MATH2023 Multivariable Calculus Problem Set 4

L2/L3 (Fall 2019)

Solution to Extra Problem Set 4

1. The unit vector along the direction of 𝐯 = 12𝐢 − 5𝐣 is

1 12 5
𝐯̂ = (12𝐢 − 5𝐣) = 𝐢− 𝐣,
√122 + 52 13 13

so the directional derivative of a function 𝑓 along this direction is defined by the limit
12 5
𝑓 (𝑥 + ℎ, 𝑦 − ℎ) − 𝑓(𝑥, 𝑦)
𝐷𝐯̂ 𝑓(𝑥, 𝑦) ≔ lim 13 13 .
ℎ→0 ℎ
2
Now 𝑓(𝑥, 𝑦) = 𝑥 + 3𝑥𝑦 + 4𝑦, so
12 2 12 5 5
[(𝑥 + ℎ) + 3 (𝑥 + ℎ) (𝑦 − ℎ) + 4 (𝑦 − ℎ)] − (𝑥 2 + 3𝑥𝑦 + 4𝑦)
13 13 13 13
𝐷𝐯̂ 𝑓(𝑥, 𝑦) = lim
ℎ→0 ℎ
24 122 2 12 5 60 5
( ℎ𝑥 + 2 ℎ ) + 3 ( ℎ𝑦 − ℎ𝑥 − 2 ℎ2 ) + 4 (− ℎ)
13 13 13 13 13 13
= lim
ℎ→0 ℎ
9 36 20
= 𝑥+ 𝑦− .
13 13 13

2. The directional derivative of 𝑓 at (1, 1, 1) is the largest along the direction of ∇𝑓(1, 1, 1), which is given by
2 2 2
∇𝑓(1, 1, 1) = 〈𝑓𝑥 , 𝑓𝑦 , 𝑓𝑧 〉(1, 1, 1) = 〈 , , 〉 (1, 1, 1) = 〈2, 2, 2〉.
𝑥 𝑦 𝑧

The unit vector along this direction is

1 1 1 1
𝐯̂ = 〈2, 2, 2〉 = 〈 , , 〉.
√22 + 22 + 22 √3 √3 √3
The directional derivative of 𝑓 along this direction is
1 1 1
𝐷𝐯̂ 𝑓(1, 1, 1) = ∇𝑓(1, 1, 1) ⋅ 𝐯̂ = 〈2, 2, 2〉 ⋅ 〈 , , 〉 = 2√3.
√3 √3 √3

3. (a) Note that 𝐢 + 𝐣 and −2𝐣 are not unit vectors. From the given directional derivatives of 𝑓 we have
1 1
2√2 = 𝐷𝐢+𝐣 𝑓(1, 2) = ∇𝑓(1, 2) ⋅ (𝐢 + 𝐣) and − 3 = 𝐷−2𝐣 𝑓(1, 2) = ∇𝑓(1, 2) ⋅ (−2𝐣) ,
√12 + 12
⏟ ⏟
2
unit vector unit vector

so
∇𝑓(1, 2) ⋅ (𝐢 + 𝐣) = (2√2) (√12 + 12 ) = 4 and ∇𝑓(1, 2) ⋅ (−𝐣) = −3.
Therefore the derivative of 𝑓 along the direction of −𝐢 − 2𝐣 is

1 1
𝐷−𝐢−2𝐣 𝑓(1, 2) = ∇𝑓(1, 2) ⋅ (−𝐢 − 2𝐣) = ∇𝑓(1, 2) ⋅ [(−𝐣) − (𝐢 + 𝐣)]
√(−1)2
⏟ + (−2)2 √5
unit vector

1
= [∇𝑓(1, 2) ⋅ (−𝐣) − ∇𝑓(1, 2) ⋅ (𝐢 + 𝐣)]
√5
1 7
= (−3 − 4) = − .
√5 √5

Page 1 of 12
MATH2023 Multivariable Calculus Problem Set 4
L2/L3 (Fall 2019)

(b) Let 𝐯̂ = 𝑎𝐢 + 𝑏𝐣 be the required unit vector. Then we require that 𝐷𝐯̂ 𝑓(1, 2) = √2, so we have the system

(𝑎𝐢 + 𝑏𝐣) = √2
{∇𝑓(1, 2) ⋅ .
𝑎2 + 𝑏 2 = 1
Note that 𝑎𝐢 + 𝑏𝐣 = 𝑎(𝐢 + 𝐣) + (𝑎 − 𝑏)(−𝐣), so from the first equation we have
𝑎[∇𝑓(1, 2) ⋅ (𝐢 + 𝐣)] + (𝑎 − 𝑏)[∇𝑓(1, 2) ⋅ (−𝐣)] = √2,
i.e. 𝑎(4) + (𝑎 − 𝑏)(−3) = √2. Now the system becomes

(−3)(𝑎 − 𝑏) = √2
{4𝑎 + .
𝑎2 + 𝑏 2 = 1

1 1 7 1
On solving this system we get (𝑎, 𝑏) = (− , ) or (𝑎, 𝑏) = ( , ), so
√2 √2 5√2 5√2

1 1 7 1
𝐯̂ = − 𝐢+ 𝐣 or 𝐯̂ = 𝐢+ 𝐣.
√2 √2 5√2 5√2

(c) Let 𝐯̂ = 𝑎𝐢 + 𝑏𝐣 be the required unit vector. Then similar to (b) we solve the system

4𝑎 + (−3)(𝑎 − 𝑏) = 4
{ .
𝑎2 + 𝑏 2 = 1
This time the system has no real solutions, and so there is no such direction along which the derivative of 𝑓 at
(1, 2) is 4.

4. The two given surfaces are level surfaces of the functions

𝑓(𝑥, 𝑦, 𝑧) = 𝑥 3 + 3𝑥 2 𝑦 2 + 𝑦 3 + 4𝑥𝑦 − 𝑧 2 and 𝑔(𝑥, 𝑦, 𝑧) = 𝑥 2 + 𝑦 2 + 𝑧 2
respectively, whose gradient vectors are
∇𝑓(𝑥, 𝑦, 𝑧) = 〈3𝑥 2 + 6𝑥𝑦 2 + 4𝑦, 6𝑥 2 𝑦 + 3𝑦 2 + 4𝑥, −2𝑧〉 and ∇𝑔(𝑥, 𝑦, 𝑧) = 〈2𝑥, 2𝑦, 2𝑧〉
respectively. It is easy to verify that the point (1, 1, 3) lies on both surfaces, so the two vectors
∇𝑓(1, 1, 3) = 〈13, 13, −6〉 and ∇𝑔(1, 1, 3) = 〈2, 2, 6〉
are normal vectors to the two given surfaces respectively, at the point (1, 1, 3). Now a direction vector 𝐯 of the
required tangent line has to be perpendicular to both normal vectors, so we may take
𝐢 𝐣 𝐤
𝐯 = ∇𝑓(1, 1, 3) × ∇𝑔(1, 1, 3) = |13 13 −6| = 〈90, −90, 0〉.
2 2 6
Together with the point (1, 1, 3), we see that the required line has parametric equations
𝑥 = 1 + 90𝑡
{𝑦 = 1 − 90𝑡
𝑧=3
where 𝑡 ∈ (−∞, +∞). (Of course one may replace 90𝑡 by just 𝑡.)

