You are on page 1of 318

Appendix A

Problem A.1
(a)

1 1 0 4  1 7 
A B =

 

 2 1 1 3  1 11
0 4  1 1 8 4 
B A= 



1 3   2 1 7 4 

A B  B A
(b)

1 2  0 4  2 10
AT B  



1 1  1 3 1 7 

(c)

1 1 0 4  1 5 
A B  



 2 1 1 3  3 4 

(d)

1 1 0 4  1 3
A B  



 2 1 1 3  1 2

(e)

det A  (1)(1)  (1)(2)  1
det B  (0)(3)  (1)(4)  4

(f)

 1 1
adj A  

 2 1

(g)

A1 

adj A  1 1

det A  2 1

Verify:

 3 4 
adj B  

 1 0 
B 1 

adj B  3 / 4 1 

det B  1/ 4 0 

 1 1 1 1 1 0 
A1 A  



 2 1  2 1 0 1 
 3 / 4 1  0 4  1 0 
B 1 B  



 1/ 4 0  1 3  0 1 

1

Problem A.2

(a)

1 2 3
A  3 2 1
1 0 1

1 
b   2
3 

det A  8

1 2 3 1 
D   A b   3 2 1 2
1 0 1 1 
subtract row 1 from row 3

1 2 3 1 
D  3 2 1 2 
0 2 2 0 

det A  8

multiply row 1 by 3 and subtract from row 2

1 2 3 1 
D  0  4  8  1
0  2  2 0 

det A  8

multiply row 2 by ½ and subtract from row 3

1
1 2 3

D  0  4  8  1
0 0 2 1/2 

det A  8

Thus,

2
3   x1   1 
1
0 4 8   x  =  1 

  2  
0 0
2   x3  1/ 2 
1

1

2

4

2x3   x3 

4 x2  8x3  4 x2  8(1/ 4)  1  x2  1/ 4
x1  2x2  3x3  x1  2(1/ 4)  3(1/ 4)  1

2

 x1  3/ 4
x  3 / 4  1/ 4 1/ 4

T

(b)

x  A1b
 2 2 2 
 2 2 2 


adj A  4 8 4 
1
A 

det A
8
 1/ 4
x   1/ 4
 1/ 4

(c)

T

 1/ 4
  1/ 4
 1/ 4

1/ 4 1/ 2 
1/ 4  1
1/ 4 1/ 2 

1/ 4 1/ 2  1 
 3/ 4 



1/ 4  1  2   1/ 4 
 1/ 4 
1/ 4 1/ 2  1 

1 2 3
6
x1   2 2 1  det A 
 3/ 4
8
1 0 1
1 1 3
2
x2  3 2 1  det A 
  1/ 4
8
1 1 1

1 2 1 
2
x3  3 2 2   det A 
 1/ 4
8
1 0 1 

Problem A.3

z1  3x1  x3
z2  x1  x2  x3

3 0 1
or z  Ax , A  1 1 1
0 2 1

z3  2 x2  x3
Solve x3  z1  3x1
z2  x1  x2  z1  3x1  x2  2 x1  z1

3

1

1

1

1

3

2

2

2

2

2

1

1

3

2

2

2

x2  z3  x3  z3  z1  x1

Then

z2  z3  z1  x1  2 x1  z1

x1  z1  2z2  z3
Now

x3  z1  3x1  z1  3( z1  2z2  z3 )
x3  2z1  6z2  3z3

and,

1

1

2

2

x2  z3  x3

x2  z1  3z2  2z3

or,

1
1 2

x  1 3 2  z
 2 6 3 

Also,

x  A1z

1
 1 1 2 
 1 2
 adj A  


1
A 
  2 3 6    1 3
2 

 det A 
 1 2
 2 6 3 
3 
T

1
This checks with the result obtained using algebraic manipulation.

Problem A.4

x   xT x 

1/ 2

x1  12  12  22 

1/ 2

 6  2.45

x2  12  02  22 

1/ 2

 5  2.24

1 
x x2  1 1 2 0   5
 2
T
1

4

1 
1 0 2 


x1 x  1  1 0 2  1 0 2
 2
 2 0 4
T
2

1 0 0  1 
1 




x Ax1  1 1 2 0 2 0  1   1 2 8 1   19
0 0 4  2
 2
T
1

T

Let x3   x31 x32 x33  . Then x1T x3  x31  x32  2 x33
Let x31  x32  1 . Then x1T x3  0  x33  1
x3  1 1  1 is orthogonal to x1 .
T

1
det  x1 x2 x3   det 1

 2

1 1
0 1  3
2 1

  0

x1 , x2 and x3 are linearly independent.

Problem A.5

 1 1
A   1 0
 2 3

2
1 
1 

Using minors of the second row,
det A  (1)(1)1 2

(1)(1)3 2

1 2
1 2
 (0)(1) 2 2
3 1
2 1

1 1
2 3

 1(1  6)  0  (1)(3  2)  8
Using minors of the third column,

5

det A  (2)(1)13

1 0
1 1
 (1)(1) 23
2 3
2 3

(1)(1)33

1 1
1 0

 2(3  0)  (1)(3  2)  (1)(0  1)  8

Problem A.6
(a)

0 1 
A

1 4 

 
A  I  
1

1 
4   

det( A   I )   (4   ) 1   2  4 1  0
Eigenvalues are 1  2  5 and 2 = 2  5 .
T

The eigenvector v1  v11 v12  corresponding to 1 is given by
1
v11 
0 1  v1 
1 4   2   2  5  2 
v1 

 v1 

Let v11  1

 1 
v



1
 v12  2  5
2  5 

T

The eigenvector v2  v12 v22  corresponding to 2 is
1
v12 
0 1  v2 
1 4   2   2  5  2 
v2 

 v2 

(b)

Let v12  1

 1 
v



2
 v22  2  5
2  5 

2 0 0
A  0 3 0  det( A   I )  (2   )(3   )(2   )  0
0 0 2
 eigenvalues are 1  2, 2  3
and 3  2 .
The eigenvector v1 corresponding to 1  2 is

6

Say v3  1 0 1 . T Problem A. v31 . and v32  0. Consider the asymmetric matrix  11   a21 a22  Its eigenvalues are the roots of the equation (  a11 )(  a22 )  a12 a21  0. or 2  2  (a11  a22 )  a11a22  a12 2  0 The discriminant of this quadratic equation is D = (a11  a22 )2  4(a11a22  a122 )  (a11  a22 )2  a122 Since D is the sum of the squares of two real numbers. it cannot be negative. Therefore a12  a the eigenvalues are real. v33 arbitrary. or 7 .7 a A 2x2 symmetric matrix has the form  11  a12 a12  a22  The eigenvalues are the roots of the equation (  a11 )(  a22 )  a12  0.1  2 0 0   v1   0 3 0  v 2   2    1 0 0 2  v13   v11  2v11  2v11  v11  arbitrary  2 2 2 2 v1   3v1  2v1  v1  0  v13  2v13  2v13  v13  arbitrary   Say v1  1 0 1 T The eigenvector v2 corresponding to 2  3 is 1  2 0 0   v2   0 3 0  v 2   3    2 0 0 2   v23   v12  2v12  3v12  v12  0  2 2 2 2 v2   3v2  3v2  v2  arbitrary  v23  2v23  3v23  v23  0   Say v2   0 1 0 T The eigenvector v3 corresponding to 3  2 is (by comparison with v1 ) .

1 1 0  A   2 3 1  1 0 1  Multiply row 1 by –2 and add to row 2. 1 1 0  A  0 1 1  0 1 1  (1) Multiply row 2 by 1 and add to row 3. Then. Then. “Nonlinear Analysis in chemical Engineering. (2) A 1 1 0   0 1 1  0 0 2  Now. and multiply row 1 by –1 and add to row 3.” McGraw Hill. we must have D < 0.A. NY. Thus an asymmetric matrix whose elements satisfy the condition (a11  a22 )2  4a12a21  0 has complex eigenvalues. 2  (a11  a22 )  a11a22  a12a21  0 The discriminant of this quadratic equation is D = (a11  a22 )2  4(a11a22  a21a12 ) = (a11  a22 )2  4a12 a21 For the eigenvalues to be complex. 1980.8 The LU decomposition algorithm used here is from B. Finlayson. Problem A. 8 .

x1  x2  1  x1  2 x   2  1 0 T Problem A. 2 x3  0  x3  0 From row 2.UA (2) 1 1 0  1 0 0    0 1 1  . and L   2 1 0  0 0 2 1 1 1  1 2 3  1 0 0 adj L   1 L   0 1 1   2 1 0  det L 0 0 1  3 1 1  T 1  1 0 0  1   1  ˆb  L1b   2 1 0 1    1        3 1 1   2  0 1 1 0   x1   1 Ux  bˆ  0 1 1   x2    1 0 0 2  x3   0 From row 3.9 g 2  x1 x2  4 g1  x12  x22  8. x2  x3  1  x2  1 From row 1. T Iteration 1: 9 .  g / x g1 / x2   2 x1 2 x2  J  1 1  x1  g 2 / x1 g 2 / x2   x2 Starting point x1   0 1 . Note: superscript denotes iteration number.

46475.03236 .47059 x23  x22  x22  2. x22  30.72059 2. The solution is x  1.44118  J3     2.96990.03594. x24  0.27410  4. x25  2. x24  2.00355 We stop now because g15 and g25 are “small enough”. x23  0.03236 Iteration 5: g15  0.25 /17 x13  x12  x12  2.50541 g 23  2. x22  x12  x12  4.72145  4.5 x12  x11  x11  4 .72059 Iteration 3: g13  5.58976 x14  2.56383 g 24  0.96990 2.47059  x13  0.13083 Iteration 4: g14  0.01168 4.01099 g 25  0.26166 J4     2.13083 2.09847 x15  1.25 g12  14  8 9 J    4.00584 x14  0.5 4 2  g12  J x    2  g2  2 2  x12  26 /17. x12  3.94118 5.5 Iteration 2: g12  28.00584. The exact solution is  2 2 .g11  7 g 12  4  g1  J 1x1    11   g2  0 2  J1    1 0   x11  4. T T 10 .

then T 8 8  .  4 4 cannot be used as a starting point for the Newton-Raphson method.If x1   4 4 is the starting point. T 11 . Thus. In this case x11 and x12 J1    4 4 cannot be uniquely determined.

y*  33.21 cm x*  204  8(33.21 cm.21) 33. 9. of variables = 2 No.1 Minimize: f(x.y) = xy Subject to: (x-8) (y-12) = 300 Total no.21 cm Neglecting the physically unrealizable negative value. y)  y  12 f (16 y  204)( y  12)  (8 y 2  204 y )  0 y ( y  12)2 (16 y  204)( y 12)  (8 y 2  204 y)  0 8 y 2 192 y  2448  0 y  33.14 cm 1 .SOLUTIONS MANUAL CHAPTER 1 Problem 1. of equality constraints = 1 No. of degrees of freedom =1 Independent variable: y Solution: Eliminate x using the equality constraint xy 12x  8 y  96  300 204  8 y x y  12 8 y 2  204 y f ( x.21  12 x*  22.

21 = 33.21 w* = 14. of equality constraints = 1 No. of variables = 2 No. Minimize: f = b2 + 4bh Subject to: b2h = 1000 Total no.Alternative Solution: Minimize: area = (w + 8) ( z + 12) St. wz = 300 area = wz + 8z + (2w + 96)  300  = 300 + 8z 12   +96  z   1  d(area) = 0 = 8 + 3600   = 0  z  2 z = 450 z* = + 21.21 Problem 1. we just need to minimize the surface area of the inside of the box. of degrees of freedom = 1 Independent variable = b b>0 h>0 Solution: Eliminate h using the equality constraint h 1000 b2 f  b2  4000 b 2 .14 y* = 12 + 21.14 = 22.14 x* = 8 + 14.2 Since thickness is uniform.

the volume is used.35 Note: It is easier to maximize ½ of the rectangle as it is symmetric. of degrees of freedom = 1 Independent variable: b 2 2 Solution: A  b(10  (b / 2)2 )  10b  b3 / 4 dA / db  10  3b2 / 4  0 b*  3. h > 0.)  maximum h*  10  (3.6 cm d 2 f  3  2  2  8000 / b  0 at b  12. of variables = 2 No. 3 .3 cm Note: Another viewpoint. If by material.65 (not reqd.)  minimum.65 d 2 A / db2  6b / 4  0 at b*  3.67 A*  (3.6) 2 h*  6.65)(6.6  (not reqd.67)  24. t Problem 1. and b>0.3 Maximize: A = bh  b Subject to: h = 10-(b/2) and     2 Total no. of equality constraints = 1 No. then the volume of a side is (b) (t) (h) and of the bottom is (b) (t) (b) so that the f objective function would be . Let t = thickness of material.6 cm 1000 h*  (12.df 4000  2b  2  0 db b 2b3  4000  0 b  12.65/ 2)2  6.  db  b*  12.

x2  x0  2h Let f  B0  B1 ( x  x0 )  B2 ( x  x0 )( x  x1 ) f ( x0 )  f0  B0  B1 ( x0  x0 )  B2 ( x0  x0 )( x0  x1 ) B0  f0 f ( x1 )  B0  B1 ( x1  x0 )  B2 ( x1  x0 )( x1  x1 )  f1 f  B0 f1  f0  B1 = 1 x1  x0 h f ( x2 )  B0  B1 ( x2  x0 )  B2 ( x2  x0 )( x2  x1 )  f 2 B2  f 2  B0  B1 ( x2  x0 ) f 2  2 f1  f 0  ( x2  x0 )( x2  x1 ) 2h 2 Problem 1. of degrees of freedom = 1 Independent variable: x To avoid using the square root.Problem 1. minimizing d is the same as minimizing D  d 2  x 2  y 2  x 2  (2 x 2  3x  1) 2  4 x 4  12 x3  14 x 2  6 x  1 dD / dx  16 x3  36 x 2  28 x  6  0 8 x3  18 x 2  14 x  3  0 4 . of variables = 2 No.4 Let x1  x0  h.5 Minimize d  x2  y 2 Subject to y  2x2  3x  1 Total no. of equality constraints = 1 No.

of degrees of freedom = 1 x*   ( x0  x1 ) B2  B1 x0  x1 f1  f 0    h  2 B2 2  f 2  2 f1 _ f 0  5 .4544355 .341 minimum.25. A cubic equation has the form x3  a1 x 2  a2 x  a3  0 Let p  (3a2  a1 ) / 3. If we have f  C0  C1 x  C2 x 2 .209) is closest to the origin.375 p  0. then the cubic equation has one real root and two complex conjugate roots. The real root is x1  A  B  a1 / 3 where A  3 (q / 2)  R B  3 (q / 2)  R For our problem. “Chemical Engineers Handbook”. x*  0. B = -0. 0. p.H. q  (27 a3  9a1a2  2a1 )27 2 3 and R  ( p / 3)3  (q / 2)2 . q  0.H.09375. Note: You can use a least squares method too.341 and y*  0. this is equivalent to C0  B0  B1 x0  B2 x0 x1 C1  B1  B2 ( x0  x1 ) C2  B2 df / dx  B1  B2  ( x  x0 )  ( x  x1 )   0 Total no. of variables = 1 No equality constraints No.You need solutions of a cubic equation (Ref: R.2063 x103 Since R > 0.75. there is one real root.341.. A = 0. Perry and C.0625. Anilton.0458311.2-9.209 d2 D / dx2  48x2  72x  28  9. R  2. a2  14 / 8  1. If R>0. a1  18 / 8  2. 5th ed. or use a computer code.03 at x  0. a3  3/ 8  0. (-0.

sin   1/ 3. d2V / d 2  4 R 3  0  minimum At cos  2 / 3. f has a maximum at x* if B2  0. or cos = 2/3 and sin   1/ 3 d2V / d 2  2R3 (2sin3   7cos2  sin  ) At cos  0. of degrees of freedom = 1 Independent variable =  Solution: Eliminate r and h using the equality constraints. of variables = 3 No.6 Maximize: V  r 2 h Subject to: r = R cos  h = 2R sin  0 <  < /2 Total no. V  2R3 cos2  sin  dv / dB  2R 3  2 cos  sin  ( sin  )  cos3    0 2cos sin 2   cos3   0 2cos (1  cos 2  )  cos3   0 cos (2  3cos 2  )  0 cos  0 and sin   1. Problem 1.sin   1. of equality constraints = 2 No. d2V / d 2  8R3 / 3  0  maximum  2  1  V *  2R3      3   3  6 . f has a minimum at x* if B2  0.d2 f / dx2  2B2 .

nA > 20 nB > 0 nC > 0 (physical requirement) Problem 1.7 0<x<1 f(x) > 0 Problem 1. of trucks of type C Objective function Minimize f = 2100nA + 3600nB + 3780nC (ton-mile/day) Constraints 1. 10.000 nB + 23.4x1-1. of trucks of type A nB= no.000 nA + 20.8 Let nA = no.000 nC < 600.30 7 . nA + 2nB + 2nC < 145 (drivers) 3.000 ($) 2. nA + nB + nC < 30 (trucks) 4. of trucks of type B nC = no.66 + 91.5x3-0.9 Minimize f(x) = 19.8x2-1.V*  4 3 3 R 3 Problem 1.47 + 16.

4 (dependent) C4 is in the objective function (independent) 3. say x1 (or x2). Objective function. x2  0 No. Maximize C4 2. Equality constraints V i  20 so that Vi 20   q  71 i Material balances 8 .05 x0 > x1 > x2 > x3 x2 > 0 x1 > 0 Problem 1. 3. variables: 2 x1 and x2 One equation reduces the number of independent variables to 1. 2. Variables C1.10 Minimize: f ( x)  4x1  x22 12 Subject to: 25  x12  x22  0 10 x1  x12  10 x2  x22  34  0 ( x1  3) 2  ( x2  1) 2  0 x1 . C2. Problem 1.Constraints 0 < x3 < 0.11 1. C3 1.

2.10765  p*  0 inequality T*  0 1750. Problem 1.40 + (350 – T)1.C1  C2   2 kC1 n C2  C3   3 kC2 n C3  C4   4 kC3 n 4.286 equality 235. T p = p* since water condenses Constraints: p* n*   0.7 psia 1.0  T  273.7) psia.12 1.. Think of the solution of the two ODE’s -. Inequality constraints i  0 Ci  possibly Problem 1.01 pT nT and log10 p*  8. One technique of solution would be to apply NLP to the above statement.9 Variables p*. Objective function Minimize F = p*1.01(14. 9 .t is fixed. solve for T.15 Let pT = 14. Introduce this value into the Antoine eq. (This procedure implies 2 equality constraints exist as the problem has no degrees of freedom). and then calculate F. Another technique would be to assume p* is at its bound so that p* = 0.13 (a) The independent variable is not time but temperature (via the k’s).

14 (a) The problem consists of (at constant T and p) 10 . and directly corresponds to temperature. This insight eliminates the exponential terms and simplifies the structure of the problem. u2  k2 2 simplifies the optimization problem to: Maximize: y2 (1. (d) The inequality constraint is T < 282F.0) y1  (u  u 2 / 2) y1 Subject to: y2  uy1 y1 (0)  1 .(b) The dependent variables are A and B. as for example: - get analytical solution of A and B vs T and minimize convert ODE’s to difference equations (constraints) and minimize approximate solution via collocation (constraints) and minimize introducing the following transformations: y1  A A0 y2  B A0 u  k1 . (c) Equality constraints are the 4 equations (including initial conditions). Problem 1. y2 (0)  0 0u 5 Note that the control variable u(t) is the rate constant k1. Also implicit are T > 0 A>0 B>0 t>0 (e) Any answer is ok.

(b) For C + D  A + B nA n Kx  T nC nT nB nT nD nT The element balances are based on At start (bk) C 1 H CO H2O - CO2 H2 At equilibrium O 1 n *CO C H - n* CO O 2 1 - 2n *H2O n *H2O - - - n *CO2 - 2n *CO2 - - - - 2n *H2 0 Total moles n* = Total moles = 2 n*CO  n*H 2O  n*H 2  n*CO2 As variables use xi or ni (either are ok) C O H balance: 1 = nCO  nCO balance: 2 = 2nCO  nCO  nH O balance: 2 = 2nH  nH O 2 2 2 2 2 11 .Minimize: G   ( io  RT l n p  RT l n xi )ni i   = RT ln p +   io  RT  ln xi  (ni ) constant i  i    n n  RT ln K x ni  in i subject to the element balances:  aik ni  bk for each of the elements k = 1 … M i and inequality constraints ni  0 with ni  xi n.

15 k 1 k 1   k     P3 k  kPV 1 1  P2 W   2   k  1  P1   P2     1 1     k  1  k      p p3   dW kPV k  1 p k  1 3 1 1  2           0 dP2 k  1  k  p1   p1  k  p2   p22      1 k  1 k  p2   1   p3   p3        2   0  p1   p1   p2   p2  p2 1/ k p  1 1 2 k 1 2 1 1 p1 k  p3 1 k 1 p2 k 1   p1 p3  k 2 0 1 p2  p1 p3 2 p2  p1 p3  (1)(4) = 2 atm Problem 1.Kx  xi  xCO2 xH 2  xCO xH 2O nCO2 N H 2 nCO N H 2O ni so that nT cancels nT Problem 1.1  2  0 dP P P*  300 bbl/day 12 .1P  9000 ($ / bbl) P dC 9000  0.16 (a) C  50  0.

17 Basis: 1 hr Heat balance for the gas: q= m Cp T q  (3000) (0.2 P  0  0 dp 250 P*   1250 bbl / day 0.1P     P  bbl  day   (c) df  300  50  0. Annual cooling water cost ($) 13 .45 x104 Btu/hr Heat balance for cooling water q  m(1) (To  80)  m(To  80) Btu/hr so m  q 70  80 For the heat exchanger q  UATLM  8 ATLM Btu/hr so A  where TLM  q 8TLM (195  To )  (90  80) in 0 F  195  To  ln    90  80  Basis: 1 yr.(b) f  300  50  0.1P  9000 ($ / bbl) P 9000  $  Pbbl   f   300  50  0.2 (d) They are different because you can sell more Problem 1.3) (195  90)  9.

8  0.8  0.5 L  0.4  1000  2.2 D4.1421 1m1.9063x103 TLM dc  1.4615  0.5)   8TLM  5.2 D0.8 L dc  1.1421C2  2m2.2      (24 x 365)  To  80   62.2V 0.5C1D 0.45x104    (0.2  0. Problem 1.5 L  0.8 L The cost function in terms of D is now C  C1D1. 9. p  60 lb/ft 3 .1538 m0.5 L  0.2 D 5.682C2  2 m2.5C1 D 0.8  0.081C2 m3  2 LD 6  0 dD D opt C   0.42 lb/ft hr) .3077 For   1cP(2.2 D4.5 L  C2mP /  where P  2 V 2 L / Df 0.2 V  4m /  D2 f  Substituting the expression for f and V into that for P.45 x104   1  0.366ft. we get P  0.046 0.6533 x103  To  80 Annual fixed charges for the exchanger ($)  9.8 L = 0 dD 14 .638  2   C1  0.18 (a) C  C1D1. D opt  0.

3174  0.5ft/hr For   0.   50 lb/ft 3 D opt  0.317 m0.0476 0.2 lb / ft hr).7618 (b) 0.387ft.42 lb/ft hr).2ft/hr For   10cP(24.m.0478 L C1  1.828C1 0.4755m0.m.366 m0.0634 0.5   for C in $/hr 13 2 2  C2  2.317   dln p   .2382  04784 m0.0317 C  1  dln Dopt  Dopt S    0.   60 lb/ft 3 D opt  0.6ft/hr Problem 1.42363 x103 $ / hr ft 2.7097 x10 $hr /ft lb   For   1cP(2.444  0.484 lb/ft hr).384ft.2cP(0.1587 C  D  0.C opt 2  d ln Dopt  SDopt    0.377ft .   80 lb/ft 3 D opt  0.317   d ln    .882  2   C1  From this.8413 C2  0.0317 opt V C   1.Solving this equation for D.112  0.666  0. 0. V opt  1363.6688  0.1587 C2  0. V opt  895. we get C  D  0.1587  0.444  0.2596C1 0.317 m0.6367  1   C2  opt C opt  0. V opt  1151.882  2   0.19 0.C2 15 .

666  0.828C1 0.1587   2 dln C  2   .20 The variables selected could be times.2596C1 0.1587 C2  0. 2.  .m C where T1  0.4784 m0.666  0.m.0478 L   ln C opt  1   opt (0.8413 C2 0.  .2382 opt SCC2  0.C2 C  ln C opt  1   opt (0.4755m0.666T1  0.6688  0.1587T1  0.0478m0. Let Xij be the number of batches of product i (i = 1.6688T2 )   ln m   . 3) produced per week on unit j (j = A.C2 C Copt  ln C opt  1   opt (0.m C opt  0.42 lb/ft hr) and T2  0.444   dln m   .m.  .C2  dln D opt  SCDopt   0.2596C1 0.6688 L For   60lb/ft 3 and   1cp(2. dln D opt  SmDopt    0.2382T2 )   ln C2   .8413 C2  0.666 C2 0.C2 C   ln C opt  1   opt (0.4784T2 )   p   .0476 L 0.  .163 Problem 1.C).1587  0. 16 .4755  0.0476m0.0478T2 )    ln    .476 opt  0.B.7618 S Copt S Copt Sm Copt SC2 0.828C1 0.476T1  0.2382  0.7618 S C S C SmC opt   0. but the selection below is easier to use. We want to maximize the weekly profit.4784  0.4755T1  0.476 opt  0.

B. 3 .1X 3C  5 and non-negativity constraints xij > 0 . Units: batch/week X 3 A  X 3B  X 3C  20 (none on 1 and 2) Hours available on each unit  0. i = 1.8 X 1 A  0. Let rain fall for T hours at a stretch (T should be specified). j = A.21 We have to minimize the pumping rate subject to the constraint that the basin cannot overflow.3 X 3 A  20 hr week 0.3 X 2 B 0.8hr  X batch      batch  week  Unit A 0.2 X 2 A  0. 2.2 X 1C  10  0. The volume of rain during this period is A(a  bT 2 ) in3 The maximum amount of water that can be treated during a time period T is PmaxT in3 17 . C Problem 1.Objective function: Units: ($/batch) (batch/week) = $/week: Maximize: f ( X )  20( X1A  X1B  X1C )  6( X 2 A  X 2 B  X 2C )  8( X 3 A  X 3B  X 3C ) Subject to: Sales limits.4 X 1B  0.

A(a  bT 2 )  PmaxT V The minimum Pmax is therefore given by A(a  bT 2 )  PmaxT  V 1 Pmax   A(a  bT 2 )  V  T or Of course. we must have Pmax > 0 Problem 1.t (feed conversion) x = xi (r.z ) = 0 (initial conversion profile) Cp T T k 1   T  VG  r t z R 2 r r  r    (H )  k f (1  x)  kr x   T / dr  0 at r  0.Thus. all r . t (symmetry) kT / r  U (T  To )  0 at r  1.22 Assume: (i) first order reactive. z) at t  0 (initial temp. all z. profile) Inequality constraints: 18 . all z. t (impenetrable wall) x = x0 at z = 0. t (symmetry) x / r  0 at r  1. L. t (feed temperature) T  Ti (r. all r. t (heat transfer to jacket) T  Tin  0 at z  0. all z. )dr 1 0 Equality constraints: x DAB 1   x  x  2  k f (1  x)  kr x  0  r  V t R r r  r  z x / r  0 at r  0. all z. (ii) flat velocity profile Objective function: maximize  rx(r.

t Tmax  T  0 at all r .x  0 all r . t T  0 all r .5 L + 200 DL (a) C  250 L(2.23 C  C1D1.3077 For   1cP(2.5 L  C2mP /  where P  (2 V 2 L / D) f f  0.1538 m0. t To  To min To  To max Problem 1.   60 lb/ft 3 .366 ft.5C1 D 0.005 V  4m /   D 2 Substituting the expressions for P.24 C = 7000 + 25002.5  200 L D is the absolute sensitivity 19 .081C2 m3  2 LD 6  0 dD D opt C   0.5 L  0.4615  0.016C2m3  2 D5 L dC  1.638  2   C1  0. z.5 L  0. we obtain C in terms of D: C  C1D1. z .5) D1. Problem 1. f and V into the cost function. z . Dopt = 0.42 lb/ft hr) .

