Engineering Math II Midterm Exam 1

2022/04/22, 18:30–21:00

1. a) Let V be an inner product space, W be its finite-dimensional subspace, and W⊥

be the orthogonal complement of W. Show that V = W ⊕ W⊥ .
Sol) See p.372, Theorem 6.6.
b) Consider the inner product space V = {f : [0, ∞) → R, f is piecewise-continuous}
with ˆ ∞
⟨f, g⟩ = e−t f (t)g(t)dt.
0

Use the Gram-Schmidt process to construct an orthonormal basis for P2 (R) ≤ V

from the standard ordred basis S = ˆ {1, x, x2 }. (Hint: One can use the fact
∞
Γ(n + 1) = n! for n ∈ N, where Γ(s) = ts−1 e−t dt for s > 0.)
ˆ ∞ 0

Sol) ||1||2 = e−t dt = 1. ∴ u1 = 1.

0 ˆ ∞
Consider v2 = x − ⟨x, 1⟩1. Since ⟨x, 1⟩ = te−t dt = Γ(2) = 1, v2 = x − 1. And
ˆ ∞ ˆ ∞ 0

||x − 1||2 = (t − 1)2 e−t dt = (t2 − 2t + 1)e−t dt = Γ(3) − 2Γ(2) + Γ(1) = 1.

0 0
∴ u2 = x − 1. ˆ ∞
Consider v3 = x2 − ⟨x2 , x − 1⟩(x − 1) − ⟨x2 , 1⟩1. Since ⟨x2 , x − 1⟩ = t2 (t −
ˆ ∞ 0

1)e−t dt = Γ(4) − Γ(3) = 4, and ⟨x2 , 1⟩ = t2 e−t dt = Γ(3) = 2, v3 = x2 −

0 ˆ ∞
4(x − 1) − 2 = x2 − 4x + 2. And ||x2 − 4x + 2||2 = (t2 − 4t + 2)2 e−t dt =
ˆ ∞ 0

(t4 − 8t3 + 20t2 − 16t + 4)e−t dt = Γ(5) − 8Γ(4) + 20Γ(3) − 16Γ(2) + 4Γ(1) = 4.
0
1 2
Setting u3 = (x − 4x + 2), we’re done.
2
(Of course, there is a fancier way to compute the norms than this brute-force
calculation. Anyway...)

2. a) Construct the 2 × 2 Householder matrix sending (5 12)t to (13 0)t .

1
Sol) Since (5 12) − (13 0) = (−8 12), we can set u = √ (−2 3)t . So the
! 13
!
1 0 2 −2
Householder matrix is Hu = I2 − 2uut = − (−2 3)
0 1 13 3
! ! !
1 0 2 4 −6 1 5 12
= − = (=: Q).
0 1 13 −6 9 13 12 −5
!
5 5
b) Give the QR-decomposition of the matrix A = . That is, find an
12 −1
orthogonal matrix Q and an upper-triangular
! matrix R satisfying A = QR.
13 1
One can easily verify that QA = (=: R). Since Q−1 = Qt = Q, one has
0 5
A = QR.
3. Consider the system of linear equations given by Ax = b, where
! !
1 1 1 1 1 15
A= , b= .
1 −1 1 −1 1 3

a) Use Gauss elimination to find the general solution of Ax = b, and use the gradient
(and the Hessian if necessary) of ||x||2 to find the minimal solution.!
1 0 1 0 1 9
Sol) The reduced row-echelon form of (A|b) is .
0 1 0 1 0 6
Let x = (x1 x2 x3 x4 x5 )t and set x3 = t1 , x4 = t2 , x5 = t3 . Then
x = (−t1 − t3 + 9 − t2 + 6 t1 t2 t3 )t , so ||x||2 =: g(t1 , t2 , t3 ) = 2t21 + 2t22 + 2t23 +
2t1 t3 − 18t1 − 18t3 − 12t2 + 117.
Since ∇g = (4t1 +2t3 −18, 4t2 −12, 4t3 +2t1 −18), ∇g = 0 only when (t1 , t2 , t3 ) =
(3, 3, 3) Obviously this unique critical pointprovidesthe minimal solution x =
4 0 2
t ′′
 
(3 3 3 3 3) because the Hessian of g, g = 0 4 0

 is positive-definite.
2 0 4
b) Let u be a solution of AA∗ u = b. Compute s = A∗ u and explain that s is
actually the minimal solution, which is independent of the choice of u.
Sol) For the minimality of s and the independence
! of choice
! of u, see p.387,
5 1 15
Theorem 6.13. In this problem AA∗ u = u= (= b). So u = (3 0)t
1 5 3
is uniquely determined. Thus s = A∗ u = (3 3 3 3 3)t is what we want.
c) Give the singular value decomposition of A, and use this to compute A† , the
Moore-Penrose pseudoinverse of A.
Sol) AAt has two positive eigenvalues λ1 = 6, λ2 = 4, and their correspond-
1 1
ing eigenvectors are w1 = √ (1 1)t , w2 = √ (−1 1)t , respectively. So v1 =
2 2
1 ∗ 1 1 ∗ 1
√ A w1 = √ (1 0 1 0 1) and v2 = √ A w2 = √ (0 − 1 0 − 1 0)t form
t
6 3 4 2
an orthonormal subset of R5 . Let P = (v1 |v2 |v3 |v4 |v5 ), Q = (w1 |w2 ), where
{v1 , . . . , v5 } is the orthonormal basis for R5 obtained by extending {v1 , v2 } via
the Gram-Schmidt process. Hence the singular value decomposition of A can be
√ !
6 0 0 0 0
represented as A = QΣP t , where Σ = √ . Moreover the SVD
0 4 0 0 0
√ √
of A has much simpler form, say A = 6w1 v1t + 4w2 v2t . From this, we can
easily find the Moore-Penrose peudoinverse of A, say
!t
1 1 1 1 1
† 1 1
A = √ v1 w1t + √ v2 w2t = 6
1
4 6 4 6
.
6 4 6 − 14 1
6 − 41 1
6

