Partial Derivatives

1 Functions of two or more variables

In many situations a quantity (variable) of interest depends on two or

more other quantities (variables), e.g.

Figure 1: b is the base length of the triangle, h is the height of the triangle, H is the height of
the cylinder.

The area of the triangle and the base of the cylinder: A = 12 bh

The volume of the cylinder: V = AH = 12 bhH

The arithmetic average x̄ of n real numbers x1, . . . , xn

1
x̄ = (x1 + x2 + · · · + xn)
n
We say
A is a function of the two variables b and h.
V is a function of the three variables b, h and H.
x̄ is a function of the n variables x1, ..., xn.
1
The expression z = f (x, y) means that z is a function of x and y;
w = f (x, y, z); u = f (x1, x2, . . . , xn).
y z

(x,y,z)
(x,y) f f

x y w
x z

Figure 2: A function f assigns a unique number z = f (x, y), or w =

f (x, y, z) to a point in (x, y)-plane or (x, y, z)-space.

The independent variables of a function may be restricted to lie in

some set D which we call the domain of f , and denote D(f ). The
natural domain consists of all points for which a function defined
by a formula gives a real number.

Definition. A function f of two variables, x and y, is a rule that

assigns a unique real number f (x, y) to each point (x, y) in some set
D in the xy-plane.
A function f of n variables, x1, ..., xn, is a rule that assigns a unique
real number f (x1, ..., xn) to each point (x1, ..., xn) in some set D in
the n-dimensional x1...xn-space, denoted Rn.

Definition. The graph of a
function z = f (x, y) in xyz-space is a

set of points P = x, y, f (x, y) where (x, y) belong to D(f ).

In general such a graph is a surface in 3-space.

2
Examples. Find the natural domain of f , identify the graph of f as
a surface in 3-space and sketch it.

1. f (x, y) = 0;

2. f (x, y) = 1;

3. f (x, y) = x;

4. f (x, y) = ax + by + c;

5. f (x, y) = x2 + y 2;

p
6. f (x, y) = 1 − x2 − y 2;

p
7. f (x, y) = 1 + x2 + y 2;

p
8. f (x, y) = x2 + y 2 − 1;

p
9. f (x, y) = − x2 + y 2;

3
2 Level curves

If z = f (x, y) is cut by z = k, then at all points on the intersection we

have f (x, y) = k.
This defines a curve in the xy-plane which is the projection of the in-
tersection onto the xy-plane, and is called the level curve of height
k or the level curve with constant k.
A set of level curves for z = f (x, y) is called a contour plot or
contour map of f .

4
Examples.

1. f (x, y) = ax + by + c;
2. f (x, y) = x2 + y 2;
p
3. f (x, y) = 1 − x2 − y 2;
p
4. f (x, y) = 1 + x2 + y 2;
p
5. f (x, y) = x2 + y 2 − 1;
p
6. f (x, y) = − x2 + y 2;

7. f (x, y) = y 2 − x2. It is the hyperbolic paraboloid (saddle surface).

Figure 3: The hyperbolic paraboloid and its contour map.

5
There is no “direct” way to graph a function of three variables. The
graph would be a curved 3-dimensional space ( a 3-dim manifold if it
is smooth), in 4-space. But f (x, y, z) = k defines a surface in 3-space
which we call the level surface with constant k.

Examples.
1. f (x, y, z) = x2 + y 2 + z 2;

2. f (x, y, z) = z 2 − x2 − y 2;

Figure 4: Level surfaces of f (x, y, z) = z 2 − x2 − y 2

6
3 Limits and Continuity

1.0

0.8
f(x)
0.6

0.4

0.2

a x
0.5 1.0 1.5 2.0

There are two one-sided limits for y = f (x).

y

1.5 C5

1.0 C6 (x,y) C4

0.5 C3
(a,b)
C1
x
0.5 1.0 1.5 2.0

C2

For z = f (x, y) there are infinitely many curves along which one can
approach (a, b).
This leads to the notion of the limit of f (x, y) along a curve C.
If all these limits coincide then f (x, y) has a limit at (a, b), and the
limit is equal to f (a, b) then f is continuous at (a, b).
7
4 Partial Derivatives

Recall that for a function f (x) of a single variable the derivative of f

at x = a
f (a + h) − f (a)
f 0(a) = lim
h→0 h
is the instantaneous rate of change of f at a, and is equal to the slope
of the tangent line to the graph of f (x) at (a, f (a)).
y

f(x)

(a,f(a))

x
a

Figure 5: Equation of the tangent line: y = f (a) + f 0(a)(x − a).

