1

Implicit Functions

Reading [Simon], Chapter 15, p. 334-360.

1.1

Examples

So far we were dealing with explicitly given functions

y = f (x1 , ..., xn ),
like y = x2 or y = x21 x32 .
But frequently the dependence of endogenous variable y on exogenous
variables (x1 , ..., xn ) can be given in a form
G(x1 , ..., xn , y) = c.
If for each (x1 , ..., xn ) this equation determines a corresponding value of y,
we say that the endogenous variable y is an implicit function of exogenous
variables (x1 , ..., xn ).
Example. Suppose G(x, y) = 4x + 2y 5. Then the equation
4x + 2y 5 = 0
expresses y as an implicit function of x. This implicit function can be written
explicitly as
y = 2.5 2x.
Example. Suppose G(x, y) = xy 2 3y ex . Then the equation
xy 2 3y ex = 0
yields an explicit function
y=

1
(3 + 9 + 4xex ).
2x

By the way, there is another one

y=

1
(3 + 9 4xex ).
2x

Example. Suppose G(x, y) = y 5 5xy + 4x2 . Then the equation

y 5 5xy + 4x2 = 0
yields an implicit function y = y(x) which can not be written in a form of
explicit formula because there is no general formula for equations of order
5.
1

Example. Suppose G(x, y) = x2 + y 2 1. The corresponding equation

x2 + y 2 1 = 0
determines, as we know, the unit circle.
It is evident that
(x1 = 0, y1 = 1), (x2 = 0, y2 = 1), (x3 = 1, y3 = 0), (x4 = 1, y4 = 0)
all lay on this circle, that is all four are particular solutions of this equations.
1. Around (x1 = 0, y1 = 1) this equation determines the explicit function

y = 1 x2 ,
whose domain can be enlarged to x (1, 1).
2. Around (x2 = 0, y2 = 1) this equation determines the explicit
function

y = 1 x2 ,
whose domain can be enlarged to x (1, 1).
3. Around (x3 = 1, y3 = 0) this equation determines the explicit function
q

x=

1 y2,

whose domain can be enlarged to y (1, 1).

4. Around (x4 = 1, y4 = 0) this equation determines the explicit function
q

x = 1 y2,
whose domain can be enlarged to y (1, 1).

1.2

Implicit Function Theorem for R2

So our question is: Suppose a function G(x, y) is given. Consider the equation
G(x0 , y0 ) = c.
Does there exists a function y = y(x) defined on some interval (x0 , x0 + )
such that G(x, y(x)) c?
Or
Does there exists a function x = x(y) defined in an interval (y0 , y0 + )
such that G(x(y), y) c?
Such a function y = y(x) (or x = x(y)) is called implicit function defined
by the equation G(x, y) = c around the point (solution) (x0 , y0 ). The graph
of implicit function must be a locus the level curve G(x, y) = c.
2

This picture shows that y(x) does not exist around the point A of the
level curve G(x, y) = c (note that x = x(y) does not exist around D).
Why so? What is wrong, with A? Because around A the level curve
G(x, y) = c can not pass the vertical line test (the horizontal line test for D).
Note that the tangent line at A is vertical, and this means that the gradient
at A is horizontal, and this means that Gy (A) = 0. This is wrong with A!
And what is wrong with D?
Theorem 1 Suppose a point (x , y ) R2 is a particular solution of G(x , y ) =
c and G
(x , y ) 6= 0. Then the equation
y
G(x, y) = c
determines the function y = y(x) defined on some interval I = (x , x + )
about the point x such that
(a) G(x, y(x)) c for all x I;
(b) y(x ) = y ;
G

(x ,y )

x
(c) y 0 (x ) = G
.
(x ,y )
y

Proof. We prove the implication (a), (b) (c). So, suppose we have
y(x) such that y(x ) = y and G(x, y(x)) = c for all x I. Differentiating
G(x, y(x)) = c with respect to x at x we obtain
dx G
dy
G
(x , y(x ))
+
(x , y(x )) (x ) = 0,
x
dx
y
dx
or

G
G
(x , y ) +
(x , y ) y 0 (x ) = 0,
x
y

and this gives (c).