Page 2 of 12
MATH2023 Multivariable Calculus Problem Set 4
L2/L3 (Fall 2019)

1 1 1 1
5. From the given formula = + + , we obtain
𝑅 𝑅1 𝑅2 𝑅3

1 𝜕 1 1 1 𝜕 1 1 1 𝜕 1 1 1
− 2
𝑑𝑅 = ( + + ) 𝑑𝑅1 + ( + + ) 𝑑𝑅2 + ( + + ) 𝑑𝑅3
𝑅 𝜕𝑅1 𝑅1 𝑅2 𝑅3 𝜕𝑅2 𝑅1 𝑅2 𝑅3 𝜕𝑅3 𝑅1 𝑅2 𝑅3

1 1 1
=− 2 𝑑𝑅1 − 2 𝑑𝑅2 − 𝑑𝑅3 .
𝑅1 𝑅2 𝑅3 2
This equation of differentials suggests the approximation
1 1 1 1
− 2
Δ𝑅 ≈ − 2 Δ𝑅1 − 2 Δ𝑅2 − 2 Δ𝑅3 .
𝑅 𝑅1 𝑅2 𝑅3
Now applying the given formula with the measurements 𝑅1 = 25, 𝑅2 = 40 and 𝑅3 = 50, we obtain
1 200
= , 𝑅=
1 1 1 17
+ +
25 40 50
and the maximum absolute error of this calculated 𝑅 is
𝑅2 𝑅2 𝑅2 𝑅2 Δ𝑅1 𝑅2 Δ𝑅2 𝑅2 Δ𝑅3
Δ𝑅 ≈ 2 Δ𝑅1 + 2 Δ𝑅2 + 2 Δ𝑅3 = ( ) + ( ) + ( )
𝑅1 𝑅2 𝑅3 𝑅1 ⏟𝑅1 𝑅2 ⏟𝑅2 𝑅3 ⏟𝑅3
percentage percentage percentage
error in 𝑅1 error in 𝑅2 error in 𝑅3

200 2 200 2 200 2

( ) ( ) ( ) 1
≈ 17 (0.5%) + 17 (0.5%) + 17 (0.5%) = .
25 40 50 17

6. Since 𝑓 is continuous on the given closed and bounded region, we look for the absolute extremum values of 𝑓 by
comparing the values of 𝑓 at critical points and boundary points.
 Critical points: Since
∇𝑓(𝑥, 𝑦) = 〈𝑦, 𝑥〉
is 𝟎 only when (𝑥, 𝑦) = (0, 0), this is the only critical point of 𝑓. At this critical point we have
𝑓(0, 0) = −2.
 Boundary points: The boundary ellipse 𝑥 + 4𝑦 2 = 8 can be parametrized by
2

𝑥 = √8 cos 𝑡
{
𝑦 = √2 sin 𝑡
where 𝑡 ∈ [0, 2𝜋]. Along this boundary ellipse we have 𝑓(𝑥, 𝑦) = 𝑓(√8 cos 𝑡 , √2 sin 𝑡), and if we let
ℎ(𝑡) = 𝑓(√8 cos 𝑡 , √2 sin 𝑡) = (√8 cos 𝑡)(√2 sin 𝑡) − 2
= 4 sin 𝑡 cos 𝑡 − 2 = 2 sin 2𝑡 − 2
for 𝑡 ∈ [0, 2𝜋], then
ℎ′ (𝑡) = 4 cos 2𝑡 ,
𝜋 3𝜋 5𝜋 7𝜋
and ℎ′ (𝑡) = 0 only when 𝑡 = or 𝑡 = or 𝑡 = or 𝑡 = . The values of 𝑓 at the points
4 4 4 4
corresponding to these four values of 𝑡 are
𝑓(2, 1) = 0, 𝑓(−2, 1) = −4, 𝑓(−2, −1) = 0 and 𝑓(2, −1) = −4.
Among all these five values, we find that the absolute maximum value of 𝑓 is 0, which occurs at (2, 1) and at
(−2, −1); and the absolute minimum value of 𝑓 is −4, which occurs at (−2, 1) and at (2, −1).

Page 3 of 12
MATH2023 Multivariable Calculus Problem Set 4
L2/L3 (Fall 2019)

Alternative solution: [One can also treat the boundary case using Lagrange’s multiplier.]
 Critical points: Since ∇𝑓(𝑥, 𝑦) = 〈𝑦, 𝑥〉 is 𝟎 only when (𝑥, 𝑦) = (0, 0), this is the only critical point of 𝑓.
At this critical point we have 𝑓(0, 0) = −2.
 Boundary points: The boundary ellipse can be described by the equality constraint
𝑔(𝑥, 𝑦) = 𝑥 2 + 4𝑦 2 − 8 = 0.
To find the possible local extrema of 𝑓 subject to this equality constraint, we solve the system

𝑓𝑥 (𝑥, 𝑦) = 𝜆𝑔𝑥 (𝑥, 𝑦) 𝑦 = 𝜆(2𝑥)

{𝑓𝑦 (𝑥, 𝑦) = 𝜆𝑔𝑦 (𝑥, 𝑦) , i.e. {𝑥 = 𝜆(8𝑦) .
𝑔(𝑥, 𝑦) = 0 𝑥 2 + 4𝑦 2 = 8

From the first two equations we get 𝑥 = 8𝜆(2𝜆𝑥). Since 𝑥 ≠ 0 (otherwise if 𝑥 = 0 then the first equation
1
would imply 𝑦 = 0, and the third equation would become 0 = 8 which is absurd), we have 𝜆2 = .
16

1 𝑥 𝑥 2
If 𝜆 = , then the first equation gives 𝑦 = , so the third equation becomes 𝑥 2 + 4 ( ) = 8, which gives two
4 2 2

1 𝑥
solutions (𝑥, 𝑦) = (2, 1) or (−2, −1). If 𝜆 = − , then the first equation gives 𝑦 = − , so the third
4 2

𝑥 2
equation becomes 𝑥 2 + 4 (− ) = 8, which gives two other solutions (𝑥, 𝑦) = (2, −1) or (−2, 1). The
2

values of 𝑓 at these four points are

𝑓(2, 1) = 0, 𝑓(−2, 1) = −4, 𝑓(−2, −1) = 0 and 𝑓(2, −1) = −4.
Among all these five values, we find that the absolute maximum value of 𝑓 is 0, which occurs at (2, 1) and at
(−2, −1); and the absolute minimum value of 𝑓 is −4, which occurs at (−2, 1) and at (2, −1).

1 1
7. (a) We have ∇𝑓(𝑥, 𝑦) = 〈𝑦 − ,𝑥 − 〉. Since ∇𝑓(𝑥, 𝑦) = 𝟎 only when (𝑥, 𝑦) = (1, 1), this is the only critical
𝑥2 𝑦2

point of 𝑓. Now

𝑓𝑥𝑥 (𝑥, 𝑦) = 2/𝑥 3

{𝑓𝑥𝑦 (𝑥, 𝑦) = 1 .
𝑓𝑦𝑦 (𝑥, 𝑦) = 2/𝑦 3

Since 𝑓𝑥𝑥 (1, 1) = 2 > 0 and 𝐷(1, 1) = (2)(2) − (1)2 = 3 > 0, 𝑓 has a local minimum at (1, 1).