Consider f (x, y). If we fix y = b where b is a number from the domain

of f then f (x, b) is a function of a single variable x and we can calculate
its derivative at some x = a. This derivative is called the partial
derivative of f (x, y) with respect to x at (a, b) and is denoted by
∂f (a, b)
fx(a, b) or by
∂x
∂f (a, b) d 
 f (a + h, b) − f (a, b)
fx(a, b) = = f (x, b)  = lim
∂x dx x=a h→0 h

∂x ∂y
If f (x, y) = x then = 1 , and if f (x, y) = y then =0
∂x ∂x
8
Geometrically, given the surface z = f (x, y), we consider its intersec-
tion with the plane y = b which is a curve. This curve is the graph of
the function f (x, b), and therefore the partial derivative fx(a, b) is the
slope of the tangent line to the curve at (a, b, f (a, b))

Equation of the tangent line: x = t, y = b, z = f (a, b)+fx(a, b)(t−a)

We call fx(a, b) the slope of the surface in the x-direction at

(a, b)

9
Similarly, if we fix x = a where a is a number from the domain of f
then f (a, y) is a function of a single variable y and we can calculate
its derivative at some y = b. This derivative is called the partial
derivative of f (x, y) with respect to y at (a, b) and is denoted by
∂f (a, b)
fy (a, b) or by
∂y
∂f (a, b) d 
 f (a, b + h) − f (a, b)
fy (a, b) = = f (a, y)  = lim
∂y dy y=b h→0 h

∂x ∂y
If f (x, y) = x then = 0 , and if f (x, y) = y then =1
∂y ∂y
The intersection of the surface z = f (x, y) with the plane x = a is
a curve which is the graph of the function f (a, y), and therefore the
partial derivative fy (a, b) is the slope of the tangent line to the curve
at (a, b, f (a, b))

Equation of the tangent line: x = a, y = t, z = f (a, b)+fy (a, b)(t−a)

We call fy (a, b) the slope of the surface in the y-direction at
(a, b)
10
 If we allow (a, b) to vary, the partial derivatives become functions of
two variables:

a → x , b → y and fx(a, b) → fx(x, y), fy (a, b) → fy (x, y)

f (x + h, y) − f (x, y) f (x, y + h) − f (x, y)

fx(x, y) = lim , fy (x, y) = lim
h→0 h h→0 h

Partial derivative notation: if z = f (x, y) then

∂f ∂z ∂f ∂z
fx = = = ∂x f = ∂x z , fy = = = ∂y f = ∂y z
∂x ∂x ∂y ∂y

Example.
p
3
2x2 − 3xy 2 + 3 cos(2x + 3y) − 3y 3 + 18
z = f (x, y) = ln
2
Find fx(x, y), fy (x, y), f (3, −2), fx(3, −2), fy (3, −2)

For w = f (x, y, z) there are three partial derivatives fx(x, y, z), fy (x, y, z),
fz (x, y, z)

Example.
p
f (x, y, z) = z 2 + y − x + 2 cos(3x − 2y)
Find
fx(x, y, z), fy (x, y, z), fz (x, y, z),
f (2, 3, −1), fx(2, 3, −1), fy (2, 3, −1), fz (2, 3, −1)

11
In general, for w = f (x1, x2, . . . , xn) there are n partial derivatives:
∂w ∂w ∂w
, , ... ,
∂x1 ∂x2 ∂xn

Example. q
r= x21 + x22 + · · · + x2n
Find
∂r ∂r ∂r ∂r ∂r
, , , , , n ≥ 9, i ≤ n
∂x1 ∂x2 ∂x9 ∂xi ∂xn−1

Second-order derivatives: fxx, fxy , fyx, fyy

fxx
%
fx → fxy
%
f
&
fy → fyx
&
fyy

Notation
2 2
   
∂ f ∂ ∂f ∂ f ∂ ∂f
fxx = = , fxy = =
∂x2 ∂x ∂x ∂y∂x ∂y ∂x
∂ 2f ∂ 2f
   
∂ ∂f ∂ ∂f
fyx = = , fyy = 2 =
∂x∂y ∂x ∂y ∂y ∂y ∂y
fxy and fyx are called the mixed second-order partial
derivatives. fx and fy can be called first-order partial derivative.