In fact this Theorem states that may be y can not be solved as an
function of x explicitly but it is possible to find the derivative y 0 (x )
3

and approximate the value of the unknown function y(x) around

small interval of x :
y(x + x) = y(x ) + y 0 (x ) x = y(x )

G
(x , y )
x
G
(x , y )
y

x.

Example. Consider the equation

G(x, y) = x2 3xy + y 3 7 = 0,
one solution of this equation is (x = 4, y = 3) (check this!). Estimate the
solution corresponding to x1 = 4.3.
Solution. Since
G
(4, 3) = (3x + 3y 2 )|(4,3) = 12 + 27 = 15 6= 0
y
the Theorem tells us that there exists a solution y(x) around x = 4 s.t.
y(4) = 3 and
G
(4, 3))
1
x
y 0 (4) = G
= .
15
(4, 3))
y
Then, by linear approximation we obtain
y1 = y(x1 ) y + y 0 (x ) x = 3 +

1
0.3 = 3.02.
15

By the way, G(4.3, 3.2) = 0.0756 6= 0, that is our point but our point(x1 , y1 ) =
(4.3, 3.2) does not lay exactly on the level curve G(x, y) = 0, but y(4.3) 3.2
is just an approximation!

1.3

Implicit Function Theorem for Several variables

Theorem 2 Suppose a point (x1 , ..., xk , y ) Rk+1 is a particular solution

of
G(x1 , ..., xk , y ) = c
and

G
(x1 , ..., xk , y )
y

6= 0. Then the equation

G(x1 , ..., xk , y) = c

determines the function y = y(x1 , ..., xk ) defined on an open ball B (x ) about

the point x = (x1 , ..., xk ) such that
(a) G(x1 , ..., xk , y(x1 , ..., xk )) c for all x B x ;
(b) y(x1 , ..., xk ) = y ;
(c)

y
xi

(x1 ,...,x ,y )

k
i
= x
.
G
(x ,...,x ,y )
y

Proof. We prove the implication (a), (b) (c). So, suppose we have
y(x1 , ..., xk ) such that G(x1 , ..., xk , y(x1 , ..., xk )) c for all x B x and
y(x1 , ..., xk ) = y . Differentiating G(x1 , ..., xk , y) = c with respect to xi at x
we obtain
G
G
y
(x , y(x ))
(x , y(x )) +
(x ) = 0,
xi
y
xi
solving from this equation

y
(x )
xi

we obtain (c).

Example. Consider the equation

G(x, y, z) = x3 + 3y 2 + 4xz 2 3z 2 y 1 = 0.
Does this equation define z = z(x, y) as a function of x and y
(a) in a neighborhood of A = (x = 1, y = 1)?
(b) in a neighborhood of B = (x = 1, y = 0)?
(c) in a neighborhood of C = (x = 0.5, y = 0)?
If so, compute zx and zy at this point. Besides, if x increases to 0.6 and
y decreases to 0.2, estimate the corresponding change in z.
Solution. The partial derivative Gz is
Gz (x, y, z) = 8xz 6yz.
(a) At point A = (1, 1): the given equation in this case looks as
1 + 3 + 4z 2 3z 2 1 = 0, 3 + z 2 = 0,
so no solution (x0 , y0 , z0 ) in this case.
(b) At point B = (1, 0): the given equation in this case looks as
1 + 0 + 4z 2 0 1 = 0, z 2 = 0, z = 0,
so the solution in this case is (x0 , y0 , z0 ) = (1, 0, 0). But what about the
implicit function z = z(x, y) around B?
3
2
2 3z 2 y1)
Let as check Gz = (x +3y +4xz
= 8xz 6yz at this point:
z
Gz (1, 0, 0) = (8xz 6yz)|1,0,0) = 0,
so the Implicit Function Theorem does not work at B.
(c) At point C = (0.5, 0): the given equation in this case looks as
0.53 + 0 + 2z 2 0 1 = 0, 2z 2 = 1 0.53 , z 2 = 0.4375000000,
z = 0.6614378278, z 0 = 0.6614378278,
so we have two solutions in this case is (x0 , y0 , z0 ) = (0.5, 0, 0.6614378278)
and (x0 , y0 , z0 ) = (0.5, 0, 0.6614378278).
5

