Mathematics for Economics Anthony Tay

20. Multivariable Linear Approximations, and Differentials

The concepts of differentials and linear approximations can be extended to functions of many variables.

20.1 Linear Approximations The linear approximation to a function z = f ( x, y ) at point

( x, y ) = ( x0 , y0 ) is the tangent plane to the function at that point. This tangent plane is the plane that

passes through the point ( x0 , y0 , z0 ) , where z0 = f ( x0 , y0 ) , and has the same derivatives there as the

function in every direction (and in particular the x- and y- directions).

The equation of a plane is, in general,

p ( x, y ) = α + β x + δ y

We want the tangent to satisfy

p ( x0 , y0 ) = f ( x0 , y0 ) ,

p′x ( x0 , y0 ) = f x′ ( x0 , y0 ) ,

p′y ( x0 , y0 ) = f y′ ( x0 , y0 ) .

Therefore
p′x= β= f x′ ( x0 , y0 ) and p′y= δ= f y′ ( x0 , y0 ) .

α + f x′ ( x0 , y0 ) x0 + f y′ ( x0 , y0 ) y0 =
Also, p ( x0 , y0 ) = f ( x0 , y0 ) , which gives

α =f ( x0 , y0 ) − f x′ ( x0 , y0 ) x0 − f y′ ( x0 , y0 ) y0

Substituting all this into p ( x, y ) above gives

p ( x=
, y) f ( x0 , y0 ) + f x′ ( x0 , y0 )( x − x0 ) + f y′ ( x0 , y0 )( y − y0 )

We use this function as the linear approximation to f ( x, y ) at the point ( x0 , y0 ) . That is to say, for all

( x, y ) near (x0, y0), we make the approximation

f ( x, y ) ≈ p ( x=
, y) f ( x0 , y0 ) + f x′ ( x0 , y0 )( x − x0 ) + f y′ ( x0 , y0 )( y − y0 ) .

20-1
Mathematics for Economics
Example 20.1.1 Find the linear approx. for f ( x, y )= 1 + x + y at the point ( x, y ) = (0,0) . We

have f (0,0) = 1 , and as

1
=f x′ = f y′ ,
2 1+ x + y

so that f=′
x (0,0) f=′
y (0,0) 1/ 2 . The linear approximation to the function at (0,0) is thus

1 1 1
1
z =+ ( x − 0) + ( y − 0) =+
1 ( x + y) .
2 2 2
The function (curved surface) and the linear approximation (tangent plane) is shown below.

Remarks: The linear approximation to the function f ( x, y ) at the point ( x, y ) = ( x0 , y0 ) can be written

in matrix notation as

f ( x, y ) ≈ p ( x=
, y) f ( x0 , y0 ) + f x′ ( x0 , y0 )( x − x0 ) + f y′ ( x0 , y0 )( y − y0 )
= f ( x0 , y0 ) +  f x′ ( x0 , y0 ) f y′ ( x0 , y0 )   x − x0 
y − y 
 0

= f ( x0 , y0 ) + f0′ (x − x0 )

 f x′   x x 
where f0′ =   evaluated at the point ( x0 , y0 ) , x =   , and x 0 =  0  . We can generalize to
f
 y ′  y  y0 
functions of more variables in the obvious way.

The quadratic approximation is given (in matrix notation) by

1
f ( x, y ) ≈ q ( x=
, y) f ( x0 , y0 ) + f0′ (x − x0 ) + ( x − x 0 )′ H 0 ( x − x 0 )
2
 f xx′′ f xy′′ 
where H 0 is the hessian matrix  evaluated at ( x0 , y0 ) . (Proof omitted)
 f yx′′ f yy′′ 

20-2
Mathematics for Economics
20.2 Differentials Recall that linear approximations can be expressed in terms of changes. Let

y = f ( x) be the function to be approximated, and consider a small change in x from x = x0 to

x x0 + dx . The actual change in f is f ( x0 + dx) − f ( x0 ) . The linear approximation to this change is

=
f ′( x0 )dx .
p ( x0 + dx) − p ( x0 ) =

We denote the linear approximation to the actual change as “ df ”, i.e.,

df = p ( x0 + dx) − p ( x0 ) = f ′( x0 ) dx .