12
Example.
π π
z = 2ey− 2 sin x − 3ex− 4 cos y
Find
∂z ∂z ∂ 2z ∂ 2z ∂ 2z ∂ 2z
, , , , , ,
∂x ∂y ∂x2 ∂x∂y ∂y 2 ∂y∂x
∂z π π ∂z π π ∂ 2z π π ∂ 2z π π
( , ), ( , ), ( , ), ( , )
∂x 4 2 ∂y 4 2 ∂x∂y 4 2 ∂y∂x 4 2

Equality of mixed partial derivatives

Theorem. Let f be a function of two variables. If fxy and fyx are

continuous on some open disc, then fxy = fyx on that disc.

Higher-order derivatives

Third-order, fourth-order, and higher-order derivatives are obtained by

successive differentiation.
3
 2  3
 2 
∂ f ∂ ∂ f ∂ f ∂ ∂ f
fxxx = 3 = , fxyy = =
∂x ∂x ∂x2 ∂y 2∂x ∂y ∂y∂x
∂ 4f ∂ 3f
 
∂
fxyxz = =
∂z∂x∂y∂x ∂z ∂x∂y∂x

For higher-order derivatives the equality of mixed partial derivatives

also holds if the derivatives are continuous.

In what follows we always assume that the order of partial derivatives

is irrelevant for functions of any number of independent variables.

13
5 Differentiability, differentials and local linearity

For f (x, y), the symbol ∆f , called the increment of f , denotes the
change
∆f = f (a + ∆x, b + ∆y) − f (a, b)
For small ∆x, ∆y
∆f ≈ fx(a, b)∆x + fy (a, b)∆y
Definition. A function f (x, y) is said to be differentiable at (a, b)
provided fx(a, b) and fy (a, b) both exist and

∆f − fx(a, b)∆x − fy (a, b)∆y

lim p =0
(∆x,∆y)→(0,0) (∆x)2 + (∆y)2

For f (x, y, z)
∆f = f (a + ∆x, b + ∆y, c + ∆z) − f (a, b, c)
For small ∆x, ∆y, ∆z
∆f ≈ fx(a, b, c)∆x + fy (a, b, c)∆y + fz (a, b, c)∆z
and f (x, y, z) is differentiable at (a, b, c) if

∆f − fx(a, b, c)∆x − fy (a, b, c)∆y − fz (a, b, c)∆z

lim p =0
(∆x,∆y,∆z)→(0,0,0) (∆x)2 + (∆y)2 + +(∆z)2

Theorem. If a function is differentiable at a point, then it is contin-

uous at that point.

Theorem. If all first-order derivatives of f exist and are continuous

at a point, then f is differentiable at a point.
14
 Differentials

If z = f (x, y) is differentiable at (a, b) we let

dz = fx(a, b)dx + fy (a, b)dy
denote a new function with dependent variable dz and independent
variables dx, dy. It is called the total differential of z (or f ) at
(a, b). It is a linear function of dx and dy.
Note that ∆z ≈ dz if ∆x = dx and ∆y = dy
If we allow (a, b) to vary, the differential becomes a function of four
variables, dx, dy, x, y:

a → x , b → y ⇒ dz = fx(x, y)dx + fy (x, y)dy

Definition. If f (x, y) is differentiable at (a, b) then

L(x, y) = f (a, b) + fx(a, b)(x − a) + fy (a, b)(y − b)
is called the local linear approximation of f at (a, b).
Its graph is the tangent plane to the surface z = f (x, y) at (a, b, f (a, b))
p
Example. f (x, y) = x2 + y 2. Compute f (3.04, 3.98), and esti-
mate the error if a calculator gives f (3.04, 3.98) ≈ 5.00819

If w = f (x, y, z), the total differential of w (or f ) at (a, b, c) is

dw = fx(a, b, c)dx + fy (a, b, c)dy + fz (a, b, c)dz
or if a → x , b → y , c → z
dw = fx(x, y, z)dx + fy (x, y, z)dy + fz (x, y, z)dz
The local linear approximation of f at (a, b, c) is
L(x, y, z) = f (a, b, c)+fx(a, b, c)(x−a)+fy (a, b, c)(y−b)+fz (a, b, c)(z−c)

