Static Optimization

Un-Constrained Optimizations
1. Find and characterize the stationary point of
𝑓(𝑥, 𝑦) = 𝑥 3 +𝑦 3 − 3𝑥 − 2𝑦 given 𝑥 > 0 𝑦>0
Solution
FOC:
𝑓 ′ 𝑥 = 3𝑥 2 − 3 = 0
Hence 𝑥 = ±1 implying 𝑥 = 1 𝑠𝑖𝑛𝑐𝑒 𝑥 > 0
𝑓 ′ = 3𝑦 2 − 2 = 0
2 2
Hence 𝑦 = ±√3 implying that 𝑦 = √3 𝑠𝑖𝑛𝑐𝑒 𝑦 > 0

2
So, (x,y)=(1, √3)

Characterizing the critical point

𝐹𝑥𝑥 = 6𝑥, 𝐹𝑦𝑦 = 6𝑦 𝑎𝑛𝑑 𝐹𝑥𝑦 = 0
6𝑥 0
𝐻𝑒𝑠𝑠𝑖𝑎𝑛 𝑚𝑎𝑡𝑟𝑖𝑥 = [ ]
0 6𝑦
6𝑥 > 0, 36𝑥𝑦 > 0 𝑡ℎ𝑒 ℎ𝑒𝑠𝑠𝑖𝑎𝑛 𝑖𝑠 𝑝𝑜𝑠𝑖𝑡𝑖𝑣𝑒 𝑑𝑒𝑓𝑖𝑛𝑖𝑡𝑒 𝑖𝑚𝑝𝑙𝑦𝑖𝑛𝑔 𝑡ℎ𝑎𝑡 𝑓(𝑥, 𝑦) 𝑖𝑠 𝑐𝑜𝑛𝑣𝑒𝑥

2
𝑡ℎ𝑒 𝑝𝑜𝑖𝑛𝑡 (1, √ ) 𝑖𝑠 𝑎 𝑔𝑙𝑜𝑏𝑎𝑙 𝑚𝑖𝑛𝑖𝑚𝑢𝑚
3

2. Find and characterize the stationary point of

𝑓 = 𝑥12 + 𝑥22 + 3𝑥32 − 𝑥1 𝑥2 + 2𝑥1 𝑥3 + 𝑥2 𝑥3
Solution
FOC
𝑤. 𝑟. 𝑡 𝑥1 : 2𝑥1 −𝑥2 + 2𝑥3 = 0
𝑤. 𝑟. 𝑡 𝑥2 : 2𝑥2 −𝑥1 + 𝑥3 = 0
𝑤. 𝑟. 𝑡 𝑥3 : 6𝑥3 + 2𝑥1 + 𝑥2 = 0
Stationary point (𝑥1 , 𝑥2 , 𝑥3 ) = (0,0,0)
Characterizing the critical point
𝐹11 = 2, 𝐹22 = 2, 𝐹33 = 6 𝐹12 = −1 𝐹13 = 2 𝑎𝑛𝑑 𝐹23 = 1
 2 −1 2
𝐻𝑒𝑠𝑠𝑖𝑜𝑛 𝑚𝑎𝑡𝑟𝑖𝑥 = [−1 2 1]
2 1 6
𝐿𝑒𝑎𝑑𝑖𝑛𝑔 𝑝𝑟𝑖𝑛𝑐𝑖𝑝𝑎𝑙 𝑚𝑖𝑛𝑜𝑟
𝐷1 = 2
𝐷2 = 3
𝐷3 = 4
𝐷𝑘 > 0
𝑡ℎ𝑒 ℎ𝑒𝑠𝑠𝑖𝑎𝑛 𝑖𝑠 𝑝𝑜𝑠𝑖𝑡𝑖𝑣𝑒 𝑑𝑒𝑓𝑖𝑛𝑖𝑡𝑒 𝑖𝑚𝑝𝑙𝑦𝑖𝑛𝑔 𝑡ℎ𝑎𝑡 𝑓(𝑥1 , 𝑥2 , 𝑥3 ) 𝑖𝑠 𝑐𝑜𝑛𝑣𝑒𝑥
𝑎𝑛𝑑 𝑡ℎ𝑒 𝑝𝑜𝑖𝑛𝑡 (0,0,0)𝑖𝑠 𝑎 𝑔𝑙𝑜𝑏𝑎𝑙 𝑚𝑖𝑛𝑖𝑚𝑢𝑚

