1082 ®M4-(ßF¡É( sãÊU
1. (10%) Read the following statements. Answer ”True” if it is correct. Answer ”False” if it is incorrect.
(a) (2%) True A is an n× m matrix. The dimension of A’s column space equals the dimension of A’s row space.
(b) (2%) True A is an n× n matrix. If detA=0, then 0 is an eigenvalue of A.
(c) (2%) True A is an n× m matrix. The rank of A is less than or equal to n. The rank of A is less than or equal to m.
(d) (2%) False The set S=⎧⎪⎪⎪
⎨⎪⎪⎪⎩
⎡⎢⎢⎢
⎢⎢⎣ 1 2 3
⎤⎥⎥⎥
⎥⎥⎦ ,
⎡⎢⎢⎢
⎢⎢⎣
−1 3
−2
⎤⎥⎥⎥
⎥⎥⎦ ,
⎡⎢⎢⎢
⎢⎢⎣ 1 7 4
⎤⎥⎥⎥
⎥⎥⎦
⎫⎪⎪⎪⎬⎪⎪
⎪⎭ is a basis for R3.
(e) (2%) False B is a k× n matrix. Suppose that there is a x0≠ 0, x0∈ Rn such that Bx0= 0. Then B is not full rank.
2. (18%) Consider the quadratic form f(x, y, z) = −2x2+ 2xy − 2y2− 5z2. (a) (3%) Express f in the form of
f(x, y, z) = (x, y, z) M ⎛
⎜⎝ x y z
⎞⎟
⎠, where M is a real-valued symmetric 3× 3 matrix. Find M.
(b) (6%) Find the eigenvalues of M and their corresponding eigenvectors.
(c) (4%) Diagonalize M . That is, find a nonsingular matrix P and a diagonal matrix D such that P−1M P= D.
(d) (5%) Use Sylvester’s criterion to determine the definiteness of M , and thus f .
Solution:
(a) M =⎛
⎜⎝
−2 1 0
1 −2 0
0 0 −5
⎞⎟
⎠. (3pts) (b)
0=det ⎛
⎜⎝
−2 − λ 1 0
1 −2 − λ 0
0 0 −5 − λ
⎞⎟
⎠ characteristic polynomial
=(−5 − λ) ((−2 − λ)2− 1) expand
=(−5 − λ) (λ2+ 4 λ + 3) expand
= − (λ + 5) (λ + 1) (λ + 3) factor
λ= − 5, −1, −3 solve (3pts)
(1) λ= −5 Ô⇒ ⎛
⎜⎝
−2 + 5 1 0
1 −2 + 5 0
0 0 −5 + 5
⎞⎟
⎠
⎛⎜
⎝ x y z
⎞⎟
⎠=⎛
⎜⎝ 0 0 0
⎞⎟
⎠
Ô⇒ eigenvector is⎛
⎜⎝ x y z
⎞⎟
⎠= t⎛
⎜⎝ 0 0 1
⎞⎟
⎠for any t∈ R with t ≠ 0. (1pt)
(2) λ= −3 Ô⇒ ⎛
⎜⎝
−2 + 3 1 0
1 −2 + 3 0
0 0 −5 + 3
⎞⎟
⎠
⎛⎜
⎝ x y z
⎞⎟
⎠=⎛
⎜⎝ 0 0 0
⎞⎟
⎠
Ô⇒ eigenvector is⎛
⎜⎝ x y z
⎞⎟
⎠= s⎛
⎜⎝ 1
−1 0
⎞⎟
⎠for any s∈ R with s ≠ 0. (1pt)
(3) λ= −1 Ô⇒ ⎛
⎜⎝
−2 + 1 1 0
1 −2 + 1 0
0 0 −5 + 1
⎞⎟
⎠
⎛⎜
⎝ x y z
⎞⎟
⎠=⎛
⎜⎝ 0 0 0
⎞⎟
⎠
Ô⇒ eigenvector is⎛
⎜⎝ x y z
⎞⎟
⎠= r⎛
⎜⎝ 1 1 0
⎞⎟
⎠for any r∈ R with r ≠ 0. (1pt)
(c) P =⎛
⎜⎝
0 1 1
0 −1 1
1 0 0
⎞⎟
⎠(2pts) Ô⇒ P−1M P =⎛
⎜⎝
−5 0 0
0 −3 0
0 0 −1
⎞⎟
⎠= D. (2pts) (d) M is negative definite (2pts), and thus f , since (3pts)
−2 < 0, det ( −2 1
1 −2 ) =3> 0, det M = −15 < 0.