For arbitrary x , we have df = f ′( x) dx . The quantities ‘ df ’ and ‘ dx ’ are called the differentials of f

and x respectively. If we write the function as y = f ( x) , we can write dy = f ′( x)dx .

For functions of two variables, taking the linear approximation of f ( x + dx, y + dy ) at any point

( x, y ) gives
f ( x + dx, y + dy ) ≈ f ( x, y ) + f x′ ( x + dx − x) + f y′ ( y + dy − y )

= f ( x, y ) + f x′dx + f y′dy
Thus,
f ( x + dx, y + dy ) − f ( x, y ) ≈ f x′ ( x + dx − x) + f y′ ( y + dy − y )

= f x′dx + f y′dy

We call f x′dx + f y′dy the total differential of f and give it the symbol df (or dz if the function is

written z = f ( x, y ) ). That is,

dz
= f x′dx + f y′dy

This formula can be used to approximate changes in z as a result of small changes in x and y by the

amounts dx and dy respectively.

NOTE again the conceptual difference between the (partial) derivatives and the (total)

differential. The partial derivatives are the slopes of the function in particular directions. The differential

is made up of four items, the two partials and the two quantities dx and dy . It is a formula for taking

the linear approximation to the change in the function value when x and y are increased by dx and

dy respectively.

20-3
Mathematics for Economics
Example 20.2.1 Earlier, we considered the function z = f ( x, y )= 1 + x + y , and found that
1
=f x′ = f y′
2 1+ x + y
Therefore, the differential is
dx dy 1
dz = + = (dx + dy ) .
2 1+ x + y 2 1+ x + y 2 1+ x + y

At the point ( x, y ) = (0,0) , we have f=′

x (0,0)
′
y (0,0) 1/ 2 , so
f=
1 1
dz
= dx + dy
2 2

′ ′ 1 1
At the point ( x, y ) = (1, 2) , we have f=
x (1, 2) y (1, 2)
f= = , so
2 1+1+ 2 4

1 1
dz
= dx + dy
4 4

As an illustration, consider a change in ( x, y ) from (0,0) to (0.1,0.1) , i.e., dx = 0.1 . At (0,0) ,

= dy

( x, y ) f=
f= (0,0) 1 . Because

f (0.1,0.1) = 1 + 0.1 + 0.1 = 1.0954 (4 dec pl.),

the actual change in z is 0.0954. Using the differential formula, at (0,0) , we have
dx dy 0.1 0.1
dz = + = + = 0.1
2 1+ x + y 2 1+ x + y 2 1+ 0 + 0 2 1+ 0 + 0

which is our linear approximation to the actual change. The error is 0.0046.

For functions of n-variables, all these ideas remain valid: if z = f ( x1 , x2 ,..., xn ) , then

dz = df = f1′dx1 + f 2′dx2 + ... + f n′dxn .

Example 20.2.2 Determine the total differential for z = x y2 + x3. We have

∂z ∂z
= y 2 + 3 x 2 , and = 2 xy .
∂x ∂y

Therefore dz =( y 2 + 3 x 2 ) dx + 2 xy dy .

20-4
Mathematics for Economics
20.3 Rules for Differentials Although differentials and derivatives are different concepts, they

contain the same information, and we can in fact differentiate expressions using the language of

differentials. That is, we can use “rules for differentials” and differentiate expressions using these rules.
e.g. If z = xy , then=
dz y dx + x dy .

∂z ∂z
Proof: ∂z / ∂x =y , ∂z / ∂y =x . Since=
dz dx + dy , we have=
dz y dx + x dy .
∂x ∂y

Many of you are in fact familiar with this, and other similar expressions, and some of you were taught

to memorize rules for differentiation using these expressions. (Now you know that these expressions

are not simply mnemonics, i.e. memorization tricks, but have real mathematical meaning.)