Equality Constraints
1.
𝑚𝑎𝑥𝑖𝑚𝑖𝑧𝑒 𝑓(𝑥, 𝑦) = 2𝑥𝑦
𝑠𝑢𝑏𝑗𝑒𝑐𝑡 𝑡𝑜 𝑔 = 3𝑥 + 4𝑦 = 12
𝐿 = 2𝑥𝑦 − 𝜆(3𝑥 + 4𝑦 − 12)
𝑑𝐿 3𝜆
= 2𝑦 − 3𝜆 = 0 ℎ𝑒𝑛𝑐𝑒 𝑦 =
𝑑𝑥 2
𝑑𝐿
= 2𝑥 − 4𝜆 = 0 ℎ𝑒𝑛𝑐𝑒 𝑥 = 2𝜆
𝑑𝑦
𝑑𝐿
= 3𝑥 + 4𝑦 = 12
𝑑𝜆
ℎ𝑒𝑛𝑐𝑒 6𝜆 + 6𝜆 = 12
3
𝜆 = 1 𝑎𝑛𝑑 ℎ𝑒𝑛𝑐𝑒 𝑥 = 2 𝑎𝑛𝑑 𝑦 =
2
With 𝜆 = 1, 𝐿 = 2𝑥𝑦 − 3𝑥 − 4𝑦 + 12
𝐿𝑥𝑥 = 2, 𝐿𝑦𝑦 = 2, 𝐿𝑥𝑦 = 0
2 0
𝐻𝑒𝑠𝑠𝑖𝑜𝑛 𝑚𝑎𝑡𝑟𝑖𝑥 = [ ]
0 2
𝐿𝑒𝑎𝑑𝑖𝑛𝑔 𝑝𝑟𝑖𝑛𝑐𝑖𝑝𝑎𝑙 𝑚𝑖𝑛𝑜𝑟
𝐷1 = 2
𝐷2 = 4
𝐷𝑘 > 0
3
𝑡ℎ𝑒 ℎ𝑒𝑠𝑠𝑖𝑎𝑛 𝑖𝑠 𝑝𝑜𝑠𝑖𝑡𝑖𝑣𝑒 𝑑𝑒𝑓𝑖𝑛𝑖𝑡𝑒 𝑖𝑚𝑝𝑙𝑦𝑖𝑛𝑔 𝑡ℎ𝑎𝑡 𝑓(𝑥, 𝑦) 𝑖𝑠 𝑐𝑜𝑛𝑣𝑒𝑥 𝑎𝑛𝑑 𝑡ℎ𝑒 𝑝𝑜𝑖𝑛𝑡 (2, ) 𝑖𝑠 𝑎 𝑔𝑙𝑜𝑏𝑎𝑙 𝑚𝑖𝑛𝑖𝑚𝑢𝑚
2
2.
𝑀𝑎𝑥 𝑓 = (𝑥, 𝑦, 𝑧) = 2𝑥 + 𝑦
𝑔1 = 𝑥 + 𝑦 + 𝑧 = 1
7
𝑔2 = 𝑥 + 𝑦 2 + 𝑧 2 =
4
𝐿 = 2𝑥 + 𝑦 − 𝜆1 (𝑥 + 𝑦 + 𝑧 − 1) − 𝜆2 (𝑥 + 𝑦 2 + 𝑧 2 )
𝐿𝑋 = 2 − 𝜆1 − 𝜆2 = 0 … … … … … … … . (𝑖)
𝐿𝑌 = 1 − 𝜆1 − 2𝜆2 𝑦 = 0 … … … … … … … . . (𝑖𝑖)
𝐿𝑧 = −𝜆1 − 2𝜆2 𝑧 = 0 … … … … … … … … . (𝑖𝑖𝑖)
