CAS EC 505 Mathematics for Economics

III. Multivariate Calculus

Bjorn Persson

Boston University

Fall 2023

Bjorn Persson (Boston University) Calculus Fall 2023 1 / 47

Open and Closed Sets I

Open and closed sets

Let X Rn . For x 2 X , and ε > 0 :

Bε (x ) = fy 2 X : d (x, y ) < εg

is an open neighborhood of x

A set Oi X is open (in X ) if for each x 2 Oi , there exists a scalar

ε > 0 such that Bε (x ) Oi

A set B is closed (in X ) if its complement B C = A is open

De…ne the union of all open subsets of X as:

O = fOi : Oi Rn open and Oi Xg

Bjorn Persson (Boston University) Calculus Fall 2023 2 / 47

Open and Closed Sets II

The interior points of a set X Rn are:

int (X ) = [i Oi = O

A set C is closed (in X ) if its complement C C = O is open

Bounded set
A set X 2 Rn is bounded from above if there is a number M 2 R
such that kx k M for all x 2 X

The set of boundary points of a set X Rn is:

n o
∂X = x 2 Rn : each Bε (x ) has some y 2 int (X ) and some y 2 X C

Note that X = int (X ) [ ∂X

Bjorn Persson (Boston University) Calculus Fall 2023 3 / 47

Open and Closed Sets III

A closed set contains all its boundary points

In Rn , all sets of the form (ai , bi )n are open sets, all sets [ai , bi ]n are
closed

A set can be neither open nor closed, e.g. (ai , bi ]n

The sets Rn and ∅ are both open and closed in Rn

Compactness
Theorem (Heine-Borel): A set K 2 Rn is compact i¤ it is closed and
bounded

Bjorn Persson (Boston University) Calculus Fall 2023 4 / 47

Open and Closed Sets IV

Example
Are the following sets compact?

X = (x, y ) 2 R2 : x + y < 1
Y = (x, y ) 2 R2 : 1 < x < 1, and y = 0

In Rn , all sets of the form (ai , bi )n are open sets, all sets [ai , bi ]n are
closed

A set can be neither open nor closed, e.g. (ai , bi ]n

All …nite sets in Rn are closed - they are the …nite union of the sets
n
i [ i [ ai , ai ]

The sets Rn and ∅ are both open and closed in Rn

Bjorn Persson (Boston University) Calculus Fall 2023 5 / 47

Open and Closed Sets V

Properties of open sets

X and ? are open in X

the intersection of a …nite collection of open sets is open

the union of an arbitrary family of open sets is open

Properties of closed sets

X and ? are closed in X

the intersection of an arbitrary family of closed sets is closed

the union of a …nite collection of closed sets is closed

Bjorn Persson (Boston University) Calculus Fall 2023 6 / 47

Open and Closed Sets VI

Example
Consider the following sets in R :

A = \n 2N [n, n + 1]
1 n
B = [ n 2N ,
n+1 n+1

Bjorn Persson (Boston University) Calculus Fall 2023 7 / 47

Sequences in Euclidean Space I

De…nition
A sequence in a set X Rn is a mapping f : N ! X often denoted:

xk = fxk gk∞=1 = fx1 , x2 , ...g

Bounded sequence
A sequence xk is bounded if there exists an M 2 R such that
kxk k M for all k

Converging sequence
A sequence xk converges to a point x 0 if for every ε > 0 there exists
an N 2 N such that:

xk 2 Bε x 0 for all k > N

Bjorn Persson (Boston University) Calculus Fall 2023 8 / 47

Sequences in Euclidean Space II

Limit
A sequence xk that converges to a point x 0 said to have a limit at
x = x0 :
lim fxk g = x 0
k !∞

Accumulation point
A sequence xk has an accumulation point y 0 if x = y 0 in…nitely often

Theorem
A sequence may have multiple accumulation points, but at most one
limit

Right limit and left limit

Bjorn Persson (Boston University) Calculus Fall 2023 9 / 47

Continuity I
De…nition of continuity
A function f : X ! Y is continuous at x 2 X if for every nbh B of
y = f (x ) exists a nbh A of x such that f (x 0 ) 2 B for 8x 0 A \ X