15
6 The Chain Rule

Recall
dy dy dx
y = f (x(t)) ⇒ =
dt dx dt
because
dy dx
∆y ≈ ∆x , ∆x ≈ ∆t
dx dt

Let z = f (x, y) and x = x(t), y = y(t). Then z = f (x(t), y(t)) is a

function of the single variable t.
∂z ∂z dx dy
∆z ≈ ∆x + ∆y , ∆x ≈ ∆t , ∆y ≈ ∆t
∂x ∂y dt dt
and therefore
dz ∂z dx ∂z dy
= +
dt ∂x dt ∂y dt

p
Example. z = 4 − x2 − y 2, x = 1 + cos t, y = sin t

16
Similarly, if w = f (x, y, z) and x = x(t), y = y(t), z = z(t). Then
w = f (x(t), y(t), z(t)) is a function of the single variable t, and
dw ∂w dx ∂w dy ∂w dz
= + +
dt ∂x dt ∂y dt ∂z dt
In general, if w = f (x1, x2, . . . , xn) and x1 = x1(t), x2 = x2(t), ... ,
xn = xn(t), then
n
dw ∂w dx1 ∂w dx2 ∂w dxn X ∂w dxi
= + + ··· =
dt ∂x1 dt ∂x2 dt ∂xn dt i=1
∂xi dt

Implicit differentiation

Let z = f (x, y) and y = y(x). Then

dz ∂f dx ∂f dy ∂f ∂f dy
= + = +
dx ∂x dx ∂y dx ∂x ∂y dx
dz
Suppose y(x) is such that f (x, y(x)) = const. Then, dx = 0 and
∂f ∂f dy dy fx
+ =0 ⇒ =− if fy 6= 0
∂x ∂y dx dx fy

Example. The lemniscate is defined by the equation

0.6
0.4
0.2
0.0
-0.2
-0.4
-0.6
2 2 2 2 2 2
(x + y ) = 2a (x − y ) -2 -1 0 1 2

Find dy/dx.

17
The chain rule for partial derivatives

1. Let y = f (x) and x = x(u, v)

Then y = f (x(u, v)) is a function of u and v, and
dy ∂x ∂x
∆y ≈ ∆x , ∆x ≈ ∆u + ∆v
dx ∂u ∂v
Thus,
∂y dy ∂x ∂y dy ∂x
= , =
∂u dx ∂u ∂v dx ∂v

2. Let z = f (x, y) and x = x(u, v), y = y(u, v)

Then x = f (x(u, v), y(u, v) is a function of u and v, and
∂z ∂z ∂x ∂x ∂y ∂y
∆z ≈ ∆x+ ∆y , ∆x ≈ ∆u+ ∆v , ∆y ≈ ∆u+ ∆v
∂x ∂y ∂u ∂v ∂u ∂v
Thus,
∂z ∂z ∂x ∂z ∂y ∂z ∂z ∂x ∂z ∂y
= + , = +
∂u ∂x ∂u ∂y ∂u ∂v ∂x ∂v ∂y ∂v

3. Let w = f (x, y, z) and x = x(u, v), y = y(u, v), z = z(u, v)

∂w ∂w ∂x ∂w ∂y ∂w ∂z ∂w ∂w ∂x ∂w ∂y ∂w ∂z
= + + , = + +
∂u ∂x ∂u ∂y ∂u ∂z ∂u ∂v ∂x ∂v ∂y ∂v ∂z ∂v

4. Let w = f (x1, ..., xn) and x1 = x1(u1, ..., um), ... , xn = xn(u1, ..., um)
n
∂w X ∂w ∂xi
= , α = 1, . . . , m
∂uα i=1
∂x i ∂u α

18
 ∂z ∂z
Example. Find ∂u and ∂v where
x
z = cos sin 2y ; x = 3u − 2v , y = u2 − 2v 3
2

Example. The wave equation: Consider a string of length L that is

stretched taut between x = 0 and x = L on an x-axis, and suppose
that the string is set into vibratory motion by “plucking” it at time
t = 0. The displacement of a point on the string depends both on x
and t: u(x, t). One-dimensional wave equation for small displacements

∂ 2u 2
2∂ u
−c =0
∂t2 ∂x2
Show that

u(x, t) = f (x − ct) + g(x + ct)

is a solution to the equation. In fact it is the general solution.