(b) We have ∇𝑓(𝑥, 𝑦) = 〈3𝑥 2 − 𝑦 2 − 2𝑥, 2𝑦(1 − 𝑥)〉. Now ∇𝑓(𝑥, 𝑦) = 𝟎 if the system

3𝑥 2 − 𝑦 2 − 2𝑥 = 0
{
2𝑦(1 − 𝑥) = 0

holds. The second equation gives either 𝑦 = 0 or 𝑥 = 1.

2
 If 𝑦 = 0, then the first equation becomes 3𝑥 2 − 2𝑥 = 0, which gives 𝑥 = 0 or 𝑥 = .
3
2
 If 𝑥 = 1, then the first equation becomes 1 − 𝑦 = 0, which gives 𝑦 = 1 or 𝑦 = −1.
2
Therefore the only critical points of 𝑓 are the four points (0, 0), ( , 0), (1, 1) and (1, −1). Now
3

𝑓𝑥𝑥 (𝑥, 𝑦) = 6𝑥 − 2
{𝑓𝑥𝑦 (𝑥, 𝑦) = −2𝑦 .
𝑓𝑦𝑦 (𝑥, 𝑦) = 2 − 2𝑥

Page 4 of 12
MATH2023 Multivariable Calculus Problem Set 4
L2/L3 (Fall 2019)

 Since 𝐷(0, 0) = (−2)(2) − (0)2 = −4 < 0, 𝑓 has a saddle point at (0, 0).
2 2 2 4 2
 Since 𝑓𝑥𝑥 ( , 0) = 2 > 0 and 𝐷 ( , 0) = (2) ( ) − (0)2 = > 0, 𝑓 has a local minimum at ( , 0).
3 3 3 3 3
 Since 𝐷(1, 1) = (4)(0) − (−2)2 = −4 < 0, 𝑓 has a saddle point at (1, 1).
 Since 𝐷(1, −1) = (4)(0) − (2)2 = −4 < 0, 𝑓 has a saddle point at (1, −1).

2 +𝑦 2 ) 2 +𝑦 2 ) 2 +𝑦 2 )
(c) We have ∇𝑓(𝑥, 𝑦) = 〈[1 − 2𝑥(𝑥 − 𝑦)]𝑒 −(𝑥 , [−1 − 2𝑦(𝑥 − 𝑦)]𝑒 −(𝑥 〉. Since 𝑒 −(𝑥 is never
zero, ∇𝑓(𝑥, 𝑦) = 𝟎 if the system

1 − 2𝑥(𝑥 − 𝑦) = 0
{
−1 − 2𝑦(𝑥 − 𝑦) = 0

holds. Adding the two equations we get 𝑥 2 = 𝑦 2 , so either 𝑥 = 𝑦 or 𝑥 = −𝑦. But 𝑥 = 𝑦 is impossible
since the equations will become 1 = 0 and −1 = 0 which are absurd, so we must have 𝑥 = −𝑦. The first
1 1
equation now becomes 1 − 2𝑥(2𝑥) = 0, so 𝑥 = or 𝑥 = − . Therefore the only critical points of 𝑓 are
2 2
1 1 1 1
( , − ) and (− , ). We have
2 2 2 2
2 2
𝑓𝑥𝑥 (𝑥, 𝑦) = [−6𝑥 + 2𝑦 + 4𝑥 2 (𝑥 − 𝑦)]𝑒 −(𝑥 +𝑦 )
2 2
{𝑓𝑥𝑦 (𝑥, 𝑦) = [2𝑥 − 2𝑦 + 4𝑥𝑦(𝑥 − 𝑦)]𝑒 −(𝑥 +𝑦 ) .
2 +𝑦 2 )
𝑓𝑦𝑦 (𝑥, 𝑦) = [−2𝑥 + 6𝑦 + 4𝑦 2 (𝑥 − 𝑦)]𝑒 −(𝑥

1 1 1 1 2
1 1 1 1
 Since 𝑓𝑥𝑥 ( , − ) = −3𝑒 −2 < 0 and 𝐷 ( , − ) = (−3𝑒 −2 ) (−3𝑒 −2 ) − (𝑒 −2 ) = 8𝑒 −1 > 0 , 𝑓 has a
2 2 2 2

1 1
local maximum at ( , − ).
2 2

1 1 1 1 2
1 1 1 1
 Since 𝑓𝑥𝑥 (− , ) = 3𝑒 −2 > 0 and 𝐷 (− , ) = (3𝑒 −2 ) (3𝑒 −2 ) − (−𝑒 −2 ) = 8𝑒 −1 > 0, 𝑓 has a local
2 2 2 2

1 1
minimum at (− , ).
2 2

8. Let 𝑔(𝑥, 𝑦) be the square of the distance between (3, 0, 0) and the point (𝑥, 𝑦, 𝑓(𝑥, 𝑦)) on the graph of 𝑓, i.e.
𝑔(𝑥, 𝑦) ≔ (𝑥 − 3)2 + (𝑦 − 0)2 + (𝑓(𝑥, 𝑦) − 0)2
= (𝑥 − 3)2 + 𝑦 2 + (𝑥 2 + 𝑦 2 )2
for every (𝑥, 𝑦) ∈ ℝ2 . Our goal is to find the absolute minimum of 𝑔 (if it has one).

The domain of 𝑔 is the whole plane ℝ2 , which has no boundary points. So if 𝑔 were to have an absolute
minimum, it must be a critical point. Now
∇𝑔(𝑥, 𝑦) = 〈𝑔𝑥 (𝑥, 𝑦), 𝑔𝑦 (𝑥, 𝑦)〉 = 〈4𝑥 3 + 4𝑥𝑦 2 + 2𝑥 − 6, 2𝑦(2𝑥 2 + 2𝑦 2 + 1)〉,
so ∇𝑔(𝑥, 𝑦) = 𝟎 only when 𝑦 = 0 and 𝑥 = 1, and so (1, 0) is the only critical point of 𝑔. The second partial
derivatives of 𝑔 are

𝑔𝑥𝑥 (𝑥, 𝑦) = 12𝑥 2 + 4𝑦 2 + 2

{𝑔𝑥𝑦 (𝑥, 𝑦) = 8𝑥𝑦 ,
𝑔𝑦𝑦 (𝑥, 𝑦) = 4𝑥 2 + 12𝑦 2 + 2

so at the critical point (1, 0) we have 𝑔𝑥𝑥 (1, 0) = 14 > 0 and

𝑔𝑥𝑥 𝑔𝑥𝑦 2
𝐷(1, 0) = |𝑔
𝑥𝑦 𝑔𝑦𝑦 | (1, 0) = (14)(6) − 0 = 84 > 0.