Let as check Gz = 8xz 6yz

Gz (0.5, 0, 0.6614378278) = 2.645751311 6= 0,
Gz (0.5, 0, 0.6614378278) = 2.645751311 6= 0,
so it both cases - it does.
Now we compute zx and zy at this point:
x
zx (0.5, 0, 0.6614378278) = G
(0.5, 0, 0.6614378278),
Gz
Gx (0.5,0,0.6614378278)
2.500000000
Gz (0.5,0,0.6614378278) = 2.645751311 = 0.9449111825.

and

y
zy (0.5, 0, 0.6614378278) = G
(0.5, 0, 0.6614378278),
Gz
Gy (0.5,0,0.6614378278)
2.500000000
Gz (0.5,0,0.6614378278) = 1.312500000 = 0.4960783708.

Now we are ready to estimate the corresponding change in z if x increases

to 0.6 and y decreases to 0.2:
z = zx x + zy y =
0.944911182 0.1 + (0.4960783708) (0.2) = 0.1937067924.
Similarly for another solution (0.5, 0, 0.6614378278).
Phhhhh!

1.4

Regular Points

For a given smooth enough function G(x, y) the equation G(x, y) = c defines
the smooth curve, the level curve. Suppose a point (x , y ) lays on this curve,
i.e. is a solution of this equation.
1. If for (x , y ) one has

G
(x , y ) 6= 0
y

then the locus of level curve G(x, y) = c around (x , y ) can be thought of as

the graph of a function y = y(x), and the slope of this curve is
G
(x , y )
x
G .
(x , y )
y

2. If for (x , y ) one has

G
(x , y ) 6= 0
x
then the locus of level curve G(x, y) = c around (x , y ) can be thought of as
the graph of a function x = x(y), and the slope of this curve is

G
(x , y )
y
(x , y ).
G
x

3. If for (x , y ) one has

G
(x , y ) 6= 0
y

and

G
(x , y ) 6= 0
x
then the locus of level curve G(x, y) = c around (x , y ) can be thought of as
the graph of a function y = y(x) and y = y(x), and the slope of this curve
at (x , y ) with respect to x axis is:

G
(x , y )
x
G
(x , y )
y

and the slope of this curve with respect to y axis is:

G
(x , y )
y
G .
(x , y )
x

(What? We have two different slopes for one curve? Explain).

Definition 1 A point (x , y ) is called regular point of G(x, y) if
G
G
(x , y ) 6= 0, or
(x , y ) 6= 0.
x
y
If every point of G(x, y) = c is regular, then the level set G(x, y) = c is called
a regular curve.
So the Implicit Function Theorem states that at each point of regular
curve we can consider y as a function of x or x as a function of y.
Example. Consider G(x, y) = x2 , you know the graph of this function.
Its level curve G(x, y) = 0 is just the y-axes: G(x, y) = x2 = 0 x =
0. Each point of this level curve (0, y) is irregular: Gx (0, y) = (2x)|0,y) =
0, Gy (0, y) = 0|(0,y) = 0 (nevertheless this curve determines implicit function
x(y) = 0).

1.5

Tangent of the Level Curve

Theorem 3 The tangent vector to the level curve G(x , y ) = c at a regular

point (x , y ) is
(Gy (x , y ), Gx (x , y )).

Proof. Recall that the tangent vector of a curve (x(t), y(t)) is (x0 (t), y 0 (t)).
Suppose G(x , y ) = c and Gy (x , y ) 6= 0, then there exists implicit
function y = y(x) around x , i.e. G(x, y(x)) = c. Thus this level curve is
given by (x, y(x)) locally around x . Then its tangent vector at (x , y ) is
given by
Gx (x , y )
),
(x0 , y 0 (x)) = (1,
Gy (x , y )
this vector is parallel to (Gy (x , y ), Gx (x , y )).
Corollary 1 At a regular point the gradient is orthogonal to level curve.
Proof.
G(x , y ) (Gy (x , y ), Gx (x , y ) =
(Gx (x , y ), Gy (x , y ) (Gy (x , y ), Gx (x , y ) =
Gx (x , y ) Gy (x , y ) Gy (x , y ) Gx (x , y ) = 0.
Important Example. Let F (x, y) = x2 + y 2 and G(x, y) = x y. Find
a point (x , y ) on the level curve G(x, y) = 1 and c such that the curves
G(x, y) = 1 and F (x, y) = c touch each other at the point (x , y ).
Solution. The slope of tangent line of G(x, y) = 1 at (x , y ) is

