MATH 2023 • Spring 2015-16 • Multivariable Calculus

Problem Set #4 • Critical Points, Lagrange’s Multiplier

1. (F) Find all local extrema (a.k.a. critical points) of the following functions f ( x, y). Deter-
mine the nature (a local minimum, a local maximum or a saddle) of each of them using
the Second Derivative Test whenever possible. If the Second Derivative Test is inconclu-
sive, use some other methods to determine its nature.
(a) f ( x, y) = 4 + x3 + y3 − 3xy

∂f ∂f
Solution: To find critical points, we solve ∂x = 0 and ∂y = 0, i.e.

3x2 − 3y = 0
3y2 − 3x = 0

After simplification: x2 = y and y2 = x. Substitute y = x2 into y2 = x, we get:

2
x2 = x =⇒ x4 − x = 0 =⇒ x ( x3 − 1) = 0

Therefore, we have x = 0 or x3 = 1. The latter case gives x = 1.

Since y = x2 , we get y = 0 when x = 0; whereas y = 1 when x = 1. There are
two critical points:
( x, y) = (0, 0), (1, 1).
Next, compute the second derivatives:

f xx = 6x f xy = −3
f yx = −3 f yy = 6y
 
P f xx ( P) f yy ( P) f xy ( P) 2
f xx f yy − f xy ( P) P is:
(0, 0) 0 0 −3 −9 saddle
(1, 1) 6 6 −3 (6)(6) − (−3)2 > 0 local minimum

(b) f ( x, y) = x2 + 4y2 − 2x2 y + 4

Solution: (Answer only)

Critical Point f xx 2
f xx f yy − f xy Nature
(√0, 0) 2 16 local minimum
(−√ 2, 12 ) 0 −32 saddle
( 2, 12 ) 0 −32 saddle

(c) f ( x, y) = sin x cos y

∂f ∂f
Solution: To solve for critical points, we solve ∂x = 0 and ∂y = 0, i.e.

cos x cos y = 0 and − sin x sin y = 0

From the first equation, we get two cases: Case A: cos x = 0, or Case B: cos y = 0.
Case A: Since cos x = 0, we have x = π2 + kπ where k is any integer. Then
sin x = ±1 (depending on whether k is even or odd), and the second equation

Page 1
MATH 2023 Critical Points, Lagrange’s Multiplier Problem Set #4

sin x sin y = 0 then tells us that sin y = 0, and so y = mπ where m is any integer.
Critical points obtained in this case are:
π 
( x, y) = + kπ, mπ , where m and k are any integers.
2

Case B: From cos y = 0, we get y = π2 + nπ where n is any integer, then substitute

it into the second equation, we get sin x · (±1) = 0 where ± depends on whether
k is even or odd. Therefore, x = qπ where q is any integer. Critical points
obtained in this case are:
π
( x, y) = (qπ, + nπ ) where n and q are any integers.
2

To apply the Second Derivative Test, we first compute:

f xx = − sin x cos y
f xy = − cos x sin y
f yy = − sin x cos y
2
f xx f yy − f xy = sin2 x cos2 y − cos2 x sin2 y
 
P f xx f yy f xy 2
f xx f yy − f xy P is:
( π2
+ kπ, mπ ) k odd, m odd −1 −1 0 1 local max
( π2
+ kπ, mπ ) k odd, m even 1 1 0 1 local min
( π2
+ kπ, mπ ) k even, m odd 1 1 0 1 local min
( π2
+ kπ, mπ ) k even, m even −1 −1 0 1 local max
(qπ, π2 + nπ ) any integers q, n 0 0 ±1 −1 saddle

(d) f ( x, y) = x4 + y4

Solution: By solving ∇ f = 4x3 i + 4y3 j = 0, we get (0, 0) as the only critical point
2 = 0 at (0, 0), and so the Second
of f . However, one can check that f xx f yy − f xy
Derivative Test does not conclude anything.
However, this function f ( x, y) = x4 + y4 is always non-negative. Therefore,
f ( x, y) ≥ f (0, 0) = 0 for any ( x, y). In other words, (0, 0) is a local minimum
point (in fact a global minimum).