For every differentiation rule, we can derive (and use) a corresponding “differential” rule.

z ax + by where a and b are constants, then=

1. If = dz a dx + b dy ;

2. If z = xy , then=
dz y dx + x dy ;

x y dx − x dy
3. If z = , then dz = ;
y y2

4. If z = x r , then dz = rx r −1dx ;

dx
5. If z = ln x , then dz = ;
x

These rules can be used in conjunction with each other:

 dx  y
e.g. If z = y ln x , then dz =y d (ln x) + dy ln x =y   + dy ln x = dx + ln x dy .
 x  x

You can show this by computing the partial derivatives directly, and then constructing the differential

from its definition:

∂z y ∂z ∂z ∂z y
since = , = ln x , and=
dz dx + dy , we have=
dz dx + ln x dy .
∂x x ∂y ∂x ∂y x

20-5
Mathematics for Economics
Example 20.3.1 Find dz in terms of dx and dy if=z xy 2 + x3 .

Using rules for differentials gives

=dz d ( xy 2 ) + d ( x3 )
= xd ( y 2 ) + y 2 dx + 3 x 2 dx
.
= 2 xydy + y 2 dx + 3 x 2 dx
= (3 x 2 + y 2 )dx + 2 xydy

In simple examples, it is probably easier to simply take the partial derivatives, but there are times when

it is ‘cleaner’ to “take differentials”.

The convenience of “taking differentials” really comes into play with the chain rule:

Suppose z = f ( y ) , and y = g ( x) . This means that z = f ( g ( x)) , so the differential (wrt x ) is

dz = [ f ( g ( x))]′ dx . The chain rule gives us [ f ( g ( x))]′ = f ′( g ( x)) g ′( x) , so we have

dz = f ′( g ( x)) g ′( x)dx .

We can obtain this expression using differentials. From z = f ( y ) , we have dz = f ′( y ) dy , and

y = g ( x) gives us dy = g ′( x) dx . Simply substituting the latter into the former gives

= ′( y ) g ′( x)dx f ′( g ( x)) g ′( x)dx .
dz f=

The usefulness is in the fact that the expression dz = f ′( y ) dy is valid regardless of whether y is a

function of any other variable.

E.g. If z = xy 2 . Then dz = x d ( y 2 ) + y 2 dx = y 2 dx + 2 xy dy .

If x = s ln t and y= s + t 2 , then to find dz , simply take

s
= dx ln t ds + dt and = ds + 2t dt
dy
t
and substitute into the previous expression to find dz with respect to ds and dt
s
dz =y 2 dx + 2 xy dy =( s + t 2 ) 2 [ln t ds + dt ] + 2( s ln t )( s + t 2 )[ds + 2tdt ]
t
s
=( s + t 2 )[( s + t 2 ) ln t + 2 s ln t ]ds + ( s + t 2 )[( s + t 2 ) + 4 st ln t ] dt
t
In fact, from here you can read off the partial derivatives:
∂z ∂z s
=( s + t 2 )[( s + t 2 ) ln t + 2 s ln t ] and =( s + t 2 )[( s + t 2 ) + 4 st ln t ] .
∂s ∂t t

20-6
Mathematics for Economics
We can apply all these ideas to systems of equations.

Example 20.3.2 Suppose

u 2v − u = x3 + 2 y 3

and eux = vy

where u and v are endogenous, and x and y are exogenous. In other words, u and v are functions

of x and y . You can easily show that the point ( x, y, u , v) = (0,1, 2,1) satisfies both equations. What are

the partial derivatives of u and v with respect to x and to y ?

Taking differentials, we have

2u du v + u 2 dv − du= 3 x 2 dx + 2(3 y 2 dy )
eux (udx + xdu ) =vdy + ydv

Rewriting gives

(v 2u − 1) du + (u 2 ) =
dv 3 x 2 dx + 6 y 2 dy
xeux du − y dv =− ueux dx + v dy

In principle, we can solve these two equations in two unknowns (the unknowns are du and dv ) in

terms of dx , dy , u , v , x and y . The expressions will be, in this case, rather complicated (perhaps

best expressed, and left, in matrix notation.

Instead of obtaining the general solution, which we are not looking for, we apply this to the

point ( x, y, u , v) = (0,1, 2,1) . We get

3 du + 4 dv = 0 dx + 6 dy
0 du − 1 dv =− 2 dx + 1 dy

So =
dv 2dx − dy and
du = (1 / 3)[−4dv + 6dy ] = (1 / 3)[−4(2dx − dy ) + 6dy ] = (1 / 3)[−8dx + 10dy ]

In other words, at the point ( x, y, u , v) = (0,1, 2,1) ,

∂v / ∂x =2 , ∂v / ∂y =−1 , ∂u / ∂x =−8 / 3 , and ∂u / ∂y =10 / 3 .