The following are equivalent:

f is continuous at x 0
8ε > 0 9δ > 0 such that:

x0 x <δ) f x0 f (x ) < ε

for every sequence fxt g from X with limit:

x 0 : fyt g = ff (xt )g

is a sequence in Y with limit y = f (x 0 )

Bjorn Persson (Boston University) Calculus Fall 2023 10 / 47

Continuity II

An operational de…nition
A function f : X ! Y is continuous at x = x 0 if:
f (x ) is de…ned at x = x 0
f (x ) has a limit y at x = x 0
f (x 0 ) = y

Bjorn Persson (Boston University) Calculus Fall 2023 11 / 47

Weierstrass’Theorem I

Theorem
A continuous function de…ned on a compact set K Rn achieves
global extrema:

f (x ) = max(min)f (x ) for at least one x 2 K

x 2K

The continuous image of a compact set is compact

Su¢ ciency theorem

Example
Let f : K R ! R be de…ned by:

f (x ) = 4x x 2 on K = [0, 3]

Bjorn Persson (Boston University) Calculus Fall 2023 12 / 47

Multivariate Functions I
De…nition
Recall that a function from X to Y is a rule f that assigns a unique
value f (x ) 2 Y to each x 2 X

Let f : X ! Y and suppose that X Rn and Y Rm

If m = 1, f is scalar-valued and if m > 1, f is vector-valued

Examples
A utility function maps consumption vectors x 2 Rn into real
numbers:
u : Rn ! R
A production function maps n inputs into m outputs:

F : Rn ! Rm

Bjorn Persson (Boston University) Calculus Fall 2023 13 / 47

Scalar-valued Functions I

De…nition
A function f : Rn ! R is real-valued (or scalar-valued)

Recall that the graph of a function f : X ! Y is the geometric image

of the function:

Gf = f(x, y ) 2 X Y : y = f (x )g

The graph of a univariate scalar-valued function f : R ! R

Gf = (x, y ) 2 R2 : y = f (x )

is a subset of R2 and provides a lot of information about the function

In higher dimensions, picturing the graph is di¢ cult or impossible

Bjorn Persson (Boston University) Calculus Fall 2023 14 / 47

Scalar-valued Functions II

One procedure is to restrict the function in some dimension, and then

consider it in this restricted domain

Example
The function f : R2 ! R given by f (x1 , x2 ) = y can be restricted by
…xing x1 = x 1 . This gives a one-dimensional function with graph in
the (x2 , y ) plane:
f (x 1 , x2 ) = y
Repeating this for various levels of x1 can provide a good picture of
the whole function

Example
The expenditure function e : Rn++ R ! R+ restricted to R++

Bjorn Persson (Boston University) Calculus Fall 2023 15 / 47

Level Sets I

De…nition
For a function f : Rn ! R, a level set (contour set) consists of all
vectors in the domain where f is constant:

Lf (c ) = fx 2 Rn : f (x ) = c for some c 2 Rg

For a function of two variables f : R2 ! R, the level sets are images

in the plane R2

Note how the inverse of the function plays a large part here:

Lf (c ) = (x1 , x2 ) 2 R2 such that f 1

(c ) = fx1 , x2 g

Level sets are projections of f into the plane

Bjorn Persson (Boston University) Calculus Fall 2023 16 / 47

Level Sets II

Examples
Sketch some level sets of the following functions:
f : R2 ! R+ de…ned by f (x1 , x2 ) = x12 + x22

g : R2 ! R de…ned by g (x1 , x2 ) = x12 x22

h : R2 ! R de…ned by h (x1 , x2 ) = x12 x22

k : R2++ ! R+ de…ned by: k (x1 , x2 ) = x1a x21 a , where a 2 R++

l : R2 ! R de…ned by: l (x1 , x2 ) = min fx1 , x2 g

Bjorn Persson (Boston University) Calculus Fall 2023 17 / 47

Di¤erentiation I

One variable
Consider a one-variable real-valued function f : R ! R. The
derivative of f at x 0 (if it exists) is de…ned:

f x 0 + ∆x f (x 0 )
f 0 (x 0 ) = lim 2R
∆x !0 ∆x

Equivalently, f is di¤erentiable at at x 0 is if there exists an a¢ ne

function
g ( x ) = a0 + a1 x
such that for x = x 0 + ∆x :
f (x ) g (x )
lim =0
∆x !0 ∆x