19
7 Directional Derivatives and Gradients

Suppose we need to compute the rate

of change of f (x, y) with respect to
the distance from a point (a, b) in
some direction. Let ~u = u1~i + u2~j
be the unit vector that has its initial
point at (a, b) and points in the
desired direction. It determines a line in the xy-plane:
x = a + s u1 , y = b + s u2
where s is the arc length parameter that has its reference point at (a, b)
and has positive values in the direction of ~u.

Definition. The directional derivative of f (x, y) in the direction

of ~u at (a, b) is denoted by D~uf (a, b) and is defined by
d 
D~uf (a, b) = [f (a + s u1 , b + s u2)]  = fx(a, b) u1 + fy (a, b) u2

ds s=0

provided this derivative exists.

Analytically, D~uf (a, b) is the

instantaneous rate of change of
f (x, y) with respect to the distance
in the direction of ~u
at the point (a, b).

Geometrically, D~uf (a, b) is

the slope of the surface z = f (x, y)
in the direction of ~u
at the point (a, b, f (a, b)).

20
Generalisation to f (x, y, z) (and f (x1, . . . , xn)) is straightforward.

Definition. Let ~u = u1 ~i + u2 ~j + u3 ~k be a unit vector.

The directional derivative of f (x, y, z) in the direction of ~u at
(a, b, c) is denoted by D~uf (a, b, c) and is defined by

d 
D~uf (a, b, c) = [f (a + s u1 , b + s u2 , c + s u3)] 

ds s=0
= fx(a, b, c) u1 + fy (a, b, c) u2 + fz (a, b, c) u3

Example. Find D~uf (2, 1) in the direction of ~a = 3~i + 4 ~j

 
1 2/3 p
f (x, y) = ln e 3 12 sin(x − 2y) + 8y 2 − x3 − 6x2y + 32
2
Answer: D~uf (2, 1) = −5/3

21
The gradient

Note that
D~uf = fx u1 + fy u2 + fz u3 = (fx ~i + fy ~j + fz ~k) · (u1 ~i + u2 ~j + u3 ~k)

Definition. Let ~ei be the standard orthonormal coordinate basis of

Rn, so that ~r = ni=1 xi~ei.
P

The gradient of f (x1, · · · , xn) is defined by

n
~ (x1, · · · , xn) =
X ∂f (x1, · · · , xn)
∇f ~ei
i=1
∂x i

In particular

~ (x, y) = fx(x, y)~i + fy (x, y) ~j

∇f

~ (x, y, z) = fx(x, y, z)~i + fy (x, y, z) ~j + fz (x, y, z) ~k

∇f

The symbol ∇ ~ is read as either “nabla” (from ancient Hebrew) or “del”

(it is inverted ∆).

~ (a, b)·~u ,
D~uf (a, b) = ∇f ~ (a, b, c)·~u ,
D~uf (a, b, c) = ∇f ~ ·~u
D~uf = ∇f

~ r =
p
Example. Find ∇r; x2 + y 2 + z 2 and D~ur(1, 1, 1) in the
direction of ~a = ~i + 2 ~j + 2 ~k.
22
Properties of the gradient

~ (a, b) · ~u = |∇f
D~uf (a, b) = ∇f ~ (a, b)| |~u| cos θ = |∇f
~ (a, b)| cos θ

Since −1 ≤ cos θ ≤ 1, if |∇f~ (a, b)| =

6 0 then the maximum value of
~ (a, b)| and it occurs when θ = 0, that is, when ~u is
D~uf (a, b) is |∇f
~ (a, b).
in the direction of ∇f

Geometrically, the maximum slope of the surface z = f (x, y) at

~ (a, b)|.
(a, b) is in the direction of the gradient and is equal to |∇f

~ (a, b)| = 0 then D~uf (a, b) = 0 in all directions at (a, b).

If |∇f
It occurs where the surface z = f (x, y) has a relative maximum or
minimum or a saddle point.

23
 ~ (x1, . . . , xn)| cos θ, these properties
Since D~uf (x1, . . . , xn) = |∇f
hold for functions of any number of variables.

Theorem. Let f be a function differentiable at a point P .

~ = ~0 at P then all directional derivatives of f at P are 0.
1. If ∇f
~ 6= ~0 at P then the derivative in the direction of ∇f
2. If ∇f ~ at P
~ | at P .
has the largest value equal to |∇f
~ 6= ~0 at P then the derivative in the direction opposite to
3. If ∇f
~ at P has the smallest value equal to −|∇f
that of ∇f ~ | at P .