Page 5 of 12
MATH2023 Multivariable Calculus Problem Set 4
L2/L3 (Fall 2019)

Thus 𝑔 attains local minimum at (1, 0) by the second derivative test, and the local minimum value is
𝑔(1, 0) = (1 − 3)2 + 02 + (12 + 02 )2 = 5.

Finally we divide the whole plane ℝ2 into two parts, the closed disk 𝑥 2 + 𝑦 2 ≤ 3 and its exterior 𝑥 2 + 𝑦 2 > 3.
 The closed disk 𝑥 2 + 𝑦 2 ≤ 3 is a closed and bounded set, so 𝑔 has an absolute minimum inside it. At
2 2
each boundary point (𝑥, 𝑦) of the disk we have 𝑥 + 𝑦 = 3, so 𝑔(𝑥, 𝑦) ≥ 9 > 𝑔(1, 0). Therefore 𝑔
attains absolute minimum (inside the disk) at (1, 0).
 For every (𝑥, 𝑦) in the exterior of the disk, since 𝑥 2 + 𝑦 2 > 3, we still have 𝑔(𝑥, 𝑦) > 9 > 𝑔(1, 0).
Therefore 𝑔 attains absolute minimum (over the whole plane ℝ2 ) at (1, 0), and the shortest distance from the
point (3, 0, 0) to the graph of 𝑓 is √𝑔(1, 0) = √5.

9. Each of the given functions 𝑓 has a critical point at (0, 0), at which 𝑓(0, 0) = 0.
(a) Since 𝑓(𝑥, 𝑦) = 𝑥 2 𝑦 2 ≥ 0 for every (𝑥, 𝑦), it follows that 𝑓 attains absolute minimum at (0, 0). (In
particular, it is a local minimum of course.)
(b) For every non-zero real number 𝑡 near 0, we have 𝑓(𝑡, 𝑡) = 𝑡 6 > 0 but 𝑓(𝑡, −𝑡) = −𝑡 6 < 0, so it follows
that 𝑓 has a saddle point at (0, 0).
(c) Since 𝑓(𝑥, 𝑦) = −𝑥 4 𝑦 4 ≤ 0 for every (𝑥, 𝑦), it follows that 𝑓 attains absolute maximum at (0, 0). (In
particular, it is a local maximum of course.)
(d) For every non-zero real number 𝑡 near 0, we have 𝑓(𝑡, 0) = 𝑡 2 > 0 but 𝑓(0, 𝑡) = −𝑡 4 < 0, so it follows
that 𝑓 has a saddle point at (0, 0).

2 +𝑦 2 ) 2 +𝑦 2 ) 2 +𝑦 2 )
10. We have ∇𝑓(𝑥, 𝑦) = 〈2𝑥(1 − 𝑥 2 − 𝑦 2 )𝑒 −(𝑥 , 2𝑦(1 − 𝑥 2 − 𝑦 2 )𝑒 −(𝑥 〉. Since 𝑒 −(𝑥 is never zero,
∇𝑓(𝑥, 𝑦) = 𝟎 if the system

2𝑥(1 − 𝑥 2 − 𝑦 2 ) = 0
{
2𝑦(1 − 𝑥 2 − 𝑦 2 ) = 0

holds.
 If 𝑥 2 + 𝑦 2 = 1, then both equations automatically hold.
 If 𝑥 2 + 𝑦 2 ≠ 1, then the equations become 2𝑥 = 0 and 2𝑦 = 0, so 𝑥 = 𝑦 = 0.
Therefore the only critical points of 𝑓 are (0, 0) and all the points on the unit circle 𝑥 2 + 𝑦 2 = 1. Now
2 2
𝑓𝑥𝑥 (𝑥, 𝑦) = [(2 − 4𝑥 2 )(1 − 𝑥 2 − 𝑦 2 ) − 4𝑥 2 ]𝑒 −(𝑥 +𝑦 )
2 2
{𝑓𝑥𝑦 (𝑥, 𝑦) = [−4𝑥𝑦 − 4𝑥𝑦(1 − 𝑥 2 − 𝑦 2 )]𝑒 −(𝑥 +𝑦 ) .
2 +𝑦 2 )
𝑓𝑦𝑦 (𝑥, 𝑦) = [(2 − 4𝑦 2 )(1 − 𝑥 2 − 𝑦 2 ) − 4𝑦 2 ]𝑒 −(𝑥
 Since 𝑓𝑥𝑥 (0, 0) = 2 > 0 and 𝐷(0, 0) = (2)(2) − (0)2 = 4 > 0, 𝑓 has a local minimum at (0, 0).
 For each point (𝑥, 𝑦) on the circle 𝑥 2 + 𝑦 2 = 1, we have
𝐷(𝑥, 𝑦) = (−4𝑥 2 𝑒 −1 )(−4𝑦 2 𝑒 −1 ) − [(−4𝑥𝑦)𝑒 −1 ]2 = 0,
so the second derivative test cannot be applied here .

Page 6 of 12
MATH2023 Multivariable Calculus Problem Set 4
L2/L3 (Fall 2019)

However, we may observe that the values of 𝒇 are constant along each circle centered at the origin. In
2 2 −𝑟
particular, for each 𝑟 ≥ 0, if 𝑥 + 𝑦 = 𝑟, then 𝑓(𝑥, 𝑦) = 𝑟𝑒 . Now let 𝑔: [0, +∞) → ℝ be the function of
one variable defined by
𝑔(𝑟) = 𝑟𝑒 −𝑟 .
Then 𝑔′ (𝑟) = (1 − 𝑟)𝑒 −𝑟 = 0 only when 𝑟 = 1, and 𝑔′′ (1) = −𝑒 −1 < 0, so 𝑔 attains local maximum at 1.
Moreover, since 𝑔(0) = 0, 𝑔(𝑟) ≥ 0 for every 𝑟, and lim 𝑔(𝑟) = 0, we see that 𝑔 attains absolute maximum
𝑟→+∞
at 1 and attains absolute minimum at 0. Therefore 𝑓 attains absolute maximum on every point on the circle
2 2
𝑥 + 𝑦 = 1, and attains absolute minimum at (0, 0).

11. We have
𝑓𝑚 (𝑚, 𝑏) = 2(𝑚𝑥1 + 𝑏 − 𝑦1 )𝑥1 + 2(𝑚𝑥2 + 𝑏 − 𝑦2 )𝑥2 + ⋯ + 2(𝑚𝑥𝑛 + 𝑏 − 𝑦𝑛 )𝑥𝑛
= 2(𝑥1 2 + 𝑥2 2 + ⋯ + 𝑥𝑛 2 )𝑚 + 2(𝑥1 + 𝑥2 + ⋯ + 𝑥𝑛 )𝑏 − 2(𝑥1 𝑦1 + 𝑥2 𝑦2 + ⋯ + 𝑥𝑛 𝑦𝑛 )
and
𝑓𝑏 (𝑚, 𝑏) = 2(𝑚𝑥1 + 𝑏 − 𝑦1 ) + 2(𝑚𝑥2 + 𝑏 − 𝑦2 ) + ⋯ + 2(𝑚𝑥𝑛 + 𝑏 − 𝑦𝑛 )
= 2(𝑥1 + 𝑥2 + ⋯ + 𝑥𝑛 )𝑚 + 2𝑛𝑏 − 2(𝑦1 + 𝑦2 + ⋯ + 𝑦𝑛 ).
For simplicity let’s denote
𝑊 ≔ 𝑥1 2 + 𝑥2 2 + ⋯ + 𝑥𝑛 2
𝑋 ≔ 𝑥1 + 𝑥2 + ⋯ + 𝑥𝑛
{
𝑌 ≔ 𝑦1 + 𝑦2 + ⋯ + 𝑦𝑛
𝑍 ≔ 𝑥1 𝑦1 + 𝑥2 𝑦2 + ⋯ + 𝑥𝑛 𝑦𝑛
(these are just fixed given numbers). Now ∇𝑓(𝑚, 𝑏) = 〈2𝑊𝑚 + 2𝑋𝑏 − 2𝑍, 2𝑋𝑚 + 2𝑛𝑏 − 2𝑌〉 = 𝟎 if the system