Bjorn Persson (Boston University) Calculus Fall 2023 18 / 47

Di¤erentiation II

Example
f (x ) = 2x 2 + x at x 0 = 1
Two variables
Consider a two-variable real-valued function f : R2 ! R. The partial
derivatives of f with respect to x evaluated at x 0 :

f x10 + ∆x1 , x20 f (x10 , x20 )

f1 x 0 = lim
∆x1 !0 ∆x1
f x1 , x2 + ∆x2
0 0 f (x10 , x20 )
f2 x 0 = lim
∆x2 !0 ∆x2

The derivative of f at x 0 is de…ned as the a¢ ne function:

g (x ) = a0 + a1 x1 + a2 x2

Bjorn Persson (Boston University) Calculus Fall 2023 19 / 47

Di¤erentiation III

Notation
Various forms:
∂f (x 0 )
fi (x 0 ) = Di f x 0 =
∂xi
Geometric representation
The partial derivatives can geometrically be represented by tangents

Computation
Find the partials of f : R2 ! R and g : R2 ! R de…ned by

f (x ) = 5x12 x2 + 3x1 e 2x2

g (x ) = x1 x2 + x13 ln x2

Bjorn Persson (Boston University) Calculus Fall 2023 20 / 47

Di¤erentiation IV
Tangent planes
Let f : R2 ! R be continuously di¤erentiable and consider a point x 0
in the domain

Suppose an in…nitesimal movement from x 0 to x 0 + ∆x

The tangent plane of f : R2 ! R at x 0 is an a¢ ne mapping in x :

g (x ) = f x 0 + f1 x 0 ∆x1 + f2 (x 0 )∆x2

The gradient
The vector of (…rst-order) partial derivatives is known as the gradient
and is often written:
f1 x 0
= rf (x 0 )
f2 ( x 0 )

Bjorn Persson (Boston University) Calculus Fall 2023 21 / 47

Di¤erentiation V
Using the gradient, we note that the tangent plane is the best a¢ ne
approximation of f around x = x 0 :

g (x ) = f (x 0 ) + rf (x 0 ) x x0

Example
Find the best a¢ ne approximation of f (x ) at x 0 = (1, 1) where
f : R2 ! R is de…ned as above:

f (x ) = 4x12 x2 + 2x1 x22

The chain rule of di¤erentiation

In one-variable calculus, two (or more) functions
f :X R!Y R and g : Y R !Z R can be combined into
one:
h=f g :X !Z
Bjorn Persson (Boston University) Calculus Fall 2023 22 / 47
Di¤erentiation VI

The function h can then be written:

h (x ) = f (g (x ))

If both f and g are di¤erentiable, the chain rule of di¤erentiation

states that the derivative of h is:

h0 = f 0 g 0

This generalizes naturally into higher dimensions

A continuously di¤erentiable function f : R2 ! R de…ned by

z = f (x ) where x : R ! R2 are di¤erentiable can be written:

f (x (t )) = z (t )

Bjorn Persson (Boston University) Calculus Fall 2023 23 / 47

Di¤erentiation VII

Then the chain rule states that:

z 0 (t ) = f1 ( ) x10 (t ) + f2 ( )x20 (t )

The total derivative of the composite function z with respect to t is

the sum of the changes in z caused by t through the functions x1 (t )
and x2 (t )

Bjorn Persson (Boston University) Calculus Fall 2023 24 / 47

Di¤erentiation VIII

Example
Let f (x1 , x2 ) = x12 x2 x22 , where x1 (t ) = t 2 and x2 (t ) = 2t :

z 0 (t ) = 2x1 x2 (2t ) + x12 2x2 (2)

= 4x1 x2 t + 2x12
4x2
2 4
= 4t (2t )t + 2t 8t = 10t 4 8t

The chain rule generalizes to many variables. Let f : Rn ! R and

x : Rm ! R be di¤erentiable functions. Then:
n m
(f x )t = fx xt = ∑ ∑ fi x j
i =1 j =1

Bjorn Persson (Boston University) Calculus Fall 2023 25 / 47

Di¤erentiation IX
Directional derivative
The directional derivative indicates how a di¤erentiable function
varies as a result of an in…nitesimal movement in an arbitrary
direction in the domain