Example. The point P = (2, 3, −1)

p
f (x, y, z) = 2xy + 3z 4 − 6 cos(3x − 2y)

24
Gradients are normal to level curves and level surfaces

Level curve C: f (x, y) = k.

Let C be smoothly parametrised as x = x(s), y = y(s) where
s is an arc length parameter. The unit tangent vector to C is

dx dy
T~ (s) = ~i + ~j
ds ds
Since f (x, y) is constant on C we expect DT~ f (x, y) = 0. Indeed

~ · T~ = (fx ~i + fy ~j) · ( dx ~i + dy ~j)

DT~ f (x, y) = ∇f
ds ds
dx dy d ~ ⊥ T~
= fx + fy = f (x(s), y(s)) = 0 ⇒ ∇f
ds ds ds
~ (a, b) 6= ~0 then
Thus if (a, b) belongs to the level curve, and ∇f
~ (a, b) is normal to T~ at (a, b) and therefore to the level curve.
∇f

25
Definition. A vector is called normal to a surface at (a, b, c) if it is
normal to a tangent vector to any curve on the surface through (a, b, c).

Level surface σ: F (x, y, z) = k

Let C, smoothly parametrised as x = x(s), y = y(s), z = z(s)
be any curve on σ through (a, b, c). The unit tangent vector to C is

dx dy dz
T~ (s) = ~i + ~j + ~k
ds ds ds
and DT~ F (x, y, z) is

~ · T~ = (Fx ~i + Fy ~j + Fz ~k) · ( dx ~i + dy ~j + dz ~k)

DT~ F (x, y, z) = ∇F
ds ds ds

dx dy dz d ~ ⊥ T~
= Fx + Fy + Fz = F (x(s), y(s), z(s)) = 0 ⇒ ∇F
ds ds ds ds

~ (a, b, c) is normal to T~ at (a, b, c) and therefore to σ.

Thus, ∇F

26
Tangent planes
Consider a level surface σ: F (x, y, z) = k,
and let P = (a, b, c) belong to σ.
~ (a, b, c) is normal to tangent
Since ∇F
vectors to curves on σ through P ,
all these tangent vectors belong to one
and the same plane.
This plane is called the tangent plane
to the surface σ at P .
To find an equation of the tangent plane
we use that if we know a vector ~n normal
to a plane through a point ~r0 = a~i + b ~j + c ~k
then an equation of the plane is
~n · (~r − ~r0) = 0 ⇔ n1(x − a) + n2(y − b) + n3(z − c) = 0
because ~r − ~r0 is parallel to the plane and therefore normal to ~n.
Choosing ~n = ∇F~ (a, b, c), we get the equation of the tangent plane to
the level surface σ at P = (a, b, c)

Fx(a, b, c)(x − a) + Fy (a, b, c)(y − b) + Fz (a, b, c)(z − c) = 0

The line through P parallel to ∇F~ (a, b, c) is perpendicular to the

tangent plane, and is called the normal line to the surface σ at
P . Its parametric equations are
x = a + Fx(a, b, c)t , y = b + Fy (a, b, c)t , z = c + Fz (a, b, c)t

Example. 4x2 + y 2 + z 2 = 18 at (2, 1, 1).

Tangent plane, normal line, the angle the tangent plane makes with
the xy-plane?
27
Tangent planes to z = f (x, y)

The graph of a function z = f (x, y) can be thought of as the level

surface of the function F (x, y, z) = f (x, y) − z with constant 0.
We find
1. the gradient
~ (a, b, c) = fx(a, b)~i + fy (a, b) ~j − ~k ,
∇F c = f (a, b)

2. the equation of the tangent plane to the surface z = f (x, y) at

(a, b, f (a, b))
fx(a, b)(x − a) + fy (a, b)(y − b) − (z − c) = 0 ⇒
z = f (a, b) + fx(a, b)(x − a) + fy (a, b)(y − b)
that is the local linear approximation of f at (a, b),
3. the parametric equations of the normal line to the surface
z = f (x, y) at (a, b, f (a, b))
x = a + fx(a, b) t , y = b + fy (a, b) t , z = f (a, b) − t

Example. Consider the surface

 
1 2/3 p
z = f (x, y) = ln e 3 12 sin(x − 2y) + 8y 2 − x3 − 6x2y + 32
2
1. Find an equation for the tangent plane and parametric equations
for the normal line to the surface at the point P = (2, 1, z0) where
z0 = f (2, 1).
2. Find points of intersection of the tangent plane with the x-, y-
and z-axes. Sketch the tangent plane, and show the point P on it.
Sketch the normal line to the surface at P .