Page 2
MATH 2023 Critical Points, Lagrange’s Multiplier Problem Set #4

2. (FF) Give an example of a C2 function f ( x, y) such that:

2 = 0 at (0, 0).
(a) (0, 0) is a saddle point and f xx f yy − f xy

Solution: One of many examples: f ( x, y) = x4 − y4 . Along the x-axis, the graph

is f ( x, 0) = x4 which is concave up. However, along the y-axis, the graph
is f (0, y) = −y4 which is concave down. It is straight-forward to verify that
f xx (0, 0) = f yy (0, 0) = f xy (0, 0) = 0.

2 = 0, f
(b) (0, 0) is a local minimum, f xx f yy − f xy xx > 0 and f yy = 0 at (0, 0).

Solution: One of many examples: f ( x, y) = x2 + y4 . It must be non-negative,

and f (0, 0) = 0. Therefore, the origin must be a minimum point (in fact the
absolute minimum). It is straight-forward to verify that f xx (0, 0) = 2 whereas
f yy (0, 0) = f xy (0, 0) = 0.

2 = 0, f
(c) (0, 0) is a local maximum, f xx f yy − f xy xx = 0 and f yy < 0 at (0, 0).

Solution: One of many examples: f ( x, y) = − x4 − y2 .

3. (F) Using Lagrange’s Multipliers, find the maximum and minimum values of the given
function subject to the given constraint:
(a) f ( x, y) = x + 2y subject to x2 + y2 = 4

Solution: Let g( x, y) = x2 + y2 , then the constraint is the level-set g( x, y) = 4.

The Lagrange’s Multiplier system is:

∂f ∂g
=λ 1 = λ · 2x
∂x ∂x
∂f ∂g
=λ 2 = λ · 2y
∂y ∂y
g( x, y) = 4 x 2 + y2 = 4

From the first equation, we must have λ 6= 0 (otherwise if λ = 0, then 1 = 0).

Therefore, by rearranging the first and second equations, we get:

1 1
= 2x and =y
λ λ
Therefore, 2x = y and combine with the third equation, we get:
r r
2 2 2 2 4 4 4
x + (2x ) = 4 =⇒ 5x = 4 =⇒ x = =⇒ x = or − .
5 5 5
Since y = 2x, the candidate points are:
√ √ √ √
( x, y) = ( 4/5, 2 4/5), (− 4/5, −2 4/5).

Page 3
MATH 2023 Critical Points, Lagrange’s Multiplier Problem Set #4

Substitute them into f ( x, y), we get:

r r ! r
4 4 4
f , 2 =5
5 5 5
r r ! r
4 4 4
f − , −2 = −5
5 5 5

The former is the maximum, the latter is the minimum.

(b) f ( x, y) = x − y subject to x2 + y2 = 20 + 3xy

Solution: Note that the constraint is not yet in the level-set form. By rearrange-
ment, we get:
x2 + y2 − 3xy = 20
Therefore, we let g( x, y) = x2 + y2 − 3xy and then the constraint becomes the
level-set g( x, y) = 20. We need to solve the Lagrange’s Multiplier system ∇ f ( x, y) =
λ∇ g( x, y) and g( x, y) = 20:

1 = λ(2x − 3y)
−1 = λ(2y − 3x )
2 2
x + y − 3xy = 20

Solutions are: ( x, y) = (−2, 2), (2, −2). Substitute them into f , we get:

f (−2, 2) = −4, and f (2, −2) = 4.

The maximum is 4 and the minimum is −4.

(c) f ( x, y, z) = xyz subject to x2 + 2y2 + 4z2 = 9

Solution: Let g( x, y, z) = x2 + 2y2 + 4z2 , then the constraint is g( x, y, z) = 9. The

Lagrange’s Multiplier system is:

yz = λ · 2x
xz = λ · 4y
xy = λ · 8z
x + 2y + 4z2 = 9
2 2

The strategy of solving this system is as follows: Suppose the third equation
xy = λ · 8z 6= 0, then by dividing the first equation by the third one gives:

yz 2λx z x
= =⇒ = =⇒ 4z2 = x2 .
xy 8λz x 4z

Also, divide the second equation by the third one gives:

xz 4yλ z y
= =⇒ = =⇒ y2 = 2z2
xy 8zλ y 2z

Page 4
MATH 2023 Critical Points, Lagrange’s Multiplier Problem Set #4

Then, substitute x2 = 4z2 and y2 = 2z2 into the fourth equation gives: 4z2 + 4z2 +
4z2 = 9. This gives the solutions for z, which can be used to determine the values
of x and y.
Next, we need to find the solutions for the case when the third equation is zero,
i.e. xy = λ · 8z = 0.
We skip the detail of computations here. Below are the results:
( x, y, z) f ( x, y, z)
(±3, 0, 0) 0
(0, 0, ±√ 3/2) 0
(0, ±3/ 2, 0) 0q
√ √ √
(± 3, ± 3/2, ± 3/2) ± 2 32 3
q q
The maximum is 32 32 and the minimum is − 32 32 .

(d) f ( x1 , x2 , . . . , xn ) = x1 + 2x2 + . . . + nxn subject to x12 + x22 + . . . + xn2 = 1

Solution: Let g( x1 , . . . , xn ) = x12 + . . . + xn2 , then the constraint becomes the level-
set g = 1. The Lagrange’s Multiplier system is:
∂f ∂g
=λ 1 = 2λx1
∂x1 ∂x1
∂f ∂g
=λ 2 = 2λx2
∂x2 ∂x2
.. ..
. .
∂f ∂g
=λ n = 2λxn
∂xn ∂xn
g ( x1 , . . . , x n ) = 1 x12 + . . . + xn2 = 1
It is a very large system. However, it is helpful to note that λ cannot be zero
(otherwise from the first equation we would have 1 = 0). By rearrangement, we
can get:
1 2 n
x1 = , x2 = , . . . , xn = .
2λ 2λ 2λ
Substitute them into the constraint equation, we get:
 2  2  n 2
1 2
+ +...+ =1
2λ 2λ 2λ
After simplification, we get:
12 + 22 + . . . + n 2 p
= 1 =⇒ 2λ = ± 12 + 22 + . . . + n2 .
(2λ)2
1 2 n
Therefore, the candidate points are: x1 = 2λ , x2 = 2λ , . . . , xn = where
√ 2λ
2λ = ± 12 + 22 + . . . + n2 .

1 2 n 12 + 22 + . . . + n2
f ( x1 , . . . , x n ) = +2· +...+n· = ±√
2λ 2λ 2λ 12 + 22 + . . . + n2
2 +22 + ... + n2 2 2 2
The minimum is − √1 and the maximum is √1 +2 +...+n .
12 +22 +...+n2 12 +22 +...+n2

Page 5
MATH 2023 Critical Points, Lagrange’s Multiplier Problem Set #4

4. (FF) The rationale behind Lagrange’s Multipliers method is that if a function f achieves
its maximum or minimum on the constraint g = c at points P, then the level sets of f and
g at P must be tangent to each other, and so ∇ f ( P) is parallel to ∇ g( P). Therefore, we
solve the system:
∇ f ( P) = λ∇ g( P) and g( P) = c
to locate all such P’s.
However, another way to determine whether ∇ f ( P) and ∇ g( P) are parallel is by their
cross product:

∇ f ( P) k ∇ g( P) if and only if ∇ f ( P) × ∇ g( P) = 0.

By solving the vector equation ∇ f ( P) × ∇ g( P) = 0 for P (instead of using Lagrange’s

Multiplier), try to redo Problems #3(a)(b)(c).
What is the limitation of this method when compared to Lagrange’s Multiplier?

Solution: Take (a) as an example, in which we have f ( x, y) = x + 2y and g( x, y) =

x2 + y2 . By direct computation, their gradients are:

∇ f = i + 2j, ∇ g = 2xi + 2yj

and their cross product is given by ∇ f × ∇ g = (−4x + 2y)k. By setting ∇ f × ∇ g = 0,

we get −4x + 2y = 0, and so: y = 2x.
constraint equation x2 + y2 = 4, we have x2 + (2x )2 = 4, whose solutions
Using the √
are x = ± 4/5. Therefore, using y = 2x, we get:
√ √ √ √
( x, y) = ( 4/5, 2 4/5), (− 4/5, −2 4/5)

which are exactly the same as in #3(a).

Similar approaches can be applied to solve #3(b)(c).
However, since cross product is not defined in dimensions higher than 3, we cannot
apply this approach to maximize/minimize a function for 4 variables or higher.