More speci…cally, consider how the function f : R2 ! R varies in

direction v 2 R2 at a point x 0 =2 R2

A directional function φ : R ! R associated with f at x 0 is:

φ (t ) = f (x (t )) = f x 0 + tv

where t 2 R is a parameter

The chain rule gives the change in φ resulting from a movement by t

in direction v :

φ 0 ( t ) = f1 x 0 + tv v1 + f2 x 0 + tv v2
Bjorn Persson (Boston University) Calculus Fall 2023 26 / 47
Di¤erentiation X

The directional derivative can be expressed as the inner product of

the gradient and the direction vector:

Dv f x 0 = rf (x 0 ) v

This is a generalization of the derivative

Properties of the directional derivative

Using the cosine theorem, we can write the directional derivative:

Dv f x 0 = r f x 0 v = rf x 0 kv k cos α

where α is the angle between rf x 0 and v

Bjorn Persson (Boston University) Calculus Fall 2023 27 / 47

Di¤erentiation XI

Questions
In what direction v is the directional derivative the largest?

In what direction v is the directional derivative zero?

Example
Let f (x ) = x12 + 2x22 and suppose x 0 = (1, 2)

In what direction v should we move to get the quickest increase of f ?

The quickest decrease of f ?

Keep f constant?

Bjorn Persson (Boston University) Calculus Fall 2023 28 / 47

The General Case I

Partial derivative
The partial derivative in the general case of f : Rn ! Rm is de…ned
similarly. The partial derivative of f j wrt xi evaluated at x 0 is:

f j xi0 + ∆xi , x 0 i f j xi0 , x 0 i

fi (xi0 ) = lim
∆xi !0 ∆xi

More generally, a function f : Rn ! Rm is di¤erentiable at x 0 2 Rn

if 9 a¢ ne function g : Rn ! Rm and for 8ε > 0 9δ > 0 such that:

kf (x ) g (x )k
8x 2 Bδ (x0 ) ) ε
kx x0k

The best a¢ ne approximation g (x ) is given by:

g (x ) = f (x 0 ) + J (x 0 )(x x0)

Bjorn Persson (Boston University) Calculus Fall 2023 29 / 47

The General Case II

Here, J x 0 is an m n matrix of …rst-order partial derivatives

known as the Jacobian evaluated at x 0 :
0 1 0 1
f1 ( x ) fn1 (x 0 )
B .. .. .. C
J x0 = @ . . . A
f1m (x 0 ) fnn (x 0 )

Example
Find the Jacobian at x 0 = (1, 1) for:

f (x ) = 2x12 + 4x1 x2 , ln x1 x23

Bjorn Persson (Boston University) Calculus Fall 2023 30 / 47

Second-order Derivatives and the Hessian Matrix I

Partial derivatives of the second order

Again consider the special case where f : R2 ! R and suppose f is
di¤erentiable

The second-order partial derivative of f with respect to x1 at x 0 is:

f1 x10 + ∆x1 , x20 f1 x10 , x20

f11 x 0 = lim
∆x1 !0 ∆x1

if it exists

A function f : R2 ! R has at most four second-order partial

derivatives

Bjorn Persson (Boston University) Calculus Fall 2023 31 / 47

Second-order Derivatives and the Hessian Matrix II
The Hessian
The second-order partial derivatives are often grouped in a matrix:

f11 x 0 f12 x 0
r2 f x 0 = H (x 0 ) =
f21 x 0 f22 x 0

known as the Hessian matrix of f evaluated at x 0

Example
Find the Hessian matrix of f (x ) = 5x12 x2 + 3x1 e 2x2

Young’s Theorem
If f is twice continously di¤erentiable the Hessian is symmetric:

H x0 = HT x0

Bjorn Persson (Boston University) Calculus Fall 2023 32 / 47

Second-order Derivatives and the Hessian Matrix III

The properties of the Hessian matrix is used to classify stationary

points (investigate curvature)

Second-order partial derivatives extend naturally into higher

dimensional space

Example
Let f : Rn ! R be a twice di¤erentiable at x 0 2 Rn . Then the
Hessian evaluated at x 0 is an n n matrix:
0 1
f11 x 0 f1n (x 0 )
B .. .. .. C
H (x 0 ) = @ . . . A
0
fn1 (x ) 0
fnn (x )

Bjorn Persson (Boston University) Calculus Fall 2023 33 / 47

The Implicit Function Theorem I
Implicit functions
Level sets have an interpretation as implicit relationships between
variables in the domain

Let f : Rn ! R and consider the expression:

f (x ) = 0

Example
Let f 1 (x ) = 2x1 + x2 = 0 and f 2 (x ) = x12 + x22 = 1

When is it possible to de…ne a function implicitly from a given

relation?