3. (20%) Find the maximum value of the function f(x, y) = x2+ 2y2 subject to the constraints x+ 2y ≤ 9, x2+ y2≥ 16, x ≥ 0 and y≥ 0.
(a) (2%) Could both constraints x+ 2y ≤ 9 and x2+ y2≥ 16 be binding?
(b) (2%) Check whether the Kuhn - Tucker version NDCQ is satisfied.
(c) (2%) Write down the Kuhn - Tucker version Lagrangian function ˜L.
(d) (3%) Write down the Kuhn - Tucker version first order conditions.
(e) (4%) Is there a solution such that the constraint x2+ y2≥ 16 is binding?
(f) (5%) Find solution(s) such that the constraint x+ 2y ≤ 9 is binding.
(g) (2%) Find the maximum value.
Solution:
Let g1(x, y) = x + 2y and g2(x, y) = −x2− y2. Suppose that(x∗, y∗) is the maximizer of the problem.
(a) We want to find(x, y) satisfies x + 2y = 9 and x2+ y2= 16. Since x = 9 − 2y,
(9 − 2y)2+ y2= 16 ⇒ 65 − 36y + 5y2= 0 ⇒ it has no real solution becase 362− 65 × 20 < 0.
(2 points).
(b) We have that∇g1(x, y) = ⟨1, 2⟩ and ∇g2(x, y) = ⟨−2x, −2y⟩.
From (a), we know that g1≤ 9 and g2≤ 16 can not both binding.
Suppose tht g1(x∗, y∗) = 9. Since ∇g1(x, y) = ⟨1, 2⟩, (∂g∂x1), (∂g∂y1), (∂g∂x1 ∂g∂y1) all have rank 1. (1 point).
Suppose that g2(x∗, y∗) = 16, then (x∗)2+ (y∗)2= 16 ⇒ x∗ and y∗ can not be both 0. If x∗> 0 but y∗= 0, then (∂g∂x2 = (2x∗) has rank 1.
If x∗= 0 but y∗> 0, then (∂g∂y2 = (2y∗) has rank 1.
If x∗> 0 and y∗> 0, then (∂g∂x1 ∂g∂y1) has rank 1. (1 point)Therefore it satisfies Kuhn-Tucker version NDCQ.
(c) ˜L(x, y, l1, l2) = x2+ 2y2− l1(x + 2y − 9) − l2(−x2− y2+ 16).(2 points).
(d) At the maximizer (x∗, y∗), there exist l1∗≥ 0, l∗2≥ 0 satisfy
⎧⎪⎪⎪⎪⎪⎪⎪⎪
⎪⎪⎪⎪⎪
⎪⎪⎪⎪⎪
⎪⎨⎪⎪⎪⎪⎪
⎪⎪⎪⎪⎪
⎪⎪⎪⎪⎪
⎪⎪⎪⎪⎩
∂ ˜L
∂x = 2x∗− l∗1+ 2x∗l∗2≤ 0...(1)
∂ ˜L
∂y = 4y∗− 2l1∗+ 2y∗l∗2≤ 0...(2) x∗ ∂ ˜∂xL = 0 = x∗⋅ (2x∗− l1∗+ 2x∗l2∗)...(3) y∗ ∂ ˜∂yL = 0 = y∗⋅ (4y∗− 2l1∗+ 2y∗l∗2)...(4)
∂ ˜L
∂λ1 = −(x∗+ 2y∗− 9) ≥ 0...(5)
∂ ˜L
∂λ2 = (x∗)2+ (y∗)2− 16 ≥ 0...(6) l∗1∂λ∂ ˜L
1 = 0 = l∗1⋅ (x∗+ 2y∗− 9) = 0...(7) l∗2∂λ∂ ˜L
2 = 0 = l∗2⋅ [(x∗)2+ (y∗)2− 16] = 0...(8)
If you write down one correct equation, you get 0.5 point. If you write down all correct equations, you get 3 points.