5 D  2000  250 D  200 DL  D (b) The relations for sensitivity are the same. Problem 1.25 ln C  a0  a1 ln S  a2 (ln S ) 2 dln C  dC dS dS  a1  a2 (2)(ln S ) C S S dC C C  0  a1  2a2 (ln S ) dS S S or (a) a1  2a2 ln S or (b) a1 1  ln or S  e  a1 / 2 a2 2a2 S dC / C  a1  2a2 ln S dS / S Problem 1. the constraints limit the feasible region of application.26 Refer to Section 1.7 of the text.C D C    250 L(2.5) D1.5  200 L   is the relative sensitivity 2. 20 .

then the model is linear.2 (a) (b) Nonlinear Linear if vx is independent of vy.4 2. then the model is nonlinear. (b) Steady state (except on start up and shut down) (c) Linear Problem 2.1 (a) (b) The model is linear if the ratios h1 / S1 1Cp1 and h1 / 2Cp2 S2 plus V are independent of temperature.CHAPTER 2 Problem 2.1 (a) (b) 2. nonlinear Problem 2.6 Total variables (2 streams + Q): 2(C+2)+1 = 2C+5 1 . If they are not.1 (a) (b) 2.3 2. A plug flow mode is also possible. Problem 2.2 (a) (b) Unsteady state Unsteady state Steady state Steady state Problem 2. otherwise. it is nonlinear. Else.5 (a) A distributed parameter model would be best. If D is independent of concentration.2 (a) (b) Distributed Distributed Lumped Distributed Problem 2.

In some instances. of degrees of freedom = 2C+5-(C+1) = C + 4 Conventional specifications are the variables in the entering stream (C+2) and the temperature and pressure of the exit stream.Constraints: independent material balances: C energy balance : 1 no. Q may be specified rather than the temperature of the exit stream. Problem 2.7 2 .

0109 -0.3035 log x/log y 1.59 1.0245 0. y = a xb is a good functional relationship.9 68.66 Since  log x/ log y is nearly constant.0570 0.9 81. a good functional relation is y= a bx (c) x 2 4 8 16 32 64 128 y 0. 3 . a linear fit is best: y = ax + b (b) x 2 5 8 11 14 17 y 94.1295 0.0102 Since  log y/x is nearly constant.0125 -0.3 74.60 1.68 1.7 64.2000 0.Alternative Analysis (a) x 1 2 3 4 y 5 7 9 11 y/x 2 2 2 Since y/x is a constant.67 1.0 log y/x -0.0855 0.0115 -0.71 1.8 87.0113 -0.0370 0.

8 4 .Problem 2.

074   8 7200 8160000       12777.Problem 2.1217  0.001673 x) 5 .698        0. a good model is ln Y     x or Y  e   x E=  (   x  ln Y ) i 2 i E /   2 (   xi  ln Yi )  0 E /   2 xi (   xi  ln Yi )  0  n  xi      ln Yi     2     xi  xi       xi ln Yi  7200     11.   0.1217.9 A plot of the date looks like: Since the data seems to lie on a straight line after the transformation.001673 Y  exp (0.

11  e 3 xi yi = 1.0552 C1 = -1.0524 = 1.-1. C1 and C2 4 E=  (C i 1 0  C1e3 xi  C2 e 3 xi  yi ) 2 4 E / C0  2 (C0  C1e3 xi  C2e 3 xi  yi )  0 i 1 4 E / C1  2 e3 xi (C0  C1e3 xi  C2e 3 xi  yi )  0 i 1 4 E / C2  2 e 3 xi (C0  C1e3 xi  C2e 3 xi  yi )  0 i 1  n  3 xi  e   e3 xi  n=4  e3 xi  e6 xi  e 3 xi  e 6 xi  yi  e3 xi  e6 xi n  e3 xi  C0    yi   n   C1     e3 xi yi   e6 xi  C2    e3 xi yi  = 8527.00248 =6  e3 xi yi = 8951.Problem 2.10 (a) To find Co .0551 If C0 is set equal to zero.10466 Solution of the set of 3 linear equations gives C0 = 2.6 = 65823128 = 1.302 x10-4 C2 . then E   (C1e3 xi  C2e3 xi  yi )2 E / C1  2  e3 xi (C1e3 xi  C2e3 xi  yi )  0 E / C2  2  e3 xi (C1e3 xi  C2e3 xi  yi )  0 6 .

595 a2 = 4.41005   6 14  a    2.3592 x10-4 C2 = 1. Then E =  (b1  ln xi  a2 xi  ln yi ) 2 E / b1  2 (b1  ln xi  a2 xi  ln yi )  0 E / a2  2  xi (b1  ln xi  a2 xi  ln yi )  0  n  x  i   xi   b1    ln yi   ln xi     xi2   a2    xi ln yi   xi ln xi   4 6   b1   13. then ln yi  ln a1  ln xi  a2 xi Let ln a1  b1 .101385 (b) If yi  a1 xi e a2 xi .6026759   2   b1 = 10.  e6 xi   n n   C1    e3 xi      e6 xi  C2    e3 xi yi   yi  Solution of this set of equations gives C1 = 1.5435487 7 .167834  a1 = 26051.

11 Problem 2.686198x + 0.384431 + 0.12 Y = 0.Problem 2.0731065 x2 8 .

1.325737x3 Problem 2. Form the objective function 2.77758x .33ci ) 2 dF  2 ( i    1.Y = 0. Apply a NLP code Problem 2.33Ci )(1)  0 solve to get d 9 .14 F=  ( i    1.87157 + 3.13 Apply least squares as the objective function (y Minimize exp  b0  b1e  cxi ) 2 Subject to x > 0 1.83175 x2 + 0.

15 xT xb  xT x b  ( xT x) 1 xT y P  xT x    x  x2  x  x2  x3  x 2   11 31 111    x 3    31 111 451   x 4  111 451 1959    y  105. set the derivative = 0. You need two independent variables to even obtain an orthogonal design.6   x y    xy   3601    x 2 y  1460  T 8.93 x + 8. the design is not orthogonal. and then differentiate.24 Is the Design Orthogonal? No.24  b   2.33  ci 29.98 Problem 2.95 x2 –2. 10 .95  y = 0.04)   1.93 0.optimal value of    i  1.33(15. Our problem has only one (x). and solve the same equation for .94  1.98 5 5 or introduce the data into the equation and sum 5 linear equations squared.  = 1.

5813  1.16 b  ( xT x)1 xT Y       x        b0  b  b1  b2  x0 x1 1 1 1 1 1 1 1 1 160 160 160 160 200 200 200 200 x2  4  1  5     1 10   7    11  7 Y    24   1     26  1  35     7  38   7  Use a computer to get  91.Problem 2.17 n E   (   xi  yi ) 2 i 1 n E  2 (   xi  yi )  0  i 1 n n i 1 i 1 n    xi   yi  0 n 1     yi    xi  n  i 1 i 1  n 11 .33 b  0.4588  Problem 2.

5  4. Y   0  1 x1   2 x2  3 x12   4 x22  5 x1 x2 12 .8 x1 1 0.5 -0.55 x2 Problem 2.8 78.5 x1  0.0 78.9 91.866 0 x 1 1 1 1 1 1 4 The last point has a weight of 4 because it is the average of 4 data points.866 0 -0866 -0. we get p1 -1 1 -1 1 For  n    p1  p  2 p2 -1 -1 1 1 Y 24 44 4 20 y  0  1 p1  2 p2 .Problem 2.18 Let p1  x1  3 x  260 and p2  2 2 20 When the Y values for a given ( p1 . 1  9.866 0.19 y 96.  2  11 Y  23  9 p1  11 p2 Y  152.6 64.5 0 x2 0 0.5 0.7 76.5 -1 -0. p2 ) are averaged. we have  p1  p1 2  p1 p2  p2    0    Y    p1 p2   1     p1Y  2  p2    2    P2Y   4 0 0    0  92  0 4 0      36    1   0 0 4   2   44  0  23.7 54.

E  (C1xA  C2 s  Y )2 13 . 7 Min E =  w ( i i 1 0 1 x1i   2 x2i   3 x12i   4 x22i  5 x1i x2i  Yi )2 The answer is  0  91.2  5  6.06   90 0 4500  C2  39  C0  0.8 1  16.1. (b) If Yp  0 for xA  s  0.09 0  C1   0.378  3  16. without a least-squares fit.68 Problem 2.3 0. C3  0.0  3 0.20 (a) E = (C0  C1 x A  C2 s  Yp ) 2  n    xA  s   xA  xA  sx A 2  s  C0    Yp      x A s  C1     x AYp   s 2  C2    sYp    0.986 E  106. then C0  0 .3 90  C0  1. C2  0.Then. one can also solve three equations in the three unknowns.483  2  3.5  4  17.3333.00667 In this case.

H.E / C1  2  x A (C1 x A  C2 s  Y )  0 E / C2  2  s (C1 x A  C2 s  Y )  0   x A 2  x A s  C1    x AY     2    x A s  s  C2    sY  0  C1  0. balances (C.667. N) energy balance fixed O2/N2 ratio in air zero concentration of components in fuel air : : : 4 1 1 : : 4 4 14 .22 There are six components.09   0 4500  C2  39   C1  0. of degrees of freedom = (5c+11) – (2c+3) = 3c + 8 Problem 2. Total variables (5 streams + Q) : 5 (C+2)+1 = 5 C+11 Constraints: independent material balances : energy balance : equilibrium relationships : T same in each phase : P same in each phase : C 1 C 1 1 2c+3 No. Therefore. C2  0.008667 Problem 2. the total number of variables is (3 streams + Q): 3(C+2) + 1 = 25 Constraints: (1) (2) (3) (4) matl. and not the same as any of the other streams.06  0. O.21 Assume that the feed is a single phase stream.

of variables for 7 streams: 7(C+2) = 35 Constraints: (1) (3) 3 independent material balances for each piece of equipment : one energy balance for each piece of equipment : specification of F0 .23 Total no. wA 0 . WA4 (4) and F5/F4 P-V-T relationship for each stream (2) : : 12 4 5 7 28 No. Problem 2. The last variable would be the CO/CO2 ratio. Problem 2. air or flue gas flowrate.(5) (6) (7) flue gas specification of % excess air specification of % N2 in fuel specification of three temperatures : : : : 1 1 1 3 20 No.24 Objective function: Minimize total cost per year Ctot  (Cs  Ct )r  Cop or minimize present value of total costs  N  Ctot  Cs  Ct    Cop  F  i 1  where r = capital recovery factor F = discount factor 15 . Three pressures must be specified. and one extensive variable – either feed. WA 0 . of degrees of freedom = 25-20 = 5. of degrees of freedom = 35-28 = 7.

Subject to: Cs  2.4(273  Tg ) Q  U0 A0 Tin Q  Wg Cp(T2  T1 ) U0  0.5V 2  g fLsV 2  g  2g 2 gdi f = 0.023(3600  gVdi /  ) 0.95hio hio  hi (di / do ) hi di / k  0. of shell for triangular pitch P  1.053 kcal /(m)(hr)(o C ) T n  (T0  T1 )  T0  T2 )  T T  n 0 1   T0  T2  T1  1200O C T2  350O C 16 .14 kg m / h k  0.25d 0 pitch  4n  1  np     3  Ct  150 A0 A0  d 0 nL 1/ 2 Cop  Wg Pg /(102 g )(3600) (8400 hr $ (0.8 (C p  / k ) 0.06 ) yr kwh Wg  25000 kg/hr 1.4 C p  0.24 kcal/kg oC   0.023 2 Y  4Wg /(3600  g di n) Pg  Pg  273MW /  22.5Ws Ws  DzLs  s D  P (n p  1)dia.

00 9372.35 7923.1(1.000  i (1  i ) n   0.54 16274.24 2824.54 for each payment each year or use P  100.32 51588.08 14795.87 61693.55 8682.47 79231.13 7087.1 P = 100.54 16274.54 16274.54 16274.05 Interest ($) 10000.54 16274.28 70879.1)10   (1  i) n  1  P   (1.38 28245.54 16274.11 40472.1)10  1  (100000)     = $16274.46 86823.51 1 .00 93725.99 6169.54 16274.51 1479.54 16274.33 5158.000 and the present value of the sum of future payments must equal 100. 000  F1 F2   1 (1  i) (1  i)2  F10 (1  i)10 The schedule of interest and payments is: Year 1 2 3 4 5 6 7 8 9 10 Amount Paid annually ($) 16274.81 4047.54 Year ending principal balance ($) 100000.54 16274.CHAPTER 3 Problem 3.

15 (1.15)5  1 0.94 5 (1.000 – 5.Problem 3.05% Problem 3.15) (1.825. 000 15.15)5 or P  15.825. 000   $304. 000 550.15) (1.4 Split the problem into three parts: the present value of $550. the expense of $25.2 Io = $10000 . 000 550. 000   $304. i = 12% Year 1 2 3 4 5 6 7 8 9 10 NPV = (a) -1000 -1000 -1000 -1000 -1000 -1000 -1000 -1000 -1000 -1000 -5650 (b) -800 -1400 -1200 -1000 -1000 -1000 -900 -900 -900 -900 -5779 Problem 3.15)5 2 . 000     2 2 3 1.15) (1.000 at the end of year 2 (a negative amount).000 received at the end of 5 years.3 10.15(1. 000 15. 000 15.15)4 15.15) (1.15)5  25.15) (1. and 5 dividends each of $15. P  15.94 2 (1.000 received at the end of each year for 5 years. 000 (1. 000 25.000 (1+i)8 = 0 which yields i = 9.

6 i = 0.52. The improvement should not be implemented.127)  394.005) 24  350(1.005 12 F0  200(1  0.5 The time line is Answer -$980.06 i  0.005)12  250(1.34 F1  350(1  0.42 Answer: $899.1967)  239.005)36  200(1.Problem 3.50 F2  250(1  0. Problem 3.0617)  265.27 Problem 3.7 This is a problem in which the payments are not uniform n F   Fk (1  i ) n  k 1 applies k 1 3 .

and add the two parts to get the answer.An iterative solution is needed.0% Date Payment Aug Sep Oct Nov Dec $3600  Payments Interest Principal $1400. but the benefit is received at end of the year only $6740. The interest is charged and the payments made 24 times a year. Problem 3. You can split the problem into one initial payment of $775 and a subsequent series of equal payments of $50.00 1374.21 n (1  i) (1.13 F $2696.13 P   $2544. The answer given is 58 payments (29 months).10 The statement is true for the same interest rates.67 2251.44 Problem 3.33 1348. For example: 4 .00 2225.52 $6740.10 1295.9 i = 14.48 40% of the interest paid is a benefit.40) = 2696.63 2277.32 (.0117)5 Problem 3.90 2304.8 The time line is Answer: $717.37 1322.32 $2200.

359 $164.10(1.000 $23.12)30 = $-177. Year 1 2 3 4 Balance Due at Beginning of Year.425 $8.41 From the values of F you can split each payment into interest and principal.641 $94.000 $235.11 You borrow $300. and find that the sum of the interest payments over the 30 years is higher than that over 15 years. You plan to pay in equal annual end-of-year installments.10)  F15  131.1030  1  1000  F30  30   0.641 Problem 3.12(1.721 j1 5 .604 Total Payment $ $94.10)  (b) F30  106.037 Interest Payment $ $30.12)30  1 0. Problem 3. Fill in the following table.1015  1  1000  F15  15   0.105 $78.12) j (1.253 $86.10(1.641 $94.536 $16.12 Plan A: 1 (1  i)n  1  1 (1  i)n i(1  i)n 30  30 NPV of 30 year costs = -160000 + 2200  (1.037 Principal Payment $ $64.216 $86.641 $71.04 For 15 years  1. $ $300.000 for 4 years at an interest rate of 10% per year.(a) For a $1000 mortgage at 10% paid over 30 years (F is the annual payment)  1.641 $94.

288 Plan B is favored because its NPV of costs is higher (less negative) than that for plan A Problem 3.12) (1.13 The total annual cash flows for each reactor.12) (1.12)30 (salvage value) 10   2500 j j1 (1.000 5600 30 yrs plus 2500 for 30 yrs 58.000 5600 5600 10 yrs 34. and the present values of the cash flows and total costs are tabulated below: 6 .12) j  21 (1.12)20 (capital costs) 4500 20 4500 30 4500   j j j j 1 (1.12) j 11 (1.12) (operating costs) 5600 5600 5600   10 20 (1.Plan B: 4500 34.12) 30 (58000)   (ditch costs) = $-160.12)10 (1.000 20 yrs 34.000 NPV of 30 year costs = 34000  34000 34000  (1.

Glass-lined reactor: (all flows in $) Note: -2400 + 1700 = -700 Year 1 2 3 4 5 6 7 8 9 10 Operating costs -700 -930 -700 -700 -1260 -1600 -1600 -1600 -1600 +2400 PV (i = 0.1) -636 -769 -526 -478 -782 -903 -821 -746 -679 +925 PV (i = 0.2) -583 -646 -405 -338 -506 -536 -447 -372 -310 +388 -5415 -3755 (including salvage value) 7 .

5) (6000) = 9000 to 13.2.000 per year.15 For after tax profits of $10. Problem 3.With the installed cost included.39% 8 .  5415  24000  $  29415  PV   3755  24000  $  27755 for 10 years Cast iron reactor: (all flows in $) Note: -1440-730=-2170.75% For after tax profits of $12.14 Project A has the largest rate of return. -2170 + 800 = -1370 year 1 2 3 4 Operating costs -2170 -2170 -2170 -1370 PV (i = 0. the value of all costs at the initial time is for i  0.1) -1973 -1793 -1630 -936 -6332 PV (i = 0. the value of all costs at the initial time is for i  0.1.  6332  7200  $  13532   for 4 years  5232  7200  $  12432  You have to calculate operating costs for another 4 year period followed by a 2 year period for a total of 10 years. 20 12000  (1  i) j 1 j  100000  0  i  10. 20 10000  (1  i) j 1 j  100000  0  i  7. for i  0. because most of the cash flow is returned early in the life of the project. Problem 3.1.500 gives a value smaller than that for the gas lined reactor.000 per year. The cast iron reactor is favored for either interest rate. and is discounted less. but you can see that adding roughly (1.2) -1808 -1507 -1256 -661 -5232 With the installed cost included. for i  0.2.

000 per year.746 kw 1 hp 1 actual power 0.98% i / i s / s For a + 20% error.39  7.031 PV of operating cost 6460 Operating costs are more substantial.7 theo.16 Installed capital cost = $200/hp Operating cost = $0.For after tax profits of $8. RS = (10.264 = $6460 PV of capital cost 200   0.264 Basis: 1 hp Installed capital cost = $200 Assume the pump efficiency corrects the $0. Operating cost $0.75  1.04 to actual cost. 9 .04 / kwhr  = 70% i = 10% Basis: 8000 hr/yr operation Assume life = 5 years Then. r = 0.70 12000 10000) /10000 Problem 3.75) / 7. power .04 Kwh 8000 hr yr 5 yr 0. 20 8000  (1  i) j 1 j  100000  0 Relative sensitivity =  i  4.

15) j j1 n PV of maintenance costs = $3760(1.000 $300(1. we solve the following equation for n: 10 .18 Let n be the payback period.17 The ratio is R  where I n  m 1 C j j  n 1 (1  i )  I = initial investment C = annual cash flow n = no.08) j  (1.Problem 3. of years to build the facility m = life of facility (years) R 1 17  j 2 106 150000 (1  i ) j i  11.08) j  (1. PV of initial investment = $10.6% 106 150000 (1  i) j i  8. R 1 19  j 4 Problem 3.15) j j1 n PV of savings = To find the payback period.5% For n = 2. R 1 18  j 3 For n = 3.0% 106 150000 (1  i) j i  9.

715 1  0 n   (1.0648)n  1  10000  3460  0 n   (0.0648)  n = 3. Problem 3.08) j n 3760(1.0648)(1.2 years.0648) 10000    (1.15) j (1.000 U Btu hr Overall heat transfer coefficient change gives Q 11 .300(1.08) j  0 (1.19 Objective function: Basis : 100 ft 2 C = Annual cost = -Energy savings  Capital Costs (in $/yr) (b) (a) Installation cost $0. = 75 t (75t) (0.5 t $ yr Heat loss savings are Q without insulation minus Q with insulation   Q = U Btu (hr) (ft2)( F) 100 ft2 (500-70)F = 43. 100 ft2 t in.30) = 21.75 (ft2) (in) (a) Capital cost per year.15) j j 1 j 1 n 10000   n 3460 0 j j1 (1.0648)    1 10000  54925.

5  0 2 (0.25  278t ) t  7.25  0. Hence we expect there to be a trade-off between capital cost and energy savings.25  0.278t   1 1   U 4 (12)(0.2t $ 0.60 (8700)(43. wsteam.25  0.11t   6 10 Btu  (0.2   21. The energy balance for the steam generator is woil C poil (400  T2 )  UA(400  T2 ) ln[150 /(T2  250)] (a) or 12 . First determine if any equality constraints exist in the problem.25  U 4 Heat savings per year  Btu $0. 000)  1.28t 249.30)     0. minimize C so set dC 0 dt 69.02in Problem 3.278t yr Constraints t>0 C>0 To get optimal t.25  0. The variables to be optimized include T2 and A as well as the amount of steam generated.278t )  yr (b) = $249.U  1 1 t  with:    0.25 0.20 (a) In this problem recognize that an exchanger of infinitely large area will maximize the energy recovery in the stream but at an exorbitant cost.25  0.278t ) (0.278t  1 1  without:  = 0.

(b) and (c). This outcome. the rate of return becomes negative. This occurs here because Io  0 for T2  400.6). leading to a maximum value at T2 = 400.064(400  T2 )  I o ln[150 /(T2  250)] (f) The maximum value of ROI must be found numerically because of the complicated expressions appearing on the right-hand side of (f). which is the same temperature as at the inlet. in fact. the investment Io is decreasing faster than is the numerator. Recognize that Eq. The profitability ratio is therefore ROI  F 0. In addition.woil C poil  UA ln[150 /(T2  250)] (b) The water converted to steam is obtained from woil C poil (400  T2 )  H v wsteam (c) where Hv = 950 Btu/lb and wsteam = lb/h. you can see that the ratio of F/Io may yield unrealistic results for an optimum. The capital cost is (dropping the “oil” subscript): I o  (25)( A)   150  ln   100  T2  250  25wC p (d) The annual credit for the value of the steam is  h   $    Btu    F  2x106    wC p (400  T2 )   8000    year   Btu     h     $  F  [0. is an unrealistic answer. If T2 > 400. Therefore we can express all costs in terms of T2 and with the aid of Eqs. (c) relates T2 and wsteam. (b) relates the variables T2 and A. (f) constrains the area to be above a minimum size. as T2  400. At the optimum an extremely high rate of return occurs (r = 9. The optimum is at T2 = 400F. From the above example. hence they are not independent. since it suggests the optimum return consists of an exchanger with infinitesimal area! Why does this result occur? The difficulty with using ROI as an objective function is that nothing in Eq. of course. Consider 13 . Eq. which can be found by applying L’Hopital’s rule to the above expression for ROI when T2 = 400F. corresponding to A  0.016]  wC p (400  T2 ) year   (e) Note that wCp for the oil appears in the expressions for both F and Io and thus cancels.

reformulating the problem using the net present value (NPV) of before-tax profits as an alternative objective function.  (1  i )n  1 NPV   F  Io n   i (1  i )  or NPV  F  Io r Since r is fixed by the assumptions about i and n.016) wC p (400  T2 )  25r wC p U ln 150 T2  250 (k) Note that wCp is a common factor of the two terms and will not be involved in calculating T2. (b) gives: f  (0. an equivalent criterion is r  NPV  F  rI o Note that this modified objective function (r  NPV) is equivalent to the use of the annualization factor (repayment multiplier) to obtain the capitalization charge. An example arises in optimizing pipe size to minimize pump operating costs and pipe investment costs. then F is negative.016 dT 100 T2  250 (l) T2  250  15. Instead of maximizing r  NPV .62r (m) 14 . Eq. (b) Let us use the net present value analysis to determine the optimum value of T2. In problems in which you seek to minimize only costs rather than maximizing profit because there is no stated income. Assume an interest rate for capital of 15 percent and a period of 10 years. Use of NPV means that a rate of return on the capital is specified. We can differentiate Eq. you can minimize (r  NPV) . (k) and set df/dT2 = 0: df 25r 1 0  0. The objective function for net present value (to be maximized with respect to T2) is f  F  rI o  2x106 wC p (400  T2 )(8000)  r  25  A($ / year) (j) By elimination of A in terms of T2 .

ROI = 0. rather than 9. r. Remember that while higher values of ROI can be obtained by selecting T2 closer to 250F. should be meaningful because it is guaranteed that $100 million will be committed to plant investment. if the heat transfer area costs were assumed to be Io = I’o + 25 A (I’o is the fixed installation cost for the exchanger). i = 15 percent in Table 3. Problem 3.22 Refer to P3.0. therefore. and the ratio of the value of steam to the cost-per-unit area for the heat exchanger. I1 + I2 + I3 in this case is a constant value ($100 million). Another case which gives a meaningful answer for ROI occurs when several projects are considered simultaneously with a constraint on total capital expenditure. The answer is 10. there are some conditions under which this criterion can be employed effectively to obtain a reasonable value for the optimum. depends on U. Problem 3. Decisions made on the basis of the internal rate of return often favor investment in smaller facilities rather than large plants because the ratio of profit to investment is optimized. Set P = 0. a 3. less than one year. (f). maximization of ROI leads to the meaningless solution obtained previously.1 approach (somewhat lower than the normal approach temperatures of 5 to 10F recommended in design manuals).If r = 0. While the rate of return on investment (F/Io) did not lead to meaningful results. The optimum value of A is 2905 ft2.1). To calculate the annual profit before taxes.5. Problem 3.23 The payback period is calculated as follows: 15 .6 obtained earlier for Eq. we compute the value of F = (2 x 106)(wCp)(400 – T2)(8000). then an overall rate of return for the three projects. If $100 million is to be invested among three projects. when calculated.1F. For example.280. The optimum. according to the analysis in this example. then maximizing F/Io would yield a more realistic result for T2. then T2 = 253. so the original investment is $72. but in general the statement is correct. The payout is. The optimal approach temperature.2 (n = 10.625. defined as (F1 + F2 + F3)/(I1 + I2 + I3). can be formulated. hence we simply optimize F1 + F2 + F3. which would be $176.13%. Note that at T2 = 400F. In fact.21 The last sentence is not clear.

