THE UNIVERSITY OF HONG KONG

DEPARTMENT OF MATHEMATICS

MATH2822 Mathematical Methods for Actuarial Science II

Assignment 2 Solutions

1. (a) We find the partial derivatives of both sides of F (x, y, z) = 0 with respect to x. By the
chain rule, this gives
∂x ∂y ∂z
Fx + Fy + Fz = 0.
∂x ∂x ∂x
∂x ∂y ∂z Fx
Since = 1 and = 0, by rearranging the terms, this gives =− .
∂x ∂x ∂x Fz

(b) Since Fx = ey − 9x2 and Fz = 2yz + 2, using part (a), we have

∂z Fx (1, 0, 1) −8
∣ =− =− = 4.
∂x (x,y,z)=(1,0,1) Fz (1, 0, 1) 2

2. We have ∇f (x, y) = (2xy+3y−6x−4, x2 +3x+2). We solve ∇f (x, y) = (0, 0). From x2 +3x+2 = 0,
we have (x + 1)(x + 2) = 0, and hence x = −1, −2.

ˆ If x = −1, the equation 2xy + 3y − 6x − 4 = 0 becomes y + 2 = 0, which gives y = −2. Thus,

(−1, −2) is a critical point.
ˆ If x = −2, the equation 2xy + 3y − 6x − 4 = 0 becomes −y + 8 = 0, which gives y = 8. Thus,
(−2, 8) is a critical point.

Next, we compute fxx = 2y − 6, fxy = 2x + 3 and fyy = 0.

ˆ For the critical point (−1, −2), we have fxx fyy − fxy
2
= (−10)(0) − (1)2 = −1 < 0. So
(−1, −2) is a saddle point.
ˆ For the critical point (−2, 8), we have fxx fyy − fxy
2
= (10)(0) − (−1)2 = −1 < 0. So (−2, 8)
is a saddle point.

3. For the interior points satisfying x2 + y 2 < 1, we compute ∇f (x, y) = (6x2 , 4y 3 ). We solve
∇f (x, y) = (0, 0) to get the only solution x = y = 0. Thus, (0, 0) is the only critical point.

For the boundary points satisfying x2 + y 2 = 1, we have y 2 = 1 − x2 . This implies

f (x, y) = 2x3 + (y 2 )2 = 2x3 + (1 − x2 )2 .

Define g(x) = 2x3 + (1 − x2 )2 where −1 ⩽ x ⩽ 1. Then we have

g ′ (x) = 6x2 − 4x(1 − x2 ) = 2x(x + 2)(2x − 1).

1
The solutions to g ′ (x) = 0 are x = 0, −2, . The solution x = −2 is rejected since −1 ⩽ x ⩽ 1. For
2
each x, we can solve for y using x2 + y 2 = 1. Therefore,√ the possible points of global extrema
1 3
among the boundary points are (x, y) = (0, ±1), ( , ± ) , (−1, 0), (1, 0).
2 2

1
 Now, we compare the following images:
√
1 3 13
f (0, 0) = 0, f (0, ±1) = 1, f ( ,± )= , f (−1, 0) = −2, f (1, 0) = 2.
2 2 16

It is clear that the global maximum is 2 (attained at (1, 0)), and the global minimum is −2
(attained at (−1, 0)).

4. (a) By definition, we have

f (hs, ht) − f (0, 0) 3(hs)3 − 2(hs)(ht)2 3s3 − 2st2

Du f (0, 0) = lim = lim = lim = 3s3 − 2st2
h→0 h h→0 h((hs)2 + (ht)2 ) h→0 s2 + t2

since s2 + t2 = ∥u∥ = 1.

(b) No. Note that ∇f (0, 0) = (D(1,0) f (0, 0), D(0,1) f (0, 0)) = (3, 0). If f were differentiable,
Then
Du f (0, 0) = ∇f (0, 0) ⋅ u = (3, 0) ⋅ (s, t) = 3s
for any unit vector u = (s, t) by proposition 2.6. This contradicts with the result of part
1 1
(a). For example, the directional derivative at (0, 0) in the direction of ( √ , √ ) is
2 2
1 1
√ ≠3⋅ √ .
2 2 2

(c) This means we need to minimize f (s, t) = 3s3 −2st2 under the constraint g(s, t) = s2 +t2 = 1.
We have ∇f (s, t) = (9s2 − 2t2 , −4st) and ∇g(s, t) = (2s, 2t) ≠ (0, 0) (unless s = t = 0 which
is rejected). So we solve
9s2 − 2t2 = 2λs,
{
−4st = 2λt.
The second equation gives t = 0 or −2s = λ.
ˆ If t = 0, then from s2 + t2 = 1, we obtain s = ±1. So (s, t) = (±1, 0) are two solutions.
ˆ If −2s = λ, then the first equation becomes√ 9s2 − 2t2 = −4s2 , i.e. 13s2 = 2t2 √
= 2(1 − s2 ).
2 13
This implies 15s2 = 2, and hence s = ± . Correspondingly, t = ± from
15 15
s2 + t2 = 1 (where the two signs are independent).
Now, we compare the following images:
√ √ √ √ √ √
⎛ 2 13 ⎞ 4 2 ⎛ 2 13 ⎞ 4 2
f (±1, 0) = ±3, f ,± =− , f − ,± = .
⎝ 15 15 ⎠ 3 15 ⎝ 15 15 ⎠ 3 15

Therefore, the global minimum is −3, attained at (−1, 0). This means f decreases the
most rapidly in the direction of (−1, 0).

5. We have fx (x, y) = 2ex−3y + 2x and fy (x, y) = −6ex−3y + 3. So fx (3, 1) = 8 and fy (3, 1) = −3.
The linearization of f (x, y) at (3, 1) is

L(x, y) = f (3, 1) + fx (3, 1)(x − 3) + fy (3, 1)(y − 1) = 14 + 8(x − 3) − 3(y − 1).

It follows that an approximation to f (3.1, 1.1) is L(3.1, 1.1) = 14 + 8(0.1) − 3(0.1) = 14.5.

2
Next, we compute fxx = 2ex−3y + 2, fxy = −6ex−3y and fyy = 18ex−3y . By Taylor’s theorem,
there exists c ∈ (0, 1) such that the error term is
1
E = [fxx (3 + 0.1c, 1 + 0.1c)(0.1)2 + 2fxy (3 + 0.1c, 1 + 0.1c)(0.1)(0.1)
2
+ fyy (3 + 0.1c, 1 + 0.1c)(0.1)2 ]
= 0.01(e−0.2c + 1) − 0.06e−0.2c + 0.09e−0.2c
= 0.04e−0.2c + 0.01.

Thus, we have ∣E∣ = 0.04e−0.2c + 0.01 < 0.04 + 0.01 = 0.05 since 0 < c < 1.