(e) If(x∗)2+ (y∗)2= 16, then x∗+ 2y∗≠ 9.
(7) ⇒ l∗1= 0.
(2) ⇒ 2y∗(2 + l2∗) ≤ 0. Since l∗2≥ 0 and y∗≥ 0, y∗= 0. It implies that x∗= 4.
(3) ⇒ l2∗= −1. It is a contradiction. Therefore (x∗)2+ (y∗)2≥ 16 can not be binding. If you get one of correct x∗, y∗, l∗1, l∗2, you get one point.
(f) If x∗+ 2y∗= 9, then (x∗)2+ (y∗)2> 16.
(8) ⇒ l∗2= 0.(1 point)
(3), (4) ⇒ { x∗(2x∗− l∗1) = 0 y∗(4y∗− 2l∗1) = 0.
If x∗= 0, then y∗= 92⇒ l∗1= 9. It implies that (x∗, y∗, l∗1, l2∗) = (0,92, 9, 0).(1 point) If y∗= 0, then x∗= 9 ⇒ l∗1= 18. It implies that ((x∗, y∗, l∗1, l2∗) = (9, 0, 18, 0).(1 point)
If x∗ > 0, y∗ > 0, then l1 = 2x∗ = 2y∗ ⇒ x∗ = y∗. Thus 3x∗ = 9 ⇒ x∗ = 3 ⇒ l∗1 = 6. Thus ((x∗, y∗, l∗1, l∗2) = (3, 3, 6, 0).(2 points)
4. (22%) Consider a C2 utility function U(x1, x2) such that ∂x∂U1 > 0 and ∂x∂U2 > 0. We want to maximize U(x1, x2) under constraints P1x1+ P2x2≤ I, x1≥ 0, and x2≥ 0, where P1> 0, P2> 0 are unit prices and I > 0 is the budget.
(a) (2%) Write down the usual version Lagrangian function.
(b) (2%) Show that the usual version NDCQ is satisfied.
(c) (3%) List the usual version first order conditions.
(d) (3%) From the first order conditions show that at the maximizer the constraint P1x1+ P2x2≤ I must be binding.
(e) (6%) Suppose that the first order conditions has a solution (x∗1, x∗2) with x∗1 > 0, x∗2 > 0. Write down the bordered Hessian matrix at(x∗1, x∗2). List the condition which guarantees that U(x∗1, x∗2) is a local maximum value.
(f) (6%) Now P1, P2and I are parameters. Then the maximum value of U depends on P1, P2, and I, which is denoted by ˜U(P1, P2, I). Compute ∂P∂ ˜U1, ∂P∂ ˜U
2, and ∂ ˜∂IU. Determine the signs of these partial derivatives.
Solution:
(a) L(x1, x2, λ1, λ2, λ3) = U(x1, x2) − λ1(P1x1+ P2x2− I) + λ2x1+ λ3x2. (: λ2x1+ λ3x2 c λ2x1+ λ3x2 ë −λ2x1− λ3x2 c ) (b) Let g1(x1, x2) = P1x1+ P2x2, g2(x1, x2) = −x1, g3(x1, x2) = −x2.