Gx (x , y )
Gy (x , y )

and the slope of tangent line of F (x, y) = c at (x , y ) is

Fx (x , y )
.
Fy (x , y )

So we can find (x , y ) from the system of equations

Gx (x ,y )
Fx (x ,y )

Gy (x ,y ) = Fy (x ,y )

G(x , y ) = 1

which in our case looks as

y
2x

x = 2y

x y =1

The solution gives (x = 1, y = 1) and (x = 1, y = 1). In both cases

c = F (1, 1) = F (1, 1) = 2.
8

1.6

Implicit Function Theorem and Marginal Rate of

Substitution

Suppose G(x , y ) = c, and let us step from x to x + x. How shall we

change y in order to stay on the same level set? That is for which y we
shall have G(x + x, y + y) = c?
In other words, what combinations of linear movements x and y from
(x , y ) lead to no change in G?
If we have the locus of level curve G(x, y) = c explicitly as y = y(x) the,
its clear, y = y(x + x) y .
But what if we have not explicit y(x)?
By linear approximation
G(x + x, y + y) G(x , y )

G
G
(x , y )x +
(x , y ))y.
x
y

So for no change, i.e. for G(x + x, y + y) G(x , y ) 0 we need

G
G
(x , y )x +
(x , y )y 0.
x
y
Thus

G
(x , y )
y
x
G
,
x
(x , y )
y

so the direction of no change of G at (x , y ) is the direction of tangent line

of the level curve. Besides, we can answer the above question: if we change
x by x what should be the change y of y in order to keep
G(x + x, y + y) = G(x , y )?
As we see we have
y

G
(x , y )
x
x.
G
(x , y )
y

In economics the slope of the tangent line to the level curve G(x, y) = c
at (x , y ) is called marginal rate of substitution MRS, it is given by the
derivative of implicit function y = y(x)
G

(x , y )
.
(x , y )
y

x
M RS = G

MRS measures, in a marginal sense, how much should one increase y to

compensate the loss of one unit of x to keep the same level of G.
That is it, but if you need, here is
Economical Interpretation. This use of the Implicit Function Theorem is the natural approach when studying the slope of an indifference curve
9

of a utility function and the slope of an isoquant of a production function,

since in these situations we are interested in which directions to move to keep
the function constant.
The level curve of a utility function U (x, y) is called an indifference curve
of U .
Its slope at (x , y ) is called the marginal rate of substitution (MRS) of
U at (x , y ) since it measures, in a marginal sense, how much more of good
y the consumer would require to compensate for the loss of one unit of good
x to keep the same level of satisfaction.
By the Implicit Function Theorem, the MRS at (x , y ) is:
U

(x , y )
(x , y )
y

x
U

Similarly, if Q = F (K, L) is a production function, its level curves are called

isoquants and the slope - FK /FL of an isoquant at (Ko, Lo) is called the
marginal rate of technical substitution (MRTS). It measures how much of
one input would be needed to compensate for a one-unit loss of the other
unit while keeping production at the same level.
Example. Consider a function f (x, y) = x2 ey .
(a) What is the slope of the level set at x = 2, y = 0?
(b) In what direction should one move from the point (2, 0) in order to
increase f most quickly? Express your answer as a vector of length 1.
(c) Suppose at (2, 0) the variable x is changed to 2.5. Estimate the
corresponding change of y = 0 which substitutes this change of x, that is
the output remains the same.
Solution.
f
f
(x, y) = 2xey ,
(x, y) = x2 ey ,
x
y
thus
f
f
(2, 0) = 4,
(2, 0) = 4.
x
y
(a) So the slope of the level curve at (2, 0) is 1.
(b) The function f increases most rapidly in the direction of gradient
f (2, 0) = (4, 4).
The suitable vector of the length 1 is ( 12 , 12 ).
(c) f (2 + x, 0 + y) f (2, 0) f
(2, 0) x +
x
= 4 0.5 + 4 y = 0, y = 0.5.
The same using MRS:

f
(2, 0)
y

fx (2, 0)
4
y
M RS =
= = 1,
x
fy (2, 0)
4
thus y = M RS x = 1 x = 0.5
10

1.7

System of Implicit Functions

First warming up. Start with a system of linear equations

a11 x1

a x
21 1

...