2𝑊𝑚 + 2𝑋𝑏 − 2𝑍 = 0
{
2𝑋𝑚 + 2𝑛𝑏 − 2𝑌 = 0

holds. Note that

𝑛𝑊 − 𝑋 2 = (1 + 1 + ⋯ + 1)(𝑥1 2 + 𝑥2 2 + ⋯ + 𝑥𝑛 2 ) − (𝑥1 + 𝑥2 + ⋯ + 𝑥𝑛 )2 > 0
by Cauchy-Schwarz inequality, so on solving the above system we get

𝑛𝑍 − 𝑋𝑌 𝑊𝑌 − 𝑋𝑍
𝑚= and 𝑏= ,
𝑛𝑊 − 𝑋 2 𝑛𝑊 − 𝑋 2
𝑛𝑍−𝑋𝑌 𝑊𝑌−𝑋𝑍
and so ( , ) is the only critical point of 𝑓. Now for every (𝑚, 𝑏) ∈ ℝ2 , we have
𝑛𝑊−𝑋 2 𝑛𝑊−𝑋 2

𝑓𝑚𝑚 (𝑚, 𝑏) = 2𝑊
{𝑓𝑚𝑏 (𝑚, 𝑏) = 2𝑋 .
𝑓𝑏𝑏 (𝑚, 𝑏) = 2𝑛

Since 𝑓𝑚𝑚 (𝑚, 𝑏) = 2𝑊 = 2(𝑥1 2 + 𝑥2 2 + ⋯ + 𝑥𝑛 2 ) > 0 and 𝐷(𝑚, 𝑏) = (2𝑊)(2𝑛) − (2𝑋)2 = 4(𝑛𝑊 − 𝑋 2 ) > 0
𝑛𝑍−𝑋𝑌 𝑊𝑌−𝑋𝑍
for (𝑚, 𝑏) ∈ ℝ2 , it follows in particular that the critical point ( , ) is a local minimum of 𝑓. In terms
𝑛𝑊−𝑋 2 𝑛𝑊−𝑋 2
of the given numbers 𝑥1 , 𝑥2 , … , 𝑥𝑛 , this local minimum occurs at
𝑛 ∑𝑛𝑘=1 𝑥𝑘 𝑦𝑘 − (∑𝑛𝑘=1 𝑥𝑘 )(∑𝑛𝑘=1 𝑦𝑘 ) (∑𝑛𝑘=1 𝑥𝑘 2 )(∑𝑛𝑘=1 𝑦𝑘 ) − (∑𝑛𝑘=1 𝑥𝑘 )(∑𝑛𝑘=1 𝑥𝑘 𝑦𝑘 )
𝑚= and 𝑏= .
𝑛 ∑𝑛𝑘=1 𝑥𝑘 2 − (∑𝑛𝑘=1 𝑥𝑘 )2 𝑛 ∑𝑛𝑘=1 𝑥𝑘 2 − (∑𝑛𝑘=1 𝑥𝑘 )2

Page 7 of 12
MATH2023 Multivariable Calculus Problem Set 4
L2/L3 (Fall 2019)

12. (a) The statement is true.

If 𝑓 is continuous at (𝑎, 𝑏), then the limit lim 𝑓(𝑥, 𝑦) exists, the value 𝑓(𝑎, 𝑏) is well-defined, and
(𝑥,𝑦)→(𝑎,𝑏)
the limit equals the value, i.e. lim 𝑓(𝑥, 𝑦) = 𝑓(𝑎, 𝑏).
(𝑥,𝑦)→(𝑎,𝑏)

(b) The statement is false.

Even if 𝑓 is discontinuous at (𝑎, 𝑏), lim 𝑓(𝑥, 𝑦) may still exist. As long as the value 𝑓(𝑎, 𝑏) is either
(𝑥,𝑦)→(𝑎,𝑏)
undefined or does not equal to the limit lim 𝑓(𝑥, 𝑦), we still have 𝑓 to be discontinuous at (𝑎, 𝑏).
(𝑥,𝑦)→(𝑎,𝑏)

(c) The statement is false.

It may happen that (𝑎, 𝑏) is a saddle point of 𝑓.

(d) The statement is false.

It may happen that (𝑎, 𝑏) is not even a critical point of 𝑓. If we assumed further that 𝑓𝑥 (𝑎, 𝑏) and 𝑓𝑦 (𝑎, 𝑏)

2
are both zero, then the statement would be true because 𝐷(𝑎, 𝑏) = 𝑓𝑥𝑥 (𝑎, 𝑏)𝑓𝑦𝑦 (𝑎, 𝑏) − (𝑓𝑥𝑦 (𝑎, 𝑏)) < 0.

13. (a) Let 𝑔(𝑥, 𝑦, 𝑧) = 𝑥 2 + 2𝑦 2 + 4𝑧 2 − 9. To find the possible local extrema of 𝑓 subject to 𝑔(𝑥, 𝑦, 𝑧) = 0, we
solve the system
𝑓𝑥 (𝑥, 𝑦, 𝑧) = 𝜆𝑔𝑥 (𝑥, 𝑦, 𝑧) 𝑦𝑧 = 2𝜆𝑥
𝑓𝑦 (𝑥, 𝑦, 𝑧) = 𝜆𝑔𝑦 (𝑥, 𝑦, 𝑧) 𝑥𝑧 = 4𝜆𝑦
, i.e. { .
𝑓𝑧 (𝑥, 𝑦, 𝑧) = 𝜆𝑔𝑧 (𝑥, 𝑦, 𝑧) 𝑥𝑦 = 8𝜆𝑧
{𝑔(𝑥, 𝑦, 𝑧) = 0 𝑥 2 + 2𝑦 2 + 4𝑧 2 = 9

3 3
 If 𝑥𝑦 = 8𝜆𝑧 = 0, then we obtain the six solutions (𝑥, 𝑦, 𝑧) = (±3, 0, 0) or (0, ± , 0) or (0, 0, ± ).
√2 2

The values of 𝑓 at each of these six points are all 0.