Ð⇀∇g1= (P1, P2) ≠ Ð⇀0 , Ð⇀∇g2= (−1, 0) ≠ Ð⇀0 , Ð⇀∇g3= (0, −1) ≠ Ð⇀0 . (1pt: ú Ð∇g⇀ 1, Ð⇀∇g2, Ð∇g⇀ 3)
At the maximizer at most two of the constraints are binding and any two of {Ð⇀∇g1, Ð⇀∇g2, Ð∇g⇀ 3} are linearly independent. Hence the NDCQ is satisfied. (1pt)
(c) First Order conditions : At the maximizer(x∗1, x∗2) there are λ∗1, λ∗2, λ∗3 s.t
⎧⎪⎪⎪⎪⎪⎪⎪⎪
⎪⎪⎪⎪⎪
⎪⎪⎪⎨⎪⎪
⎪⎪⎪⎪⎪
⎪⎪⎪⎪⎪
⎪⎪⎪⎪⎩
∂L
∂x1 = ∂x∂U1− λ∗1P1+ λ∗2= 0⋯ 1O
∂L
∂x2 = ∂x∂U2− λ∗1P2+ λ∗3= 0⋯ 2O λ∗1(P1x∗1+ P2x∗2− I) = 0⋯ 3O λ∗2x∗1= 0⋯ 4O
λ∗3x∗2= 0⋯ 5O P1x∗1+ P2x∗2≤ I⋯ 6O
x∗1≥ 0, x∗2≥ 0, λ∗1≥ 0, λ∗2≥ 0, λ∗3≥ 0⋯ 7O 1pt 1O+ 2O
1pt 3O+ 4O+ 5O 1pt 6O+ 7O
(d) If P1x∗1+P2x∗2< I, then 3O ⇒ λ∗1= 0. (1pt) Thus 1O ⇒ ∂x∂Ui(x∗1, x∗2)+λ∗2= 0 ⇒ ∂x∂Ui(x∗1, x∗2) = −λ∗2≤ 0 contradiciton!
(2pts)
Hence P1x∗1+ P2x∗2= I.
(e) Since only the constraint P1x1+P2x2≤ I is binding at (x∗1, x∗2), the bordered Hessian matrix is H =⎛
⎜⎝
0 P1 P2
P1 Lx1x1 Lx1x2
P2 Lx2x1 Lx2x2
⎞⎟
⎠ (1pt)
Because Lx1x1= Ux1x1, Lx1x2= Ux1x2, Lx2x2= Ux2x2, the bordered Hessian matrix is⎛
⎜⎝
0 P1 P2
P1 Ux1x1 Ux1x2 P2 Ux2x1 Ux2x2
⎞⎟
⎠(1pt) There are 2 variables(x1, x2) and one binding constraint.
We need to check the last(2 − 1) leading principal minor which is det⎛
⎜⎝
0 P1 P2
P1 Ux1x1 Ux1x2 P2 Ux2x1 Ux2x2
⎞⎟
⎠. (1pt) det H= −P1(P1Ux2x2− P2Ux2x1) + P2(P1Ux2x1− P2Ux1x1)
= −P12Ux2x2+ 2P1P2Ux1x2− P22Ux1x1∣
(x∗1,x∗2). (2pts) If det⎛
⎜⎝
0 P1 P2
P1 Ux1x1 Ux1x2
P2 Ux2x1 Ux2x2
⎞⎟
⎠ RRRRR RRRRR RRR(x∗1,x∗2)
= −P12Ux2x2+ 2P1P2Ux1x2− P22Ux1x1 > 0, then U(x∗1, x∗2) is a local maximum
value. (1pt)
(f) By the envelope theorem
∂ ˜U
∂P1 = ∂L
∂P1 = −λ∗1x∗1< 0
∂ ˜U
∂P2 = ∂L
∂P2 = −λ∗1x∗2< 0
∂ ˜U
∂I = λ∗1> 0 (1pt for ∂P∂ ˜U
1 = −λ∗1x∗1, 1pt for the sign< 0, 1pt for ∂P∂ ˜U2 = −λ∗1x∗2, 1pt for the sign< 0, 1pt for ∂ ˜∂IU = λ∗1, 1pt for the sign> 0.)
÷× (f) TH. ∂P∂ ˜U1 ë −λ1x1 ( *)∂P∂ ˜U2 = −λ1x2( *)∂ ˜∂IU = λ1( *) c
5. (18%) Suppose that in the following week you have 12 hours each day to study for the final exams of Calculus 4 and English. Let C be the number of hours per day spent studying for Calculsu 4 and E be the number of hours per day spent studying for English. Let your grade point average from these two courses be GP A= f(C, E) =23(√
C+√ 2E).