5. (F) A closed rectangular water tank is to be made with three different materials. The
top part will be made by a thin material which costs $1 per cm2 . The four sides will use
stronger material which costs $2 per cm2 . To support the weight of water, the bottom of
the tank has to be made with very durable and strengthened material which costs $5 per
cm2 .
Suppose the volume of the water tank is to be 96cm3 . What dimensions of the tank will
minimize the cost of construction?

Solution: (Sketch) Let l be the length, w be the width and h be the height.

minimize: f (l, w, h) = |{z}

wl + 2(2hl ) + 2(2wh) + 5(wl ) = 6wl + 4hl + 4wh
| {z } | {z }
top sides base
subject to: g(l, w, h) = lwh = 96

Page 6
MATH 2023 Critical Points, Lagrange’s Multiplier Problem Set #4

Answer: (l, w, h) = (4, 4, 6) is the only solution to the Lagrange’s Multiplier system. It
must give the minimum, since the maximum does not exist. Given any fixed volume,
we can always construct a rectangular cube with arbitrarily large surface area.

6. (F) Find the point(s) on the cone:

q
z= x 2 + y2

that is/are closest to the point (1, 3, 1).

Solution: (Sketch) To ease computations, we rewrite the cone equation as:

z2 = x2 + y2 , or equivalently z2 − x2 − y2 = 0.

Let g( x, y, z) = z2 − x2 − y2 , then g = 0 is the constraint.

minimize: f ( x, y, z) = ( x − 1)2 + (y − 3)2 + (z − 1)2

subject to: g( x, y, z) = z2 − x2 − y2 = 0
 √ √ √   √ √ √ 
10+ 10 3(10+ 10) 1+ 10 10− 10 3(10− 10 1− 10
Answer: ( x, y, z) = 20 , 20 , 2 , 20 , 20 , 2 .
p
The second point is rejected since the original cone equation z = x2 + y2 requires
that z ≥ 0. It is straight-forward to check that:
√ √ √ !
10 + 10 3(10 + 10) 1 + 10 11 √
f , , = − 10
20 20 2 2

 √ √ √ 
3(10+ 10) 1+ 10
Therefore, the required point is 10+20 10 , 20 , 2 and the distance from this
q √
point to (1, 3, 1) is given by 11
2 − 10.

7. (FF) Consider the surface given by the equation

x2 y2 z2 = 1.

(a) Show that for any point ( a, b, c) on the surface, its tangent plane does not contain the
origin.

Solution: Let g( x, y, z) = x2 y2 z2 , then the surface is a level-set g( x, y, z) = 1. To

find the normal vector to the surface at ( a, b, c), we compute:

∇ g( x, y, z) = 2xy2 z2 i + 2x2 yz2 j + 2x2 y2 zk

Therefore, the normal vector to the required tangent plane can be taken to be:

n = ab2 c2 i + a2 bc2 j + a2 b2 ck

Page 7
MATH 2023 Critical Points, Lagrange’s Multiplier Problem Set #4

As the plane passes through ( a, b, c), the equation of the tangent is then given by:

ab2 c2 x + a2 bc2 y + a2 b2 cz = ab2 c2 · a + a2 bc2 · b + a2 b2 c · c

As ( a, b, c) lies on the surface x2 y2 z2 , we can assume that all of a, b and c are non-
zero (otherwise if any of them are zero, we would get 0 = 1). After simplification,
we get:
x y z
ab2 c2 x + a2 bc2 y + a2 b2 cz = 3a2 b2 c2 =⇒ + + = 3.
a b c
By substituting ( x, y, z) = (0, 0, 0), we get LHS = 0 whereas RHS = 3. Therefore,
(0, 0, 0) is not contained on the tangent plane.

(b) Find all points ( a, b, c) on the surface such that the tangent plane at these points are
closest to the origin.