25 Since all alternatives have acceptable individual IRR’s. Going from C to D costs $5.000 + $3.100(P/A.000 that generates $2.10) = 0. The incremental IRR of this is 22. Problem 3.162 net savings $162.000 investment with annual return of $3.000  30 months $1000 Problem 3.000 going from B to C. C remains the preferred case. Thus C is the preferred alternative. B becomes the preferred case.000 Problem 3. Going from C to E involves an investment of $15.100.PBP  cost of investment cash flow per period $30. 000   investment $1. You then calculate the IRR of the incremental investment of $5.4%>18%.4%.7% so C becomes the preferred case. start with the one with the lowest-investment (A) and look at the incremental return on incremental investment in going to the next-larger investment alternative (B).i%. this is 42. This would be $12. Choose Alternative D3 because it has the lowest negative PV.000 but the return of $500 per year is insufficient to justify that investment. Problem 3.27 (a) Return an investment (ROI) = 0. Because 22. The IRR of this is 14.900 per year. which you reject because it is less than 18%.2%.26 All of the outflows are negatives. 000. 000 or 16. This calculation can be done either with sequential cash flow entry or by trial-and-error solving the equation NPW = $12.24 The return on investment in percent is calculated as follows: ROI  net income (after taxes) 100 cost of investment $5000 (100)  10% $50.2% 16 .

000 (cost) Rotary Air 17 .15 per year under one assumption: Oil 10 1 k k 1 (1  i ) PV1  1.00 r (1  i )k PV1  $9. 000 i  . 000  1. For depreciation use the MACR table or just 10% per year. 758.1 106 (1  i)10  1 Problem 3.574.888.15 of costs From Table 3.1 r  0.200 so 1 1   5.000 i(1  i)10   r  .10 (b) 162. 000  106  PV k k 1 (1  i ) IRR :  i F i  I o (1  i ) n  1 162. Interest is .162 so i  10% from Table 3.28 Find the present value of each option (use cost as the criterion).

63 smallest I o $2.517. 000 Too close to choose Alternate Solution Assume return on investment means ROI as in text ROI  net income after taxes ($/yr)-assumed constant initial investment ($) NI net income sales costs    I0 initial investment initial invesment initial investment If sales are fixed. 000  1. 000 i = .888.15 1  5.570. the smallest costs/initial investment will have the biggest ROI Oil NI $1. 000   0.00 r PV2 $10. 000 18 . 000 Rotary Air NI $1.517.83 Io $1. 000   0. 420. 000.10 1 k k 1 (1  i ) PV  2. 420.

229.000 + $7. 19 .000 = -$140.29 (a) What is the PV of your base case? $43. Another solution: Calculate the interest rate given present value and payments.035 (c) Sketch the PV sensitivity diagram for these two variables below.084 and the PV for +50% annual savings to be $128.000 = -$140. Problem 3. What is the PV at +50% life? $72. PV = -$140.000 + $25.000 + $40.087. PV = -$140.257.ROI may be < 15% depending on sales.194 + $205.342 (b) You calculate the PV of –50% annual savings to be -$42.000 + $25.000 + $40. The PV at –50% life is -$8. To which of the two variables is the decision most sensitive? Savings.000 + $12.539.745 + $170.

Problem 4.CHAPTER 4 Problem 4.1 (a) continuous over – < x <  (b) discrete (c) continuous over 0 < xs < 1 and 0 < xD < 1 if xs and xD are mole fractions.5 (a) Minimize:  x1  f ( x )  [3 2 1]  x2   x3  or f ( x )  a T x 1 .3 n may be treated as a continuous variable when small changes in n do not affect the average unit cost significantly.4 R  100(1  t )  S  (V  F / n)  100(1  t )  Sn  (Vn  F ) I /n I dR 100(1  t )( S  V )   function of n dn I Thus. Problem 4. and n is a discreet variable.2 In all cases. R increases linearly with n and there is no stationary maximum. This happens when: (i) n is very large. Problem 4. Problem 4. The same is true when n is discreet. so that average unit cost  V (limit n  ) (ii) F is very small (limit F  0). i is a continuous variable.

6 The function has the form f ( x)  a  xTb  x Tcx Let a=3 x   2 2 1 x   1 b    c     3 1 6   x2  x   2 1   x1  [x1 x2 ]   [(2 x1  x2 )( x1  3x1 )]  1     1 6  x2   x2  2 .Subject to: (b)  x1   2 3 1   1 0  g ( x)   x  2  1 5    1 1 3  x     3 or g ( x )  Dx  c  x1  Maximize: f ( x )  [5 10 12]  x2     x3  or f ( x)  a T x Subject to:    200   15 10 10  x   1  0   g( x)   1 0 0   x2     0    2 0   x3     0 0    0 0 3   x1  h( x )  [10 25 20]  x2   300  x3  or g ( x )  Dx  c h( x )  Ex  d Problem 4.

7 (a) 3 . 2 x1  x1 x2  x1 x2  6 x2 2 2 or  3 1 3/ 2  1  f ( x )  1 x1 x2   1 2 1   x1  3/ 2 1 6   x2  Problem 4.

(b) (c) 4 .

654053 right hand side of equation 4.1.Problem 4.65405 100) Values of x1 0.000632 3.654053  x1*    and x*2    0  0.99996 4. The Newton-Raphson method was used to check the roots of the Equation (4) from different starting points (0.6541 0.0000003 x2 = -0. 0 0.05. They all go to x1 = 0.8 f ( x )  2 x1  x2  x1 x2  4 x1 x2  3 3 2 2 2 f ( x ) 2 2  6 x1  2 x1 x2  4 x2  0 x1 (1) f ( x ) 2  2 x2  2 x1 x2  4 x1  0 x2 (2) 2 x1 2 x1  1 (2) x2   (3)  2 x1   2 x1  3x  x1   2 0 2  2   1  x1   1  x1  2 2 1  5  5  x1 3x1  6 x1  3x1  4 x1  4 x1  4 x1  0 4 2 3 3  x1 3x1  6 x1  3x1  4  0 x1  0 (4) 3 x2  0 is one solution 3x1  6 x1  3x1  4 5 3 Only one other solution exists.9161783 5 .9161783 Thus.01.65405 0.

60308 2. The contours in the x1=x2 plane will yield the results calculated above.85557  hence this x is a local minimum that is the global minimum An alternate solution is to plot f(x) vs. check the Hessian matrix. 4 2 The eigenvalues are (0   )(2   ) 16  0  2  2  16  0  4  4  4(16)  2  1  16 2 one is positive and one is negative.916173  . x1 and x2. and use the graph to reach a conclusion. hence x is a saddle point.Next.5274 1.654053  x *    0. For 0.60308  H   is positive definite 1.5)T 6 . Problem 4. For x*  0 12 x  2 x2 H  1  4 x1 x2  4 2 4 x1 x2  4   2 x2  2  0 4 =   is not positive or negative define.9 (1) x  [5 2 10]T g 2 to g 4 satisfied 5(15) + 10(2) + 10(10) = 195 < 200 g1 10(5) + 25(2) + 20(10) = 300 satisfied h1 is satisfied (a) is a feasible point (b) is an interior point (2) x = [ 10 2 7.

g1 is not satisfied (a) hence not a feasible point (b) is an exterior point (3) x = [ 0 0 0]T g2 to g4 satisfied (boundary points) (a) h1 is not satisfied.g2 to g4 satisfied 15(10) + 10(2) + 10(7. not an interior or exterior point Problem 4. hence not a feasible point (b) g1 = 0 hence is a boundary point.10 (a) 7 .5) = 245 > 200 .

(b) (c) 8 .

Problem 4.11 (a) h1 ( x)  x1  x2  x3  0 2 (b) 2 2 h2 ( x)  x1  x2  x3  0 Feasible region is the origin g1 ( x)  x1  x2  2  0 2 (c) g 2 ( x)  x1  x2  4  0 9 .

9x10 -4 max ok h1 ( x 2 )  (0.6x10 -3 max ok 10 .h1 ( x)  x1  x2  3 2 (d) 2 g1 ( x)  x1  x2  2  0 g 2 ( x)  x1  0 g3 ( x)  x2  0 Problem 4.219) 2  1  0? min =1.12 For the first constraint: h1 ( x1 )  (0.0772) 2  1  0? = 3.535) 2  3(0.947) 2  2(0.207) 2  3(0.5x10-3 probably ok For the second constraint: h2 ( x1 )  5(0.534) 2  2(0.207)  3( 0.947)  5(0.0772)  6  0? =1.

an ellipse.9]T (not required) 6 and 4 are  0.13 The point (1.14 If the problem is a convex programming problem. Problem 4. The question is: how can it be shown to be the global minimum for the problem? Some ways are: (1) Start a numerical search from several starting points. f is both convex and strictly convex.534)  5(0.535)  3(0. Then ascertain if any other minimum exists by examining the graph.15 f  2 x1  2 x1 x2  3x2  7 x1  8 x2  25 2 (a) 2  4 x  2 x2  7  4 2 f   1 H    2 6   2 x1  6 x2  8  4 2  20  0  H is positive definite 2 6 Thus.219)  6  0? -3 =2x10 min probably ok Problem 4. that is if f(x) is convex g(x) are concave (form a convex set) Another possibility is if f(x) is unimodal in the feasible region.3  0. 11 . 1) can be proven to be a local minimum by the methods described in Chapter 8. a circle. x*  [1. in the x1-x2 plane along with the function g(x) = x12 + x22.h2 ( x 2 )  5(0. (2) Plot the contours of f(x). and locate the local minimum at x* = [1 1]T. Problem 4.

17 f  e x1  e x2 e x1  e x1  f  x  H  2 0 e  0  e x2  H is positive definite.  2 x3   0 0 2  strictly convex f  e x1  e x2 (c) e x1  e x1 0  2  f  is pos. f  5e5 x . 4  strictly convex f  x1  x2  x3 2 (b) 2 2  2 x1  2 0 0   2  f   2 x2   f   0 2 0  is pos. def. def. Thus. in the finite plane  x2  x2  0 e e     strictly convex in the finite plane  f  Problem 4.  2 f  25e5 x  2 f  0 for    x   f is both convex and strictly convex.(b) f  e5 x .16 f  ( x1  x2 )2  x2 (a) 2  2( x  x )   f  1 2   2( x1  x2 )  2 x2  2 2f  2 2 is pos. def. Problem 4. 12 . and thus convex. because ex > 0 everywhere on the finite real axis. f is strictly convex over finite values of x1 and x2.

and 0    1 Let f L  (1   ) x1   x2 and f R  (1   ) x1   x2 (a) x1  0. x2  0. This means that (1   ) x1   x2  f L  f R  function is convex. f R  (1   ) x1   x2  (1   )( x1 )   x2 . )x1 +  x2 involves subtraction of (1. We have (1   ) x1  0 and  x2  0 . f L  ( x2 )   ) x2  f R f is convex for 0 <  <1. and f is convex for 0 <  < 1. x2  0. fL must be less than fR because the term (1. f is convex. We have f L   x2  f R  f is convex (c) x1  0.Problem 4. ) |x1| from  x2. x2 . f L  (1   ) x1   x2  [(1   )( x1 )   ( x2 )] f L  (1   )( x1 )   ( x2 ). x2  0. x2  0. f L  (1   ) x1  (1   ) x1  f R For   1. (b) x1  0. Thus fL < fR. 13 .18 We will show that (1   ) x1   x2  (1   ) x1   x2 for all x1 . (d) x1  0. For   0. f R  (1   ) x1   x2  (1   )( x1 )   ( x2 )  f L .

It is also a closed region as seen from the figure. is strictly concave. Problem 4.  g1   Problem 4. -2  g2 is concave. -2  g1. the two constraints form a convex region. and thus concave functions.  1 0 0 14 .19 All the constraints are linear. 0.  0 2  2 x2  g2  x1  x2  1  0 2  1  0 0  g 2   H2     eigenvalues are 0. g2 is  0 2   2 x2  concave because it is linear. -2  g1 is strictly concave.20 g1  ( x1  x2 )  9  0.21 g1  x1  x2  9  0 2 2  2 x1   2 0  g1   H1     eigenvalues are –2.Problem 4. Thus. The region is thus convex.  g3 is concave and convex. 0 2   2 x2  g3   x1  x2  1  0  1 0 0 g3    H 3    eigenvalues are 0. g 2   x1  x2  1  0 2 2  2 x1   2 0   2 g1     eigenvalues are –2.

Thus, the region is convex. That it is closed can be seen from the figure.

Problem 4.22

f ( x)  ln x1  ln x2
1
x 
f ( x)   1 
1
x 
 2

ln xi is not defined for xi  0

 1
 x 2
H ( x)   1

 0


0 

1 
 2
x2 

For any xi  0, H ( x) is neg. def. At xi  0, H ( x) is undefined, but the conclusion
is that f(x) is concave.

15

Problem 4.23

Problem 4.24
If the region is convex, all the constraints must be concave, i.e., the Hessian
matrix must be neg. def.
(a)

h( x)  x1  x2  9
2

2

h
h
2h
2h
 2 x1
 2 x2
2
2
2
2
x1
x2
x1
x2
2 0
H 
 is pos def. And hence h(x) is not
0 2
concave (it is convex).

16

Answer:
No.

Problem 4.25
The Hessian of the each term of  (x) is pos def. or pos. semi-def. If the term is
convex. The Hessian of a sum of convex terms is the sum of the individual Hessians,
hence the matrix of  (x) is convex. For example:

f ( x)  x1  x2
2

2

H of the first term alone is

2 0
0 0

H of the second term alone is

0 0 
0 2 

and the sum of the two H’s is

2 0
0 2 .

Problem 4.26
Note the solution is wrong if x = 0.
Determine if f(x) is convex and g1 ( x)  0 is convex.
The Hessian matrix of f(x) is

17

 400
 x 3x
H ( x)   1 2
 200
 2 2
 x1 x2

200 
2 2
x1 x2 
 is f(x) convex?
400 
3 
x1 x2 

The principal minors are

 400  400   200  200  
det H =  3 
 2 2  2 2   > 0
3  
 x1 x2  x1 x2   x1 x2  x1 x2  

400
400
and 3  0 for xi  0 so f(x) is convex
3
x1 x2  0
x1 x2
Is the constraint g < 0 convex?
The Hessian matrix of g(x) is

 600
 x 3x
H g ( x)   1 2
 300
 2 2
 x1 x2

300 
2 2
x1 x2 

600 
3 
x1 x2 

 3.6x105  9x104 
det H g  
0
4 4
x1 x2

600
 0 for xi  0
3
x1 x2

600
 0 for xi  0
3
x1 x2

so that g(x) is convex
The xi  0 are convex functions . Consequently, the problem is a convex
programming problem for xi  0. as xi  0, g  , but the function asymptodically is
convex.

18

Problem 4.27
(a)

1 0 
0 1 

positive definite

(b)

1 1
1 1

neither

(c)

 oil 3 / 2  which is not positive definite,
oil 0 
 3 1  not symmetric hence change to 3 / 2 1  so that neither is the answer



(d)

 1 2
 2 1 

1 


 neither
(1  4)  3  0 

Problem 4.28

1 1 1 
A  1 1 1 
1 1 0 
Use principal minors:
det A = 0

1 1 
delete 1st col. and row A1  

1 0 

The elements on the main diagonal are positive: 1, 1, 0
Thus, none is the answer
Or

Use eigenvalues

1
1 
1  

det  1
(1  )
1   0
 1
1
(0  ) 

(1   )[(1   )(0   ) 1] 1[1(0   ) 1  1[1  (1   )]  0

1  0


2  2.73  same conclusion
3  0.73

19

Problem 4.29

f ( D, h)  Dh 

2

D2

(h  D) 
f ( D , h )   

D

1 1 
H ( D, h)   

1 0 

H is not pos. def. so no minium exists (except at  limits).
The eigenvalues of H are

1 5 1 5
,
, hence one is +
2
2

and the other is -.
Based on the above
a.
b.
c.
d.
e.

neither
is continuous
neither
does not
The trivial constraints (bounds) are linear and concave

g

4

g 

D 2 h  400  0
 2 Dh 


4  D2 

H

 2h 2 D  h  0
4  2 D 0  but 0  0

det H = -4D2 and is negative always
so that H is concave
Thus, the constraints do form a convex region (they all must be concave).

Problem 4.30
Basis: 1 lb mol feed
Income:

50 (0.1 + 0.3xA + 0.001S + 0.0001 xAS)(1) [$]

Expenses: Assume the cost of the additive is $/1b mol feed, not additive
Additive: (2.0 + 10xA + 20xA2) (1) [$]

20

and form a convex region.286 P2 0.0 x 10-6S2) (1) [$] Steam: f = (5 + 15xA + 0.286  0.375x10-6 (almost zero but not zero).05S + 0.286 0.286 P2 f / P2  0.005xAS) – (2.003S + 2.286 1. (c) The search region is linear because the constraints 0 < xA < 1 S>0 are linear.286 P3  2 f / P2  0.0 + 0.286  ( P4 / P3 ) 0.5526 P3 P2 0.0 x10-6S2 + 0.714 0.(1.31 f  ( P2 / P1 )0.0 + 0.286 1.286 0.047  4.714 P2  2 f / P3  0.003S + 2.005xAS (a) (5  40 xA  0.0x10  (b) H is negative definite.2042 P3 2  1.714 1.005 4.714 P2 P2 2. hence concave.286 21 .0x10-6S2) -(2.286  P3 0.0 + 10xA + 20xA2) f = 1 + 5xA – 20xA2 + 0.286 P2 1.714  0.005S )   6 (.005  2f H  6  . hence concave.047S – 2.0x10 S  0.005 x A )    f   40 0.286 1.286 with P1 = 1 atm and P4 = 10 atm.3678 P3 0.286 0.0818P3 0.7106 P3 2.286  0.932 P3  2 f / P2P3  0.286  ( P3 / P2 ) 0.286  0. The eigenvalues are –40 and –3.286 P3 0.2042 P2 2 f / P3  0. Problem 4. this becomes f  P2 0.

1. 1  P3  10. f ( x) is convex 3 x1 x2 x1 x2 (b) g ( x )  2 x2  (a) 300 1  0 x1 x2 300   600   3 2 2  x x x1 x2  1 2 2    g ( x)   300 600   2 2  3  x1 x2   x1 x2 600 600  3  0 and   0 for xi  0 and asymptodically as xi  0.3678 P3 0.714 2 Problem 4. and f is not convex over this entire range.714  0.1057  0. Therefore. P2 ( 2 f / P2 )  0.572 This has its lowest value at P2  10 atm. H is not positive semi-definite over the range 1  P2  10. 3 x1 x2 x1 x2 so that g(x) is convex 22 . P3  1 atm.3678P3 2 P2 0.286 P2 0. 2  2 f / P2  0.286 ( 2 f / P2 )  0.286 2.32 (a) f ( x)  100 x1   400  x 3x 1 2 H ( x)    200  2 2  x1 x2 200 x1 x2 200  2 2 x1 x2   400   x1 x2 3   400  400   200  200    2 2  2 2   det (H) =  3  3    x1 x2  x1 x2   x1 x2  x1 x2   400 400  0 and 3  for xi  0 and asymptodically as xi  0.2042  0.714 P2 1.2042 P2 2 1.  2 f / P2 2  2 f / P2P3  H   2 2 2  f / P2P3  f / P3  For convexity. must have  2 f / P2  0 (as well as some other conditions also).

33 (a) f ( x ) 3  x1  x1 x1 f ( x )  2 x2 x2  2 f ( x)  2 2 x2  2 f (x ) 2  3x1  1 2 x1  2 f ( x) 0 x1x2  3x12  1 0  H ( x)    2   0 23 . The following figure (with 300 / x1 x2 changed to 30 / x1 x2 to reduce the scale) illustrate the surface for g ( x)  1 . In P. Problem 4. 4.concave  Also xi  0 are   convex  but the constraint region is not a convex region because g ( x)  0 has to be a concave function for the region to be convex. Note that the region above g ( x)  1 is not convex.26 g ( x)  1.

1+1 4 (b) no: h1 ( x ) is not satisfied (c) yes: x T  [2 2] lies in the interior of the inequality g2 ( x* )  x1  x2  2  2 Problem 4.38 2 f / Pn  224.875ny 1.5ny.5  125.25ny 0.2    74 553. all diagonal elements must be positive.2P  19.46592 y 1. hence the answer is no.6 0  For H to be positive definite.75 y 0. Thus f is not convex at P = 99.5 (R  1) 2 f / R2  (4.875ny 1.5  5.38 )(5000  23P)2 2 f / Rn  9.5  23.2 y.62 where y = 5000R – 23PR + 5000 – 23P Differentiation gives 2 f / P2  529(4.38 )( R  1)2 2 f / PR  30. 24 .5  330.71616 y 1.7 169. we have  2 f / P 2  2 f/ PR  2 f / Pn    H   2 f / PR  2 f/ R 2  2 f / Rn   2 f / Pn  2 f/ Rn  2 f / n 2    74 12.832 y 0. It is therefore.f ( x ) is not a convex function for all x. which is not the case here.5  5.6   12.9  3.2  (112. R  8 and n  55.34 f  14720(100  P)  6560R  30.25 y 0.125ny 1.9 169.5 (5000  23P) 2 f / n2  0 For P  99.38 )(R  1) x (5000  23P)  224. not convex in some small neighborhood of the optimum.2PR  6560  30. R = 8 and n = 55.46592 y 1.

35 If the search is started in the vicinity of point A. Repeat the above analysis with ci in the sum. If f(x) is continuos for a  x  b and f '( x) is positive for a  x  b.Problem 4. Problem 4. f "( )  0.36 f() is a continuous function because it is the sum of continuous functions. hence f is convex. The region is not convex because the line segment AB does not entirely lie within the region. We have n f ( )   x (i )   i 1  (  x(1) )  (  x(2) )   (  x( k ) ) ( x( k 1)   )  ( x( k 2)   )   ( x( n)   ) k  k   x (i )  i 1 n x (i )  (n  k ) i  k 1 Differentiation gives f '( )  k  (n  k )  2(k  n / 2) which is negative for k < n/2 and positive for k > n/2. On the interval x( k )    x( k 1) . (Assume ci  0 ) 25 . then f (b)  f (a). The corresponding fact occurs for f '( x) being negative. even though the endpoints A and B lie inside the region. it is likely to terminate at point L which is a local maximum. The global maximum is at point G.

(c) f  x 4  2 x 2  1 f '  4 x3  4 x  4 x( x 2  1)  0 f "  12 x 2  4 f "(1)  8  0 f "(0)  4  0 f "(1)  8  0 x  1 is a minimum x  0 is a maximum x  1 is a minimum (d) f  x1  8 x1 x2  x2 2  x*  1. 1 2  2 x  8 x2  * T f   1   0  x  [0 0] 2 x  8 x  2 1  2 -8 Eigenvalues are (  2) 2  82  0 H =  or  =  6.9  0. 26 . and x  [0 0]T is a saddle point.Problem 4.25  0 x  3 / 4 is a maximum (b) f   x3  3x 2  x  5 f '  3x 2  6 x  1  0  x*  1  2 / 3 f "  6x  6 f "(1  2 / 3)  4. 0. 3 / 4 f "  12 x 2  6 x f "(0)  0 x  0 is a saddle point (inflection) f "(3 / 4)  2. 10 -8 2 H is indefinite.9  0. x  1  2 / 3 is a minimum f "(1  2 / 3  4.37 (a) f   x 4  x3  20 f '  4 x3  3x 2  x 2 (4 x  3)  0 x*  0. x  1  2 / 3 is a maximum.

Problem 4. 0 6  H is positive definite. 0) 27 . T Problem 4. Eigenvalues are  =2. and x   -1 -1 is a minimum.38 (a) f  x1  2 x1  3x2  6 x2  4 2 2 2 x  2   1 f   1 0  x*      1 6 x2  6  2 0 H    .39 f ( x )  2 x1 x2  2 x2  x1 2 2 3 f ( x ) 2  4 x1 x2  3x1 x1 2 f  4 x2  6 x1 2 x1 2 f  4 x1 x2 x1 f ( x ) 2  2 x1  4 x2 x2 2 f  4 x1 x1x2 2 f  4 2 x2 at (0. 6.

 f ( x )  1 rising ridge toward x1   x  [0 f ( x )  2 1]T x  [0  1]T f ( x )  2 Problem 4. Evaluate f(x) on both sides of zero to see how the value of f(x) changes.40 f ( x )  3x1 x2 3 9 x12 x2  f ( x )    3  3x1  18 x1 x2 9 x12  H ( x)    2 0   9 x1 principal minors: 18 x1 x2 is not always definite unless xi > 0 det H  (18x1 x2 )(0)  81x1 (concave function) 4 Get eigenvalues (18x1 x2   )(0   )  81x1  0 4 (18 x1 x2 )  (18 x1 x2 ) 2  4(1)(81x1 )  2 4  depends on the values of x1 and x2. 0) 9 x12 x2  (f ( x )  [0]    yields x1  x2  0. x  [1 0]T x  [1 0]T f ( x )  1  declining ridge toward x1  . x2  anything) 3  3x1  28 . but at the stationary point (0. or x1  0.(4 x  6 x1 ) 4 x1  0 0  H  2   4  0  4  4 x1 1  0 2  4 So that probably case 10 or 11 is the outcome.

i  0 . the value of  depends on the value of x1 Problem 4.42 (a)   12   6  u ( r )  4         r   r   at the stationary point 12 6  du du   1   1  0   4   12  *  *  6  *  *   0 dr r* dr r* r  r  r  r    1/ 6 1   * r 2 (b)  r *  21/ 6  12 6  d 2u   1   1   4  156  *  *2  42  *  2  dr 2 r* r  r  r  r *    72     2   0 as long as  > 0 3 2   29 .41 f ( x)  10 x1  x1  10 x2  x2  x1 x2  x1  34 2 2 3 10  2 x1  x2  3x12  f ( x )    10  2 x2  x1  (2  6 x1 H ( x)    1 1 2  Find two eigenvalues (in terms of x1) (2  6 x1   )(2   )  1  0  2  (4  6x1 )  (4 12 x1 )  0 (4  6 x1 )  4  6 x1 ) 2  4(1)(4  12 x1 )  2 Solve for  in terms of x1. hence some degenerate surface occurs Problem 4.

f ( x) is not differentiable at x*  0 where the minimum of f ( x) is located. x  2 is the minimum.44 No. Problem 4. dz / dx  2x  4  0 that differs from x2  4x  16  0. or x  a is misleading.The Lennard-Jones potential has a minimum at r * (c)  1  2 1  u ( r )  4          2  2  * Lennard-Jones potential Problem 4. and is thus not the minimum. 30 .43 The solution of y  ( x  a)2  0. For z  x2  4x  16. x  2  j 3 does not satisfy 2 x  4  0. The necessary condition is dy / dx  2( x  a)  0 that coincidentally corresponds to the solution of the equation.

and 4 .Problem 4.  2   36  0. H     6 0 12 6  i  12 and 0 . and 1/ 2  1 is a saddle point.  2     36  0. 2.  2  66  24 62 60  24. H     6 6 12 6  i  12 and 6 are greater than 0 .  6 0 H is positive definite. 04 21  7. 14 31 . and [0 0]T is a minimum 12 6  T  1 . 6 0 At x*  1/ 2 x H is indefinite.45 (a) f  6 x1  x2  6 x1 x2  3x2 2 3 2 12 x1  6 x2  f   2 0 3 x  6 x  6 x  2 1 2 0 1/ 2  x*    or   0  1  6  12 H    6 6 x2  6  12 6  T at x*   0 0 . T (b) f  3x1  6 x1  x2  6 x1 x2  x3  2 x3  x2 x3  x2 2 2 2 6 x1  6  6 x2   f   2 x2  6 x1  x3  1  0  4 x3  x2  1   4 /17   x  21/17  1/17  * 6 6 0  H  6 2 1  0 1 4 i  6.