(a) (6%) Solve the optimization problem : Maximize f(C, E) under the constraint C + E = 12.
Use the bordered Hessian matrix to verify that the solution is indeed a local maximum.
(b) (6%) To assure that you obtain certain grades for Calculus 4 and English individually, you impose inequality con- straints C≥ 5 and E ≥ 4.
Solve the optimization problem : Maximize f(C, E) under constraints C + E = 12, C ≥ 5, and E ≥ 4.
(c) (6%) By the meaning of multipliers, estimate the maximum value of GP A when the constraints are C+ E = 12.5, C≥ 5.5 and E ≥ 4.1.
Solution:
(a) Define L(x, y, µ) = 23(√ C+√
2E) − µ(C + E − 12)
⎧⎪⎪⎪⎪⎪
⎨⎪⎪⎪⎪⎪⎩
∂L
∂C =13√1C− µ = 0
∂L
∂E =√32√1E− µ = 0
∂L
∂µ = C + E − 12 = 0
⇒ C= 9µ12
E= 9µ22
and C+ E = 12 ⇒ µ =16, C= 4, E = 8.
(4pts for correct answers C= 4, E = 8, µ = 16) The bordered Hessian matrix is ˆH=⎛
⎜⎜⎝
0 1 1
1 ∂C∂22L ∂E∂C∂2 L 1 ∂C∂E∂2 L ∂E∂22L
⎞⎟⎟
⎠ (1pt) At(C, E) = (4, 8)
Hˆ =⎛
⎜⎝
0 1 1
1 −481 0 1 0 −961
⎞⎟
⎠.
det ˆH= 961 +481 > 0. (1pt) Hence f(C, E) is a local maximum.
(b) Define L(C, E, µ, λ1, λ2) = 23(√ C+√
2E) − µ(C + E − 12) − λ1(−C + 5) − λ2(−E + 4) (1pt) constraints are
h(C, E) = C + E = 12 g1(C, E) = −C ≤ −5 g2(C, E) = −E ≤ −4 Ð⇀∇h = (1, 1), Ð∇g⇀ 1= (−1, 0), Ð⇀∇g2= (0, −1). NDCQ is satisfied.
(2pts)
⎧⎪⎪⎪⎪⎪⎪⎪⎪
⎪⎪⎪⎪⎪
⎨⎪⎪⎪⎪⎪⎪⎪
⎪⎪⎪⎪⎪
⎪⎩
∂L
∂C = 3√1C− µ + λ1= 0⋯ 1O
∂L
∂E = 3√√2E− µ + λ2= 0⋯ 2O λ1(−C + 5) = 0⋯ 3O λ2(−E + 4) = 0⋯ 4O C+ E = 12⋯ 5O
λ1≥ 0, λ2≥ 0, C ≥ 5, E ≥ 4.
We disscuss cases{λ1> 0, λ2> 0}, {λ1= 0, λ2> 0}, {λ1> 0, λ2= 0}, {λ1= λ2= 0}
(å4 cases q3: BúcºãF/ Öhè cases Lc 1 2) 1. {λ1> 0, λ2> 0} : 3O 4O 5O contradiciton.
2. {λ1= 0, λ2> 0} : 4O 5O ⇒ E = 4, C = 8 O ⇒ µ =1 6√12
O ⇒2 √62−6√12+ λ2= 0 ⇒ λ2< 0 contradiction.
3. {λ1> 0, λ2= 0} : 3O 5O ⇒ C = 5, E = 7, 2O ⇒ µ =13√
2 7
O ⇒ λ1 1= µ −3√15 =13√
2
7−3√15 > 0.
Ans : (C∗, E∗, µ∗, λ∗1, λ∗2) = (5, 7,13√
2 7,13
√2
7−3√15, 0).