am1 x1

+
+
...
+

...
...
...
...

+ a1n xn
+ a2n xn
...
...
+ amn xn

+ b11 y1
+ b21 y1
...
...
+ bm1 y1

+
+
...
+

...
...
...
...

+ b1m ym
+ b2m ym
...
...
+ bmm ym

=
=
...
=

c1
c2
...
cm

Is it possible to express the (endogenous) variables y1 , y2 , ... , ym in terms of

(exogenous) variables x1 , x2 , ... xn ? Answer is yes if

b11

det ...
bm1

...
...
...

b1m
...
6 0.
=
bmm

Now turn to general problem. Suppose m functions Fi (x1 , ..., xn , y1 , ..., ym ), i =

1, ..., m, (i.e. Fi : Rn+m R, i = 1, ..., m) are given. We consider a system
of equations

F1 (x1 , ..., xn , y1 , ..., ym ) = c1

................
,
(1)

Fm (x1 , ..., xn , y1 , ..., ym ) = cm

and suppose a point (x1 , ..., xn , y1 , ..., ym

) Rn+m is a solution.
(a) Does there exist functions y1 (x1 , ..., xn ), ... , ym (x1 , ..., xn ) in some
neighborhood of x = (x1 , ..., xn ) such that

Fi (x1 , ..., xn , y1 (x1 , ..., xn ), ..., ym (x1 , ..., xn )) ci , i = 1, ..., m,

and yi (x1 , ..., xn ) = yi , i = 1, ..., m?
(b) How to compute partial derivatives

yi
(x1 , ..., xn )?
xj

Theorem 4 If the determinant of Jacobian matrix

F1
y1

...
Fm
y1

F1
... y
m
... ...

Fm
... ym

) is nonzero, then there exist functions

evaluated at (x1 , ..., xn , y1 , ..., ym

y1 (x1 , ..., xn ), ... , ym (x1 , ..., xn )

defined on a ball about (x1 , ..., xn ) satisfying the conditions
Fi (x1 , ..., xn , y1 (x1 , ..., xn ), ..., ym (x1 , ..., xn )) ci , i = 1, ..., m,
and yi (x1 , ..., xn ) = yi , i = 1, ..., m.
11

yi
Furthermore, the derivatives x
can be solved from the system of linear
k
equations

F1
F1
y1
F1
...
y1
ym
x
xk
k

=
... ...
...
...
(2)

...

Fm
F1
ym
Fm
...
y1
ym
xk
xk

The solution of this system can be computed by

F1
y1

=
...

y1
xk

...

Fm
y1

ym
xk

1 F1
xk

...

F1
... y
m
... ...

F1
... y
m

Fm
xk

or by Cramers rule (Here all the matrices evaluated at (x1 , ..., xn , y1 , ..., ym
)).

Sketch of Proof. Let us differentiate our system (1) by xk :

F
1

xk

F2

+
+
...
+

xk

...

Fm
xk

y1
x
+
k
y1
xk +
...
...
y1
Fm
xk +
y1
F1
y1
F2
y1

...
...
...
...

m
y
= 0
xk
ym
xk = 0
...
... ...
Fm ym
xk = 0
ym

F1
ym
F2
ym

+
+
...
+

and this system is exactly (2).