4a2 a3  a4  0 . and 3  6 2 1  102 x* is a sadlepoint 0 1 4 (c) f  a0 x1  a1 x2  a2 x1  a3 x2  a4 x1 x2 2  a0  2a2 x1  a4 x2  f   0  a1  2a3 x2  a4 x1   2a H  2  a4 2  2a0 a3  a0 a4   4a a  a 2  2 3 4 *  x   2a0 a2  a0 a4   2   4a2 a3  a4  a4  2a3  H must be pos. a3  0. 4 a2 a3  a4  0 . x* is a saddle point. 2 Problem 4. For x* to be a minimum. must have a2  0.46 The necessary and sufficient conditions are 1) f(x) is twice differentiable     f  2( x2  5)  ok x2  2 2   f  f   2  2 2 2 x1 x2  f  2( x1  8) x1 2) f ( x* )  0 * 2( x1  8)  0  x1  8    *  ok 2( x2  5)  0 x2  5  32 . to get a minimum and neg. 2 For x* to be a maximum. Otherwise. def.6 6 0 H is indefinite. a3  0. must have a2  0. to get a maximum. def.

2 0 H   0 2 which is pos.48 f ( x )  2 x1  x2  x1 x2  4 x1 x2  3 3 2 2 2 33 . def.3) H ( x* )is pos def.47 The stationary points of f ( x)  1 4 1 2 x  x are obtained from f ' ( x)  0  x3 1  0 4 2 Factor to get: x( x  1)( x 1)  0 x  0. ok Problem 4. x  1 Solutions: To identify status of these points determine f "( x) at each point x  0 : 0  1  1 a max x  1: 3  1  2 a min x  1: 3  1  2 a min f "( x)  3x 1 * Problem 4. x  1.

654. x2 )  (0.123 This is a saddle point For the other point 9. x2 )  (0.60 2.7 34 .86  det( H )  24. 0)    4 2 ( )(2   ) 16  0 or  2  2 16  0 The eigenvalues are   2  4  64  2  68 2 1  5.53 1. 0.916) 12 x1  2 x2 2 H ( x)    4 x1 x2  4 4 x1 x2  4   2 2  2 x1  0 4 H (0.123  2  3.654. 0) Another solution: ( x1 . 0.df 2 2  6 x1  2 x1 x2  4 x2 dx1 df 2  2 x2  2 x2 x1  4 x1 dx2 For an optimal solution df 2 2  0  6 x1  2 x1 x2  4 x2 dx1 df 2  0  2 x2  2 x2 x1  4 x1 dx2 An obvious solution: ( x1 .60  H (0.916)    1.

49  x 2  f ( x)   x 2  e x 1    x  0 2 x always+  0  x  1 f '( x)   2 x always   e x 1 always  1 x    Look at f '( x) and note that there is one minimum. 2  0 the function is strictly convex at (0.916) Problem 4. 35 . plot the function In either case you can determine that the function is not unimodel. -0.53   )(2.39  24.70  0 1  12.86   )  2.40  9.The eigenvalues are (9.56  0 or  2 12.89 2 2 .39  7.654. Or.

def. 36 . Introduce x* into H(x) 0   20.917  H ( x* )    57.50 Is x*  [0. def.51 f ( x)  x3 Differentiate f ( x) f '  3x2 x  0 f '  3x2 x  0 f "  6x x  0 f "  6 x x  0 f '''  6 f "  6 x0 x0 But because the derivatives are decontinuous at x*= 0.6   0 H(x*) is neg.Problem 4. you cannot demonstrate that the necessary and sufficient conditions are met because the derivative is not defined at x* = 0. even though the function is twice differentiable..8]T a maximum of f ( x )  x1  12 x2  15 x1  56 x2  60 4 3 2 See if  f ( x)is convex (4 x 3  30 x1 )   f ( x )    1 2   (36 x2  56)  12 x 2  30 0   H ( x)    1  0 72 x2   Apparently  f  0 at the proposed solution! Hence x* is not a maximum. but x* is not a maximum even if H is neg. at the point.87  0. Problem 4.

37 .

5x2 A local maximum occurs between x  0 and x  1 .1 f ( x)  e* 1. Consequently if you start to bracket the local minimum at values of x  1 and use reasonable step sizes.CHAPTER 5 Problem 5. you can bracket the local minimum. 1 . you will most likely proceed to x   (the global minimum). but if you start with values of x before the local maximum occurs.

5x2 The minimum will be reached from any starting point. 2 .Problem 5.2a (a) f ( x)  e*  1.

Problem 5.5( x2  1)( x  1) No matter where you start. the minimum (at  ) cannot be bracketed.2b (b) f ( x)  0. 3 .

2c (c) f ( x)  x3  3x Because both a local minimum and a global minimum (at  ) exist. the remarks in P5. 4 .1 apply here.Problem 5.

Problem 5. various starting points and step sizes will yield different results. 5 . and one maximum.2d (d) f ( x)  2 x2 ( x  2)( x  2) Because this function has two minima.

8x  1 A large scale figure shows one minimum.31x4  8.89x2  6. 6 .2e (e) f ( x)  0.1x6  0.Problem 5.29x5  2. but a small scale figure shows many minima and maxima exist.33x3  12. Starting near x = 0 you would reach the local minimum shown in both figures.

def. You can never have a maximum as f " is pos. f ""  24 The figure for the second derivative looks as follows Bracket the minimum f  ( x 1)4 7 . at all x except x = 1. so you have a minimum as f "" is positive.Problem 5.3 Use the analytical derivative to get the solution by which the numerical methods can be checked. Minimize : f  ( x  1) 4 f '  4( x  1)3 x  1 is a solution of f '  0 f "  12( x  1)2 f '''  24( x  1) The fourth derivative is an even number.

0625 0.5  1)4  2(0  1)4  (0  0.52 It is better to use a bracket value instead of x0 = 0.05 6.99835 x 6  0.05  1)3 1 4(1.5  1)4  /(2)(0.5  1)4  (0  0.99336 x 4  0.05  0.95 4(1.99959 x8  0.5  1)4  / 0.9967 x5  0.5)  0 (0  0.99917 x 7  0.95  x  1.25x10-6 0.9738 x 2  0.9867 x3  0.05 Newton’s method (using finite differences instead of analytical derivatives) (a) If you use h  0.05  x p  0.05 x1  1. say use x0  0.5 and x0  0 at the start.95 6.9999 x10  0.5 0.5 0.25x10-6 1.99995 etc.95  1)3  /(1.0625 1 0 1.x f 0 1 A bracket is: 0.05  1)3  4(0.95 x1  0. xmin  1 (b) Secant (Quasi-Newton) method x*  x q  f  ( x  1)4 f '  4( x  1)3 f '( x q )  f '( x q )  f '( x p )  /( x q  x p ) xq  1.95) 8 . the relation is: x x 1 0  f ( x  h )  f ( x  h )  / 2h   f ( x  h)  2 f ( x)  f ( x  h) / h2 (0  0.9998 x9  0.

x2 . 3. The function was not unimodel.0 Iter.598 0. It has a minimum but a.600]T.0 Point to be discarded x1 x5  0.0625 x3  2. Numerical and round off errors gives nonsense numbers.833 f5  7.64  1. x3 x2 . The final interval is [0. Problem 5.5 1.7x104 x2 9 . 0. c. The bracketing procedure at the start is not successful in bracketing a minimum.2 x  x2 The precise values at the solution will depend on the method used. A bracket on the derivative of f (+ and -) is not maintained.0 f3  1.4 f  6.0 f 4  0.x2  1 The minimum xmin = 1 Problem 5.5917.5 f 2  0. Problem 5.0 x2  0. Points used 1 x1 .6 f  ( x 1)4 x opt 2 2 2 2 2 2 1  ( x2  x3 ) f1  ( x3  x1 ) f 2  ( x1  x2 ) f3     2  ( x2  x3 ) f1  ( x3  x1 ) f 2  ( x1  x2 ) f 3  x1  0. and f = 6.0 f1  1. x4 2 xopt fopt x4  1.6049]. x3 .9999 with x* = [0. b. The problem has no minimum 2.

6x107 x7 6 x3 .9950 8 x3 .x3 . x4 . x4 .9800 f8  1.0 10 . x10 x11  0. x4 . x5 x6  0.7 f  ( x  1) 4 g  4( x  1)3  g  w z  x opt  x2   2  ( x2  x1 )  g 2  g1  2w  z  3( f1  f 2 ) /( x2  x1 )  g1  g 2 w  ( z 2  g1 g 2 )1/ 2 x1  0. x4 . x7 x8  0. x6 x7  0.9900 f9  1.6x10 6 x6 5 x3 . x8 x9  0.2x10 10 x9 x*  0. x9 x10  0.9600 f 7  2.5 x2  2. x4 . x4 .9194 f 6  4.9975 Problem 5.0x108 x8 7 x3 .9975 3 f10  6.2x105 x5 4 x3 .

x2 xopt Point discarded x1 .982 x7  1. Problem 5.0 f3  81. Points used 1 2 x1 . x7 x7 .5 f1  0.Iter. x3 x3  1.0494 x3 4 5 x4 . The difficulty arises because x1 .0002 x4 6 7 8 9 10 x5 x6 x7 x9 x10 x11 x* = 1.0625 Fitting a cubic equation through these four points gives f  9 x3  54. x2 .8 f ( x  1)4 x1  1.2287 x4  0. x11 x6  0. x9 x9 .0 x3  4.25  0 This quadratic equation does not have real roots.5 x  115.8780 x2 x1 3 x3 . x3 and x4 do not bracket the minimum. x4 x5  1. x5 x5 .9905 x9 = 0.0009 x12 = 1.0002 Problem 5.0084 x8  0.5 x2  3.9994 x10 = 1.0 x4  4.0625 f 2  16.75 x 2  115. x10 x9 . and the problem cannot be solved. x6 x6 . x8 x7 .25 x  80 df / dx  27 x 2  109.0 f 4  150.9 Minimize: f ( x)  2x3  5x 2  8 x 1 Information about the problem (not required) 11 .0028 x11 = 1.

08 1 f "( x ) 26 2 (b) 1 Secant (Quasi-Newton) method xA  2.5 xA  xB f '(1.67.636 0 0 4  (1  1.304) Error caused by round off 2  1. f '(2)  4 positive At   use them as they bracket  1 xB  1 .636  4  1. 6 Newton’s method x1  x 0  f '( x 0 ) 6  10  1  (2)  3 = 10 12  10 f "( x ) f '( x1 ) 24 x x   3  2.5)  1.f '( x)  6 x 2  10 x f "( x)  12 x  10 is pos def.301 x 2  1. f '(1.301 negative. if x  10 12 f '( x)  0  (6x 10) x yields as solutions x  0 and x  the latter is a minimum for x  1 (a) 10  1.50 negative  the derivative value of 0 2  x1  x 0  4 f '( x 0 A )  2  1.5) f '( x A )  f '( x B ) 2  1. and let xB  0.636 (c) f ( x)  2x3  5x2  8 x 1 For polynomial approximation (use a quadratic function) start with 3 points possibly evenly spaced that bracket the minimum 12 .30 4  (0.636)  0. keep xA  2.

f(x) Start at x = 1 x=2 x = 1.25c -12 = a + 2b + 4c min x = - b 13 13    1.5 -12.00b  4c  13 .5   x  1.25c   12.63 -12.5 Thus 1 < x < 2 brackets the min of f(x). Step 1 Solve the quadratic f ( x)  a  bx  cx2 using the 3 above points -11 = a + b + c -12.0 f ( x) x  1. x3  2.62 Save x1  1.00 Solve 12.66c  solve for b and c.5b  2.63 2c 214 8 f ( x)  12.5b + 2. and get 12.63b  2.5  a  1.5 -11 -12 -12.63.00  a  2. x2  1.63 x  2  and repeat -12.5 = a + 1.63  a  1.50.

10 f ( x)  1  8 x  2 x 2  10 3 1 4 4 5 1 2 x  x  x  x 3 4 6 6 f '  8  4x 10x2  x3  4x4  x5  (1  x)2 (2  x)3  0 at x  1 and x  2 a. so x  1 is a saddle point.190  c.104 146. Newton method x k 1  x k  x1  2  f '( x k ) f "( x k ) 16  1.x*   b 2c and continue to improve the values of x* . 12. f "  (1  5x)(1  x)(2  x)2  0 for x  1 and x  2 f "'  2(2  x)(5  4 x 10 x2 )  0 for x  1. so x = 2 is a maximum.7 Quadratic interpolation 14 . Problem 5. but f” = 0 for x = 2 f ""  6(1  16x  10x2 ) is negative at x = 2.910 176 x 2  1. b.66  1.

Cubic interpolation: Initial prints: 1. 2. h  0.9)  [2  (2)](1)  [(2)  0](14.73)  2  (0  2)(95.047619 3.9 3 4 5 6 f2  1 f3  1  8(2)  2(2)2  x1  10 3 1 4 4 5 1 6 (2)  (2)  (2)  (2)  14. 6 x 3. 10 1 4 1 f1  1  8(2)  2(2) 2  (2)3  (2) 4  (2)5  (2) 6  95.001 Converged in one iteration to x*  3 Quasi-Newton (Secant) method: x0  1. converges in one iteration to x*  3 Finite differences Newton method: x0  1.11 (a) f  x2  6x  3 (i) (ii) (iii) (iv) (v) Newton’s method: x0  1.73 3 4 5 6 1  (02  22 )(95.00000 1 0. x2  2.x*  2 2 2 2 2 2 1  ( x2  x3 ) f1  ( x3  x1 ) f 2  ( x1  x2 ) f3    2  ( x2  x3 ) f1  ( x3  x1 ) f 2  ( x1  x2 ) f3  Starting with x1  2 . x1  5. Converged in one iteration to x*  3.73)  and repeat Problem 5.028532 3. Quadratic interpolation: Started with x1  1. pick x2  0 and x3  2 arbitrarily.583 0 convergence is linear Here 15 . Converged in one iteration to x*  3.599 0 2 12.9)  [22  (2) 2 ](1)  [(2) 2  02 ](14. x3  5. 5.

1

(b)

x k 1  x*

and

x k  x*

2

x k 1  x*
x k  x*

2

f  sin x
(i)

Newton’s method.
x

1

2

5.0
4.86369
0.526
1.83
4.71122
0.008
0.051
4.71239
0
0
The rate of convergence is superlinear.
(ii)

Finite difference Newton method.
1
x

2

5.0
4.70419
0.029
0.100
4.71239
0
0
The rate of convergence seems to be quadratic, but there are two few points to be
certain.
(iii)

Quasi-Newton (Secant) method. Initial points were x1  3, x2  5
x

1

4.55457
4.71338
0.006
4.71238
0.010
4.71239
0
Rate of convergence is linear.
(iv)

0.04
3.06
0

Quadratic interpolation. Initial points were x1  3, x2  5, x3  5.5
x

1

4.65058
4.68558
0.433
4.71261
0.206
4.71247
0.364
4.71239
0
Rate of convergence is linear.
(v)

2

2

7.02
0.306
0.165
0

Cubic interpolation. Initial points were x1  3, x2  4, x3  5, x4  5.5.

16

x

1

2

4.74334
4.70481
0.245
7.91
4.71219
0.026
3.48
4.71239
0
0
Rate of convergence is superlinear.
(c)

f  x4  20 x3  0.1x
(i)

Newton’s method.
x

1

20.0
18.312
0.662
17.191
0.661
16.448
0.661
15.956
0.660
15.631
0.660
15.416
0.659
15.275
0.660
15.181
0.657
15.119
0.656
15.079
0.661
15.052
0.654
15.034
0.647
15.022
0.636
15.015
0.667
15.010
0.643
15.006
0.556
15.004
0.600
15.003
0.667
15.002
0.500
15.001
0
15.001
The rate of convergence is linear.
(ii)

2

0.132
0.200
0.302
0.456
0.691
1.05
1.59
2.40
3.64
5.60
8.38
12.7
19.3
31.7
45.9
61.7
120
222
250
0

Finite difference Newton method. h  0.001
2
1
x
20.0
16.667
0.333
0.067
15.278
0.167
0.100
15.010
0.036
0.129
15.000
0
0
The rate of convergence is quadratic.

17

(iii)

Quasi-Newton (Secant) method. Initial points are x1  10, x2  20.
2
1
x
12.000
13.421
0.526
14.240
0.481
14.652
0.457
14.845
0.444
14.931
0.442
14.970
0.426
14.987
0.414
14.994
0.417
14.997
0.400
14.999
0
The rate of convergence is linear.

(iv)

Quadratic interpolation. Initial points are x1  10, x2  16, x3  20.
2
1
x
14.031
14.920
0.08
14.957
0.53
14.995
0.10
14.998
0.25
14.999
0
The rate of convergence is linear

(v)

0.175
0.305
0.602
1.28
2.87
6.27
14.3
34.7
80.0
0

0.085
6.72
2.27
62.5
0

Cubic interpolation. Initial points are x1  10, x2  13, x3  16, x4  20
2
1
x
14.974
14.989
0.423
16.27
15.000
0
0
The rate of convergence is linear. There are not enough points to tell whether it is
superlinear.

18

Problem 5.12

$0.03
(150000)
x
4500
 500  0.9 x 
x
4500
dC / dx  0  0.9  2
x
x  70.71 hp.

C  $500  $0.9 x 

Problem 5.13
At least squares fit gives

E  74.764  0.0853R  1.551x103 R2  7.613x106 R3  9.605x109 R4
1  ton 
1  lb 
 $ 
Coal cost = 7 
x

x

 x5
 ton  2240  lb  14000  Btu 
x

 Btu 
 hr 
R
300 R
x 2544.43 
 x8550    7284 $ / yr.
E
E
 hr.hp 
 yr 

Fixed cost = 14000 + 0.04R2 $/yr.
 $  7284( R / E )  14000  0.04 R 2
Total cost 

(5)(300)(0.01R)
 yr  hp 

C

485.6 933.33

 2.67x103 R
E
R

Introducing the least squares expression for E into the expression for C and minimizing
gives
C *  11.44
R*  244.09
E *  69.7

19

Problem 5.14

t f  [Pc A2 /  M 2c]xc exp(axc  b)
L et c1  Pc A2 /  M 2c
Then:

t 'f  c1 exp(axc  b)(1  axc )
t "f  ac1 (2  axc ) exp(axc  b)
Newton’s method gives
xck 1  xck  (t 'f / t "f )

1  axc
a (2  axc )
Substituting numerical values,
 xck 

xck 1  xck 

1  3.643xc
7.286  13.2714 xc

k

xc

0
1
2
3
4
5
6

0
0.1372
0.2332
0.2791
0.2884
0.2888
0.2888
xc  0.2888 , and
*

t f  1634 minimum
*

Problem 5.15
The authors of the paper cited report T *  453K , but several procedures in this book indicate that
the problem does not have a realistic solution to get the minimum cost as a function of T,
probably because the function for  is incorrect.

20

Problem 5.16
The comment is true

21

Let  be the step-size. The number of function evaluations is (21)5 = 4084101 Problem 6. Then. T f   2  3 df / d   2  3  0 T    3/ 2 1 . then there are 21 values for each variable.CHAPTER 6 Problem 6.2 f ( x )  x1  x1 x2  x2  3x1 2 2 (a)  2 x  x  3  2 f ( x )   1 2   0  x*     1  x1  2 x2   2 1 H    H is positive definite. The search directions must be chosen appropriately. so x* is a minimum. it is a global minimum. (c) (d) (e) A univariate search will be a good method because the function is quadratic and well scaled. starting from  0 0 .1 If each step is 1/20 of the interval. 1 2  (b) Since f(x) has only one stationary point. the next point is   0 .

1633 or  1.1)  (1)   y3  2 2 2  (0. (-1.1732.1)  (1)]2  [2.0577. Thus. Then the required simplex has its four vertices at (-1. Problem 6. (-1. This is analogous to the method of conjugate directions. the z-coordinate.1732]2  [ z4  (2)]2  (0. -2) and (-1. -2). 2.1. -2) and (-1.2. x4  x1  x2  x3  1. z4 is given by [(1. 2. 2. x3 is 1.1732.P1  [3/ 2 0]T If we start from [0 4]T .1 3 y4  y1  y2  y3  2. 2. 2. -2).2.1. 2. -2).1732 Select y3 = 2. and  is the step size. 2. and one vertex at (-1. -2. The y coordinate is given by (1.1 . We may place the second vertex at (-1.2) 2 y3  1. By symmetry. Let one face of the regular tetrahedron (an equilateral triangle) be parallel to the x-y plane with one vertex at (-1.2 units long. 2.8367.0577 3 Then. 2.1732.3 We need a regular tetrahedron with each side 0. then P2  [ 4]T f   2  4  42  3   2  7  16 df / d  2  7  0    7 / 2 P2  [7 / 2 4]T (f) See the figure. the x-coordinate of the third vertex.1. Say z4 = -2.8628 or 2. The x and y coordinates of the centroid of this triangle are the x and y coordinates. (-1. -2). -2). the equilateral triangle has vertices (-1.1633. 2. -2).1633) 2 .2.2)2 z4  2. A line joining P1 and P2 passes through the optimum. -2).0577  2.

....... j  n1.. For example........... Coordinates for a Set of Simplex Vertices n coordinates of each point Point j  2.. and so forth. j 1 0 0 0 0 0 0 2 p q q q q q 3 q p q q q q ....... ....... the regular simplex is a regular tetrahedron (four points)... for another starting vertex such as (-1... the triangle given in Figure 1 has the following coordinates: 3 . j  3..... for two variables a regular simplex is an equilateral triangle (three points). you have to translate these values. Regular simplexes for two and three independent variables. 0...... 0)..... n n+1 q q p q q q a n 2 a n 2 q q  n 1  n 1  n 1 1 q q p q q p   a = distance between two vertices Note: The table starts at (0.. j  4. -1)....... 2... as indicated in Fig 1.. j  n.. for three variables. j  1. For example.Recall that regular polyhedrons in En are simplexes. (1) indicates the highest value of f(x)... The arrow points in the direction of greatest improvement. for n = 2 and a = 1..........

is formed.23 discard x2 . through the centroid of the simplex.00 f ( x3 )  10.965 The objective function can be evaluated at each of the vertices of the simplex. composed of the remaining old points and the one new point. Say x3  [1.00 f ( x2 )  13. x2 and x3 form an equilateral triangle. x4  [1. termed a reflection.i 1 2 3 0 0. and a new simplex. and a projection made from the point yielding the highest value of the objective function. B. plus rules for reducing the size of the simplex and for preventing cycling in the vicinity of the extremum. located along the projected line at the proper distance from the centroid.965 0.Vertex x1. Point A is deleted.8660 0. Stage 3: x5 is the reflection of x3 in the line joining x1 and x4 . Problem 6. always deleting the vertex that yields the highest value of the objective function.8660 1. permit a derivative-free search in which the step size on any stage k is fixed but the direction of search is permitted to change. Continuation of this procedure. Stage 2: x4 is the reflection of x2 in the line joining x1 and x3 .259 0.i x2.5]T Stage 1: f ( x1 )  4. x5  [1 0]T 4 .259 0 0.4 x1  [1 1]T x2  [1 2]T Select x3 so that x1 .23 discard x3 .5]T f ( x4 )  4. point A in Figure 1.

556  0.f ( x5 )  1 discard x2 .134 0. x2 . Stage 4: x6 is the reflection of x2 in the line joining x1 and x5 . is the reflection of x3 in the plane containing x1.556]T 3 x3  x5  2 xC x5  2 xC  xs x5  [7 / 9 8 / 9  7 / 9]T 5 . The next point. x4 is 1 xC  ( x1  x2  x4 )  [0. The centroid of the equilateral triangle formed by x1 . x5 .5]T f ( x6 )  0. x6  [0. And so on.768 discard x1 . x2 and x4 . Problem 6.5 x1  [0 0 0]T f ( x1 )  4 x2  [4 / 3  1/ 3  1/ 3] f ( x2 )  7 x3 [1/ 3  4 / 3  1/ 3]T f ( x3 )  10 x4 [1/ 3  1/ 3 f ( x4 )  5 T 4 / 3]T x3 is dropped.222  0.

7 f ( x )  x1  x2  2 x3  x1 x2 2 2  2 x1  x2  f ( x )   2 x2  x1   4 x3  2  2 1 0  H   1 2 0   0 0 4 H is positive definite. Problem 6. and the stationary point [0 0 0]T is a minimum. b and c which satisfy this equation gives s2 orthogonal to s1 . then ( s1 )T H s0  0 s1  [1 2 0]T (say) Step 1: start at x 0 and minimize f along the s0 direction. If s1 is conjugate to s0 with respect to H. so s1 cannot have a conjugate direction with respect to H. for s2 to be orthogonal to s1 .6 (a) s1  1 / 3 1 / 3 1 / 3    Let s2  [a b c]T Then. The optimum step size is 6 . s2T s1  1 (a  b  c)  0 3 Any values of a.Problem 6. s2  [1 1 0]T (No unique solution) (b) f ( x)  x1  2 x2  x1 x2 2 1  x2  f ( x )   4 x2  x1  0   0 1 0 H   1 4 0   0 0 0 H is not positive definite. Let x0  [1 1 1]T and s0  [1 0 0]T . Say.