4. {λ1= 0, λ2= 0} : 1O 2O 5O ⇒ C = 4, E = 8 contradiction to the constraint C ≥ 5.
Hence the maximum value is f(5, 7) = 23(√ 5+√
14).
(c) The maximum value for new constraints can be approximated by
f(5, 7) + µ∗× (12.5 − 12) + (−λ∗1) × (5.5 − 5) + (−λ∗2) × (4.1 − 4) (4pts)
= f(5, 7) + 0.5(µ∗− λ∗1) − 0.1λ∗2= 2 3(√
5+√
14) + 1 6√
5 (2pts)
6. (12%) Suppose that f(x, y, z) is a C2function and P = (1, 2, −3) is a critical point of f which means that fx= fy= fz= 0 at P . If at point P , fxx= 1, fyy = 1, fzz = −3, fxy = −1, fxz = 3, and fyz = 3, determine whether f(1, 2, −3) is a local extreme value.
(a) (1%) Write down the Hessian matrix of f at P which is denoted by H.
(b) (3%) Is H positive definite, negative definite, or indefinite? Is f(1, 2, −3) a local maximum, local minimum, or a saddle point?
(c) (3%) Now we want to find the extreme value of f under the constraint x2+xy+xz−3z = 9. Write down the Lagrangian functionL(x, y, z, µ). Find µ∗ such that ∂x∂ L = ∂y∂ L = ∂z∂ L = 0 when (x, y, z, µ) = (1, 2, −3, µ∗).
(d) (5%) On the constraint set x2+ xy + xz − 3z = 9, is f(1, 2, −3) a local maximum, local minimum, or a saddle point?
Solution:
(a) H=⎛
⎜⎝
1 −1 3
−1 1 3
3 3 −3
⎞⎟
⎠(1pt)
(b) The leading principal minors of H are H1= 1, H2= ∣1 −1
−1 1 ∣ =0, H3=RRRRR RRRRR RRR
1 −1 3
−1 1 3
3 3 −3
RRRRR RRRRR
RRR= −36 (2pts)
∵H1⋅ H3= −36 < 0 ∴ H is indefinite. f(P ) is a saddle point. (1pt) (c) L(x, y, z, µ) = f(x, y, z) − µ(x2+ xy + xz − 3z − 9) (1pt)
At(1, 2, −3),
∂L
∂x =∂f
∂x(1, 2, −3) − µ × 1 = −µ
∂L
∂y =∂f
∂y(1, 2, −3) − µ × 1 = −µ
∂L
∂z =∂f
∂z(1, 2, −3) − µ × (−2) = 2µ when µ= 0, ∂L∂x =∂L∂y = ∂L∂z = 0. Hence µ∗= 0. (2pts)
(d) S∶ g(x, y, z) = 9 where g(x, y, z) = x2+ xy + xz − 3z, Ð∇g(1, 2, −3) = (1, 1, −2).⇀
At (x, y, z, µ) = (1, 2, −3, 0), Lxx = fxx = 1, Lxy = fxy = −1, Lxz = fxz = 3, Lyy = fyy = 1, Lyz = fyz = 3, Lzz= fzz= −3. (1pt) The bordered Hessian is
Hˆ =
⎛⎜⎜
⎜⎝
0 1 1 −2
1 1 −1 3
1 −1 1 3
−2 3 3 −3
⎞⎟⎟
⎟⎠ (1pt)
There are 3 variables and 1 constraint. Hence we need to check the last 3− 1 leading principal minors of ˆH.
Hˆ3=RRRRR RRRRR RRR
0 1 1
1 1 −1
1 −1 1 RRRRR RRRRR
RRR= −4 < 0 (1pt) Hˆ4=RRRRR
RRRRR RRRRR RRR
0 1 1 −2
1 1 −1 3
1 −1 1 3
−2 3 3 −3
RRRRR RRRRR RRRRR RRR
= −36 < 0 (1pt)
Hˆ3 and ˆH4 both have the same sign as (−1)1. Thus H is positive definite on the constraint set. f(1, 2, −3) is a local minimum on S. (1pt)