Example. One solution of the system
x3 y z = 1
x + y2 + z3 = 6
is (x = 1, y = 2, z = 1). Estimate the corresponding x and y when z = 1.1.
Solution. Take F1 (x, y, z) = x3 y z, F2 (x, y, z) = x + y 2 + z 3 .
Evaluating the whole Jacobian at (x = 1, y = 2, z = 1) we obtain
F
1

x
F2
x

F1
y
F2
y

The determinant

F1
z
F2
z

F
1

x
F2

F1
y
F2
y

3x2 y x3 1
1
2y 3z 2

6 1 1
1 4 1

6 1

=
= 23 6= 0,

1 4

so the Implicit Function Theorem asserts the existence of a solution x(z) and
y(z) as functions of exogenous variable z.
Now calculate derivatives x0 (1) = xz (1) and y 0 (1) = yz (1):
F
1

x
F2
x

F1
y
F2
y

x0 (1)
y 0 (1)
12

1
F
z
F2
z

evaluating we obtain a system of linear equations

6 1
1 4

x0 (1)
y 0 (1)

1
3

7
and the solution gives x0 (1) = 23
, y 0 (1) = 19
.
23
Now we are ready to estimate x(1.1) and y(1.1) by linear approximation:

x(1.1) x(1) + x0 (1) 0.1 = 1 +

7
0.1 = 1.03,
23

19
0.1 = 1.91.
23
Finally we obtain the estimation of new solution (x = 1.03, y = 1.91, z = 1.1).
y(1.1) y(1) + y 0 (1) 0.1 = 2 +

Example. For the system

xz 3 + y 2 v 4 = 2,

xz + yvz 2 = 2

(x = 1, y = 1, z = 1, v = 1) is a solution.
(a) Can you estimate a new solution which correspond to y = 1.1 and
v = 1.2?
(b) Can you estimate a new solution which correspond to x = 1.1 and
y = 1.2?
Solution. Take F1 (x, y, z) = xz 3 + y 2 v 4 , F2 (x, y, z) = xz + yvz 2 .
Evaluating the whole Jacobian at (x = 1, y = 1, z = 1, v = 1) we obtain
F
1

x
F2
x

F1
y
F2
y

F1
z
F2
z

F1
v
F2
v

=
(1,1,1,1)

z 3 2yv 4
3xz 2
4y 2 v 3
2
z vz
x + 2yvz yz 2

1 2 3 4
1 1 3 1

=
(1,1,1,1)

(a) The determinant

F1
x
F2

F1
z
F2
z

(1,1,1,1)

1 3
=
1 3

= 0,

so the Implicit Function Theorem does not alow to express x and z as functions of Y and v.
(b) The determinant

F1
z
F2

F1
v
F2
v

(1,1,1,1)

3 4

=
3 1

= 9 6= 0,

so the Implicit Function Theorem asserts the existence of solutions z(x, y)

and v(x, y) as functions of exogenous variables x and y.
Now we calculate derivatives zx , zy , vx , vy at (1, 1, 1, 1).
13

For zx and vx we have the system

F1
z
F2
z

F1
v
F2
v

zx (1, 1)
vx (1, 1)

1
F
x
2
F
x

evaluating we obtain a system of linear equations

3 4
3 1

zx (1, 1)
vx (1, 1)

1
1

, vx (1, 1) = 0.
and the solution gives zx (1, 1) = 1
3
Similarly, for zy and vy we have the system

F1
z
F2
z

F1
v
F2
v

zy (1, 1)
vy (1, 1)

1
F
y
F2
y

evaluating we obtain a system of linear equations

3 4
3 1

and the solution gives zy (1, 1) =

zy (1, 1)
vy (1, 1)
2
,
9

2
1

vx (1, 1) = 1.

14

Inverse Function*

Suppose f : X Y is a function from X to Y . This function is called

invertible if there exists the inverse function g : Y X such that
g(f (x)) = x

and

f (g(y)) = y

for each x X and y Y . In other words the composition g f coincides

with the identity map idX : X X and the composition f g coincides with
the identity map idY : Y Y .
Examples
1. The map f : R R given by f (x) = 2x + 4 is invertible and its
inverse is the map g : R R given by g(y) = 0.5y 2. Indeed,
g(f (x)) = g(2x + 4) = 0.5(2x + 4) 2 = x,
and
f (g(y)) = f (0.5y 2) = 2(0.5y 2) + 4 = y.
2. The map f : R R+ = (0, +) given by f (x) = ex is invertible and
its inverse is the map g : R+ R given by g(y) = ln y. Indeed,
g(f (x)) = g(ex ) = ln ex = x, and f (g(y)) = f (ln y) = eln

= y.