28
8 Maxima and minima of functions of two variables

Definition. A function f
of two variables is said to have a
relative maximum (minimum)
at a point (a, b) if there is a disc
centred at (a, b) such that
f (a, b) ≥ f (x, y) (f (a, b) ≤ f (x, y))
for all points (x, y) that lie inside
the disc.
A function f is said to have an
absolute maximum (minimum)
at (a, b) if
f (a, b) ≥ f (x, y) (f (a, b) ≤ f (x, y))
for all points (x, y) that lie inside
in the domain of f .
If f has a relative (absolute)
maximum or minimum at (a, b)
then we say that f has a relative
(absolute) extremum at (a, b).
relative ↔ local

29
The extreme-value theorem. If f (x, y) is continuous on a closed
and bounded set R, then f has both absolute maximum and an abso-
lute minimum on R.

Finding relative extrema

Theorem. If f has a relative extremum at (a, b), and if the first-order
derivatives of f exist at this point, then
fx(a, b) = 0 and fy (a, b) = 0
Definition. A point (a, b) in the domain of f (x, y) is called a crit-
ical point of f if fx(a, b) = 0 and fy (a, b) = 0, or if one or both
partial derivatives do not exist at (a, b).

Example. f (x, y) = y 2 − x2 is a
hyperbolic paraboloid.
fx = −2x, fy = 2y ⇒ (0, 0) is critical
but it is not a relative extremum.
It is a saddle point.

30
We say that a surface z = f (x, y) has a saddle point at (a, b) if
there are two distinct vertical planes through this point such that the
trace of the surface in one of the planes has a relative maximum at
(a, b), and the trace in the other has a relative minimum at (a, b).

Example.

How to determine whether a critical point is a max or min?

31
The second partials test

Theorem. Let f (x, y) have continuous second-order partial

derivatives in some disc centred at a critical point (a, b), and let

2
D = fxx(a, b)fyy (a, b) − fxy (a, b)

1. If D > 0 and fxx(a, b) > 0, then f has a relative minimum at

(a, b).
2. If D > 0 and fxx(a, b) < 0, then f has a relative maximum at
(a, b).
3. If D < 0, then f has a saddle point at(a, b).
4. If D = 0, then no conclusion can be drawn.

Example.
f (x, y) = x4 − x2y + y 2 − 3y + 4

How to find the absolute extrema of a continuous function of two

variables on a closed and bounded set R?
1. Find the critical points of f that lie in the interior of R.
2. Find all the boundary points at which the absolute extrema can
occur.
3. Evaluate f (x, y) at the found points. The largest of these values is
the absolute maximum, and the smallest the absolute minimum.

Example.
f (x, y) = 3x + 6y − 3xy − 7 , R is the triangle (0, 0), (0, 3), (5, 0)

32
Lagrange multipliers

Extremum problems with constraints:

Find max or min of the function f (x1, . . . , xn) subject to constraints
gα (x1, . . . , xn), α = 1, . . . , m

Consider f (x, y) and g(x, y) = 0.

The graph of g(x, y) = 0 is a curve.
Consider level curves of f : f (x, y) = k.
At (a, b) the curves just touch, and thus have
~ (a, b)
a common tangent line at (a, b). Since ∇f
is normal to the level curve at (a, b), and
~
∇g(a, b) is normal to the constraint curve
~ (a, b)||∇g(a,
at (a, b), we get ∇f ~ b)

~ (a, b) = λ ∇g(a,
∇f ~ b)

for some scalar λ called the Lagrange multiplier.

Proof. Parametrise g(x, y) = 0.
Then, f (x, y) = f (x(t), y(t)) is a function of t
and its local extrema are at

d ∂f 0 ∂f 0
f (x(t), y(t)) = x + y
dt ∂x ∂y
= ∇f~ · (x0 ~i + y 0 ~j) = ∇f
~ · T~
~ and ∇g
Thus, both ∇f ~ are ⊥ to T~ .

33
In general, we introduce a Lagrange multiplier λα for each of the con-
straint gα , and the equations are
m
X
~ =
∇f ~ α.
λα ∇g
α=1

Example. Find the points on the sphere x2 + y 2 + z 2 = 36 that are

closest to and farthest from the point (1, 2, 2).

34