d) Show that s = A† b is also the minimal solution. One has to verify the solutions
in a),b),d) coincide.
Sol) For the explanation that A† b is the minimal solution, see p.442, Theorem
6.30. In this problem s = A† b = (3 3 3 3 3)t . All the solutions coincide!
4. a) Determine the values of R̂ · r̂ and θ̂ · r̂, where θ is the zenith angle.
Sol) R̂ · r̂ = sin θ, θ̂ · r̂ = cos θ.
1
b) Represent the vector field A = R̂ in terms of the cylindrical coordinates
R2
r̂, ϕ̂, k̂. (Of course, ϕ denotes the azimuthiangle.)
1 1 h
Sol) A = 2 R̂ = 2 (R̂ · r̂)r̂ + (R̂ · k̂)k̂ (∵ R̂ · ϕ̂ = 0)
R R 
1  1   1  
= 2 sin θr̂ + cos θk̂ = 3 R sin θr̂ + R cos θk̂ = p rr̂ + z k̂ .
R R (r2 + z 2 )3
1
5. a) Compute ∇2 f , where f (X) = .
R
1 1
Sol) ∇f = − 2 R̂, and ∇2 f = ∇ · (− 2 R̂) = −(sin θ)R = 0.
R R
¨
X −Y 4πR02
b) Show that 3
ds(Y ) = R̂ for |X| = R > R0 , where S 2 (R0 )
S 2 (R0 ) |X − Y | R2
is the sphere of radius ¨ R0 with the center O.
X −Y
Sol) By symmetry, ds(Y ) = f (R)R̂ for some f . And
2
S (R0 ) |X − Y |3
¨ ¨ ! ¨ ¨ !
X −Y X −Y
3
ds(Y ) ·ds(X) = · ds(X) ds(Y )
S 2 (R) S 2 (R0 ) |X − Y | S 2 (R0 ) 2 |X − Y |3
¨ ¨S (R)
= 4π ds(Y ) = (4π)(4πR02 ). On the other hand, f (R)R̂ · ds(X) =
S 2 (R0 ) S 2 (R)
4πR02
4πR2 f (R). Thus f (R) = , the conclusion follows.
R2
¨
1
c) Compute φ(X) := ds(Y ) for X outside S, the unit sphere with the
S |X − Y|
center O. (Hint: Consider ∇φ.) ¨
X −Y 4π
Sol) Actually one can easily verify that ∇φ = − 3
ds(Y ) = − 2 R̂ by
S |X − Y | R
4π 4π
b). Thus by integrating, we have φ(X) = = .
R |X|
cos2 ϕ
6. a) Let F = R̂ in D, the region between two spherical shells R = 1 and R = 2.
‹R3
Evaluate F · ds.
‹ ∂D ‹ ‹
Sol1) F·ds = F·ds− F·ds. In both integrals ds = R2 sin θdθdϕR̂.
S 2 (2) S 2 (1)
‹ ∂D
ˆ π ˆ π
sin θ cos2 ϕ 2π
So F · ds = dθdϕ = .
R R
‹ S (R)
2
 −π 0
1
∴ F · ds = − 1 (2π) = −π.
∂D 2
cos2 ϕ
 
1
Sol2) One may use the divergence theorem. Here ∇·F = 2 R2 sin θ =
R sin θ R3
‹ ˚ ˆ π ˆ π ˆ 2
R
cos2 ϕ cos2 ϕ 2
− . So F · ds = ∇ · Fdv = − R sin θdRdθdϕ =
R4 R4
ˆ 2  ∂D ˆ π ˆ πD  −π  0 1
1 1
− 2 dR sin θdθ cos2 ϕdϕ − · 2 · π = −π.
1 R 0 −π 2
b) Let F be a vector field on R3 . Show that ∇ · (∇ × F) = 0.
Sol) For a three-dimensional ˚
region D with closed orientable
‹ boundary ∂D, by
divergence theorem we have ∇ · (∇ × F)dv = (∇ × F) · ds. But by
‹ D ˛ ∂D

Stokes’s theorem, (∇ × F) · ds = F · dℓ = 0. Since D is arbitrary, one can

∂D ∅
shrink D to a point P . The result follows.
 c) For F = (xy − xz)î + (yz − xy)ĵ + (xz − yz)k̂, find a vector field G satisfying
∇ × G = F.
∂c ∂c
Sol) Consider C = c(x, y, z)î, of which ∇ × C = ĵ − k̂. Choose c(x, y, z)
∂z ∂y
∂c 1
satisfies = xz − yz, say c(x, y, z) = xyz − y 2 z. Then
∂y 2
h i  1 2

F+(∇×C) = (xy − xz)î + (yz − xy)ĵ + (xz − yz)k̂ + (xy − y )ĵ − (xz − yz)k̂ =
  2
1 2
(xy − xz)î + (yz − y )ĵ . Now consider H = h(x, y, z)k̂, of which ∇ × H =
2
∂h ∂h ∂h 1
î − ĵ. Choose h(x, y, z) satisfies = xy − xz, say h(x, y, z) = xy 2 − xyz.
∂y ∂x ∂y 2
Then one can verify that F + (∇ × C) = ∇ × H. So G := H − C is the vector
field what we want.

7. (출석문제) 응용수학에서 ‘약방의 감초’가 의미하는 것은?

Sol) Γ(s), the Gamma function.