Thus. For a quadratic function of three independent variables. 2 x1  x2  3x3  63  0 and H ( x) has to be positive definite. and all the principal minors must be >0. and satisfy 7 . or all the eigenvalues must be positive. The optimum step size is 1   [f ( x1 )]T s1 1  1 T 1 (s ) H s 2 x2  x1  1s1  [0 0 1]T The minimum is not reached in two steps.8 f ( x )  x1  x1 x2  16 x2  x3  x1 x2 x3 2 2 2  2 x1  x2  x2 x3  f ( x )  32 x2  x1  x1 x3   2 x3  x1 x2  1  x3  2  H ( x )  1  x3 32   x2  x1  x2   x1  2  s1 and s 2 are conjugate with respect to H ( x) when ( s1 )T H ( x) s 2  0 The det H > 0. x must lie on the above plane. i.e. Insert the two given vectors to get an equation in xi that must be satisfied. Problem 6. three steps will be required to reach the minimum.0   [f ( x 0 )]T s 0 1  0 T (s ) H s 2 x1  x0   0 s0  [1/ 2 1 1]T Step 2: start at x1 and minimize f along the s1 direction.

or s22  0     2 2   5 Second direction: s 2  [1 0]T (say).10 a.9 f ( x )  5 x1  x2  2 x1 x2  12 x1  4 x2  8 2 2 10 x  2 x2  12  f ( x )   1   2 x1  2 x2  4  10 2  H   is positive definite  2 2 If s is conjugate to s0  [1 0]T (the xi axle).(1  x3 )2  64 or  7  x3  9 x2  4  2  x2  2 x1  64  8  x1  8 2 2 and 128  2 x1  32 x2  2 x1 x2 (1  x3 )  2(1  x3 ) 2  0 2 2 Problem 6. (Note that you get back the original direction for a quadratic function) Problem 6. then 1 First direction: 10 2  1  ( s1 )T   10s11  2 s12  0     2 2 0 Say s1  [1  5]T For s 2 to be conjugate to s1 10 2   1  s2   8s22  0. The conditions for orthogonality are 8 .

x T y  0  yT z  0  z T x  0  solve simultaneously. then y3  1/ 2 z1  z2  1/ 2 z3  0   2 / 3z1  1/ 3 z2  2 / 3 z3  0  Let z1  1 Then z2  2 and z3  2 The vectors are 1  1    y  1  z   2 1/ 2   2  (not unique)  2 1 0 H  1 2 1  0 1 3   2 1 0 H  1 2 1  0 1 3  b. y2  1. The two directions for conjugacy are 9 . An example is  y1  [2 / 3  1/ 3  2 / 3]  y2   2 / 3 y1  1/ 3 y2  2 / 3 y3  0  y3   z1  [ y1 y2 y3 ]  z2   y1 z1  y2 z2  y3 z3  0  z3  2 / 3  [ z1 z2 z3 ]  1/ 3   2 / 3z1  1/ 3z2  2 / 3z3  0  2 / 3 Let y1  1.

z3  . y2  1.11 f ( x)  x1  x1 x2  x2  3x1  3x2 2 2 2 x  x  3  f ( x )   1 2   x2  2 x2  3 2 1 H ( x)    1 2 3 at (2. Then 16  z2  8 z3  0  3  2 z1  4 / 3z2  4 / 3z3  0  3z1  Let z1 = 1 and solve for z2 and z3.677 Problem 6.x T Hy  0   y T Hz  0  z T Hx  0  solve simultaneously to get a non unique solution. An example is:  y1   2 1 0   y1      [2 / 3  1/ 3  2 / 3] 1 2 1   y2   0  [1  2 / 3  7 / 3  y2   y3  0 1 3   y3   z1   2 1 0   z1      [ y1 y2 y3 ] 1 2 1   z2   0  (2 y1  y2 ) ( y1  2 y2  y3 ) ( y2  3 y3 )   z2   z3  0 1 3   z3   2 1 0  2 / 3  2 / 3      [ z1 z2 z3 ] 1 2 1   1/ 3   0  (2 z1  z2 ) ( z1  2 z2  z3 ) ( z2  3z3 )   1/ 3  0 1 3   2 / 3  2 / 3 Let y1  1.826 z2  . y3  7. def. 2) f ( x)     3 (pos.) 10 .

1  2 1   s1  [3  3]    1   0 gives a conjugate direction 1 2  s2   s11  [9  9]  1  = 0 s 2  9s1  9s2  0 1 1 Let s1  1   Direction is unique because for a quadratic function you can only then s2  1 have two conjugate directions. The values of elements in s are usually not unique.13 Max 1 2 2 f ( x )  x1  x2  ( x1  2 x1 x2  2 x2 ) 2 Start at x  [1 1] 1  x1  x2   1 f (1. Problem 6. f ( x)  [14 14 10]T Problem 6. and one was fixed by so.1)     1  x1  2 x2   2   s12  A second search direction is [1  2] H   2   0  s2  11 .12 f ( x )  ( x1  x2 )3 x3  x3 x2 x1 2 2 2 3( x1  x2 ) 2 x3  2 x1 x2 2 x32    2 2 f ( x )  3( x1  x2 ) 2 x3  2 x1 x2 x3    2 2 3 ( x1  x2 )  2 x1 x2 x3  at x  [1 1 1]T .

 1 2   s12  3 5  2   0  s1  or 3s1  5s2  0 2 2 Pick any s1 .07354 0.07382  T 12 . def.17981 )2  (0.991x103  0.7981 )2 df / d 1  7.8991 1.05045 T s1  f ( x1 )   0.991x103  0.8991 0  0.991x103 0.26549  0 1  0.798 x 2  x1  1s1   8.8991  1. determine s2 2 2 Problem 6. 1 1 H   is neg.17981 +0.112168 1  3.7981  T f ( x 2 )  10(8.1798  1.14 f ( x )  10 x1  x2 2 2  20 x  f ( x )   1   2 x2  x0  [1 1]T s0  f ( x0 )  [-20 -2]T x1  x 0  0 s 0  1  200 1  20  T f ( x1 )  10(1  200 )2  (1  20 )2 df / d 0  80080  404  0 x1   8.459 x 2   0.

15 f ( x)  ex1x2  2ex1  2ex2  ( x1x2 )2  x e x1x2  2e x1  2 x1 x2 2  f ( x )   2  2 xx x  x1e 1 2  2e 2  2 x1 x2  at x   0 0 . hence steepest decent will take more than two iterations. and stating that their ratio is 10.This is not the optimum ( x*  [0 0]T ). noting that they are positive. Note: The answer to the problem is easily obtained by first calculating the eigenvalues of H. f ( x )   2 2 T T Problem 6.16 f ( x)  x1  x2 2 2 f ( x )   2 x1 2 x2  T xold  3 5 T f ( xold )   6 10 T xnew  3 5    6 10  3  6 5  10  T T T f ( xnew )  (3  6 )2  (5  10 ) 2 df / d  272 136  0  136 2 136  34    1/ 2 xnew   0 0 which is the optimum T 13 . Thus more than two iterations are needed. Problem 6.

hence steepest decent search directions will require more iterations to reach the extremum than many other methods.4666 ( s1 )T Hs1 x 2  x1  1 s1   1.04591 T f ( x 0 )f ( x 0 ) s1  f ( x1 )   0 s 0   0.18 (a) f ( x )  3x1  x2 2 2 6 0  H   0 2  f ( x )   6 x1 2 x2  T x 0  1 1 T s 0  f ( x 0 )   6  2 T T f ( x 0 ) s 0 0   0 T 0  0.07142 0.1530  1.1785 ( s ) Hs x1   0.6428 T f ( x1 )   0.203x107  4.17 The direction of search calculated by the negative gradient does not point toward the extremum in poorly scaled functions. Problem 6.Problem 6. x *   0 0 T 14 .01x108  T This is very close to the true minimum.2857 T 0  T f ( x1 )f ( x1 )  0.4285 1.3775 T 1   T f ( x1 ) s1  0.

3393 T T f ( x1 )f ( x1 ) 0  T  0.05234  f ( x 0 )f ( x 0 ) s1  f ( x1 )   0 s 0   0.(b) f ( x)  4( x1  5)2  ( x2  6)2 8( x  5)  8 0  f ( x )   1 H    0 2   2( x2  6)  x 0  1 1 T f ( x 0 )   32  10 T s 0  f ( x 0 )  32 10 T T f ( x 0 ) s 0 0   0 T 0  0. and remains fixed from iteration to iteration.2859 2.  is prespecified. The gradient at x k gives the search direction.8446 T 1   T f ( x1 ) s1  0. 15 .0000262 T This is very close to the true minimum at x*  5 6 . T Problem 6.1339 ( s ) Hs x1  x 0  0 s 0  5.19 (a) Fixed step gradient: The move from a point x k to the next point x k 1 is given by x k 1 = x k + f ( x k ) .0001204 6. but  is determined at each iteration a unidimensional search to minimize f.6125 7.4666 ( s1 )T Hs1 x 2  x1  1s1  5. (b) Steepest descent: This is similar to (a) in that the search direction is given by f ( x k ) . The stepsize.

The value is 112770. def. If det H is not pos. The weighting factor depends upon the magnitude of the previous gradient. At (0. h > 0 f (h. The step-size is determined by a one dimensional search. at some (r.21 1  2  rh  10  r 2 2 r h Check to see if H is pos. Newton’s method may not converge. 2.16).22. 16 . but it must be positive definite at the minimum for the extremum to be a minimum. r )  f 2   2  h  20  r r  hr 3 f 1  2 2  2r h  r h 2 f 6  4  20  2 r r 2 f 2  3 2  2 rh  r h 2 f 2  2 3  2 hr  h r 2 f 2  2 3 2 h r h The elements on the diagonal of H are positive. and the determinant  6  2    2    r 4  20    r 2 h3      r 3h 2  2    0 ?      2 has to be positive for H to be positive definite.16).22 No.22. Problem 6. Problem 6. 2. h) during the search. for r > 0. def. 1) Problem 6. 1.20 The solution is: f  0 at (1.(c) Conjugate gradient: The new search direction is a linear combination of the gradient at the current point and the previous search direction. 1. hence Newton’s method will converge in the vicinity of (0.

Problem 6. (3) Simple method to understand (no complicated mathematics involved) and program. (4) Requires only one function evaluation per search step. Problem 6. so Newton’s method is not guaranteed to converge to minimum nor a maximum. (b) f ( x )  ( x1  5) 2  ( x2  8) 2  ( x3  7) 2  2 x1 x2  4 x1 x3 2 2 2 2 17 . From a positive starting point. as far as the number of iterations is concerned. and both yield the optimum in one step.23 Possible answers are: (1) If more than one extremum exists. Problem 6. (2) If the variables in the objective function are random variables as in experimentation. The search for a minimum can go to  as xi  . because the search direction is the same for both. the Simplex method may converge to a better local minimum than the Quasi-Newton (secant) method.24 They would both be equally fast.25 You must consider both minima and maxima (a) f  1  x1  x2  (4 / x1 )  (9 / x2 ) 1  (4 / x12 )  f ( x )    2 1  (9 / x2 )  8 / x13 0  H   3 18 / x2   0 H is not positive definite or negative definite for all x.

 2( x1  5)  4 x1 x2 2  8 x1 x32    2 f ( x )   2( x2  8)  4 x1 x2    2  2( x3  7)  8 x1 x3   2  4 x2 2  8 x32  H ( x)   8 x1 x2  16 x1 x3  8 x1 x2 2  4 x1 2 0 16 x1 x3   0  2 2  16 x3  It is hard to tell by inspection if H is positive definite for all x. by inspection of f(x) you can see that each term in the function is positive so that Newton’s method should reach a local minimum. which is the probable reason why the code fails.92.996. One exists at (-0.26 f ( x )  x1  x1 x2  x2 x1 3 2 2 3x 2  x2  2 x1 x2 2  f ( x )   1  2 2  x1  2 x1 x  6 x  2 x2 2 1  4 x1 x2  H ( x)   1  2 2 x1   1  4 x1 x2 at x *  1 1 . so that Newton’s method can be guaranteed to converge to the minimum. -6.993) with f = 24.27 f ( x )  2 x1  2 x2 2 2 18 . Problem 6. 7. Problem 6. However.0154.  4  3 H ( x* )   2   3 H is not positive definite at x0.

29 (a) Newton’s Method f ( x)  8 x1  4 x1 x2  5 x2 2 2  T f ( x )   (16 x1  4 x2 )(10 x2  4 x1 )  T  200   140   f  at (10.28 The Hessian matrix of f(x) is positive definite at the starting point. Therefore Newton’s method does not converge at all with   1 . x1  x 0   s 0   0 0 which is the optimum.f ( x )   4 x1 4 x2  T 4 0 H   0 4 The initial search direction is 0 0 1/ 4 0   4 x1    x1  s   H f ( x )     0   0  0 1/ 4  4 x2    x2  1 0 0 The step size is always   1 for Newton’s method. 10) 16 4    4 10   2 f ( x)   1 x (1) 10 16 4   200       10  4 10 140  1 16 4  1 16 4  4 10  144  4 10      19 . Only one step is needed to reach the minimum. T Problem 6. Adjusting  in the search direction will not help much. Problem 6. because f is quadratic. but does not remain positive definite as the search progresses.

2460  1. f ( x1 ) and calculate the next search direction s1   f ( x1 )  s 0  T f ( x1 ) f ( x1 )  T f ( x 0 ) f ( x 0 ) and continue. 600  0.1278 Next calculate f ( x1 ). 20 . Start with  200  s 0   f ( x 0 )     140  10   200  x1      0   10  140  Minimize exactly in the s0 direction to get  0  200 140  140   f ( x) s  0   T  s Hs 164   200  200 140      410 140  Then T  200 = 59. A computer program is needed to save user time.06x10 6 10   200 10 11.2460 x1     0.or solve 1  200 16 4  10  x1  0 140    4 10  2     10  x1  x   0 0 T solution: (b) Fletcher-Reeves Method Use an algorithm code such as shown in the text.05623        10  140  10 7.05623 1.8722   2.

2  0.7196 . Newton’s method converges to x *   0.27 7.2  0.350 0.Problem 6. (4) Requires only one function evaluation per search step.74x1010 T 0 Problem 6.2  0. f *  0. (3) Simple method to understand (no complicated mathematics involved) and program.3483 0.31 (a) Sequential Simplex Advantages: (1) If more than one extremum exists.30 (a) From both starting points. the Simplex method may converge to a better local minimum than the Quasi-Newton (secant) method. Disadvantages: (b) (1) Slow to converge (2) Not efficient for problems with many variables (3) Will not work for problems with constraints without modification Conjugate gradient Advantages: (1) Uses only first derivatives (2) Low storage required 21 . f *  0. (2) If the variables in the objective function are random variables as in experimentation.2 .6 T (b) x*  17.

0 2  0 x0    0 The maximum is at x  10 5 . Twenty independent variables makes a Simplex search not practical. but are not affected otherwise. Problem 6. The other two methods would converge more slowly. def. 0)  25 22 .Disadvantages: (c) (1) Have to reset directions after one cycle (2) Hessian may become ill-conditioned Newton’s Method Advantages: (1) Fast for reasonably scaled problems with one extremum (2) Simple algorithm Disadvantages: (1) Can perform poorly on problems with multiple extrema (2) Converges to a local extremum (as opposed to a global algorithm) (3) Requires second partial derivatives for a strict Newton method.32 Let   f ( x )  100  (10  x1 ) 2  (5  x2 ) 2  be maximized fˆ ( x)  100  (10  x1 )2  (5  x2 )2  2(10  x1 )  fˆ ( x )     2(5  x2 )  2 0 Hˆ ( x )    pos. T The minimum is at fˆ (0.

5174 1.9314 0.5174 3 0. BFGS method x ( k 1)  x ( k )   ( k ) Hˆ 1 ( x ( k ) )fˆ ( x ( k ) ) Let Hˆ (0)  I At (0.5 1. 0).9314 fˆ ( x ( k ) ) f(x) 25    14.5 26. b.8628 Or use a graphical procedure.a. 1 4 2.0348 -46.9  starting Simplex  0.6    25 14. Simplex Method Pick a suitable sized triangle.9 0. 3 5 3. 0) 1/ 2 0   20  10 x (0)   Hˆ -1fˆ ( x )  (1)     0 1/ 2   10   5  0  10  10  x (1)          0  5   5  c.4488 -26.6 Drop Point No. 23 . Newton’s method  1 Start at (0.6 46. f = -25.6 Drop Point No. Point x1(k) x2(k) 1 0 0 2 1.4488 2.

def. pick  to maximize fˆ ( x ) in the search direction  opt    20  10   1  10   2  2 0  20  20  10     0 2   10  20  f ( x )s T ( s ( k ) Hˆ ( k ) s ( k ) T (k ) (k ) 0   1  1 0   20 10 x (1)             0   2  0 1  10  5  (This is the optimum) If x(1) were not the optimum. the next stage via BFGS gives the approximate Hˆ 1  Hˆ  k but 1 (x ( k ) )x ( k )  (x k )T g ( k ) ( Hˆ ( k ) ) 1 g ( k ) (g ( k ) )T (Hˆ ( k ) ) 1   (g ( k ) )T ( Hˆ ( k ) ) 1 g ( k ) T T  T f (10. 10)  25 Alternately. Problem 6. 1 0   20   20  For   1: x (1)         0 1  10  10  fˆ (20.s (0) x (1) 0 1      (0)  0 0 ( s (0) is the negative gradient direction) 0  20 1  10 Pick a   1 or maximize in s  H 1 ( x)f ( x) direction to get  .33 f ( x)  x2  200x  104 f '( x)  2 x  200 f ''( x)  2 hence a minimum exists at x = 100 (pos. 5)   0 0 so x(1) will be 0 whatever ( Hˆ (1) ) 1 is.) 24 .

hence can stop. Let 0  10 x x= f ( x) 0 0 + 10 10 + 20 30 + 40 70 + 80 10.Bracket the minimum start at x = 0. Quadratic interpolation Pick 3 points. and bracket the minimum for all methods. bracketing minimum x f(x) 0 10.000 4.000 25 . f’ = 0.000 8. 900 900 2500 Newton’s Method x(1)  0  f '(0) 200   100 f ''(0) 2 x(2)  100  f '(100) (200  200)  100   100 f ''(100) 2 (one step for a quadratic function) Quasi-Newton (Secant) Method x (1)  0  x(1)  0  f '(0) f '(q)  f '( p) q p Let p = 0 and q = 150 200 200   100 100  (200) 2 150  0 At 100.

f ''  0 (not pos.100 0 150 2500 Fit f ( x)  a  bx  cx2 with min at x*   b 2c 104 = a 0  a  b(100)  c(100)2   b  200 2 2500  a  b(150)  c(150)   c 1 x*   200  100 2(1) Problem 6. Steepest descent At x  0. f '(0)   3x106 x(1)  0   (3x106 ) Select   1   or any other suitable choice or   min f ( x)  Select  opt (3x106 )(3x106 ) 1   6 6 (3x10 )(600)(3x10 ) 600 26 .34 f ( x)  ( x 100)3 f '( x)  3( x 100)2 f ''( x)  6( x 100) so an extremum is at x = 100 at x = 100. Start at x = 0 a. The problem has a minimum at x  . def.) f '''  6 hence x = 100 is an inflection.

d. you would have to pick another x p and xq . Newton’s method x(1)  0  f '(0) 3x106   . Quadratic interpolation Pick 3 points. bracketing minimum x f(x) 0 3x106 100 200 Fit f ( x)  a  bx  cx2 0 3x106 with minimum at df ( x) b  0 or x*   dx 2c 3x106  a 27 . To maintain + and – bracket on f '.x (1)  0  1 3x106   0.5x104  600 Select   1 x(1)  0  (3x106 )  3x106 Clearly x*goes to   b.5x104 f ''(0) 600 Same result as for steepest descent c. Let x p  0. xq  200 Quasi-Newton (Secant) method x(1)  0  f '(0)  f '(q)  f '( p)    q p   x(1)   3x106  100 3x106  (3x106 ) 200  0 At x = 100. f ' (100) = 0. The method will then proceed to  for x.

and Hˆ is positive definite and its inverse is positive definite. 2    17 If   17 . the eigenvalues of Hˆ are the roots of H  I  4   det   1   I  1 0 4     (    )2 17  0 1    17. then 1 and 2 are positive. 1 Alternate solution: Start with Hˆ 1   H   I  and proceed as above. Problem 6.123x1015  3x106  a  b(200)  c(200) 2  b  3x104 x*   3x104  1x1019   15 3.0  a  b(100)  c(100) 2  c  3. but you need to calculate H-1. Let Hˆ  H   I . Then.35 f ( x )  2 x1  2 x2  x1 x2 2 2  4 x1  x2   4 1 H     1 4   4 x2  x1   f ( x)   H is not positive definite.36 f ( x )  2 x1  4 x1 x2  x2 2 2  4 x1  4 x2   4 4  H     4 2   4 x1  2 x2   f ( x)   28 .123x10 Problem 6.

5  x1 (1  x2 ) u2  2. get the eigenvalues of H and add to them to get a pos. 1)     6 2  and det H = -12 so H is not pos. so f does not have a stationary point which is a minimum.25  x1 (1  x2 ) 2 u3  2.H is not positive definite. 1)with  given by   (12   )(2   )  36  0   14 12  0 2 or at = 0.  2  . 14  142  4(1)(12)  2 Another way: get the eigenvalues of H and calculate  .81 Problem 6. def.1  14.625  x1 (1  x2 ) 3 29 . it is meaningless to use Marquardt’s method to find one. Problem 6. approximation. Thus.81.37 f ( x )  2 x1  6 x1 x2  x2 3 2 6 x12  6 x2  f ( x )     6 x1  2 x2  12 x f ( x )   1  6 6  2  at x  1 1 T 12 6  H (1. Or. definite  Add  0 0 to H (1. As  4 4     for example   1. But a positive 4    4 ˆ definite approximation of H is Hˆ    and choose  to make H pos. def.38 f ( x )  u1  u2  u3 2 2 2 u1  1.

625  x1 (1  x23 )  (1  x23 )         f ( x )  2  2 3 2 1. we minimize –f(x): f ( x )  x1  x1  x2  x2  4 2 2 by the BFGS method.25  x (1  x )  (2 x x )  2.625  x (1  x )  (3x x )  1 2 1 1 2  1 2 1 2  1 2     Calculate H(x). Pick s 0  f ( x 0 ). at 0 1   x. s 0 (1. T T 30 . decompose H into LDLT  H  1e1e1   2e2e2 and change all negative ( 1 . Because H(x) is not positive definite: (1) (2) T At 0 1   x. you find x    2 2  T * Steps Pick x 0  1 1 .  2 ) to be positive.5  x1 (1  x2 )  (1  x2 )    2.40 (a) Maximize f ( x )   x1  x1  x2  x2  4 2 2 Instead of maximization of f ( x).  2 x1  1  f ( x )   2   2 x  1 1 1 If you set f ( x)  0.1)   1  1 .25  x1 (1  x2 2 )   (1  x2 2 )    2. The algorithm may initially use arbitrary search directions.1. T T T Or. True True Problem 6. Problem 6. add constants to the elements on the main diagonal of H(x) so that H(x) becomes positive definite. or so that the eigenvalues of H(x) become positive.5  x (1  x )  x    2.39 (a) (b) (c) False.

2 0 Pick Hˆ 0  H (1.) 0 2 1. 1     f ( x )s  1  k T  k k  2 0 ( s ) Hˆ ( x ) s  1  1    0 2  T k k  1  1    1 2 T 1 1  1 1 1 x         2 2  1 2  1 1 1  1 x  x  x     2  2 0 1 T 0 (the solution) 0  1  1 g 0  f ( x1 )  f ( x 0 )          0  1  1 g 0 (g 0 )T Hˆ 1  (g 0 )T (x 0 )  Hˆ 0 x 0 (x 0 )T Hˆ 0 ˆ 0 x0 (Δx 0 )T H  2 0  1/ 2  1  1 1   2 0 0 2  1/ 2    2  2  0 2  1  1  1     Hˆ 1      2 0   1/ 1/ 2 2     1   1     1  1      2  0 2  1/ 2   2  1/ 2  1 1 1 1 0 0          1 1 1 1 0 0  x1  1 Hˆ 1 ( x1 )f ( x1 )  0 1 1 The solution is at x    2 2  * T (as shown analytically) (b) (i) f ( x)  x1 exp( x2  x1  10( x1  x2 ) 2 ) 3 2 31 .1)    (pos. def. x1  x 0   s0  opt  1 1.

5942  T h0  x 0  ( Hˆ 0 )1 g 0  2.5 0.5942  1.   3x12  x13 (2 x1  20( x1  x2 )) exp  x2  x12  10( x1  x2 ) 2     f ( x )   3 2  x 1  20( x  x )  exp  x  x  10( x  x ) 2   1 2  1 2  1    2 1  x 0  1 1  T 1 Let Hˆ 0   0 0 1 f ( x 0 )   1  1  T T s 0  ( Hˆ 0 )1 f ( x 0 )  1 1  x1  (1    (1   )  T A search for  which minimizes f(x1) gives  = 0.5831 7. x1  1.0942 1.5942  g 0  f ( x1 )  f ( x 0 )     0.177  32 .9889 5.5942  x 0  x1  x 0  0.5.5831    5.5 1.5  f ( x1 )  1.0942  h0 (x0 )T  (x0 )(h0 )T (h0 )T g 0 x0 (x0 )T 0 1 ˆ ( H )   T (g 0 )T x0 (g 0 )T x0   g 0 )T x0  3.5  T T  2.

1).  x3       x 1 1 2   0  4 33 . 4. or if pick   1 2  x11  1  2 0  1     0 1 1  2  x2      1   1  2  2  0 The solution.5831 ( Hˆ 1 ) 1     5.5831 8. 1. Then s 0 is T s 0  f ( x 0 )   2  2  2  2 T x1  x 0   s 0 . Hˆ is invariant and positive definite    2 0    Hˆ  H   2   2  0  2  The gradient of f(x) is  2 x1  2 x  2 f ( x )     2 x3     2 x4  and at (1.9889 5. f   2 2 2 2 . 1.177  (ii ) f ( x )  x1  x2  x3  x4 2 2 2 2 The procedure is the same as for (i).

13149 k3 = 15.41 k1opt  0. This function satisfies the boundary conditions.42 The optimum is k1 = 2. a better approximation than y = a (1-x).1546  opt  1.Problem 6.355x10-5 Problem 6.44277 k2 = 3.43 Let y = a(1-x2). We want to find the value of a which minimizes 34 .717x104 Problem 6.6630 k2 opt  0.1593  = 4.

1  dy 2  4 1 F1      2 yx 2 dx  a 2  a 3 2  0  dx  8 1 * From dF1 / da  a   0.5.428 T x*  1 2 1  T x*   4 2 1  x 0  5 4 6  T T f *  0. 3 2 If a more complicated function is chosen.299 f *  0.267a1  0. say y  a1 (1  x 2 )  a2 (1  x 2 ) 2 .1048. Problem 6.219a2  2. a2  0.45 The solution is: T (a) x 0  5 5  (b) x 0  1 1 1  x*  4 2  T f *  3.12794 E11 and b* = -0. we conclude that a more complicated function improves the estimate of the minimum of the integral.464 35 .333a1  1.0469.152a2 2 2 The minimum of this function is a1  0. * * * Since F2* < F1*.133a1a2  0.5. F2  0.2993 E4 Problem 6. then F2  1. we get a*  3 /16. and F1  0.44 The optimum is Expected risk = -0.

2686 x 0   2 2 1  T x*   2 2 1  T f *  0.927 kL Q* = 179.719 (sum of squares of the errors) Problem 6.428 x 0   2 1 1  T x*   2 2 1  T f *  0. and slightly different T and Q give the same C.94089 b1*  3.2686 Problem 6.Problem 6.04917 b2*  0.88 $/kL The optimum is flat. 36 .46 The solution is (a) x 0   2 1  T x*  4 2  T x*  4 2  x 0   2 2  (b) T f *  3.48 The optimum is T * = 446.47 The optimum is a*  0.428 T f *  3.47456 f *  896.840 bbl/day C* = 17.

2 used (bbl/day) Constraints: Gasoline production 0.06 x 1+ 0. Let x1 = volume of crude no.70 x 2 < 6000 < 2400 < 12000 ($/day) (a) (b) (c) Also: x1  0 x2  0 1 . The problem to maximize the profit. 1 used (bbl/day) x2 = volume of crude no.00 x 1 + 0.09 x 2 Fuel oil production 0.24 x 1+ 0.31 x2 Kerosene production 0.60 x 2 profit = 1.7x1 + 0.1 The yields are bbl product per bbl of crude expressed as a fraction.CHAPTER 7 Problem 7.

50 x1 + 0.2 Basis: 1 run Units: x1 Ergies 1 run hr Ergies produced Let x1 = number of boxes of ERGIES per run x2 = number of boxes of NERGIES per run Maximize: profit Y = 0.60 x2 Constraints: blending time BT = (1/60) x1 + (2/60) x 2 < 14 hr cooking time CT = (5/60) x 1 + (4/60) x 2 < 40 hr packing time PT = (3/60) x 1+ (1/60) x 2 < 15 hr x 1.4 $/run 2 . x 2 > 0 optimum: x1 = 192 boxes/run x2 = 324 boxes/run Y = 290.Problem 7.

