Pao Pbo and e rt
all have positive values This has an important interpretation in terms of isoavant defined as the lows

or input combinations thatyield the same output level

When plotted in the ab Plane isoavants will appear as in Fig Il Il Inasmuch as each of them

Pertains to a fixed output level along any isoavant we must have

andrewstudynotes.com
 da Qa dat Qbdb o

which implies that the slope a an isoquant is expressible as

dblda Qala Mmff 11.40

Thus to have both Qa and Qb positive is to confine the firm's input choice to negatively stored
segments of the isoavant only In Fig 11.11 the relevant region is restricted to me shaded area
defined by two so called ridgelines

Outside the shaded area where he isoavant

have positive slopes the MP or one input must

be negative

The movement from M to N For instance

indicates that with input b held constant the
increase in input a leads us to lower isoauant
a smaller output thus Qc must be negative
Similarly from 14 to N illustrates the
andrewstudynotes.com

negativity or Qb When we confine our discussion to me shaded area each isoavant can be taken
as a function of the form be 0cal beiause for every admissible a it determines a unique b

The SoC revolves around the second Part dev TV obtainable from 11.38 Because Qa and Qb
being derivatives are themselves functions of a and b we can find Taa Tab tuba and make

Taa Po Qaa e rt Poaabe rt

it 11.411
Tube
jpg Po Gabe
rt
Po abbe
of

For the stationary value of tu to be a max its surrilient mat

Hilco Taa so whim can occur it r Aaa o

Hel 14170 TuanTubb Tu ab only ist QaaQbb Q'ab

Thus we can test using either Nij or ai whichever are more convenient

Qaa denotes the rate of change Qa C Mppa as input a change while b is rixed similarly
with a being fixed with Qba So the Soc stipulates that the MPP a bohinputs be diminishing
at chosen levels at and b Diminishing MPPa and MPPb doesnt guarantee the Soc though
andrewstudynotes.com
as Gab Qba is Possible

Just as the foe requires the isoavant to be negatively sloped at the chosen Point of input
that same isoavant to be strictly convex
combination the Soc sufficient serves to speliey
db
at the chosen level of output combination associated with me sign a da

To obtain it 11.40 must be differentiated totally with respect to a keeping in mind that
Qa and Qb are derivative functions of a and b and get on an iso avant b it a an iron ora

Qa Qa Carb Qb Qb Cai bl and b deal

The total differentiation process is mus as follows
dib
dal da Eg I Qb 9 Qa 4 11.42
Since b is a function of a on the isoquant the total derivative formula d g gives us
dQalda
3 91 391 Gba data t Qaa 11.43
21
dablda 381 abb aabfat
Substituting 11.40 into 11.43 and then substituting the latter into 11.42 we can rewrite me second

derivative
d baa andrewstudynotes.com
at aaa Qb Aba Qa Gab Qa t QbbQa Ib
Ib Qaa Qb 2Qab Qa Qb t Abb Ga 11.44
Note that he bracketed expression in the last line is a Quadratic form in the two variables
Qa and Qb to the Soc sufficient is satistied so that
Qaa co and Is Eight o
d b'da
Then the Quadratic form is negative definite This will in turn mane positive as
Qb has been constrained to be positive by me for The satisfaction or me Soc men means

hathrevant isoavant is strictly convex at the choseninput a combination

IIIIIpptose
mat meinterest is compounded Quarterly instead at io Quarter Also assume
that the production Process takes exactly a Quarter of a year The profit function becomes
Tu PoQcarb It io Paoa Paob
The Foc is now
l
PoQa CI ti o Pao 0 2117122 2217122
PoQb Cltio
17 06 14 00
Pbo o that
26h75
andrewstudynotes.com
Whitisexachy the same in Example 6 but with diererent manner or discounting

11 Comparative Statin Aspects a Optimization

Theidea is to find out how any change in theparameter will correct the equilibrium position
which in the present context recess to theoptimalvalues Theres nothing new so we'll discuss examples