 If 𝑥𝑦 = 8𝜆𝑧 ≠ 0, then dividing the first two equations by the third gives
𝑦𝑧 2𝜆𝑥 𝑥𝑧 4𝜆𝑦
= and = ,
𝑥𝑦 8𝜆𝑧 𝑥𝑦 8𝜆𝑧

so 𝑥 2 = 4𝑧 2 and 𝑦 2 = 2𝑧 2 , and the fourth equation gives

4𝑧 2 + 4𝑧 2 + 4𝑧 2 = 9

√3 √3 3 √3
so 𝑧 = or − . From this we obtain eight solutions (𝑥, 𝑦, 𝑧) = (±√3, ±√ , ± ), and the values
2 2 2 2

3 3 3 3
of 𝑓 at each of these eight points are either √ or − √ .
2 2 2 2

Since 𝑓 is continuous on the ellipsoid 𝑥 2 + 2𝑦 2 + 4𝑧 2 = 9 which is a closed and bounded set in ℝ3 , we

3 3 3 3
conclude that on this ellipsoid 𝑓 has absolute maximum value √ and absolute minimum value − √ .
2 2 2 2

Page 8 of 12
MATH2023 Multivariable Calculus Problem Set 4
L2/L3 (Fall 2019)

(b) Let 𝑔(𝑥, 𝑦) = 𝑥 2 + 𝑦 2 − 3𝑥𝑦 − 20. To find the possible local extrema of 𝑓 subject to the constraint
𝑔(𝑥, 𝑦) = 0, we solve the system

𝑓𝑥 (𝑥, 𝑦) = 𝜆𝑔𝑥 (𝑥, 𝑦) 1 = 𝜆(2𝑥 − 3𝑦)

{𝑓𝑦 (𝑥, 𝑦) = 𝜆𝑔𝑦 (𝑥, 𝑦) , i.e. {−1 = 𝜆(2𝑦 − 3𝑥) .
𝑔(𝑥, 𝑦) = 0 𝑥 2 + 𝑦 2 − 3𝑥𝑦 = 20

From the first two equations we obtain 2𝑥 − 3𝑦 = −(2𝑦 − 3𝑥), so 𝑦 = −𝑥. The third equation then gives
𝑥 2 + (−𝑥)2 − 3𝑥(−𝑥) = 20, 𝑓(𝑥, 𝑦) = −12
𝑓(𝑥, 𝑦) = −4
so 𝑥 = 2 or −2 . Accordingly we have (𝑥, 𝑦) = (2, −2) or
𝑓(𝑥, 𝑦) = 4
(−2, 2). The values of 𝑓 at these two points are 𝑓(𝑥, 𝑦) = 12
𝑓(2, −2) = 4 and 𝑓(−2, 2) = −4.
In fact 𝑓 attains local maximum at (−2, 2) and local minimum at 𝑥 2 + 𝑦 2 = 20 + 3𝑥𝑦

(2, −2) subject to the constraint, and 𝑓 has no absolute extrema.

Remark: Note that in this problem we cannot simply compare the two values (𝑓(−2, 2) < 𝑓(2, −2)) to
conclude that 𝑓 attains local minimum at (−2, 2) and local maximum at (2, −2) subject to
the given equality constraint. This is because the given constraint 𝑥 2 + 𝑦 2 = 20 + 3𝑥𝑦
defines a hyperbola in ℝ2 , which is not a bounded set.

(c) Let 𝑔(𝑥1 , 𝑥2 , … , 𝑥𝑛 ) = 𝑥1 2 + 𝑥2 2 + ⋯ + 𝑥𝑛 2 − 1. To find the possible local extrema of 𝑓 subject to the
equality constraint 𝑔(𝑥1 , 𝑥2 , … , 𝑥𝑛 ) = 0, we solve the system
𝑓𝑥1 (𝑥1 , 𝑥2 , … , 𝑥𝑛 ) = 𝜆𝑔𝑥1 (𝑥1 , 𝑥2 , … , 𝑥𝑛 ) 1 = 2𝜆𝑥1
𝑓𝑥2 (𝑥1 , 𝑥2 , … , 𝑥𝑛 ) = 𝜆𝑔𝑥2 (𝑥1 , 𝑥2 , … , 𝑥𝑛 ) 2 = 2𝜆𝑥2
⋮ , i.e. ⋮ .
𝑓𝑥𝑛 (𝑥1 , 𝑥2 , … , 𝑥𝑛 ) = 𝜆𝑔𝑥𝑛 (𝑥1 , 𝑥2 , … , 𝑥𝑛 ) 𝑛 = 2𝜆𝑥𝑛
{𝑔(𝑥1 , 𝑥2 , … , 𝑥𝑛 ) = 0 {𝑥1 2 + 𝑥2 2 + ⋯ + 𝑥𝑛 2 = 1
1 2 𝑛
The first 𝑛 equations give 𝑥1 = , 𝑥2 = , …, 𝑥𝑛 = , so the last equation gives
2𝜆 2𝜆 2𝜆

1 2 2 2 𝑛 2
( ) + ( ) + ⋯ + ( ) = 1,
2𝜆 2𝜆 2𝜆
i.e. 2𝜆 = √12 + 22 + ⋯ + 𝑛2 or −√12 + 22 + ⋯ + 𝑛2 . From this we get two solutions
1 2 𝑛
(𝑥1 , 𝑥2 , … , 𝑥𝑛 ) = (± ,± ,…,± )
√12 + 22 +⋯+ 𝑛2 √12 + 22 + ⋯+ 𝑛2 √12 + 22 + ⋯ + 𝑛2
(either all signs are + or all signs are −). The values of 𝑓 at these points are
1 2 𝑛
𝑓(𝑥1 , 𝑥2 , … , 𝑥𝑛 ) = ± (1 ⋅ +2⋅ + ⋯+ )
√12 + 22 + ⋯ + 𝑛2 √12 + 22 + ⋯ + 𝑛2 √12 + 22 + ⋯ + 𝑛2

𝑛(𝑛 + 1)(2𝑛 + 1)
= ± √ 12 + 22 + ⋯ + 𝑛 2 = ± √ .
6
Since 𝑓 is continuous on the “(𝑛 − 1)-sphere” 𝑥1 2 + 𝑥2 2 + ⋯ + 𝑥𝑛 2 = 1 which is a closed and bounded set

𝑛(𝑛+1)(2𝑛+1) 𝑛(𝑛+1)(2𝑛+1)
in ℝ𝑛 , we conclude that 𝑓 has maximum value √ and minimum value −√ there.
6 6

Page 9 of 12
MATH2023 Multivariable Calculus Problem Set 4
L2/L3 (Fall 2019)