Solution: By Q5 in Problem Set #1, the distance between the plane x

a + yb + zc = 3
and the origin (0, 0, 0) is given by:

3
d= q
1 1 1
a2
+ b2
+ c2

Since ( a, b, c) lies on the given surface, they are subject to the constraint a2 b2 c2 =
1. Therefore, the distance is also given by:

3 3
d= q =√
b2 c2 + a2 c2 + b2 c2 b2 c2 + a2 c2 + b2 c2
a2 b2 c2

To minimize d is equivalent to maximize a2 b2 + b2 c2 + c2 a2 . Let:

f ( a, b, c) = a2 b2 + b2 c2 + c2 a2 (to be maximized)
2 2 2
g( a, b, c) = a b c = 1 (constraint)

Solving Lagrange’s Multiplier system ∇ f = λ∇ g and g = 1, we can get these

candidate points:
( a, b, c) f ( a, b, c)
(−1, −1, −1) 3
(−1, −1, 1) 3
(−1, 1, −1) 3
(−1, 1, 1) 3
(1, −1, −1) 3
(1, −1, 1) 3
(1, 1, −1) 3
(1, 1, 1) 3
3
√
In all of the above cases, we have d = √3 = 3. Therefore, they are all the
maximum points for f and hence are also points on the surface which their
tangent plane closest to the origin.

Page 8
MATH 2023 Critical Points, Lagrange’s Multiplier Problem Set #4

8. (FF) Suppose
f ( x, y) = 2x2 + xy − 8x − y + 6
Let T be the triangular region (boundary included) in the xy-plane with vertices (0, 0),
(0, 3) and (3, 0).
(a) Find all the interior critical point(s) of f in the region T.

Solution: ∇ f = h4x + y − 8, x − 1i. Set ∇ f = 0, we get x = 1 and y = 4. The

only possible interior critical point for f is ( x, y) = (1, 4), which is beyond the
range of T. Therefore, there is no interior critical point in the region T.

(b) Find the maximum and minimum values of f ( x, y) when ( x, y) is restricted on the
vertical side of T, i.e. the line segment joining (0, 0) and (0, 3).

Solution: The vertical side of T is given by the equation x = 0. Substitute this

into the function, we get:
f (0, y) = −y + 6
which is a strictly decreasing function of y. Therefore, on the vertical side, the
maximum and minimum of f are attained at the end points (0, 0) and (0, 3).
Since f (0, 0) = 6 and f (0, 3) = 3. The maximum value of f is 6 and minimum
value is 3 when restricted on this vertical side.

(c) Find the maximum and minimum values of f ( x, y) when ( x, y) is restricted on the
horizontal side of T, i.e. the line segment joining (0, 0) and (3, 0).

Solution: The horizontal side has equation y = 0. Regarding x as the variable,

the function is given by:
f ( x, 0) = 2x2 − 8x + 6.
d
Since dx f ( x, 0) = 4x − 8, the only critical point is x = 2. Compute f at this critical
point and the end-points of the horizontal sides, we get: f (2, 0) = −2, f (0, 0) = 6
and f (3, 0) = 0. Therefore the max of f is 6 and the min is −2 when restricted
on the horizontal side.

(d) Write the equation of the line joining the vertices (0, 3) and (3, 0). Hence find the
maximum and minimum values of f ( x, y) when ( x, y) is restricted to the hypotenuse
of T.

Solution: The equation of the hypotenuse is x + y = 3. Let g( x, y) = x + y, then

the hypotenuse is the level set g = 3. Using Lagrange’s Multiplier:

f x = λgx 4x + y − 8 = λ
f y = λgy x−1 = λ
g( x, y) = 3 x+y = 3

From the first two equations, we get 4x + y − 8 = x − 1 (Be flexible! There is no

need to divide cases for this system!) Hence, 3x + y = 7. Combining with the
third equation x + y = 3, we get x = 2 and y = 1. The only critical point on
the hypotenuse is ( x, y) = (2, 1). Check the values of f at the critical point and

Page 9
MATH 2023 Critical Points, Lagrange’s Multiplier Problem Set #4

end-points, we get:

f (2, 1) = −1, f (3.0) = 0, f (0, 3) = 3.

Therefore, the maximum of f is 3 and the minimum is −1 when restricted to the

hypotenuse.

(e) Determine the absolute maximum and minimum of f ( x, y) over the domain T.

Solution: Since f has no interior critical point in the region T, its absolute max
and min must take place on the boundary of the region, i.e. the three sides of the
triangle. From previous parts, we got:
side max min
vertical 6 3
horizontal 6 -2
hypotenuse 3 -1
Therefore, the absolute maximum of f is 6 and the absolute minimum is −2 over
the region T.