30 P2  $. 000 FC  5.Problem 7. 000 Problem 7. 000 P2  7.50) Inequality constraints: F  10.50) P2  FA (.70)  FC (. 000 FA  5.40 F Equality constraints: equality : F  FA  FB  FC P1  FA (.3 Objective function: y  $.60 P1  $. x2 = 4. 000 P1  4.40)  FB (. and fmax = 10.60)  FB (.4 The graph is shown below and indicates that the optimal solution is at the extreme point B where x1 = 2. 3 .30)  FC (. 000 FB  5.

6 Let i j designate the constituent index i = 1 to 4 designate the grade index j = 1 to 3 (A = 1.00 x1 j j 1 3 3 3 j 1 j 1 j 1 15. 2.60 x3 j  14.2 X 1C  0.C Problem 7. 3) produced per week on unit j (j = A. B. max : f ( x)  20( X1A  X1B  X1C )  6( X 2 A  X 2 B  X 2C )  8( X 3 A  X 3B  X 3C ) subject to: sales limits: X 3 A  X 3B  X 3C  20 hours available on each unit: unit A 0.20 y1  15.4 X 1B  0. C = 3) xi yj xij designate a constituent designate a grade is bbl/day of constituent i in grade j Objective function ($/day): 3 f  16.30 y3  13.30 x2 j  14.1X 3C and non-negativity constraints: 5 X ij  0 i  1.8 X 1 A  0.2 X 2 A  0. B = 2. We want to maximize the weekly profit.75 y2  15.3 X 2 B  10 unit C 0.Problem 7.3 j  A.3 X 3 A  20 unit B 0. B.5 Let Xij be the number of batches of product i (i = 1. 2. C).90 x4 j 4 .

T > 0 Note: It would be also ok to let P + T + F = F* (fuel oil) profit f(x) = 5F* . V.8F =5(P + T + F) –8F 5 . 000 x11  0.20 y2 yj  0 xij  0 4 4 4 i 1 i 1 i 1 y1  xi1 y2  xi 2 y3  xi 3 Problem 7.000 j 1 4j  4.10 y2 x21  0.15 y1 x12  0. F. Gravity requirement 8P + 7T + 24F > 12(P + T + F).10 y2 x31  0. 000  1.8V Also: P. 000 j 1 j 1 3 x j 1 2j 3j 3 x  2.50 y1 x13  0.7 The objective function is: Max: f(x) = 5(P + T + F) –8F The constraints are: Viscosity requirement 5P + 11T + 37F > 21(P + T + F).Constraints: 3 3 x  x1 j  3. Material balances P + V = 1000 T = 0.40 y1 x22  0.

since this will decrease f the fastest. x4 is the pivotal row. Problem 7. x4  11 but x5  3(not feasible). 11/5 and 4/1. (c) The limiting value of x1 is found from the ratio test. Problem 7. It is 11/5.9 (a) Choose the largest positive coefficient in the bottom row. You can delete F*. Check the ratios to find the limiting constraint. Therefore. 6 .8 After adding slack variable x1  2 x2  x3  x4 =11 4 x1  x2  2 x3  x5 3 2 x1  x3 = 1 There is no basic feasible solution because with x1  x2  x3  0 . and the same is true with respect to the number of variables. and “5” is the pivotal element. x1 should be increased first.8V .viscosity: -16P –10T + 16F > 0 gravity: -4P –5T + 12F > 0 material balances on P  V  1000  T  P   1000 viscosity breaker: T  0.8  The number of equality constraints depends on the number of substitutions you make. Choose the smallest positive ratio from –3/2. V. (b) Part (a) designated x1 as the pivotal column. it is 11/5.

8 -8.4 11/5 1. x2 . x 3.8 Let x1 replace x4 as a basic variable. say x2 = 1. x1 < 4.Problem 7. x 4 Problem 7. say x1 = 4 7 . x 5 The non-basic variables are: x 2. step 1: step 2: step 3: step 4: divide row 2 by 5 multiply row 2 by 2 and add to row 1 multiply row 2 by –1 and add to row 3 multiply row 2 by –4 and add to row 4 The next matrix is: x1 0 1 0 0 x3 x1 x5 f x2 14/5 2/5 -7/5 2/5 x3 1 0 0 0 The basic variables are: x1. x3  0.11 The constraints are: x1  2 x2  x3  11 (a) 4 x1  x2  2 x3  3 (b) 2 x1  x3  1 (c) x1 .10 Start with the following matrix: x1 -2 5 1 4 x3 x4 x5 f x2 2 2 -1 2 x3 1 0 0 0 x4 0 1 0 0 x5 0 0 1 0 f 0 0 0 1 b 3 11 4 0 x4 2/5 1/5 -1/5 -4/5 x5 0 0 1 0 f 0 0 0 1 b 7. From (c).get x3  2 x1  1 (d) From (a) and (d ) get 3x1  2 x2  10 (e) From (b) and (d ) get x2  1. Then from (e).

.. Now solve the LP Minimize: x6 Subject to: x1  2 x2  2 x3  x4  8 3x1  4 x2  x3  x5  x6  7 x1 . there exists a basic feasible solution. Add an artificial variable x6. x3  3 . Problem 7..Then from (d). (b) The constraints are x1  2 x2  x3  7 x1  3x2  x4  4 x1  3x2  f  0 or x1  2 x2  7 x1  3x2  4 f  x1  3x2 8 . In our case. to the second equality constraint. * * * * * * Problem 7. then the original LP does not have a feasible solution. x3 = 9..12 A phase I procedure can be used to obtain a feasible basic solution. and thus. If * x6  0 . x3 = 9 is a basic feasible solution. This is a feasible solution. x6  0 The solution to this LP gives a feasible solution to the original LP provided x6  0 . x2  1.13 (a) The basic variables have negative values: x3 = -6 and x4 = -4. x1 = 4. Therefore. This violates the non-negativity constraints. x2 = 1. * x1  x4  x5  x6  0.

5 -1. x2  3. the final matrix is x1 0 2 -5 x3 x2 f x2 0 1 0 x3 1 0 0 b 0 3 -9 The solution is: x1  x3  x4  0.5 x4 0 1 0 f 0 0 1 x4 -0.67 0. x3  7. (c) x2 is the incoming variable. If x3 leaves the basic set. f *  0 * * * * 9 . the solution is unique. f *  9 * * * * Thus. x2  3. the final matrix is: x1 2 0 -5 x2 x4 f x2 1 0 0 x3 0.The problem is unsolvable.5 -1. because the optimum is unbounded. x4  5. The ratio test says that either of x3 and x4 can leave the basic set. (d) The given problem is already at an optional solution: x1  x2  0. f *  9 * * * * If x2 replaces x4.33 -1 f 0 0 1 b 3 0 -9 The solution is: x1  x3  x4  0.

x3  11/ 3. Now solve Minimize : f  x1  x2 Subject to: x1  3x2  x3  12 x1  x2  x4  1 2 x1  x2  x5  4 2x1  x2  x6  8 x1 .. f *  0 * * The Simplex method did not exhibit cycling.14 Add slack variables x3 . f *  0.We can also pivot around “3” (x4 row. x5 and x6 . x2  5 / 3. * * * * This means the any point on the line segment joining the points T T 0 0 7 5  and 0 5 / 3 11/ 3 0  is optimal Problem 7. x4 ... x2 column) to get x1 0 2 -1 x3 x2 f x2 0 1 0 x3 0 0 1 x4 -2/3 1/3 0 f 0 0 1 b 11/3 5/3 0 x1  x4  0.. 10 . x6  0 The beginning matrix is x1 x3 1 x4 1 x5 2 x6 2 f -1 x2 3 -1 -1 1 -1 x3 1 0 0 0 0 x4 0 1 0 0 0 x5 0 0 1 0 0 x6 0 0 0 1 0 f 0 0 0 0 1 b 12 1 4 8 0 This is already at the optimal point! x1  x2  0.

43 1885. x* and f* are 11 .4428571 0.24 x1  0.09 0.2903222 -1.4937143 0.2580647 -0.1 was Maximize: x1  0.07 387.60 x2  12000 which is the same as Minimize  x1  0.31x2  6000 0.09 x2  x4  2400 0. Thus a small change in the profit coefficient of x1 does not affect the optimum (Note that if the profit coefficient were 1. At the optimum.15 Problem 7.84 658.06 x1  0. and x1 (crude # 1) is a non-basic variable.5806453 x2 0.10 -13548.935484 -2.Problem 7.24 x1  0.2580647 x4 0 1 0 0 x5 0 0 1 0 f 0 0 0 1 b 8571.24 1 x2 0.86 -8571.7 x3 1 0 0 0 x4 0 1 0 0 x5 0 0 1 0 f 0 0 0 1 b 6000 2400 12000 0 Now.31x2  x3  6000 0.7 x1  0.1.7 x2 Subject to: 0.7 x2 Subject to: 0.0634285 0. x4 and x5 are slack variables.7 x1  0. x2 (crude # 2) is a basic variable.39 This last matrix gives the optimal solution. The initial matrix is (the origin is a feasible solution): x3 x4 x5 f x1 0.60 0.4285714 -0.1432256 -1.09 x2  2400 0.06 0.428571 x4 0 1 0 0 x5 0 0 1 0 f 0 0 0 1 x3 3. its shadow price would still be negative.71 9942. which implies that the optimum.43 b 19354.2571429 x2 1 0 0 0 x3 1.0857142 -0.3428571 -1.06 x1  0.60 x2  x5  12000 where x3 .7 0. use the Simplex method: x1 x4 x5 f x2 x4 x5 f x1 1 0 0 0 x1 2.31 0.2258067 -0.1148388 -0.

unchanged). Problem 7. function: constraints: x1 4 -2 5 1 x2 2 2 2 -1 x3 0 1 0 0 x4 0 0 1 0 x5 0 0 0 1 Constants 0 3 11 4 Next (a) Increase x1 first as it has largest positive coefficient. Row no. to choose is 3 11 (is the limit) 5 (2nd constraint) 11 . x1. x4 will be non-basic. the current x* would still be optimal).7. x3 . x4  0. and has a basic feasible solution (with x1 = x2 = 0): obj. (b) Column no. 5 (c) Limiting value is x1  (d) The next basis will be x2  0. comes from the first constraint that is encountered Look at each one: bs  as xs  0 2nd row 3-(-2)x1 = 0 x1 < 0 ok to increase 3rd row 11–5x1 = 0 so x1  4th row 4-x1 = 0 so x1 = 4 Row no. (e) Use elementary operations to make the x1 column a unit vector 12 . A five percent increase in the profit coefficient of x2 influences the objective function the most. (Note: if the profit coefficient x2 were 0.735 instead of 0.16 Start with the matrix that has been converted to standard canonical form by the addition of slack variables. is 1.

a basic feasible solution is given by x1  x2  x3  0. xi  0 for 1  i  5. 4 x1  8 x2  x3  x5  40. 13 .1 x1 0 2 0 3 1 4 0 x2 8  2  5  2  2  5  2 5 7  5 x3 0 x4 -4/5 x5 0 Constants -44/5 1 2 5 1 5 1  5 0 22   3  5   11 5  11  4  5  0 0 0 1 Problem 7. Now consider the following array E1 2 x1  2 x2  x3  x4 = 16 E2 4 x1  8x2  x3 = 40 E3 7 x1  12 x2  3x3  x5 -Z = 0 Clearly. x4  16.18 To solve this problem. introduce slack variables x4  0 and x5  0 so that the constraints become 2 x1  2 x2  x3  x4  16.17 An LP code gives f ( x)  6 x1  0 x2  2 x3  0 Problem 7. and x5  40.

2 This yields another basic feasible solution given by x1  x3  x5  0. Furthermore. 1 1 E8  E4  E7 x1  x2 3 8 1 1  x4  x5 6 6  4. and Z  72. we choose column 2. x2  4. the maximum value of the objective function is 72. x3  8. we choose row 2 and pivot about 8x2 . E5  E1  2E4 x1  3 1 x3  x4  x5 4 4 E6  E3 12E4 x1  3 x3 2 3  x5  Z=  60 . This solution is optimal since all cost factors (in E9 ) are negative.  6. Doing this.According to the above algorithm we do not have an optimal solution because the coefficient of each xi in equation E3 is positive. Since 12 is the largest relative cost factor. since 16 / a12  8 and 40 / a22  5 . Thus. 14 . This yields the following array: E4  1 1 x1  x2  x3 2 8 E2 8 1  x5 8  5. we obtain 4 E5 3 4 4 1 x1  x3  x4  x5 3 3 3  8. 3 E9  E6  E7 2 3  x4  x5  Z   72 2 E7   x1 Continuing on until all the variables in the Z function have negative coefficients (so the variables cannot be increased). x2  5. then we know that this solution is not optimal. the optimal solution is x1  0 x2  4 x3  8 z  72 In this case we obtain as a basic solution x1  x4  x5  0. Since 3/2 is the largest relative cost factor we choose column 3 and observe that we must pivot about (3/4) x3 . and x4  6 Since there are still positive cost factors (in E6 ).

we must perform at least one pivot operation to obtain a basic feasible solution as a starting point. First. and this is why the initial solution has to be feasible. Second. If you multiply the last two in equalities by –1.20 (a) (b) (c) True True True Problem 7. To make this an equality. 2 and j = 1. the problem must be feasible. then from the constraint  j 1 a1 j x j  b1 we obtain  j 1 a1 j x j  b1. we multiply each equation by –1 to obtain a positive constant on the right-hand side of the n inequality. the assumption was made that all bi are nonnegative. It is better to translate x1 and x2 (by addition) to get new variables for which the origin is a feasible point.19 You want to have all of the b’s (the right hand sides of the inequalities) be positive at the start. i = 1. Problem 7. a trial and error process that may require considerable time before obtaining a suitable basic solution as a starting point. Inserting this equation n into the system of equations defined by the constraint inequalities shows that the basic solution of the resulting canonical system is not feasible since we would have xn1  b1  0 as part of the basic solution.Some comments are necessary with regard to the simplex algorithm as stated. If any bi is negative. Therefore. in general.21 Let xij = tons/day of product from refinery i transported to customer j. we must subtract a non-negative variable n xn1 from the left-hand side to obtain  j 1 a1 j x j  xn 1  b1. if b1  0. Capacity constraints: 15 . 2. This suggests that it might be desirable to have some test of feasibility before applying the algorithm unless it is known a priori that a physically realizeable solution must exist. in order to apply the simplex algorithm directly. but f   . 3. Problem 7. you change the values of the b’s to positive and reverse the direction of the inequalities. This requires. In particular.

1.5 ton/day Transportation cost = 25x11  60x12  75x13 + 20x21  50x22  85x23 ($/day) Total cost = Production cost + Transportation cost (to be minimized).3.7.50/day.9. then the solution to the LP is x11  0.7 tons/day * x12 x23  0.x11  x12  x13  1. x13  0.50/day.6 x21  x22  x23  0. we cannot use P1 = $30/ton. If we use P1  $40 / ton. Total cost = $151.5 ton/day ($/day)  $40 / ton if x11  x12  x13  0.2 tons/day  0. x22  0. x13  0. x11  x12  x13  1. The solution is x11  0. Problem 7.9 x12  x22  0.8. So. x12  x23  0. x22  0.8 tons/day * x13  0. x11  x12  x13  1.1 ton/day. then the optimum is x11  0. x21 x23  0.3. If we assume that P1  $30 / ton .22 Let xij = bbl/day of stream i used to make product j 16 .3 Production cost = P1 ( x11  x12  x13 )  35( x21  x22  x23 ) where P1  $30 / ton if x11  x12  x13  0. But.1 tons/day * x22  0.3 tons/day * x21  0. x12 .7 x13  x23  0.8 Minimum demand constraints: x11  x21  0. we used the correct value for P1 .7. * * Total cost = $151. x21  0. So.5 tons/day.

i= A C S j= A B L U alkylate cat cracked gas s.50( xAU  xCU  xSU ) Subject to: (i) availability constraints xAA  xAB  xAL  xAU  4000 bbl/day xCA  xCB  xCL  xCU  2500 bbl/day xSA  xSB  xSL  xSU  4000 bbl/day (ii) RVP constraints 5xAA  8xCA  4xSA  7( xAA  xCA  xSA ) 5xAB  8xCB  4xSB  7( xAB  xCB  xSB ) or 2 xAA  xCA  3xSA  0 2xAB  xCB  3xSB  0 (a) (b) (c) (d) (e) (iii)ON constraints 94xAA  84xCA  74xSA  80( xAA  xCA  xSA ) 94 xAU  84 xCU  74 xSU  91( xAU  xCU  xSU ) 108xAB  94xCB  86xSB  91( xAB  xCB  xSB ) 108xAL  94xCL  86 xSL  87( xAL  xCL  xSL ) or (f) 14 xAA  4 xCA  6 xS A  0 (g) 3xAU  7 xCU 17 xSU  0 17 xAB  3xCB  5xS B  0 21xAL  7 xCL  xS L  0 (h) (i) (iv) Non-negativity constraints all xi j  0 17 .r.50( xAB  xCB  xSB )  4. gas aviation gas A aviation gas B leaded motor gas unleaded motor gas Maximize: profit ($/day) = 5.50( xAL  xCL  xSL ) 4.00 ( xAA  xCA  xSA ) 5.

5x1  2x2  2. Problem 7.23 Let xi  100 lb / day of material i produced or consumed i = A. x5 . we have 2 2 1 x4  x5  x6 3 3 2 1 1 1 x2  x4  x5  x6 3 3 6 x1  1 x3  x6 3 Using these to eliminate x1 . x6  0 18 .Solution: x* AB  4000 bbl/day x* CB  2500 bbl/day x* SB  4000 bbl/day all other x*i j  0 x*i j  0 Profit = $57750/day. 5 5 we get the cost of raw material ($/day) = x4  x5  1.67 x4  1.5x3 2 1 Operating cost ($/day) = x4  x5  x6 3 3 From the material balances. E.8x6 Cost of raw material ($/day) = 1.3x5  0. therefore. Maximize: 1.92 x6 3 3 The objective is to maximize the profit: Profit = total income – raw material cost .3x5  3. x2 and x3 from the cost-of-raw material expression. B.88x6 Subject to availability of raw materials: 2 2 1 x4  x5  x6  40 3 3 2 1 1 1 x4  x5  x6  30 3 3 6 1 x6  25 3 x4 .operating cost The LP is. G Total income ($/day) = 4 x4  3. C. F.

37 x7 3/2 -1/2 0 -2.67 x4 1.51 x8 0 1 0 0 x9 0 0 1 0 f 0 0 0 1 b 60 10 25 -100.37 x6 ¾ -1/12 1/3 -0.67 x5 2/3 1/3 0 1.88 x7 1 0 0 0 x8 0 1 0 0 x9 0 0 1 0 f 0 0 0 1 b 40 30 25 0 x4 is the incoming variable. The next matrix is x4 x8 x9 f x4 1 0 0 0 x5 1 0 0 -0. 19 . The inequalities are converted to equality constraints: 2 2 1 x4  x5  x6  x7  40 3 3 2 1 1 1 x4  x5  x6  x8  30 3 3 6 1 x6  x9  25 3 x4  x5  x6 gives an initial feasible solution.3 x6 ½ 1/6 1/3 0.An equivalent objective is Minimize: 1. and no F and G. x6 .88x6 Introducing the slack variables x7 .3x5  0. x9 . x7 leaves the basic set.2 This matrix gives the optimal solution x4  60 x8  10 x9  25 x5 . x7  0 The optimum distribution is to produce 6000 lb/day of E. The initial matrix is x7 x8 x9 f x4 2/3 1/3 0 1. x8 .

24 Crude a b c d e f g h i j (profit or loss of each refinery cpb) Refinery X1 X2 Y Z Available M bpd -6 -11 -7 -1 3 -7 3 0 17 -16 16 13 10 9 13 3 63 49 60 48 34 16 25 15 15 4 12 7 22 10 19 17 -2 -8 4 9 15 8 13 3 Required M bpd 30 40 50 60 30 30 20 20 10 20 20 10 30 10 200 Refinery X1 PX 1  6 X1a  3 X1b  17 X1c  10 X1d  63 X1e  34 X1 f  15 X1g  22 X1h 2 X1i  15 X ij where PX 1 = net profit of Refinery X 1 X1a = crude a used in X 1 X1 j = crude j used in X 1 Refinery X 2 PX 2  11X 2 a  7 X 2b  16 X 2c  9 X 2 d  49 X 2e  16 X 2 f  4 X 2 g  10 X 2h 8 X 2i  8 X 2 j Refinery Y PY  7Ya  3Yb  16Yc  13Yd  60Ye  25Y f  12Yg  19Yh  4Yi 13Y j Refinery Z PZ  Z a  13Zc  3Z d  48Ze  15Z f  7Z g  17Z h  9Zi  3Z j Ptot  PX 1  PX 2  PY  PZ  objective function Ptot  6 X1a  3 X1b  17 X1c  10 X1d  63 X1e  34 X1 f  15 X1g  22 X1h  2 X1i  15 X1 j  11X 2 a  7 X 2b  16 X 2c  9 X 2 d  49 X 2e  16 X 2 f  4 X 2 g  10 X 2 h  8 X 2i  8 X 2 j 7Ya  3Yb  16Yc  13Yd  60Ye  25Y f  12Yg 19Yh  4Yi  13Y j Z a  0  13Zc  3Z d  48Ze  15Z f  7Z g  17Z h  9Zi  3Z j 20 .Problem 7.

Contraints 1) equality constraints: X1a  X1b  X1c  X1d  X1e  X1 f  X1g  X 1h  X 1i  X 1 j  30 X 2 a  X 2b  X 2c  X 2 d  X 2e  X 2 f  X 2 g  X 2 h  X 2i  X 2 j  40 Ya  Yb  Yc  Yd  Ye  Y f  Yg  Yh  Yi  Y j  50 Z a  Zb  Zc  Z d  Ze  Z f  Z g  Z h  Zi  Z j  60 2) inequality constraints: X1a  X 2a  Ya  Za  30 X1b  X 2b  Yb  Zb  30 X1c  X 2c  Yc  Zc  20 X1d  X 2d  Yd  Zd  20 X1e  X 2e  Ye  Ze  10 X 1 f  X 2 f  Y f  Z f  20 X 1g  X 2 g  Yg  Z g  20 X1h  X 2h  Yh  Zh  10 X1i  X 2i  Yi  Zi  30 X 1 j  X 2 j  Y j  Z j  10 Solved by Lindo: Pmax  2540x103 21 .

22 .

23 .

25 Solution (via Berkely LP code on the web): x1  200hr f = 19.27 Solution: x   0 0 7 00 T f 7 Problem 7.  i Si j  R j for each i.26 Solution: x1  0 x3  0 x2  1 f 8 Problem 7.  j Si j  Qi 24 .28 (a) Problem formulation: Minimize  Ci j   i Ci I i Subject to: for each j .Problem 7.000 x2  300hr N=2 x3  x4  x5  0 Problem 7.

..and  Ii  1 (b) I1 . 25 .Iiinteger .. I 2 . Numerical solution: Plant A 1 Plant B 0 Fixed Charge A Fixed Charge B 700 0 A to C 200 Cost A to C 200 A to D 250 Cost A to D 750 B to C 0 B to D 0 Cost B to C Cost B to D 0 0 Production A Production B 450 0 C 200 D 250 Total cost 1650 Build only plant A.

Problem 7.29 The solution is: 26 .

10). If x2  0. If x2  0. x1  1. If x2  105 .4 The Lagrange function is L( x. Problem 8. the numerical solution is (1.  )  x1  x2   (2 x1  x2  2) 2 2 1 .2 Starting from the non-feasible point (2. a minimum). a maximum). By 4 substitution of x1   x2 .1 Starting from the non-feasible point (10. the objective function becomes 2 f   1  105 x2  2 2 f '  2 1 105 x2  (105 )(2 x2 )  0 yields x2  0 or 1  105 x2  0 so that an alternate solution is x2  105. 0). If x2  105 . x1  1 105. 2). Check f ". f  4x105 (positive definite. 0). the objective function becomes f  x2 3 f '  4 x2  0 yields x2  0 so that x1  0 4 Problem 8. the numerical solution is (0. By 2 substitution of x1  1  105 x2 .3 Add to the LHS of the equation a variable that is always positive such as x  x4 x  x4 2 x  e x4 Problem 8. f  6x105 (negative definite.CHAPTER 8 Problem 8.

From (a) and (b): 10 x1 x2  6 x2  10 x1 x2  6 x1 x1   x2 From (c) and (d): 2 2 5x1  6 x1  5x1  8  0 2 2 2 (d) x1  x2   1 2 2 . and equate the resulting equations.The necessary conditions are: L   2 x1  2  0 x1  eliminate   2 x  4 x2  0 L  2 x2    0  1  x2 (a) L  2 x1  x2  2  0  (b) Solve (a) and (b) to get x1  0. def.8 and x2  0. Then   2 x2  0.  x2 L must be pos.4 .5 f ( x )  ( x1  x2 )1/ 2 2 2 h( x )  5 x1  6 x1 x2  5 x2  8  0 2 2 L  ( x1  x2 )1/ 2   (5x1  6 x1 x2  5x2  8) 2 2 1 2 2 L / x1  ( x1  x2 )1/ 2 (2 x1 )  10 x1  6 x2  0 2 2 (a) 2 1 L / x2  ( x1  x2 ) 1/ 2 (2 x2 )  6 x1  10 x2  0 (b) L /   5 x1  6 x1 x2  5 x2  8  0 (c) 2 2 2 2 2 Divide (a) by x1 and (b) by x2 . Check to make sure the above solution is a minimum. 0 2  Problem 8. 2 0  2x L    which is pos def.80 .8 and f ( x)  0.

 )  (C0  eT  Cr )2  T 2  u( 2  C0  K ) L / C0  2(C0  eT  Cr )  u  0 L / T  2(C0  eT  Cr )eT  2T  0 L / u   2  C0  K  0 L /   2 u  0 If u = 0.  2 ) and (  x1  x2   2 1   1 . (c) is not satisfied except for x1  1 . 3 . but eq. (c) but not eq.7 f (C.or 5x1  6 x1  5x1  8  0 2 2 2 1   1 The points  . we have a saddle point. (b). T )  K  C0  0 Eliminating C using he equality constraint. Thus   0. (b).6 2 2 2 L( x1 . T )  (C  Cr )2  T 2 C0 is a constant.   1 satisfies eq.  . x2  0 satisfies eq. h(C . but eq. T )  C0  e  C  0 T g (C.  and 2  2 and the points ( 2 . T .    are closest to the origin (distance =1) 2 2  2 2 )are the farthest (distance = 2). x1 is imaginary and eq.  )  x1  x2   ( x1  1)3  x2  L / x1  2x1  3 ( x1 1)2  0 L / x2  2 x2  2 x2  2 x2 (1   )  0 (a) (b) L /   ( x1  1)3  x2  0 (c) 2 If x T  1 0  satisfies eq. Problem 8. (c) will not be satisfied. we may write the Lagrangian as L(C0 . x2 . (a) gives 2 x1  3( x1 1)2  0. (a). Problem 8. we are not interested in this case. Hence no  exists.