(d) Let 𝑔(𝑥1 , 𝑥2 , … , 𝑥𝑛 ) = 𝑏1 𝑥1 + 𝑏2 𝑥2 + ⋯ + 𝑏𝑛 𝑥𝑛 − 𝑐. To find the possible local extrema of 𝑓 subject to the
equality constraint 𝑔(𝑥1 , 𝑥2 , … , 𝑥𝑛 ) = 0, we solve the system
𝑓𝑥1 (𝑥1 , 𝑥2 , … , 𝑥𝑛 ) = 𝜆𝑔𝑥1 (𝑥1 , 𝑥2 , … , 𝑥𝑛 ) 2𝑎1 2 𝑥1 = 𝑏1 𝜆
𝑓𝑥2 (𝑥1 , 𝑥2 , … , 𝑥𝑛 ) = 𝜆𝑔𝑥2 (𝑥1 , 𝑥2 , … , 𝑥𝑛 ) 2𝑎2 2 𝑥2 = 𝑏2 𝜆
⋮ , i.e. ⋮ .
𝑓𝑥𝑛 (𝑥1 , 𝑥2 , … , 𝑥𝑛 ) = 𝜆𝑔𝑥𝑛 (𝑥1 , 𝑥2 , … , 𝑥𝑛 ) 2𝑎𝑛 2 𝑥𝑛 = 𝑏𝑛 𝜆
{𝑔(𝑥1 , 𝑥2 , … , 𝑥𝑛 ) = 0 {𝑏1 𝑥1 + 𝑏2 𝑥2 + ⋯ + 𝑏𝑛 𝑥𝑛 = 𝑐

Substituting the first 𝑛 equations into the last one, we obtain

𝑏1 𝜆 𝑏2 𝜆 𝑏𝑛 𝜆
𝑏1 ( 2
) + 𝑏2 ( 2
) + ⋯ + 𝑏𝑛 ( ) = 𝑐,
2𝑎1 2𝑎2 2𝑎𝑛 2
2𝑐
so 𝜆 = 𝑏 2 𝑏 2
𝑏 2 . From this we obtain the solution
( 1 ) +( 2 ) +⋯+( 𝑛 )
𝑎1 𝑎2 𝑎𝑛

𝑏1 𝑏𝑛
𝑐 𝑐
𝑎1 2 𝑎𝑛 2
(𝑥1 , 𝑥2 , … , 𝑥𝑛 ) = ,…,
𝑏 2 𝑏 2 𝑏 2 𝑏 2 𝑏 2 𝑏 2
( 1) + ( 2) + ⋯ + ( 𝑛) ( 1) + ( 2) + ⋯ + ( 𝑛)
( 𝑎1 𝑎2 𝑎𝑛 𝑎1 𝑎2 𝑎𝑛 )
and thus the minimum value of 𝑓 subject to the given constraint is
2 2
𝑏1 𝑏𝑛
𝑐 𝑐 𝑐2
𝑎1 𝑎𝑛
2 +⋯+ 2 = .
𝑏1 2 𝑏2 2 𝑏𝑛 𝑏1 2 𝑏2 2 𝑏𝑛 𝑏 2 𝑏 2 𝑏 2
[(𝑎1 ) + (𝑎2 ) + ⋯ + (𝑎𝑛 ) ] [(𝑎1 ) + (𝑎2 ) + ⋯ + (𝑎𝑛 ) ] ( 1) + ( 2) + ⋯ + ( 𝑛)
𝑎1 𝑎2 𝑎𝑛
𝑓 has no maxima.

Remark: To classify the critical point in (d) to be a local minimum, we need a general 𝑛-variable version
of the second derivative test which involves concepts from linear algebra:
Suppose that (𝑥1 , 𝑥2 , … , 𝑥𝑛 ) is a critical point of an 𝑛-variable function 𝑓. Then we consider
the Hessian matrix of 𝑓, which is the 𝑛 × 𝑛 matrix given by
𝑓𝑥1𝑥1 𝑓𝑥1𝑥2 ⋯ 𝑓𝑥1𝑥𝑛
𝑓 𝑓𝑥2𝑥2 ⋯ 𝑓𝑥2𝑥𝑛
𝐻 ≔ 𝑥2𝑥1 .
⋮ ⋮ ⋱ ⋮
[𝑓𝑥𝑛 𝑥1 𝑓𝑥𝑛 𝑥2 ⋯ 𝑓𝑥𝑛 𝑥𝑛 ]
(i) If 𝐻(𝑥1 , 𝑥2 , … , 𝑥𝑛 ) is positive definite (i.e. all eigenvalues of 𝐻(𝑥1 , 𝑥2 , … , 𝑥𝑛 ) are
positive), then (𝑥1 , 𝑥2 , … , 𝑥𝑛 ) is a local minimum;
(ii) If 𝐻(𝑥1 , 𝑥2 , … , 𝑥𝑛 ) is negative definite (i.e. all eigenvalues of 𝐻(𝑥1 , 𝑥2 , … , 𝑥𝑛 ) are
negative), then (𝑥1 , 𝑥2 , … , 𝑥𝑛 ) is a local maximum;
(iii) If 𝐻(𝑥1 , 𝑥2 , … , 𝑥𝑛 ) is indefinite (i.e. 𝐻(𝑥1 , 𝑥2 , … , 𝑥𝑛 ) has both positive and negative
eigenvalues), then (𝑥1 , 𝑥2 , … , 𝑥𝑛 ) is a saddle point;
(iv) Otherwise this test is inconclusive.
In (d), the Hessian matrix of 𝑓 is the constant matrix 𝐻 = diag(2𝑎1 2 , 2𝑎2 2 , … , 2𝑎𝑛 2 ), whose
eigenvalues 2𝑎1 2 , 2𝑎2 2 , … , 2𝑎𝑛 2 are all positive.

Page 10 of 12
MATH2023 Multivariable Calculus Problem Set 4
L2/L3 (Fall 2019)

14. (a) Let 𝑔(𝑥, 𝑦, 𝑧) = 𝑥 + 𝑦 + 𝑧 − 𝑎. To find the possible local extrema of 𝑓 subject to the equality constraint
𝑔(𝑥, 𝑦, 𝑧) = 0, we solve the system
𝑓𝑥 (𝑥, 𝑦, 𝑧) = 𝜆𝑔𝑥 (𝑥, 𝑦, 𝑧) 𝑦𝑧 = 𝜆
𝑓𝑦 (𝑥, 𝑦, 𝑧) = 𝜆𝑔𝑦 (𝑥, 𝑦, 𝑧) 𝑥𝑧 = 𝜆
, i.e. { .
𝑓𝑧 (𝑥, 𝑦, 𝑧) = 𝜆𝑔𝑧 (𝑥, 𝑦, 𝑧) 𝑥𝑦 = 𝜆
{𝑔(𝑥, 𝑦, 𝑧) = 0 𝑥+𝑦+𝑧 =𝑎
 If 𝜆 = 0, then we obtain (𝑥, 𝑦, 𝑧) = (𝑎, 0, 0) or (0, 𝑎, 0) or (0, 0, 𝑎). The value of 𝑓 at each of these
three points are all 0.
 If 𝜆 ≠ 0, then dividing the third equation from the first two we get 𝑥 = 𝑦 = 𝑧, so the last equation gives
𝑎
𝑥=𝑦=𝑧= . The value of 𝑓 at this point is
3

𝑎 𝑎 𝑎 𝑎 3
𝑓( , , ) = ( ) .
3 3 3 3
Since 𝑓 is continuous on the (triangular) portion of the plane 𝑥 + 𝑦 + 𝑧 = 𝑎 in the closed first octant, which

𝑎 3
is a closed and bounded set, it has maximum value ( ) (and minimum value 0) on such a set.
3

Alternative solution: [Without using Lagrange’s multipliers]