Redvied Form Solutions

From Example I of Sec 11.6 we have

Q 41195Pa and Q 4129yd

We can Produle the comparative statics properties through simple partial differentiation

These conclusions follow from the particular assumptions or the model In particular he
effects of a change in Pio and at and Pro on Q are lonseauences of the assumed technica
andrewstudynotes.com
relation on the production side In the absence of such a relation we shall have

We note that the optimal output levels such as Q d and at 73 in Example 2 all the
tonstents have disappeared once we diorerentiate them all which means that they have all

lost their identityThis means that theres a lack of generality in the use a numeric constan

But he nonuse a numerilal constants doesnt guarantee that it will be solvable using
comparative Statics

General Function Models

We can derive the comparative steric properties from the input decision Problem or Example 6

To do his we need the implicit anation theorem This time unlike the nongoal Problems we
will consider me Foc of optimization For Example 6 mis is in 11.3g

andrewstudynotes.com
Collecting the terms in 11.3g to the left of me equal signs and making explicit hat Qa and
Qb are both functions of the endogenous variables a and b we can rewrite the Foc as follow
rt
F'carb Po Pao Pbo r E PoGaCarb e Pao O s
F a bi PoPao Pbo r t Po Qb a b e ut Pbo o
The functions F and F are assumed to possess continuous partial derivatives Thus it
would be possible to apply the implicit function theorem assuming the Jacobian or the exogenous
a and b doesnt vanish The Jacobian is just the Hessian det co tu tu in Example 6

11.46

Hence if we assume that the Soc sufficient for profit max is satisfied then Hl
must be positive and so does III In that event the implicit function theorem will allow us

to write the implicit actions

at at Po Pao Pbo r th 1.4
b b Po Pao Pbo r e

as well as the pair of identities

PoKaCatt b e
rt andrewstudynotes.com
Pao O
11 48
rt
PoQb Cat b e pyo o
To study the comparative statics First take the total differential or each identity in 11.48
For the time being we'll let all the exogenous to vary so we have dat dbt dPo d Pao dPbo
dr and de k we place on the Ihs terms hat only involve dat and dbt

11 Ug

The first and second derivatives a G are all to be evaluated at the eavilibvium
at and b The losses w me dat and abt are precisely the elements of the Jacobian

To derive the specific comparative statics derivatives co which there are 10 we need to
allow a single exogenous to vary alone Suppose we let Po vary alone hen d Poto but
dPao dPbo dr de o so that the first term of the rhs will remain
only
andrewstudynotes.com
Dividing through by dPo and interpreting the ratio dat dro to be the comparative static
derivative 29 12Po and similarly for db Ideo we ran write
ut rf ut
toAca e Gabe 29 12po Ga e
re r rt
Gabe Poabb e Cab 12po Qb e

The solution by Cramer's rule

CQbaab
ft ggabblioe.mx 11.50
caaaab
It gyaaatpoe.ru
Lets analyse the signs of the comparative static derivatives in 11.50 Assuming me Sossurrilient
is satisfied then 111is positive The SoCimplies that Gaa and Qbb are negatives while he
positive Poe t
Foc implies hat Ga and Qb are is also certainly positive

Thus it Qch o we can conclude hat both 29 12Po and 26 12Po are positive implying
that an increase in my price will result in increased employment or bom inputs in
product
andrewstudynotes.com
equilibrium to Qabco then it would depend on the relative strength of the negative and

Positive force a me parenthesesed rhs

Next lets let r vary alone so all the terms on the right will vanish except dr Dividing
through by dr we obtain

59

Both of these will be negative if Qab is positive but

comparative statics derivatives

indeterminate if Gob is negative Because the symmetry a r and t 129 12 1 and

andrewstudynotes.com
Gh't de will have the same appearance to 11.51

747
a

andrewstudynotes.com

andrewstudynotes.com