371 Alternate solution: Use 2 Lagrange multipliers. so that the Lagrange multipliers exist. and the original function for f. except that the Hessian matrix of L must be positive semi-definite. Then the necessary conditions for an unconstrained function can be tested. 5.524 C  C0  eT  Cr  2  eT C  Cr  eT  2  0. A set of nonlinear equations obtained as in Step 5 of Part a still has to be solved for x*. You can substitute for V into the objective function. a set of 4 nonlinear equations obtained by setting the partial derivatives of L = 0 must be solved for x* and  * . 4 . They exist for the equality constraints.311) 2  (0.K  C0  Cr  2 eT (eT  2)  T  0 T  0. b. The gradients of the constraints are linearly independent. 3. that the eigenvalues of the H of L are negative or zero.8 a. you have to show that the Hessian matrix L is negative semidefinite at x * .311 f  (0. The constraints are satisfied (the second constraint is 3. The steps for part b are the same as Part a.98  4. The Lagrange multipliers for any inequality constraints are not involved in the problem.1 but close enough).e. 1. i. Problem 8. The objective function and constraints are twice differentiable at x* 2. 4.. and require the use of a nonlinear equation solver. 6. and get P in terms of xi. To show that x* is at a stationary point where  x L  0.524) 2  0. one for h and one for g. Additionally. This problem is as difficult to solve as the original NLP problem.

 *  15. f *  96 (b) L f    e e L  f  (24)(0.01)  0.  )  x1  x2  4 x1 x2   ( x1  x2  8) L / x1  2 x1  4 x2    0 (a) L / x2  2 x2  4 x1    0 (b) L /   x1  x2  8  0 (c) The solution to this set of equations is * * x1  4. x2 )  x1  x2  10 x1  20 x2  25  r ( x1  x2 )2 P / x1  2 x1  10  2r ( x1  x2 )  0 2 2 P / x2  2 x2  20  2r ( x1  x2 )  0 Simultaneous solution of these two equations gives 5 .5.5.Problem 8. * * f *  12.24 Problem 8.9 (a) The Lagrangian is 2 2 L( x1 .15 f * increases by 0.5 (b) f *   *e  (15)(0. x2  4.24 L  f  96. x2  2.  )  x1  x2  10 x1  20 x2  25   ( x1  x2 ) L / x1  2 x1  10    0 (a) L / x2  2 x2  20    0 (b) L /   x1  x2  0 (c) 2 2 The solution to this set of equations is x1  2.01)  0. x2 .10 (a) L( x1 .  *  24. x2 .15 (c) P( x1 .

Problem 8. we have x1  2. Thus. det ( 2 P)  4(2r  1)  0. h(x). because  2 P is positive definite Problem 8.x1  5r  5 2r  1 and x2  5r  10 2r  1 As r   . r  0 always and 2r  2  0. Also.12 (a) (b) f ( x).5 * (d) * From parts (a) and (c).5 and x2  2. P is convex.11 6 x 0 The Hessian of f ( x ) is  1   0 8 This is positive definite or indefinite depending upon the value of x1 . and g ( x) are twice differentiable Are the gradients of the binding constraints independent? 2 x  1  h ( x )   1   2 x2  1 6 . this problem is not a convex programming problem which requires that f ( x ) be convex and the equality constraint be concave. For convexity. you need the further specification that xi  0 . we have   2r ( x1  x2 ) *  (e) lim  2r ( x1  x2 )  r  2r   2r  2 2r  2   2r  2P   Now.

0. (d) The constraints are satisfied (e) The Lagrange multipliers of the inequality constraint is non negative (f) u*g (0. w* . u* )  0? L  ( x1  1)2  x2   ( x1  x2  x1  x2 )  u ( x1  x2 ) 2 2 2 2 L  2( x1  1)   (2 x1  1)  u  0 x1 L  2 x2   (2 x2  1)  u (2 x2 )  0 x2 at (0. (c) The Lagrange multipliers exist because (b) is satisfied. 0) = 0 ok (g) Is Lx(0. 0) 2    u  0     0 and u  2   u (0)  0  (h) so both (e) and (g) are satisfied Is the Hessian of L positive definite? 7 . 1  g ( x )     2 x2  Is the only solution at x T  0 0  of 2 x  1   1  c1  1   C2    0  2 x2   2 x2  1 C1  0 and C2  0? C1  (1)C2  0 so C1  C2 1  1 C1    C2    0 1 0  C1  0(C2 )  0 hence C1  0 so C2  0 Answer is yes.

14 (i) The functions are twice differentiable at x * 8 . However. Problem 8. if unique.  2 0  Is vT  v  0  0 4  No.) not pos def. Thus. is the only feasible point. Hence. Also g1 . def. because  1 vT    0 0  so v1 (1)  v2 (0)  0 so v1  0 1 vT    0 1 so v1  v2  0 hence v2  0 The answer to the problem is: No. this is a problem with three variables and four active constraints. (ii) All the four constraints are active 0   h  0  g1  0 1  0 0 0  g  1   g  0  at x* 2 3       0 0 1  but they are not linearly independent. and it is the minimum. the point  0 0 0  T does not satisfy the necessary conditions for a minimum. their intersection.13 (h) f is not twice differentiable at x* .(2  2 ) 0  2 x L   0 2u   for   0 and u  2  2 0  2 x L     0 4  (nor neg. Problem 8. g 2 and g3 are twice differentiable at x* .

vTh( x* )  4v1  2v2  0  v2  2v1 For the inactive constraint vTg ( x* )  v1  v2  0  v1  0  v1  0  2 2  v1  v T 2 Lv  v1 v2   No. (ii) h is the only binding constraint. and 4  h1 ( x* )    is linearly independent 2   (iv) L   x x   ( x1 x2  2 1 2 2 x1 2  6)  u ( x1  x2 ) L / x1  2 x1 x2   x2   x1  u  0  4  u  8 L / x2   x1   x1  u  0  2  u  4 2   2.(ii) The constraints are satisfied at x * (iii) h1 is the only active constraint. and T h  1  2  is linearly independent 9 .  2 x2   2 x L    2 x1   2 x1     2 2 =    0  2 0    x L is neither positive definite nor negative definite. x* does not meet the sufficient conditions for a minimum. u  0. g and h are all twice differentiable at x*.    > 0?  2 0  v2  We have not been able to show that vT2 Lv  0. 2 For the active constraint.15 (i) f. Problem 8. Thus.

Its gradient g1  1 1  is linearly independent.(iii) The constraints are satisfied at x*. and v '1  2v '2  1.16 (i) The functions are all twice differentiable at x*. (ii) g1 is the only active constraint. (iv) L  ( x1  2) 2  ( x2  1) 2  w( x1  2 x2  1)  u ( 2 x1 4  x2  1) 2 L / x1  2( x1  2)  w  ux1 / 2  0 L / x2  2( x2  1)  2w  2ux2  0 u = 0 because g is inactive.41  v 'T      0. Problem 8. Because of (iv) and (vi).82v '2   0. the sufficient conditions are not met. u3  0 * (iv) * * L  ln(1  x1 )  ln(1  x2 )2  u1 ( x1  x2  2)  u2 x1  u3 x2 10 .41v '1  1. u2  0. (iii) All the constraints are satisfied. v '1  0 v 'T 2 Lv '  2v '1  2v '2  0 2 2 Because of (iv) the necessary conditions are not met. g1 is the only active constraint. Thus L is not at a stationary point. (v) For the active constraint  1 v T    0  v1  2v2  2  2 0  v1  2 2 v T 2 Lv  v1 v2    2v1  2v2  0     0 2  v2  (vi) For the inactive constraint  0. There is no w which satisfies the above two equations at x*. so T u1  0.82  v '2  0.

4 x T  0.13 x T   0 0 0 T f 0 11 .17 Solutions: p8. v1  v2  0 1/(1  x1 )2  L 0  2  0 2 2 /(1  x2 )  Thus. 2 L is positive definite. Problem 8. the sufficient conditions are not met.4  f  0.8 statement for the solutions.8 See problem 8. 1/(1  x1 )  u1  u2  L   0  2 /(1  x2 )  u1  u3  u1  3/ 4  0 (v) Is vT2 Lv  0 ?  0? For the active constraints  1 v T    0  v1  v2  0  1 For the inactive constraints 1  v T    0  v1  0 0  0 vT    0 1   v2  0 Thus.80 p8. but vT2 Lv  0 as no non-zero v exists.8 0. Thus. p8.

f* = 96 Penalty function method: (a) (b) P  x1  x2  4 x1 x2  r ( x1  x2  8)2 P / x1  2 x1  4 x2  2r ( x1  x2  8)  0 P / x2  2 x2  4 x1  2r ( x1  x2  8)  0 Simultaneous solution gives 16r opt opt x1  x2  6  4r 2 lim  16r  4 r    6  4r  opt x1 (c) 2  x2 opt 4 2 lim opt lim  16r 2  32r  P  6 r  8 r  r    6  4r   6  4r  = 96 = f opt Problem 8.063x107 / year Problem 8.18 Direct substitution From h  x1  x2  8  0 we have x1  8  x2 .x11*  120 p8.24 * * * x12*  x13*  x23  x31  x32 0 * x21  20 * x22  100 * x33  170 f  $2. f  64  16 x2  2 x2 df / dx2  16  4 x2  0 and 2 d 2 f / dx2  4 2  x2  4 so that x1  4 T   4 4  is a maximum. Substituting into f.19 Lagrange multipliers: 12 .

L  f  4 minimum Penalty function method: P  x2  x1  r ( x1  x2  4)2 2 2 2 2 P / x1  2 x1  4rx1 ( x1  x2  4)  0 (a) P / x2  2 x2  4rx2 ( x  x2  4)  0 (b) 2 2 1 2 2 Multiply (a) by x2 and (b) by x1 . x2  2. w  1. There is one more case: x1  0 and x2  0 . L  f  4 minimum x1  0.L( x1 . x2  0. L  f  4 minimum x1  2. eqn (a) gives x1   4  (1/ 2r ) as r  . w)  x2  x1  w( x1  x2  4) L / x1  2 x1  2wx1  0 2 2 2 2 L / x2  2 x2  2wx2  0 L / w  x1  x2  4  0 2 2 Solution of these three equations gives four stationary points. r )  x1  6 x1  x2  9  r     x1 x2  Hessian matrix of P is 13 . but this is not a feasible point. x2 . w  1. Problem 8. w  1. x2  2. x2  0.20 (a) (b) 1 1 2 2 P( x. x1  2 The minimum is f  P  4 at x1  2. and subtract one from the other to get 4 x1 x2  0 For x1  0. x2  0. x2  2 For x2  0. w  1. eqn (b) gives x2   4  (1/ 2r ) as r  . x1  2. L  f  4 minimum x1  0.

Problem 8. (iii) An increased level of nonlinearity is introduced. In the region x1  0.22 (a) Penalty function problems: (1) min P( x)  2 x1  2 x1 x2  2 x2  6 x1  6 r ( x1  x2  2)2 (2) min P( x )  x1  3x1 x2  4  r1 (2 x1  x2  5) 2 2 2 3 2  r2  min 0. Disadvantages: (i) An initial feasible point must be located. x2  0 it always is. at the minimum of the original problem  0 0 . (ii) There is a possibility of overshooting the minimum – e.  )  x1  3x1 x2  4  w1 (2 x1  x2  5)  r1 (2 x1  x2  5) 2  w2 (5 x1  2 x2  18  2 )  r2 (5x1  2x2  18   2 )2 3 2 2 Problem 8.g. negating the pseudo constraint. 14 . w.23 (a) This formulation is called the method of moving truncations.2r  2  x 3 1 2 P     0     2r  2 3  x2  0 In general 2 P is not positive definite. so P is not convex. (b) Augmented Lagrange problems: (1) min P( x.  2 P T is not defined. r . w)  2 x1  2 x1 x2  2 x2  6 x1  6 2 2  w( x1  x2  2)  r ( x1  x2  2) 2 (2) min P( x. r . Advantages: (i) It will remain within the feasible region.21 Ans: Yes Problem 8. (5 x1  2 x2  18 2 Other penalty functions are possible – see text. However. (ii) parameter (“non parameter”) adjustment is automatic.

Because of complementary slackness. 3. 1  r / g  1  r  106 2  r / g 2  1  r  102 Problem 8. assume  *  1 .j = 1.(b) and (c) These formulations differ only in the penalty term. 2. For example. a much smaller value of r is required for the ln g term to satisfy g to within a given termination criterion. r rln g   g  1g g (r / g )  (r / g 2 )g  2g Assume that g is a tight constraint. We can examine the difference in terms of the value of r needed.24 Let X ij = million lb/year of DAB made by producer i and shipped to customer j: i. and that g is to be satisfied to within 106 . Production cost = P1 ( X11  X12  X13 )  P2 ( X 21  X 22  X 23 )  P3 ( X 31  X 32  X 33 ) where  45000  50( X ij  100 P1   44000  200( X ij  120) P2  50000 100   X i j  120 120   X ij  170 ($ /106 lb) ($ /106 lb)  39000  50( X 3 j  120 P3   46000  100( X 3 j  140) 120   X 3 j  140 140   X 3 j  200 Transportation cost  2000 X11  7000 X12  6000 X13  7000 X 21  3000 X 22 8000 X 23  6000 X 31  8000 X 32  2000 X 33 Constraints: Capacity 100  X11  X12  X13  170 15 . as 1 and 2 approach the optimal solution.

we shall linearize it about some nominal set of X 0ij .80  X 21  X 22  X 23  120 120  X 31  X 32  X 33  200 Demand X 11  X 21  X 31  140 X 12  X 22  X 32  100 X 13  X 23  X 33  170 Non-negativity all xij  0 Since the production cost is quadratic in the X ij . This gives Production cost = Q1  Q2  Q3 50000  x 0ij  50  x 0ij 2  (50000  100  x 0i j )( xi j )   for 100   x 0ij  120 Q1   0 0 2 0 20000  x ij  200  x ij  (20000  400  x i j )( xi j )  for 120   x 0ij  170   Q2  50000   x 20 j   50000  x2 j  33000  x 03 j  50  x 03 j 2  (33000  100  x 03 j )( x3 j )  for 120   x 03 j  140  Q3   0 0 2 0 32000  x 3 j  100  x 3 j  (32000  200  x 3 j )( x3 j )  for 140   x 03 j  200  The linearized transportation cost is 2000 x11  7000 x12  6000 x13  7000 x21  3000 x22 0 0 0 0 0 8000 x23  6000 x31  8000 x32  2000 x33  2000 x11 0 0 0 0 7000x12  6000x13  7000x21  3000x22 8000x23  6000x31  8000x32  2000x33 The linearized constraints are 100   x 01 j   x1 j  170   x 01 j 80   x 0 2 j   x2 j  120   x 0 2 j 120   x 03 j   x3 j  200   x 03 j 16 .

(5) Calculate the nominal x0ij ' s for the next iteration using ( x 0ij ) new = ( x0ij )old + (xij )opt . x 0 22  100. (2) Calculate numerical values for all the terms in the objective function and constraint equations which involve the x0ij ' s. assume x 011  140. then the current set x 0ij is the optimal solution. go to step 5. If not. x 0 21 . For example. x 033  170 x 012 . We now consider the LP: Minimize: (linearized production cost + linearized transportation cost) Subject to: all linearized constraints are satisfied. x 032  0 The LP is then: 17 . The solution strategy is as follows: (1) Assume a feasible x 0ij set. to maintain feasibility. x 0 23 . x 013 . x 031 . we require all xij  h (some preset quantity). (3) Solve the LP (4) If all the xij are very small (with respect to some preset tolerance). x j1  140   x 0i1  x j 2  100   x 0 2 j  x j 3  170   x 03 j all x 0ij  xij  0 In addition. (6) Go back to step 2.

 0 all xij  20 (say) The optimal solution is x11  20. x32 .097x107  78000x11  83000x12 +8200x13  57000x21  53000x22  58000x23  72000x31  74000x32 +68000x33 Subject to:  40   x1 j  30  20   x2 j  20  50   x3 j  30  x j1  0  x j 2  0  x j 3  0 x11  140 x22  100 x33  170 x12 . x33  0 The new x 0ij set is x011  120 x012  0 x013  0 x021  20 x022  100 x0 23  0 x031  0 x032  0 x033  170 This set is used to recalculate the various coefficients in the LP. x23 . x21 . x13 . x32 . x22 . x13 . x31 . x31 .Minimize: 2. x21  20 x12 . Repeated application of steps 2 through 6 gives the final solution as 18 . x23 .

0. Problem 8. 0.27 (a) Two local minima were obtained: T (1) x*   0.91878.62  1. 0. 0. f *  9.11752.91878. f *  9. Problem 8.86602 Problem 8.43  2.99307.01  .opt  120 x12 opt 0 x13 opt 0 opt  20 x22 opt  100 x23 opt 0 opt 31 0 x32 opt 0 x33 opt  170 x11 x21 x Cost = $2.39476. 0.39476.25 Solution: x*  0. 0.11752.47  .26 T The known solution is at x*  1 2 0  where f *  5.99307. 0.18 -2.995 (2) x*  0. 0.60445x1014  f *  0.051 T 19 T .16 2.063x107 /year.

472 1.847  0. # 1 1 1 2 1 6 4 3 4 f evals.52 x *   0.91 1.701  2.989 0.7661 2.96 x*   2.679 (2) f *  13.666  0.5377 3.362 1.52 x*   1.07877 x*  1.9496  2.619  2.4681  1.717 2.610 (4) f *  21.422 1.635 1. 64 64 51 42 34 40 42 50 68 48 -6 -8 -10 (b) Six local minima were found: (1) f *  0.1719 3.247 2.5 x*   2. 642 581 458 419 376 425 465 470 585 477 f evals.666 0. 106 165 427 1939 255 632 574 389 647 h evals.033  0.174  0.384 2.4859 0. 67 75 101 211 82 216 267 89 253 h evals. 169 175 141 97 93 106 110 133 187 131 h evals.604 (5) f *  86.106 (6) f *  649. # 2 2 1 1 1 1 2 2 2 2 f evals. 164 180 254 520 235 855 804 421 869 f evals.8479  0.2132  1.7401 T T T T T T n 2 4 6 8 10 -2 -4 soln.7359 (3) f *  27.n 2 4 6 8 10 -2 -4 -6 -8 -10 soln. 514 559 501 512 440 551 452 463 187 439 h evals. 30 33 43 85 39 80 160 42 161 20 .45 x*   -0.

Now h1 and h2 are not satisfied. x2 . So. set x3 = 0. solve h1 and h2 for x1 and x2. with x3 = 0.e. Thus. it is a feasible point. 21 .28 T You must start at a feasible point. the decent direction is -8. i. Phase 1: T 1 df f  f   h   h         dx3 x3  xD   xD   x3   2 x3   (4 x1  2 x2  4) 1  1 1 1  (4 x2  2 x1  6)       2 x1 5 0 1 1 1 1   2   6  2      8  0 5  0  For the variable x3 . xD   x1 x2  . Phase 2: get  : Find the minimum of the reduced objective function by an analytical approach to Set df d 0 ( x   ) d 2(0  8 )2  2(1  0 )2  (1  8 )2  2(0  8 )(1  0 )  4(0  8 )  6(1  0 )  0 d 384  64  0   1/ 6 1 x3  1  (8)  1/ 3 6 Since x3 has crossed its lower bound. and T x1 and x2 be the dependent variables (the choice is arbitary). x3  0. The point 0 1 1  satisfies h1 and h2 as well as x1 .Problem 8. Let x3 be the independent variable.

5121 x6  0.826 x1  9. If Newton’s method does not converge to a solution.4523 x8  0. Problem 8.29 Both are true.8 x1  0 x6  0 x2  0 x7  0 x3  0 x8  0 x4  0 x9  0 x5  0 x10  0 * * * * * (b) * * * * * * * * * * * * * * * x11  0 * (c) A reported solution is f *  14672.x1  x2  2  0 x1  5 x2  5  0 2 using Newton’s method.8558 x9  3.58267 x3  20 x6  6.6432 x2  4.51733 The solution is f *  17.30 (a) The solution is: f *  17286.1 x1  3.3922 x10  3. and is used to start the next iteration. The initial guess is x1 = 4/3 and x2 = 1.1518 x4  4.0877 x7  3.80 x1  0.4812 x4  0.52267 x4  20 x2  6.6023 * * * (d) * * * * * 22 . Problem 8.0243 The solution is: f *  12. The solution is a feasible point. replace x3 as the independent variable by either x1 or x2 and repeat.9191 x5  1.588 x5  13.4375 x3  2.

96 27.384 3.989 x1 1.6109 * T * * * x3  1.679 0.52 649.2132 3.7401 Problem 8.8479 -0.45 27.182 x5  0.7359 -2.166 x4  1.7514 x2  2.362 2.604 2.52 86. 23 .x5  0.5377 0.7661 -1.033 2.1719 x2 1.610 -1.666 2.31 Possible ways are (1) Minimize   g   h  where g 2 i i i are the violated constraints only.4681 1.5263 * x*  1 1 1 1 1  (e) The solution is f* 0 (f) The solution is f *  0.173 0.9496 -2.191 2.7012 * x4 1.422 -2.5 * * x3 1.2467 0.07877 13.106 -0.380 * (g) There are six reported solutions: f* * 0.635 -0.472 -0.266 -2.847 * x5 1.4859 -1.619 -0.717 -0.2415 x1  1. Change gi to equality constraints g1  4  x1  x2  x3  0 2 2 2 g2  x1  x2  16  x4  0 2 2 2 and minimize  ( hi  g i ) 2 2 from some reasonable starting point.4962 * * x3  0.506 x2  1.

L.Problem 8.33 Solution given in the problem. Problem 8. Problem 8.7L  22. this becomes (Nmin = 5) C  50 AN  0. A and L may be eliminated using equations (b) and (c) to get a cost function in terms of N only (the independent variable): 24 .34 The problem is C  C p AN  Cs HAN  C f  Cd  Cb  CL  Cx Minimize: Subject to:  L  L  1    D 1  N min / N   D min A = K (L + D) N > Nmin With numerical values.32 Solution given in the problem.000 Minimize: Subject to: L 1000 N N 5 (b) A L  1000 100 (c) N>5 a. and N. Although N is an integer we will assume it to be a continuous variable. (a) (d) The variables are A.

56254 E  4 x9  1. 000  0 dN ( N  5)2 N = 1 or N = 9 Because of constraint (d). This is a minimum N 9 C*  $38200 N *  9 plates L* = 2250 lb/hr A* = 32.90223 E  3 x4  3.75768 E  2 x3  9.36 The solution is f *  43. 000 = N 5 Thus. To obtain the minimum. 000 N 2  10.35 The solution is f *  267.5 z*  0 Problem 8. N is the independent variable. 000 N  9. 000 N  C  50 N   10   0.68197 E  4 x2  6.80709 E  2 x7  1. and A and L are dependent variables.07223 E  1 x8  2. 200 N  110.7    22. select N = 9 d 2C dN 2 b. 000 N 2  20.99958 E  3 x6  4.4945 x1  6. 000  N 5   N 5  1.5 ft2 Problem 8.51766 E  2 * * * * * * * * 25 . dC 1.5 x*  2 y*  7. 10 N   1.  500  0.

37 The problem is 10 f (a )    P(a.36359 E 0 a2  0.16220 E  1 a3  0.90799 E  1 x10  4.8408x103 26 .x5  4.110366 a1  0.29062 E  5 a5  0.91722 E  8 Problem 8.32901 E  3 a4  0.38 The solution with unscaled constraints from starting points 1 and 3 is f ( x* )  6. xi )  x Minimize: i 1 1/ 3 i   2 0 5 0      W1a  5 0 5     0 5 Subject to: W2 a  5 T where 1 W1 =  8  27   64 1 64 729 4096 1 1 1  512 4096 32768  19683 531441 14348907   262144 16777216 1073742824  T W2  125 15625 1953125 244140625 30517578135  The solution is f (a)  1.19451 E  2 * * Problem 8.

7214x104 x5  0. = -$11.1031 MM * * * * * * * * * * * * * Problem 8. func.1788x104 x7  0 lower bound x3  0.5131x10 4 x2  0.x1  0. Problem 8.40 Let f1 = tons of fuel oil to generator 1 f 2 = tons of BFG to generator 1 27 . = -$11.3686x104 x9  0.105 MM * * * * * * * * * * * * * The second best solution is: S1  10 P132  3 S2  0 P113  5 S3  0 P123  2 S4  0 P133  2 P111  1 P122  3 P121  2 P131  4 P112  4 obj.1000x104 upper bound * * * * * * * * * * Different codes gave slightly different results.6019x104 x4  0. The best and second best solutions were: Best solution: S1  7 P132  3 S2  3 P113  5 S3  0 P133  2 S4  0 P221  2 P111  1 P222  3 P121  4 P223  2 P112  4 obj.2436x104 x6  0.39 Case studies were used because local extrema were found.1328x103 x10  0. func.5795x102 x8  0.

000916 x21 2 g2  0.001832 x21 0 C  0. f 2 .2617  0.2013x21 0.00145( x110 ) 2  0.4609  0. j  1.15186  0.0029 x11 0 B  0.2556  0.g1 = g2 = x11 = x12 = x21 = x22 = tons of fuel oil to generator 2 tons of BGF to generator 2 MW from generator 1 obtained using fuel oil MW from generator 1 obtained using BFG MW from generator 2 obtained using fuel oil MW from generator 2 obtained using BFG The NLP is Minimize: f1  g1 Subject to: f1  1.00145 x11 2 f 2  1.15186 x11  0.00916( x21 ) 0 0 28 .2013  0.5742  0.7266  0.000778x 0 22 0 2 0 2 0 2 F  2.1631x12  0.2013x21  0.6992  0.15186 x11  0.002716 x12 0 D  0.001556 x22 0 E  7. the NLP is converted to an LP: Minimize: Ax11  Bx21  F Subject to: x11  x12  30  x11  x12 0 x21  x22  25  x21  x22 0 0 0 x11  x12  18  x11  x12 0 0 x21  x22  14  x21  x22 C x12  Dx22  E 0 0 x11  x12  x21  x22  50  x11  x12  x21  x22 0 0 0 0 where A  0.8008  0. 2).001358 x12 2 g1  0.1631  0.001358 x12  0. g1 and g 2 using equality constraints.000778 x22 18  x11  x12  30 14  x21  x22  25 f 2  g 2  10 x11  x12  x21  x22  50 2 Eliminate f1 .2556 x22  0.2556 x22 0. Linearizing the objective function and the constraints about some set xij (i.1631x12  0.

h. x12 . Now solve the LP again using these x0ij ' s . say.2382 * x22  12.Since the x ' s are unrestricted in sign.0235 f1  2. The solution is x11  7. and solve the LP. x21 and x22 .292 * x21  7. This process is repeated until the x ' s are less than some specified tolerance. we introduce the new variables x11  y1  y2 x12  y3  y4 x21  y5  y6 x22  y7  y8 If we restrict the absolute values of the x ' s to be less than. From the optimal solution.718 29 .7084 * x12  22.762 * f *  5. the LP is now Minimize: Ay1  Ay2  By5  By6  F Subject to: y1  y2  y3  y4  30  x11  x12 0 0 y5  y6  y7  y8  25  x21  x22 0 0  y1  y2  y3  y4  18  x11  x12 0 0  y5  y6  y7  y8  14  x21  x22 0 0 Cy3  Cy4  Dy7  Dy8  E y1  y2  y3  y4  y5  y6  y7  y8  50  x11  x12  x21  x22 0 0 0 0 y1  y2  h y2  y1  h y3  y4  h y4  y3  h y5  y6  h y6  y5  h y7  y8  h y8  y7  h and all yi  0 0 0 0 0 The strategy is to assume values for x11 . calculate the x ' s and thus the new x0ij ' s .