9. (FF) This purpose of this problem

938
isChapter
to explain why the Second Derivative Test works
13 • Functions of Several Variables
for two-variable functions. Let f ( x, y) be a C2 function with a critical point P0 ( a, b). Given
any unit direction û = u1 i + u2 j, the path:

x = a + tu1 ,
13.6
y = b + tu
Directional Derivatives and
2
Partial derivatives tell us a lot about the rate of cha
is a straight-line passing through ( a, b) in the direction of û. ever,
As such,
they dothe
not intersection
directly answer some important
curve of the vertical plane shown in the diagram below can be are as:a point 1a, b, f1a, b22 on the surfa
standing at
expressed
fx and fy tell you the rate of change (or slope) of th
z = f ( a + tu1 , b + tu2 ) parallel to the x-axis and y-axis, respectively. But y
directions from that point and find a different rate
observation in mind, we pose several questions.
z • Suppose you are standing on a surface and you w
dinate direction—say, northwest or south-southea
function in such a direction?
• Suppose you are standing on a surface and you re
(a,, b,
b, f (a,
(a b)) In which direction will it roll?
z  f (x, y) • If you are hiking up a mountain, in what direction
you want to follow the steepest path?
These questions will be answered in this section a
tive, followed by one of the central concepts of cal
x P0(a, b)
u y
Directional Derivatives
Unit vector u
Let 1a, b, f1a, b22 be a point on the surface z = f
FIGURE 13.63 xy-plane (Figure 13.63). Our aim is to find the ra
P01a, b2. In general, this rate of change is neither f

dz 
Therefore, the first derivative dt  兩u兩 ⫽measures the slope of tangent at ( a, b, f ( a, b)), whereas
 t =0 1 u or u = 8 0, 1 9 ), but it turns out to be a combinatio
2
the second derivative dt2  d z
indicates whether u ⫽ sin
2 the curve is concaveFigure
up 13.64a
or downshows the unit vector u at an an
around
t =0 nents are u = 8 u 1, u 2 9 = 8 cos u, sin u 9 . The deri
the point ( a, b, f ( a, b)). u1 ⫽ cos the line / in the xy-plane through P0 in the directio
(a) h units from P0 along /, has coordinates P1a + h co
ᐉ
Now imagine the plane Q perpendicular to the
y P(a ⫹ h cos , b ⫹ h sin ) the surface z = f1x, y2 in a curve C. Consider two
they have z-coordinates f1a, b2 and f1a + h cos u,
u
Page 10 h sin of the secant line between these points is
h
f1a + h cos u, b + h sin
P0(a, b) h cos h
u ⫽ 具u , u 典 ⫽ 具cos , sin 典 The derivative of f in the direction of u is obtaine
MATH 2023 Critical Points, Lagrange’s Multiplier Problem Set #4

(a) Using the chain rule, show that:

dz
= f x u1 + f y u2
dt
d2 z
= f xx u21 + 2 f xy u1 u2 + f yy u22
dt2

Solution: Given that z = f ( a + tu1 , b + tu2 ), then f is a function of ( x, y), and

( x, y) is a function of t. Precisely, we have: x = a + tu1 and y = b + tu2 (note that
a, b, u1 and u2 are constants). Using the chain rule, we get:

dz ∂ f dx ∂ f dy
= +
dt ∂x dt ∂y dt
d d
= f x ( a + tu1 ) + f y (b + tu2 )
dt dt
= f x u1 + f y u2
d2 z
 
d dz
=
dt2 dt dt
d

= f x u1 + f y u2 (from above)
dt
d fx d fy
= u1 + u2
dt dt 
∂ f x dx ∂ f x dy
= + u1
∂x dt ∂y dt
∂ f y dx ∂ f y dy
 
+ + u2
∂x dt ∂y dt


= f xx u1 + f xy u2 u1 (recall that x = a + tu1


+ f yx u1 + f yy u2 u2 and y = b + tu2 )
= f xx u21 + f xy u2 u1 + f yx u1 u2 + f yy u22
= f xx u21 + 2 f xy u1 u2 + f yy u22 (Mixed Partials Theorem)

Page 11
MATH 2023 Critical Points, Lagrange’s Multiplier Problem Set #4

(b) By completing-the-square, show further that:

  2  f f − f 2  
f xy
 f xx u1 + f u2 + xx yy2 xy u22
 if f xx 6= 0
 xx f xx
d2 z

  2  f f − f 2  
= f f xy xx yy xy
u21 if f yy 6= 0
dt2 
 yy f yy u1 + u2 + 2
f yy


2 f xy u1 u2 if f xx = f yy = 0


Solution: If f xx 6= 0, then:

d2 z
= f xx u21 + 2 f xy u1 u2 + f yy u22
dt2
2 f xy f yy 2
 
2
= f xx u1 + u1 u2 + u
f xx f xx 2
" 2  2 #
2 f xy f xy f xy f yy 2

2
= f xx u1 + u1 u2 + u2 − u2 + u
f xx f xx f xx f xx 2
" ! #
2 2
f xy
f xy f yy
= f xx u1 + u2 + − 2 + u22
f xx f xx f xx
" ! #
2
f xx f yy − f xy2
f xy
= f xx u1 + u2 + 2
u22
f xx f xx

For the case f yy 6= 0, we can obtain the result using symmetry argument: by
switching x with y, and u1 with u2 .
d2 z
The last case is trivial (since dt2
= f xx u21 + 2 f xy u1 u2 + f yy u22 ).

(c) The nature of the point ( a, b, f ( a, b)) can be determined by the following argument:
2 > 0 and f d2 z
i. If f xx f yy − f xy xx > 0, what can you say about dt2
? Using this observa-
tion, conclude the nature of the point ( a, b, f ( a, b)).

Solution: In this case, we have:

 
!
2 2
f xx f yy − f xy
d2 z
 
 f xy 2 

= f xx
 u 1 + u 2 + u 2
dt2 |{z} | f xx 2
f xx |{z}  
+ {z } | {z } + or 0
+ or 0 +

d2 z d2 z
and therefore dt2
≥ 0 for any u1 and u2 . The only situation with dt2
= 0 is
f xy
that both u1 + f xx u2= 0 and u2 = 0. However, it would imply u1 = u2 = 0,
which contradicts to the fact that u = u1 i + u2 j is a unit (hence non-zero)
vector.
2
Therefore, ddt2z > 0 for any unit direction u. The graph z = f ( x, y) is concave
up in all directions u, and so the critical point must be a local minimum.

Page 12
MATH 2023 Critical Points, Lagrange’s Multiplier Problem Set #4

2 > 0 and f
ii. How about if f xx f yy − f xy xx < 0?

Solution: In this case we have:

 
!
2 2
f xx f yy − f xy
d2 z
 
 f xy 2

2
= f xx  u1 + u2 + 2
u2 
dt |{z}  f xx f xx |{z}  
− | {z } | {z } + or 0
+ or 0 +

d2 z
and therefore ≤ 0. By similar discussion as in (i), the only situation in
dt2
d2 z
which dt2
= 0 is that u1 = u2 = 0 (that is impossible for a unit vector u).
2
Therefore, ddt2z < 0 for any unit vector u. The graph z = f ( x, y) is concave
down in all directions u, and so the critical point must be a local maximum.

2 > 0 and f
iii. How about if f xx f yy − f xy xx = 0?

2 > 0.
Solution: This is not possible. Otherwise we would have − f xy

2 < 0?
iv. What if f xx f yy − f xy

Solution: In the case f xx 6= 0, we have:

 
! 
2 2
f xx f yy − f xy
d2 z

f xy
u22 
 
2
= f xx  u1 + u2 + 2
dt 
| f xx f xx 

{z } | {z }
+ or 0 − or 0

and in the case f yy 6= 0, we have:

 
! 
2 2
f xx f yy − f xy
d2 z

 f xy 2

= f yy
 u 1 + u 2 + u 1
dt2 | f yy {z
 2
f yy 

} | {z }
+ or 0 − or 0

2
In either case, whether ddt2z is positive or negative depends on the choice of
u1 and u2 . In other words, the graph z = f ( x, y) is concave up in some
directions and concave down in some other directions. The critical point is a
saddle.
2
In the last case where f xx = f yy = 0, we then have ddt2z = 2 f xy u1 u2 . Therefore,
whether it is positive or negative depends only the choice of u. The critical
point is again a saddle.

Page 13