Let ℎ(𝑥, 𝑦) = 𝑥𝑦(𝑎 − 𝑥 − 𝑦). We find the maximum value of ℎ in the closed triangular region
𝑅 = {(𝑥, 𝑦) ∈ ℝ2 : 𝑥 ≥ 0 and 𝑦 ≥ 0 and 𝑥 + 𝑦 ≤ 𝑎}.
Since ℎ is continuous on the closed and bounded region 𝑅, we look for the absolute maximum of ℎ by
simply comparing the values of ℎ at critical points and boundary points.
 Critical points: The gradient vector
∇ℎ(𝑥, 𝑦) = 〈𝑦(𝑎 − 2𝑥 − 𝑦), 𝑥(𝑎 − 𝑥 − 2𝑦)〉
𝑎 𝑎 𝑎 𝑎
is 𝟎 only when (𝑥, 𝑦) = (0, 0), (0, 𝑎), (𝑎, 0) or ( , ), among which only ( , ) is an interior point of
3 3 3 3
𝑎 𝑎
𝑅. So ( , ) is the only critical point of ℎ in the interior of 𝑅, and at this critical point we have
3 3

𝑎 𝑎 𝑎 3
ℎ( , ) = ( ) .
3 3 3
 Boundary points: The boundary of 𝑅 consists of three line segments. Along the line segment joining
(0, 0) and (0, 𝑎), we consider
𝑔(𝑡) ≔ ℎ(0, 𝑡) = 0 ⋅ 𝑡 ⋅ (𝑎 − 0 − 𝑡) = 0
for 𝑡 ∈ [0, 𝑎]. 𝑔 is the constant function 0, so ℎ always takes the value 0 on this line segment.
Similarly along the line segment joining (0, 0) and (𝑎, 0), ℎ always takes the value 0 also. Finally
along the line segment joining (𝑎, 0) and (0, 𝑎), we consider
𝑔(𝑡) ≔ ℎ(𝑡, 𝑎 − 𝑡) = 𝑡(𝑎 − 𝑡)(𝑎 − 𝑡 − (𝑎 − 𝑡)) = 0
for 𝑡 ∈ [0, 𝑎], which shows that ℎ always takes the values 0 on this line segment again.

𝑎 3
Among all these values, we find that the absolute maximum value of ℎ in the region 𝑅 is ( ) , which occurs
3

𝑎 𝑎
at ( , ) (and the absolute minimum value of ℎ is 0, which occurs at every point on the boundary of 𝑅).
3 3

Page 11 of 12
MATH2023 Multivariable Calculus Problem Set 4
L2/L3 (Fall 2019)

(b) For every non-negative numbers 𝑥, 𝑦 and 𝑧, let’s denote 𝑎 ≔ 𝑥 + 𝑦 + 𝑧. According to (a) we always have
𝑎 𝑎 𝑎
𝑓(𝑥, 𝑦, 𝑧) ≤ 𝑓 ( , , ), i.e.
3 3 3

𝑎 3 𝑥+𝑦+𝑧 3
𝑥𝑦𝑧 ≤ ( ) = ( ) .
3 3
Taking cube roots on both sides we obtain

3
𝑥+𝑦+𝑧
√𝑥𝑦𝑧 ≤
3
as desired. The equality holds if and only if 𝑥 = 𝑦 = 𝑧 according to (a) again. ∎

15. Let 𝑥, 𝑦 and 𝑧 be the length of the front face, length of the side face and height of the box respectively. Then
the total cost of materials in making the box is given by
𝑓(𝑥, 𝑦, 𝑧) = 2𝑎𝑥𝑦 + 2𝑏𝑥𝑧 + 2𝑐𝑦𝑧.
Denoting 𝑔(𝑥, 𝑦, 𝑧) = 𝑥𝑦𝑧 − 𝑉, we need to find the absolute minimum of 𝑓 subject to the equality constraint
𝑔(𝑥, 𝑦, 𝑧) = 0, so we solve the system
𝑓𝑥 (𝑥, 𝑦, 𝑧) = 𝜆𝑔𝑥 (𝑥, 𝑦, 𝑧) 2𝑎𝑦 + 2𝑏𝑧 = 𝜆𝑦𝑧
𝑓𝑦 (𝑥, 𝑦, 𝑧) = 𝜆𝑔𝑦 (𝑥, 𝑦, 𝑧) 2𝑎𝑥 + 2𝑐𝑧 = 𝜆𝑥𝑧
, i.e. { .
𝑓𝑧 (𝑥, 𝑦, 𝑧) = 𝜆𝑔𝑧 (𝑥, 𝑦, 𝑧) 2𝑏𝑥 + 2𝑐𝑦 = 𝜆𝑥𝑦
{𝑔(𝑥, 𝑦, 𝑧) = 0 𝑥𝑦𝑧 = 𝑉
Multiplying the first equation by 𝑥, the second equation by 𝑦 and the third equation by 𝑧, we obtain
2𝑎𝑥𝑦 + 2𝑏𝑥𝑧 = 𝜆𝑉
{2𝑎𝑥𝑦 + 2𝑐𝑦𝑧 = 𝜆𝑉 .
2𝑏𝑥𝑧 + 2𝑐𝑦𝑧 = 𝜆𝑉
From these three equations we easily get

𝜆𝑉
2𝑎𝑥𝑦 = 2𝑏𝑥𝑧 = 2𝑐𝑦𝑧 = .
2
Now the last equation 𝑥𝑦𝑧 = 𝑉 implies that 𝑥, 𝑦 and 𝑧 are all non-zero, so dividing the equation 𝑥𝑦𝑧 = 𝑉 by
4𝑐 4𝑏 4𝑎
each component 2𝑎𝑥𝑦, 2𝑏𝑥𝑧 and 2𝑐𝑦𝑧, we obtain 𝑥 = , 𝑦= and 𝑧 = . Finally from
𝜆 𝜆 𝜆

4𝑐 4𝑏 4𝑎
( )( )( ) = 𝑉
𝜆 𝜆 𝜆
1 1 1 1
𝑎𝑏𝑐 3 𝑐2𝑉 3 𝑏2𝑉 3 𝑎2 𝑉 3
we obtain 𝜆 = 4( ) , i.e. (𝑥, 𝑦, 𝑧) = (( ) ,( ) ,( ) ), and the minimum value of 𝑓 is
𝑉 𝑎𝑏 𝑎𝑐 𝑏𝑐

1 1 1
𝑐 2 𝑉 3 𝑏 2 𝑉 3 𝑎2 𝑉 3 1
𝑓 (( ) ,( ) ,( ) ) = 6(𝑎𝑏𝑐𝑉 2 )3 .
𝑎𝑏 𝑎𝑐 𝑏𝑐

1 1
𝑐2𝑉 3 𝑏2𝑉 3
Therefore the optimal dimensions are ( ) cm for the length of the front face, ( ) cm for the length of the
𝑎𝑏 𝑎𝑐

1
1
𝑎2 𝑉 3
side face and ( ) cm for the height; and the corresponding minimum cost is $6(𝑎𝑏𝑐𝑉 2 )3 .
𝑏𝑐

Page 12 of 12