Can we represent some variable(s) in the domain in terms of others at

a given level set?

Bjorn Persson (Boston University) Calculus Fall 2023 34 / 47

The Implicit Function Theorem II

That is, can we …nd a function ϕ such that that the following holds
at x 2 Bε (x 0 ) :

f (x ) = 0 ) xm = ϕ(x1 , .., xm 1 , xm +1 , .., xn 1 )

The Implicit Function Theorem

Suppose f : M Rn ! R is continuously di¤erentiable on an open
set M 0 M, x 0 2 M 0 and let:

f (x 0 ) = 0

Bjorn Persson (Boston University) Calculus Fall 2023 35 / 47

The Implicit Function Theorem III
Suppose further that fm (x 0 ) 6= 0. Then there exists a nbh
Bε (x 0 m ) M 0 , an open set D R containing xm0 and a real-valued
di¤erentiable function ϕ : M 0 ! D such that:

xm = ϕ(x1 , .., xm 1 , xm +1 , .., xn )

and:

f (x ) = f (x1 , .., xm 1, ϕ(x1 , .., xm 1 , xm +1 , .., xn ), xm +1 , .., xn )

Furthermore, by di¤erentiating this identity with respect to xj we

obtain:
fj (x ) + fm (x ) ϕj (x ) = 0
or, equivalently:
fj (x )
ϕj (x ) =
fm ( x )

Bjorn Persson (Boston University) Calculus Fall 2023 36 / 47

The Implicit Function Theorem IV
Example in two dimensions
Let f 3 (x ) = x13 + x12 x2 2x22 10x2 = 0 and consider the point
x 0 = (2, 1)

What is x20 (2)?

Geometric interpretation
For a two-variable real-valued function f (x ) = c where x2 can be
de…ned locally as a function of x1 , the expression x20 (x10 ) gives the
slope of the level set of the function f at x 0 where f (x 0 ) = c

The equation for the tangent line at x 0 is:

f1 (x 0 ) x1 x10 + f2 (x 0 ) x2 x20 = 0

or, equivalently:
rf (x 0 ) v = 0
Bjorn Persson (Boston University) Calculus Fall 2023 37 / 47
The Implicit Function Theorem V

The level curve (or the tangent vector v ) is perpendicular to the

gradient

Example utility function

Suppose utility function u : R2 ! R de…ned by
u (x1 , x2 ) = 31 ln x1 + 23 ln x2

In what direction from (1, 1) does u increase the most?

In what direction is u = u?

Example production function

Let f : Rn ! R be a neoclassical production function:

y = g (x1 , ..., xn )

Bjorn Persson (Boston University) Calculus Fall 2023 38 / 47

The Implicit Function Theorem VI

Write this equation implicitly as:

f (x1 , x2 , ..., xn , y ) = g (x1 , ..., xn ) y =0

where f : Rn +1 ! R. Assuming that f is continuously di¤erentiable

and the Jacobian matrix has rank 1:

Dxj f (x1 , x2 , ..., xn , y ) = Dxj g (x1 , x2 , ..., xn , y ) 6= 0 for j = 1, .., n

Solving equation for xm as a function of x m and y :

xm = ϕ (x1 , .., xm 1 , xm +1 , .., xn 1 , y )

Inserting this expression into the production function gives:

g (x1 , x2 , .., xm , .., xn ) y =0

Bjorn Persson (Boston University) Calculus Fall 2023 39 / 47

The Implicit Function Theorem VII

Di¤erentiating with respect to xj gives:

Dxj g + Dxm g Dxj xm = 0

or:
Dx j g
Dxj xm = = MRS
Dxgm
Using Leibniz’notation, this can be written as the solution to:

∂f (x1 , x2 , ..., xn , y ) ∂f (x1 , x2 , ..., xn , y ) ∂xm

+ =0
∂xj ∂xm ∂xj

Bjorn Persson (Boston University) Calculus Fall 2023 40 / 47

The General Implicit Function Theorem I
The general case
Suppose that f : M Rn +m ! Rn is a di¤erentiable function on
M 0 M written:

f 1 x10 , x20 , .., xn0 , y10 , y20 , .., ym0 = 0

..
.
f n x10 , x2,
0
..., xn0 , y10 , y20 , .., ym0 = 0

Suppose further that the Jacobian matrix:

0 1 0 0 1
f1 x , y fn1 x 0 , y 0
0 0 B .. .. .. C
Jf x , y = @ . . . A
f1n (x 0 , y 0 ) fnn x 0 , y 0

is non-singular (has rank n )

Bjorn Persson (Boston University) Calculus Fall 2023 41 / 47
The General Implicit Function Theorem II

Then there exists a neighborhood Bε (x 0 , y 0 ) M 0 , an open set

0
M Rm containing y 0 and a continuously di¤erentiable function
ϕ : R ! Rm such that:
m

xi0 = ϕi (y 0 )

Moreover, we have that:

Dx f (x 0 , y 0 )Dy ϕ(y 0 ) + Dy f (x 0 , y 0 ) = 0

or, equivalently:
1
Dy ϕ ( y 0 ) = Dx f (x 0 , y 0 ) Dy f ( x 0 , y 0 )

Bjorn Persson (Boston University) Calculus Fall 2023 42 / 47

Comparative Statics I

Example 1
Supply and demand system:

S= Q S (p, t )
D = Q D (p, y )

where p is price, y is income, and t is a commodity tax. Assume the

following:

Dp Q S (p, t ), Dy Q D (p, y ) > 0

Dt Q S (p, t ), Dp Q D (p, y ) < 0

In equilibrium:
0 = Q S (p, t ) Q D (p, y )

Bjorn Persson (Boston University) Calculus Fall 2023 43 / 47

Comparative Statics II

The equilbrium condition determines p as a function of t and y :

0 = Q S (p (y , t ), t ) Q D (p (y , t ), y )

Di¤erentiating this condition and solving for Dy p (y , t ) and Dt p (y , t )

gives:

Dy Q D ( p ( y , t ) , y )
Dy p ( y , t ) = >0
Dp Q S ( p ( y , t ) , t ) Dp Q D ( p ( y , t ) , y )
Dt Q S (p (y , t ), t )
Dt p ( y , t ) = >0
Dp Q ( p ( y , t ) , t ) Dp Q D ( p ( y , t ) , y )
S

Bjorn Persson (Boston University) Calculus Fall 2023 44 / 47

Comparative Statics III

Example 2
Consider the Keynesian model:

Y = C (Y T ) + I (r ) + G
M = L(Y , r )

where Y is GDP, C consumption, T taxes, I investments, r the

interest rate, and G government spending, M money supply, and
where L is a liquidity function

Assume that 0 < C 0 < 1, I 0 < 0, LY > 0, and Lr < 0. This gives the
implicit system:

C (Y T ) + I (r ) + G Y 0
F (Y , r , G , T , M ) = =
L(Y , r ) M 0

Bjorn Persson (Boston University) Calculus Fall 2023 45 / 47

Comparative Statics IV

De…ne ϕ : R3 ! R2 as:

Y (G , M, T )
ϕ(G , M, T ) =
r (G , M, T )

Implicit di¤erentiation gives:

D(Y ,r ) F (Y , r , G , T , M )D ϕG ,M ,T + DG ,M ,T F (Y , r , G , T , M ) = 0

or:

C0 1 I0 YG YM YT 1 0 C0 0
+ =
LY Lr rG rM rT 0 0 1 0

Bjorn Persson (Boston University) Calculus Fall 2023 46 / 47

Comparative Statics V

Solving gives:
1
YG YM YT C0 1 I0 1 0 C0
=
rG rM rT LY Lr 0 0 1
or:

YG YM YT 1 Lr 0 I0 C 0 Lr
= 0
rG rM rT det D(Y ,r ) LY 0 C LY C0 + 1

Here, det D(Y ,r ) = (C 0 1 ) Lr LY I 0 > 0 by the assumptions

above

Bjorn Persson (Boston University) Calculus Fall 2023 47 / 47