A map f : X Y is called surjective if for each y Y there exists x X

s.t. f (x) = y.
A map f : X Y is called injective if for x1 6= x2 we have f (x1 ) 6= f (x2 ).
A map f : X Y is called bijective if it is surjective and injective
simultaneously.
Let us interpret these notions in terms of equation
f (x) = y
where x is considered as an unknown.
A map f : X Y is surjective iff the equation f (x) = y has a solution
for each y Y .
A map f : X Y is injective iff the equation f (x) = y has either no
solution or unique solution for each y Y .
A map f : X Y is bijective iff the equation f (x) = y has unique
solution for each y Y .
Theorem 5 A map f : X Y is invertible if and only if it is a bijection.

15

2.1

Invertible Linear maps

A linear map F : Rn Rn in fact is given by a matrix

A=

a11 a12
a21 a22
...
...
am1 am2

... a1n
... a2n
... ...
... amn

and F (v) = A v for each vector v Rn .

Theorem 6 A linear map F : Rn Rn is invertible if and only if its
matrix A is nondegenerate, that is det A 6= 0. In this case the inverse map
G : Rn Rn is given by G(w) = A1 w.

2.2

Inverse Function Theorem

Suppose now F : Rn Rn is function. As we know such a function is a

collection of functions
f1 : R n R
f2 : R n R
...
fn : Rn R,
so that
F (x1 , ..., xn ) = (f1 (x1 , ..., xn ), f2 (x1 , ..., xn ), ..., fn (x1 , ..., xn )).
The matrix which consists of partial derivatives of these functions
f1
x1

DF = ...
fn
x1

...
...
...

f1
xn

...

fn
xn

is called Jacobian of F . By DF (x ) is denoted the numerical matrix obtained

by evaluation of Jacobian DF at a vector x = (x1 , ..., xn ).
Theorem 7 Suppose F (x ) = y and the Jacobian DF (x ) is nondegenerate
matrix, then there exists an open ball Br (x ) about x and an open set U Rn
about y such that the restriction of the map F
F : Br (x ) U
is invertible (this is called locally invertible). Furthermore, the jacobian
of inverse map
G = F 1 : U Br (x )
is the inverse matrix of the Jacobian of F :
DG(y ) = (DF (x ))1 .
16

Example. Consider the function F : R2 R2 given by

F (x, y) = (x2 y 2 , 2xy).
It Jacobian looks as

DF =

2x 2y
2y 2x

and its determinant is detDF (x, y) = 4(x2 + y 2 ). By the Inverse Function

Theorem, F is locally invertible at every point except (0, 0).
Example. Show that the map F (x, y) = (x + ey , y + ex ) is everywhere
locally invertible.
Solution. The Jacobian looks as

DF =

1
ey
ex 1

so its determinant is 1 + eyx . This expression is nonzero at any (x, y).

17

Exercises
1. Consider the equation x3 + 3y 2 + 4xz 2 3z 2 y = 1. Does this equation
define z as a function of x and y:
(a) In a neighborhood of x = 1, y = 1?
(b) In a neighborhood of x = 1, y = 0?
(c) In a neighborhood of x = 0.5, y = 0?
z
z
If so, compute x
and y
at that point.
2. Consider the function F (x1 , x2 , y) = x21 x22 + y 3 .
(a) If x1 = 6 and x2 = 3, find a y which satisfies F (x1 , x2 , y) = 0.
(b) Does this equation define y as an implicit function of x1 and x2 near
x1 = 6, x2 = 3?
y
y
(c) If so, compute x
(6, 3) and x
(6, 3)
1
2
(d) If x1 increases to 6.2 and x2 decreases to 2.9, estimate the corresponding change of y.

3. Show that if for functions f (x, y) and g(x, y) one has fx = gy and
fy = gx , then level curves of f and g intersect orthogonally.
4. One solution of the system
2x2 + 3xyz 4uv = 16,
x + y + 3z + u v = 10
is x = 1, y = 2, z = 3, u = 0, v = 1. If one varies u and v near their original
values and plugs these new values into this system, can one find unique values
of x, y and z that still satisfy this system? Explain.