𝐹𝑟𝑜𝑚 (𝑖) 𝜆2 = 2 − 𝜆1
𝐼𝑛𝑡𝑜 (𝑖𝑖𝑖) − 𝜆1 − 2(2 − 𝜆1 )𝑧 = 0
−𝜆1 − 4𝑧 + 2𝜆1 𝑧 = 0
−𝜆1 − 4𝑧 − 2𝜆1 𝑧 = 0
4𝑧 = 𝜆1 (2𝑧 − 1) 𝑧 ≠ 1/2
4𝑧 4𝑧
𝑖𝑛𝑡𝑜 (𝑖𝑖) 1− − 2 (2 − )𝑦 = 0
2𝑧 − 1 2𝑧 − 1
(2𝑧 − 1) − 4𝑧 − 2(4𝑧 − 𝑧) − 4𝑧)𝑦 = 0
2𝑧 − 1 − 4𝑧 − 2(−2)𝑦 = 0
−2𝑧 − 1 + 4𝑦 = 0
4𝑦 − 1
𝑧=
2
1
𝑧 = 2𝑦 − … … … … … … … … … … … … … . (𝑖𝑣)
2
1
𝑖𝑛𝑡𝑜 𝑐𝑜𝑛𝑠𝑡𝑟𝑎𝑖𝑛𝑡𝑠 𝑥 + 𝑦 + 2𝑦 − = 1
2
3
𝑥 + 3𝑦 = 2 … … … … … … … … … . . (𝑣)
7
𝐴𝑔𝑎𝑖𝑛 𝑥 + 𝑦2 + 𝑧2 =
4
1
𝐵𝑢𝑡 𝑧 = 2𝑦 −
2
1 7
𝑥 + 𝑦 2 + (2𝑦 − )2 =
2 4
3
5𝑦 2 − 2𝑦 + 𝑥 = … … … … … … … … (𝑣𝑖)
2
3 3
𝑡𝑎𝑘𝑖𝑛𝑔 𝑥 = 2 − 3𝑦 𝑓𝑟𝑜𝑚(𝑣) 𝑎𝑛𝑑 𝑠𝑢𝑏𝑠𝑡𝑖𝑡𝑢𝑡𝑒 𝑖𝑛𝑡𝑜 𝑦 2 − 2𝑦 + 𝑥 = 2 (𝑒𝑞𝑢𝑎𝑡𝑖𝑜𝑛 𝑣𝑖)
 3 3
5𝑦 2 − 2𝑦 + ( − 3𝑦) =
2 2
2
5𝑦 − 5𝑦 = 0
5𝑦(𝑦 − 1) = 0
𝑦=1 𝑜𝑟 0
3 3
𝑠𝑜, 𝑥=− 𝑜𝑟
2 2
1 3 1
𝑎𝑛𝑑 𝑧 = 2𝑦 − = − 𝑜𝑟 −
2 2 2
3 3 3 1
(𝑥, 𝑦, 𝑧) = ( , 1, ) or (2 , 0, − 2)
2 2