If we are interested in the general expression for the derivatives, but are interested only in differentiating

with respect to x , we can simplify the general problem by setting dy = 0 . For example, in the general

case, we would simply solve

(v 2u − 1) du + (u 2 ) dv =3 x 2 dx
xeux du − y dv − ueux dx
=

20-7
Mathematics for Economics
Example 20.3.3 Consider the following macroeconomic model
C = c(Y , r ) , L = l (Y , r ) , IV = h(Y , r )
Y = C + I +G, =
I IV + I B , M = L

where f , g , and h are functions. The endogenous variables are Y (national income), r (interest

rates), L (money demand), IV (private investment), C (consumption), and I (total investment). The

exogenous variables are M (money supply), I B (public investment), G (public consumption)

If I B and M are held fixed, what is the effect of an increase in G on the endogenous variables?

Differentiating gives
dC cY′ dY + cr′ dr ,
= dL lY′ dY + lr′ dr ,
= dIV hY′ dY + hr′ dr ,
=
dY = dC + dI + dG , dI dIV + dI B ,
= dM = dL .

Substituting, gives
dY = [cY′ dY + cr′ dr ] + [(hY′ dY + hr′ dr ) + dI B ] + dG
= (cY′ + hY′ )dY + (cr′ + hr′ )dr + dI B + dG

dM
= lY′ dY + lr′ dr

Because we are interested in an increase in G only, and are holding I B and M fixed, let us set

dI
= B = 0 . Then the two equations become (with some rewriting)
dM
[1 − (cY′ + hY′ )]dY − (cr′ + hr′ )dr =
dG and lY′ dY + lr′ dr =
0

The second equation gives

lY′
dr = − dY ,
lr′

and subs. into the first gives

dG , so
[1 − (cY′ + hY′ )]dY + (cr′ + hr′ )(lY′ / lr′ ) dY =

dG , i.e.,
[1 − (cY′ + hY′ ) + (cr′ + hr′ )(lY′ / lr′ )]dY =

 lr′ 
dY =   dG
 lr′ (1 − cY′ − hY′ ) + lY′ (cr′ + hr′ ) 
 lY′ 
Subs into dr = −(lY′ / lr′ ) dY gives dr = −   dG .
 lr′ (1 − cY′ − hY′ ) + lY′ (cr′ + hr′ ) 

The other effects can be calculated from

dI dI
= =V hY′ dY + hr′ dr ,=
dL lY′ dY + lr′ dr , and=
dC cY′ dY + cr′ dr .

20-8
Mathematics for Economics
 Exercises

1. Suppose Y = AK α Lβ . Differentiate (using differentials) to show that

dY dK dL
= α +β .
Y K L

2. Assume that the equation system

x 2 + sxy + y 2 − 1 =0
x2 + y 2 − s2 + 3 =0
defines x and y implicitly as differentiable functions of s .

dx dy
(a) Differentiate the system (using differentials) and find the values of and when x = 0 ,
ds ds
y = 1, s = 2 .
(b) Find an approximate value of the change in x if s increases from 2 to 2.1 .

3. The system of equations

ln( x + u ) + uv − y 2 ev + y =
0
u 2 − xv =
v
defines u and v as differentiable functions of x and y around the point P ( x, y, u=
, v) (2,1, −1, 0) .
(i) Differentiate the system (using differentials);

(ii) Find the values of the partial derivatives u x′ , u ′y , vx′ , v′y ;

(iii) Find an approximate value of u (1.99,1.02) , i.e., find the approximate value of u at the point

( x, y ) = (1.99,1.02) .

4. Let C f (Y − T , r ) , and I = h(r ) .

Y= C + I , =

(a) Differentiate all three equations to obtain three equations relating dY , dC , dI , dT , and dr .

(b) Solve the three equations to obtain expressions for dY , dC , and dI , each in terms of dT and

dr only.

(c) Write down expressions for ∂Y / ∂T , ∂C / ∂T , ∂I / ∂T , ∂Y / ∂r , ∂C / ∂r , and ∂I / ∂r .

20-9
Mathematics for Economics