5. Check that x = 1, y = 4, u = 1, v = 1 is a solution of the system

y 2 + 2u2 + v 2 xy = 15, 2y 2 + u2 + v 2 + xy = 38.
If y increases to 4.02 and x stays fixed, does there exist a (u, v) near (1, 1)
which solves this system? If not, why not? If yes, estimate the new u and v.

6. The economy of Northern Saskatchewan is in equilibrium when the

system of equations

2xz + xy + z 2 z = 11, xyz = 6

is satisfied. One solution of this set of equations is x = 3, y = 2, z = 1,
and Northern Saskatchewan is in equilibrium at this point. Suppose that the
18

prime minister discovers that the variable z (output of beaver pelts) can be
controlled by simple decree.
a) If the prime minister raises z to 1.1, use calculus to estimate the change
in x and y.
b) If x were in the control of the prime minister and not y or z, explain
why you cannot use this method to estimate the effect of reducing x from 3
to 2.95.

7. Consider the system of equations

x + 2y + z = 5,

3x2 yz = 12

as defining some endogenous variables in terms of some exogenous variables.

a) Divide the three variables into exogenous ones and endogenous ones in
a neighborhood of x = 2, y = 1, z = 1 so that the Implicit Function Theorem
applies.
b) If each of the exogenous variables in your answer to a) increases by 0.25,
use calculus to estimate how each of the endogenous variables will change.

8. Consider the system of two equations in three unknowns: x + 2y + z =

5, 3x2 yz = 12.
a) At the point x = 2, Y = 1, z = 1, why can we treat z as an exogenous
variable and x and y are the dependent variables?
b) If z rises to 1.2, use calculus to estimate the corresponding x and y.
Exercises 15.1-15.25 from [SB].

Homework
Exercise 15.6 from [SB], Exercise 15.9 from [SB], Exercise 15.13 from [SB],
Exercise 15.22 from [SB], Exercise 15.24 from [SB].

19

Short Summary
Implicit Function
Implicit Function Theorem in R1 If G(x , y ) = c and G
(x , y ) 6= 0,
y
then y = y(x) on (x , x + ) s.t. G(x, y(x)) c, y(x ) = y and
G

(x ,y )

x
y 0 (x ) = G
.
(x ,y )
y

Implicit Function Theorem for Rn . If G(x1 , ..., xk , y ) = c and

G
(x , ..., xk , y ) 6= 0
y 1
then

y = y(x1 , ..., xn ) on B (x ) s.t.

y(x1 , ..., xk ) = y and

y
xi

G
(x1 ,...,xk ,y )
xi
G
(x1 ,...,xk ,y )
y

G(x1 , ..., xk , y(x1 , ..., xk )) c,

Implicit Function Theorem for System. If

F1 (x1 , ..., xn , y1 , ..., ym ) = c1

................
and

Fm (x1 , ..., xn , y1 , ..., ym

) = cm

F1
y1

...

Fm
y1

F1
... y
m
... ...
m
... F
ym

6= 0,
)
(x1 ,...,xn ,y1 ,...,ym

then y1 (x1 , ..., xn ), ... , ym (x1 , ..., xn ) on B (x , Y ) s.t. for all i = 1, ..., m
Fi (x1 , ..., xn , y1 (x1 , ..., xn ), ..., ym (x1 , ..., xn )) ci , yi (x1 , ..., xn ) = yi
and

yi
xk

can be solved from

F1
y1

...
Fm
y1

y1
F1
F1
... y
xk
xk
m

= ...
...
... ...
.

F1
ym
Fm
... y
xk
xk
m

Regular point of G(x, y): DG(x , y ) 6= (0, 0).

Tangent vector of the level curve G(x, y) = c at a regular point
(x , y ): (Gy (x , y ), Gx (x , y )).

At a regular point gradient is orthogonal to level curve.

G

(x ,y )

x
Marginal Rate of Substitution y M RS x = G
x.
(x ,y )
y

20