𝑓1 = −2 𝜆1 = 3 𝜆2 = −1 𝑓1 = 3 𝜆1 = 𝜆2 = 1
4𝑧 6
𝜆1 = = =3
2𝑧 − 1 3

𝜆2 = 2 − 3 = −1

𝐿𝑎𝑛𝑔𝑟𝑎𝑛𝑔𝑖𝑎𝑛 for 𝜆1 = 3 𝑎𝑛𝑑 𝜆2 = −1

7
𝐿 = 2𝑥 + 𝑦 − 𝜆1 (𝑥 + 𝑦 + 𝑧 − 1) − 𝜆2 (𝑥 + 𝑦 2 + 𝑧 2 − )
4
7
𝐿 = 2𝑥 + 𝑦 − 3𝑥 − 3𝑦 − 3𝑧 + 3 + 𝑥 + 𝑦 2 + 𝑧 2 − )
4
5
𝐿 = 𝑦 2 + 𝑧 2 − 2𝑦 − 3𝑧 +
4
𝐿𝑥𝑥 = 0, 𝐿𝑦𝑦 = 2, 𝐿𝑧𝑧 = 2 𝐿𝑥𝑦 = 0, 𝐿𝑥𝑧 = 0, 𝐿𝑦𝑧 = 0

0 0 0
𝐻𝑒𝑠𝑠𝑖𝑎𝑛 [ 0 2 0]
0 0 2
𝐴𝑙𝑙 𝑙𝑒𝑎𝑑𝑖𝑛𝑔 𝑝𝑟𝑖𝑛𝑐𝑖𝑝𝑎𝑙 𝑚𝑖𝑛𝑜𝑟𝑠 𝑜𝑓 𝑜𝑟𝑑𝑒𝑟𝑠 1, 2 𝑎𝑛𝑑 3 are 0
We check the principal minor for semi-definiteness
Δ1 = 0 𝑎𝑛𝑑 2
Δ2 = 0 𝑎𝑛𝑑 4
Δ3 = 0
𝑆𝑜, 𝑡ℎ𝑒 𝐻𝑒𝑠𝑠𝑖𝑎𝑛 𝑖𝑠 𝑝𝑜𝑠𝑖𝑡𝑖𝑣𝑒 𝑠𝑒𝑚𝑖 𝑑𝑒𝑓𝑖𝑛𝑖𝑡𝑒 𝑖𝑚𝑝𝑙𝑦𝑖𝑛𝑔 𝑓(𝑥, 𝑦, 𝑧) 𝑖𝑠 𝑐𝑜𝑛𝑣𝑒𝑥,
3 3
𝑎𝑛𝑑 𝑡ℎ𝑒 𝑝𝑜𝑖𝑛𝑡 ( , 1, ) 𝑖𝑠 𝑎 𝑔𝑙𝑜𝑏𝑎𝑙 𝑚𝑖𝑛𝑖𝑚𝑢𝑚
2 2
Show that the other point is a maximum
Inequality Constraints
1.
1
𝑀𝑎𝑥 𝑥−𝑦
2
𝑠𝑢𝑏𝑗𝑒𝑐𝑡 𝑡𝑜 𝑥 + 𝑒 −𝑥 + 𝑧 2 ≤ 𝑦
𝑥≥0
1
𝐿= 𝑥 − 𝑦 − 𝜆1 ( 𝑥 + 𝑒 −𝑥 + 𝑧 2 − 𝑦) + 𝜆2 𝑥
2
1
𝐿𝑥 = − 𝜆1 ( 1 − 𝑒 −𝑥 ) + 𝜆2 = 0 … … … … … (𝑖)
2
𝐿𝑦 = −1 + 𝜆1 = 0 … … … … … … … … … . . (𝑖𝑖)
𝐿𝑧 = −2𝜆1 𝑧 = 0 … … … … … … … … … . (𝑖𝑖𝑖)
𝜆1 ≥ 0 𝜆1 = 0 𝑥 + 𝑒 −𝑥 + 𝑧 2 − 𝑦 < 0 … … (𝑖𝑣)
𝜆2 ≥ 0 𝜆2 = 0 𝑥 > 0 … … … … … … . (𝑣)
𝜆1 = 1 … … … … … … … … … . . (𝑖𝑖)
𝑧 = 0 … … … … … … … … … … . (𝑖𝑖𝑖)
𝑥 + 𝑒 −𝑥 = 𝑦 (𝑓𝑟𝑜𝑚 𝑖𝑣)
1
𝐼𝑓 𝑥 = 0 𝑡ℎ𝑒𝑛 𝑒𝑞𝑢𝑎𝑡𝑖𝑜𝑛 (𝑖)𝑔𝑖𝑣𝑒𝑠 + 𝜆2 = 0, 𝑤ℎ𝑖𝑐ℎ 𝑐𝑜𝑛𝑡𝑟𝑎𝑑𝑖𝑐𝑡𝑠 (𝑣) ,
2

𝑇ℎ𝑢𝑠 𝑥 > 0 ℎ𝑒𝑛𝑐𝑒 𝜆2 = 0

1
𝐹𝑟𝑜𝑚 (𝑖) − ( 1 − 𝑒 −𝑥 ) = 0
2
1
𝑒 −𝑥 =
2
𝑥 = ln 2
𝐵𝑈𝑇 𝑥 + 𝑒 −𝑥 = 𝑦
1
𝑡ℎ𝑢𝑠 𝑦 = ln 2 +
2
1
The optimal point is (𝑥, 𝑦, 𝜆1 , 𝜆2 ) = ( 𝑙𝑛2, ln 2 + 2 , 1,0)

Putting a Lagrangean for 𝜆1 = 1 𝜆2 = 0

1
𝐿 = − 2 𝑥 − 𝑒 −𝑥 − 𝑧 2 check concavity and convexity
−𝑒 −𝑥 0 0
𝐻𝑒𝑠𝑠𝑖𝑎𝑛 = [ 0 0 0]
0 0 −2
𝐷1= − 𝑒 −𝑥 < 0;
𝐷2 = 0;
𝐷3 = 0.
Checking all the principal minors
∆1= − 𝑒 −𝑥 , −2 𝑎𝑛𝑑 0 ≤ 0;
∆2 = 0, 2𝑒 −𝑥 ≥ 0;
∆3 = 0.
Since (−1)𝑘 ∆𝑘 ≥ 0 then the hessian is negative semidefinite, implying that the Lagrangean with
1
equality constraints is concave, thus the optimal point (𝑥, 𝑦)(𝑙𝑛2, ln 2 + 2) is a maximum point.