The equations for the heat transfer in the heat exchangers are those found in unit operations books: Q  UATln  UAT Q  wC p t The solution is: T1  1800 F * T2  2950 F * A1  A2  A3  7050ft 2 (minimum total area) A1  556 ft 2 A2  1369 ft 2 A3  5125 ft 2 30 .885 g1  2.115 Problem 8.f 2  5.306 g2  4.41 You can use T rather than log mean T to simplify the problem.

000 y3 100.000 y3  75.000 y1  150..000 y4  300.000 y5  200.000 y2  0 y3  50..000 y1  100.000 y6  450. 2.000 y5  60.000 y6  400.000 (b) Second year expenditure: g2  0 y1  300.000 y2  35. where y j  1 means the jth project is selected.000 y4 125.000 y2  200. 000 (c) Engineering hours: g3  4000 y1  7000 y2  2000 y3  6000 y4  3000 y5  600 y6  10. 6.. j  1.000 y5  100.000 y4  50. The objective function to be maximized is f  100.000 (d) Production line is required: g 4  y1  y2  1 (e) Automation is available only with new line: g5  y2  y3  0 (f) Waste recovery option: g6  y5  y6  1 (g) The branch and bound analysis begins by solving the LP problem with no integer restrictions on the variables with the following result: 1 .000 y6 (a) subject to the following constraints: First year expenditure: g1  300..1 Define the variables as y j . and y j  0 means the jth project is omitted.CHAPTER 9 Problem 9.

The final (optimal) integer solution is: y1  1 y2  0 y3  0 y4  0 y5  1 y6  0 f  $225. all t y variables turn x variables on and off: 2 .x(i.12 y3  0.t) > 0. zero otherwise.t) = power generated by generator i in period t (MW) y(i. y4 ) in this solution are not integers.12 y4  0. Constraints: Demand must be satisfied in each period: Sum(i. 200 Note that several variables ( y1 .2) Data: cap(i) = capacity of generator i (MW). t = time period index (t = 1. y3 . The branch and bound analysis can be carried out with Excel.t)) >= d(t).3). using GAMS-like notation: Indices: i = generator index (i=1. y2 . d(t) = demand for power in period t (MW) Decision variables: x(i.00 f  $265.00 y6  0. Problem 9. 000 This indicates that the project 2 with the highest net present value is not selected because of the constraints in the problem.88 y2  0. as is expected.2. cp(i) = operating cost of generator i ($/MW) Cs(i) = startup cost of generator i ($).2 An algebraic formulation follows.t) = 1 if x(i.y1  0. Note that the first noninteger solution achieves a larger value of f than the integer solution.40 y5  1.

x(i. generator 1 is used to capacity. Even though generator 3 has the lowest startup cost. if a generator is turned on in period 1.1) + y(i.2)) The last constraint above insures that. while 3 is much higher.2) <= cap(i)*(y(i.t)) <= 1. and 2 is used to satisfy the remaining demand.1) x(i. it stays on in period 2. all i A spreadsheet model containing the optimal solution appears below. This is because generators 1 and 2 have the lowest operating costs.1) <= cap(i)*y(i. The optimal solution turns generators 1 and 2 on in period 1. Data Generator 1 2 3 Fixed Cost per Generator Demand Demand start-up MW capacity Period 1 Period 2 cost ($) in each period (MW) 2800 5 1900 2500 3500 2000 3 1700 1900 8 2900 Model Gen 1 Gen 2 Gen 3 Total MW Gen 1 Gen 2 Gen 3 Total on-off MW per 1 1 1 0 per 1 800 1700 0 2500 MW cost 13000 10200 0 23200 <= <= <= >= startup cost 2800 2000 0 objective 4800 28000 onoff*cap per 1 1900 1700 0 2500 on-off MW per 2 0 0 0 per 2 1800 1700 0 3500 <= <= <= 1 1 1 <= <= <= >= sum*ca p per 2 1900 1700 0 3500 sum of binarys 1 1 0 3 . Turn each generator on at most one time per day: Sum(t. Further. because it has by far the lowest operating cost. y(i. and does not use generator 3. its higher operating cost excludes it.

The second year nothing is bought. because their cost per MW is much higher than the 100MW. wk4 / . wk3. Production and Inventory Planning with Setup Costs and Times set definitions 4 5 Set p products / p1.4. because first year cost is smaller than second year. model parameters 4 . wk2.3 DATA Year Minimum capacity Cost 10 MW Cost 50 MW Cost 100 MW 1 2 3 4 5 780 860 950 1060 1180 280 230 188 153 135 650 538 445 367 300 700 771 640 530 430 Total current capacity Cost of new Generators 1400 0 640 530 430 3000 700 Decision on Variables and Constraints Year # of 10 MW # of 50 MW 1 2 3 4 5 size 0 0 0 0 0 10 0 0 0 0 0 50 # of 100 MW 2 0 1 1 1 100 new capacity 200 0 100 100 100 total capacity 900 900 1000 1100 1200 Comments on Solution No 10 or 50 MW generators are installed. The first year two 100 MW generators are bought.Problem 9. p2 / 9 t time periods / wk1. Purchases in years 3 to 5 are deferred as long as possible because costs are declining. A GAMS model for this problem and its solution follows. In years 3 to 5 only one 100 MW generator is installed. Problem 9.

inv(p.pinv.t) units of product p produced in week t 44 inv(p.13 14 Table pdata(*. 51 Binary variables y .t) =e= inv(p. inv(p.t) positive part of inventory 47 ninv(p. 50 Positive Variables prd.t).t) defines positive and negative inventory 57 maxtime(t) limit on production time 58 onoff(p.t))=l= tavail .t).. inv(p.p)*prd(p. 67 68 maxtime(t). sum(p.p)*y(p.t). 66 invsplit(p.p). 70 5 .t)+ 69 pdata("setup-time". 27 28 Table demand(t.t) inventory balance 55 finalinv(p) final inventory equal zero 56 invsplit(p.p).t) units of inventory of product p at end of week t 46 pinv (p. 52 53 Equations 54 invbal(p. 35 36 Scalars tavail weekly time available(hrs) / 90 / .p) 29 30 p1 p2 31 wk1 75 20 32 wk2 95 30 33 wk3 60 45 34 wk4 90 30 .t) negative part of inventory 48 y(p.ninv .p) product data 16 p1 p2 18 setup-time 6 11 19 setup-cost 250 400 20 production-time 0. 38 maxunits(p) = tavail/pdata("production-time".t-1)+prd(p. pdata("production-time".."wk4") =e= 0..t) =e= pinv(p.75 21 production-cost 9 14 22 holding-cost 3 3 23 penalty-cost 15 20 24 selling-price 25 35 26 final-inventory 0 0 .5 0.t)-demand(t.t) turn prd on or off with binary variables 59 oneprod(t) at most one product produced in any week 60 objective revenue minus all costs 61 62 invbal(p. 40 Model Definition 43 Variables prd(p.t) 1 if product p produced in wk t else zero 49 profit objective variable .t)-ninv(p. 37 Parameter maxunits(p) .. 63 65 finalinv(p).

72 73 oneprod(t). 80 81 Model prodplan / all / .t).0000 This model breaks inventory into positive and negative parts as defined by the variables pinv and ninv.p)*ninv(p. This causes the penalty cost to be incurred.. because only one product can be produced in any week. The binary variables y turn the production variables prd on and off through the constraints “onoff”.t) =l= maxunits(p)*y(p. 84 MODEL STATISTICS BLOCKS OF EQUATIONS BLOCKS OF VARIABLES NON ZERO ELEMENTS 7 6 121 S O L V E MODEL TYPE SOLVER PRODPLAN MIP XPRESS **** SOLVER STATUS **** MODEL STATUS **** OBJECTIVE VALUE SINGLE EQUATIONS SINGLE VARIABLES DISCRETE VARIABLES 35 41 8 S U M M A R Y OBJECTIVE DIRECTION FROM LINE PROFIT MAXIMIZE 83 1 NORMAL COMPLETION 8 INTEGER SOLUTION 5331. 6 .t). The optimal solution produces product 1 in weeks 1 and 3. (pdata("selling-price". If there are only 80 hours available per week.p)*pinv(p. but the backlogged demand is satisfied in the next week (The constraint that inventory is zero at the end of week 4 insures that there is no backlogged demand after the fourth week). 83 Solve prodplan using mip maximizing profit .. All of the 90 available hours are used in week one.71 onoff(p. and product 2 in weeks 2 and 4..p))*prd(p. profit =e= sum((p. 74 75 objective. and the equations invsplit.t)) =l= 1. prd(p.t) 77 -pdata("setup-cost".t)). sum(p.t) 79 -pdata("penalty-cost".p) 76 -pdata("production-cost". and to include setup costs into the objective. the problem has no feasible solution. because the costs for positive and negative inventory are different.t). The “oneprod” constraints insure that at most one product is produced in any week. They are also used to incorporate setup times into the maxtime constraints.t) 78 -pdata("holding-cost".p)*y(p. There is unsatisfied demand for product 1 in week 2 and product 2 in week 1. This model can be used to determine approximately how many hours per week are needed to permit a feasible solution by solving it for different values of the parameter “tavail”. but fewer are needed in subsequent weeks. y(p.

and y(I) = 1 if and only if x(I) is between 5 and 20. all I (1) x(I) >= 5y(I) all I (2) x(I) >= 0 all I Where y(I) is a binary variable which is 1 if x(I) is positive and zero otherwise.x(I)) <= 100 x(I) <= 20y(I) .(1+r(I))*x(I)) Subject to: Sum(I. Constraints 1 and 2 insure that y(I) = 0 if and only if x(I) = 0.Problem 9.6 7 .5 Max sum(I. Problem 9.

Problem 9.7

Problem 9.8

Nodes 4 and 5 are fathomed. Node 2 gives the MIP optimum.

8

Problem 9.9
By inspection, one can see that plant 1 has to be in operation, because plant 2
cannot satisfy the demand by itself. Thus, there are only two possibilities:
y1  1, y2  1 and y1  1, y2  0. Explicit enumeration is easy in this situation.
y1  1, y2  1. The LP to be solved is

Case I:

f  (3x11  x12  x22  2)x104
x11  x21  1
x12  x22  1
x11  x12  2
x21  x22  1
*
*
*
*
The solution is x11
 x22
 0, x12
 x21
 1,
*
4
f  3x10

Minimize:
Subject to:

Case II:

y1  1, y2  0. The LP to be solved in this case is:

f  (3x11  x12  1)x104
x11  1
x12  1
x11  x12  2
*
*
*
*
The solution is x11
 x12
 1, x21
 x22
0
f *  5x104

Minimize:
Subject to:

Case I gives the optimal solution to the problem.
Problem 9.10
The solution can be obtained by inspection. To each extractor, assign the stream
with the least cost for that extractor. The optimum pairing is
Stream
1
2
3
4
cost = 64

Extractor
4
2
3
1

9

Problem 9.11

Nodes 6 and 7 are fathomed. Node 5 gives the IP optimum.
Problem 9.12

x1  0
x3  0
x2  1
f 8

Problem 9.13

x   0 0 7 0 0

T

f 7

Problem 9.14
(a)

Problem formulation:
Minimize:

 Cij  i Ci I i

10

Subject to:

for each j,

i Sij  R j

for each i,

 j Sij  Oi

and  Ii  1
(b)

I1 , I 2

Ii integer.

Numerical solution:
Plant A
1

Plant B
0

Fixed Charge A
700

Fixed Charge B
0

A to C
200

A to D
250

B to C
0

B to D
0

Production A
450

Production B
0

Cost A to C
200

Cost A to D
750

Cost B to C
0

Cost B to D
0

C
200

D
250

Total cost
1650

Build only plant A.

11

Problem 9.15
The solution is:
Refinery 1
5000

Refinery 2
7500

X111
X112
X113
X121
X122
X123
X131
X132
X133
X211
X212
X213
X221
X222
X223
X231
X232
X233

M1
2000

500
1500
1.93E-13
5.806898
0.009896
0
2499.993
0.016408
0
0
2.76E-13
0
494.1931
1499.99
3000
0
1499.984
0.006911

b11
b12
b13
b21
b22
b23
b31
b32
b33

M2
5000

M3
4000

Meet all Demand (2)
0j=1
0j=2
0j=3
Can’t Exceed Supply (3)
-494.1737 1 = 1
-1005.826 1 = 2

k1
0.08

k2
0.06

k3
0.04

112
113

1
1

122
123

1
1

132
133

1
1

con (4)
0j=1
0j=2
0j=3
con (5)
0j=1
0j=2
1999.993j = 3

500
1500
10000
500
1500
10000
500
1500
10000

con (6)
-10000 j = 1
-7000 j = 2
-9999.993 j = 3
con (7)
0j=1
0j=2
0j=3

C111
C112
C113
C121
C122
C123
C131
C132
C133
C211
C212
C213
C221
C222
C223
C231
C232
C233

40
90
7.72E-15
0.464552
0.000594
0
199.9994
0.000984
0
0
1.66E-14
0
39.53545
89.99941
120
0
89.99902
0.000276

OBJ

669.9997

12

.

carry out the preliminary design and optimization.Conceptual Design and HYSYS.Application Example http://www. Using HYSYS.Conceptual Design is used to calculate the interaction parameters and carry out the preliminary design and optimization of the process. Reboil Ratio and Purities C-6. optimize the steady-state process.6 Optimization Setting up the Spreadsheet Comparison of Configurations Optimization C-6. The objective is to maximize the purity of the Heptane and Toluene streams coming off the top of the first and second column. the column is set up and optimized. set up the column configurations in a single flowsheet. use the Optimizer to further refine the extractive distillation process.5 High Purity Configuration Pressures Temperature Estimates Solving the Column Low Purity Configuration C-6. Using HYSYS .SteadyState and Dynamic Design. 3. using the Spreadsheet to model economic factors.Conceptual Design.SteadyState. Finally.hyprotech. estimating/specifying key process characteristics such as Reflux Ratios. feed location. 4. HYSYS . controls are added and various disturbances are introduced to test the effectiveness of the design. the equimolar feed of Toluene and Heptane is separated using Phenol as a solvent.Conceptual Design. respectively.htm Second Column First Column: Second Pass Second Column: Second Pass Effect of Reflux Ratio.1 Process Description Using HYSYS . 2.SteadyState. and product purities. Using Hyprotech's process simulation software. In HYSYS.SteadyState. In two distillation columns. a two-column extractive process is modeled from conceptual design to dynamic simulation. Using HYSYS. taking into account 2 of 90 2/24/99 10:21 AM . and develop and test a control scheme. and determine an appropriate Property Package. calculate interaction parameters.9 Bibliography C-6. Using HYSYS .com/support/examples/extract/extract. number of stages. using the specifications determined in the previous step. These are the steps: 1.8 Summary and Conclusions C-6. we can develop a conceptual design.7 Overview Sizing the Vessels Adding the Controls and Sizing the Valves Dynamic Simulation C-6.

and the bottoms mixture (solvent and less volatile component) is separated in a second distillation column. the presence of the solvent raises the vapour pressures of the key components to different degrees. In general. Column SubFlowsheet C-6.com/support/examples/extract/extract. is to maximize the separation of n-Heptane and Toluene.Conceptual Design could also be used to screen solvents based on their effect in increasing the relative volatility of n-Heptane and Toluene.htm the basic economics.2 Background HYSYS . The more volatile component is removed in the distillate. 3 of 90 2/24/99 10:21 AM .hyprotech. The non-toluene fraction is often a narrow mixture of saturated hydrocarbons. Using HYSYS.Application Example http://www.Dynamics. and for the purpose of this study will be represented by n-Heptane. set up a candidate control scheme and evaluate dynamic operability. The objective of this process. Extractive distillation is used in the petroleum industry for the separation of aromatics from non-aromatic hydrocarbons. Toluene-"non-toluene" separation is well-documented. 5. therefore. so that the relative volatility between these key components is increased.

Application Example

http://www.hyprotech.com/support/examples/extract/extract.htm

Reflux Ratio No. of Stages
10

113

15

71

20

61

Reflux Ratio and number of stages for the non-extractive equimolar separation of nHeptane and Toluene, as
predicted by HYSYS - Conceptual Design (NRTL-Ideal). The distillate and bottoms molar purities are 0.99.
Phenol is commonly used as the solvent, due to its effect in significantly increasing the volatility ratio of n-Heptane and
Toluene. Unlike other potential solvents which can also increase the volatility ratio, phenol does not form azeotropes,
and is currently inexpensive. It is not particularly dangerous, although there is some concern as to its environmental
impact.
Since n-Heptane and Toluene do not form an azeotrope, the separation can theoretically be performed without the use
of a solvent. However, the number of stages and reflux ratio is excessive, as shown in the side table. This is due to the
fact that these components have similar volatilities.
This example is set up in five parts as outlined below. Some sections can be completed independently, without
referring to previous steps. For example, if you wish to do only the Steady-State design, you need only complete steps
3 and 4, using the interaction parameters and column design as predicted in steps 1 and 2.
1. Calculating Interaction Parameters - HYSYS - Conceptual Design - page 4
2. Ternary Distillation Design - HYSYS - Conceptual Design - page 15
3. Building the Columns in HYSYS - HYSYS.SteadyState - page 26
4. Optimization - HYSYS.SteadyState - page 33
5. Dynamic Simulation - HYSYS.Dynamics - page 55

C-6.3
Calculating Interaction Parameters
Using experimental data from various sources, interaction parameters are generated using the NRTL and Peng
Robinson Property Packages. Interaction parameters for the three binary pairs are obtained separately and combined
in the binary matrix.

NRTL Interaction Parameters

4 of 90

2/24/99 10:21 AM

Application Example

http://www.hyprotech.com/support/examples/extract/extract.htm

In earlier versions of HYSYS - Conceptual Design, you must have only two components in the Fluid Package in
order to view binary TXY and XY plots.
In HYSYS - Conceptual Design, open the Fluid Package Manager and add a new Fluid Package. The Fluid Package is
defined as follows:
¨ Property Package: NRTL-Ideal
¨ Components: C7, Toluene, Phenol
Leave all other parameters (i.e. - Binary Coefficients) at their defaults.
Now we will look at the interaction parameters for the three component pairs and if necessary, regress new parameters
from experimental data.
n-Heptane-Toluene Interaction Parameters
The default interaction parameters are usually reliable, although it is important to ensure that they were regressed
under conditions similar to the current design. New interaction parameters can be regressed from experimental data
specifically chosen for the system conditions. Data can be entered manually, or can be automatically scanned from the
TRC libraries of VLE, LLE and Heats of Mixing data. The TRC database contains data for over 16000 fitted binaries.
Extensive TRC data is available for the C7-Toluene pair. Open a new Fluid Phase Experiment, select the TRC Import
button, and specify the following Scan Control options:
Data Set Type — TXY
Data Set Pressure — 101.32 kPa
Data Set Temperature — 25 °C
Pressure Tolerance — 10 kPa
Temperature Tolerance — 10 °C
Search for all data sets which include the components C7 and Toluene:

5 of 90

2/24/99 10:21 AM

Application Example

http://www.hyprotech.com/support/examples/extract/extract.htm

For more information on the Herrington Consistency Test, see the HYSYS - Conceptual Design manual.
Check the Use box for each set, then select the Read Selected Data Sets button. These sets will be imported into the
current Fluid Phase Experiment. Next, check the Herrington Thermodynamic Consistency for each set by selecting the
Consistency page tab, and pressing the Calculate Consistency button. The Herrington parameters are calculated,
and the status of each data set is displayed:

6 of 90

2/24/99 10:21 AM

Application Example

http://www.hyprotech.com/support/examples/extract/extract.htm

Set 428:
Rose, A.; Williams, E. T.
Ind. Eng. Chem., 1955, 47, 1528.
P = 101 kPa
# of points = 13
Copyright (c) by the Thermodynamics Research Center
Note that Set 428 has Herrington parameters of 4.33% and 4.91% for D and J respectively, which is well under the
“consistency limit” for isobaric data (D - J £ 10%). This set has 13 points which is sufficient for our investigation.
If we were going to regress the interaction parameters to the experimental data, we would run the Optimizer. However,
we will instead compare the experimental data to the calculated data based on the default interaction parameters. On
the Summary page of the Fluid Phase experiment, highlight Set 428, then select the Edit button. Select the Calculate
button — the XY and TXY curves will be constructed based on the default interaction parameters, and the errors will be
calculated.
The calculated data in this case is the TXY or XY data calculated using the Property Package (and current
interaction parameters), which is displayed graphically on the Plots page of the Data Set view.
The TXY plot appears as follows:

The experimental and calculated points match remarkably well, and thus it is not necessary to regress the interaction
parameters for the C7-Toluene pair.
Toluene-Phenol Interaction Parameters

7 of 90

2/24/99 10:21 AM

Application Example

http://www.hyprotech.com/support/examples/extract/extract.htm

The amount of data available for this component pair is considerably less than what was available for C7-Toluene. We
will, however, regress interaction parameters from the available TRC data set (6014).
Set 6014
Drickamer, H. G.; Brown, G.; White, R. R.
Trans. Amer. Inst. Chem. Eng., 1945, 41, 555.
P = 101 kPa
# of points = 23
Herrington D% - 21.87
Herrington J% - 28.01
Copyright (c) by the Thermodynamics Research Center
Open a new Fluid Phase Experiment, and select the TRC Import button.

There is only one data set available for the Toluene-Phenol pair. Select it by checking the Use box, then choose the
Read Selected Data Sets button.
The Data Set Notes group box on the Summary page of the Fluid Phase Experiment displays important information
related to the data set. Note that this data, obtained at 101 kPa, has 23 points.

8 of 90

2/24/99 10:21 AM

Application Example

http://www.hyprotech.com/support/examples/extract/extract.htm

Move to the Consistency page, and calculate the Herrington Consistency as you did for the C7-Toluene data.
This data set is consistent according to the Herrington test:

Now we will calculate new interaction parameters based on this experimental data.
Before running the Optimizer, compare the experimental data to the predictions made using the default interaction
parameters. Edit TRC_VLE_SET_6814 and select the Calculate button. By looking at the Plots page, it appears that
there is reasonably good agreement between the experimental data and calculated curves.

On the Errors page of the Fluid Phase Experiment view, note that the average and maximum temperature errors are
0.316% and -1.038% respectively.

9 of 90

2/24/99 10:21 AM

For Matrix Pane bij (which for NRTL in HYSYS is equivalent to aij /cij). Move to the Variables page and “free“ the parameters. Thus.htm Now we will run the Optimizer in order to obtain improved interaction parameters for this data set. note that the default Objective Function is ActivityCoeff. 10 of 90 2/24/99 10:21 AM .Application Example http://www.com/support/examples/extract/extract.hyprotech. the parameters are initially locked. On the Summary page of the Fluid Phase Experiment view. the errors will be minimized with respect to the components’ activity coefficients.

Once you start the optimizer you cannot change pages until the calculations are complete. select Free. as well.com/support/examples/extract/extract. set up the view so that you can observe the solution progress. This allows the bij parameters to vary during the optimization process. This is best done from the Optimizer page. we will observe the solution progress from the Optimizer page.Application Example http://www. place the cursor on either cell containing the “Locked” message. it is probably more useful to observe the aij parameters). then select the Run Optimizer button. although you may prefer to remain on the Variables page and watch the progress of the interaction parameters (ensure that the Parameters radio button is selected.hyprotech.htm To “free” the parameters. 11 of 90 2/24/99 10:21 AM . choose the Degrees of Freedom radio button. For this example. and from the top drop down menu. Before running the optimizer. select Matrix Pane bij from the drop down menu. Choose the Optimizer tab.

but also is the most computer intensive.251% (average) and 0.934% Change the Objective Function to Maximum Likelihood.1 b12 = b21 = 0. we will use these interaction parameters for the Phenol-Toluene pair.146 Average Error = 0.. Nevertheless.313% and 1.251% Maximum Error = -0. but the improved results may be worth it.com/support/examples/extract/extract. The following data (taken from Chang.htm Convergence is achieved quickly.061% respectively.C.2 b12 = b21 = 0.4 a21 = -60.2 a21 = -188. and restart the optimizer. 1972) is used: 12 of 90 2/24/99 10:21 AM . n-Heptane-Phenol Interaction Parameters There is no TRC data for the Phenol-Heptane component pair.010 Average Error = 0.313% Maximum Error = -1. The temperature errors are now 0. Y.01. Note. The convergence time increases when we use this function. and the errors are automatically calculated once the algorithm converges. the phenol composition errors increased. the average and maximum temperature errors are 0.061% Maximum Likelihood a12 = 824. We obtain the following interaction parameters: The bij parameters are 0.hyprotech. 1957 and Kolyuchkina et al. however. Activity Coefficients a12 = 829. that while the toluene composition errors decreased..Application Example http://www.934% (maximum). The Maximum Likelihood function is the most rigorous from a statistical point of view. We may be able to get better results using a different Objective Function.

816 .929 .965 100.900 .967 99.6 .1 .5 .6 .3 .2 .956 104.946 101.931 102.349 . 1957) Pressure = 740 mm Hg TXY Data for Phenol-Heptane (Kolyuchkina et al.962 100. as shown below for the first data set (Chang): 13 of 90 2/24/99 10:21 AM .2 .283 .186 .956 100.918 112.6 .970 100.3 .961 98.Application Example http://www.0 . select the appropriate Fluid Package (C7-Phenol).4 .957 100.881 .701 .7 .8 .535 .339 .090 .4 .720 . 1972) Pressure = 760 mm Hg Open a new Fluid Phase Experiment. choose the Add button.6 ..499 .5 .932 103.120 .com/support/examples/extract/extract. and enter the data.337 .585 .840 106.941 112.635 .947 107.2 .4 .2 .968 99.961 101.4 .2 .950 102.964 98.7 .hyprotech.960 99.960 100.960 .978 99.837 .0 .528 .736 .970 TXY Data for Phenol-Heptane (Chang.112 .233 .htm x (1 = Temperature (°C) 1 Heptane) y1 (1 = Temperature (°C) x1 (1 = Heptane) y1 (1 = Heptane) Heptane) 116.

082 a21 = 701. the following schemes will be 14 of 90 2/24/99 10:21 AM .706 b12 = b21 = 0.293 Various methods are possible for regressing the interaction parameters.htm The default interaction parameters are shown here: aij bij The interaction parameters are written as follows: 1 = C7 2 = Phenol a12 = 1120.com/support/examples/extract/extract. In this example.hyprotech.Application Example http://www.

73 K=6.490 C=. With only the first data set active. For this component set.htm used: 1.169 C=.456 C=1. Check the prediction of liquid-liquid regions. The plots shown below are the TXY diagrams for Phenol-Heptane.868 K=2.149 C=-.98 C=8.471 C=2.44 C=8. we conclude that the Maximum Likelihood Objective function results in a better fit. it is wise to regard the following: 1. With only the second data set active.3 K=73. 2. Scheme 5 uses the Maximum Likelihood Objective Function.Application Example http://www.12 K=29.4 C=.7 . Consider defining a weight of zero for outliers (data points which deviate significantly from the regressed curve).82 K=6.com/support/examples/extract/extract. If a liquid-liquid region were predicted. 2. Interaction Parameters Errors Scheme a12 a21 b12 Default 701. optimize using the Activity Coefficients Objective Function.9 Scheme 3 2052 1125 . while the figure on the right plots the Chang experimental data.6 C=-26. The figure on the left plots the Kolyuchkina experimental data.6 C=-31. Therefore.068 C=. using the Maximum Likelihood Objective function rather than the Activity Coefficients Objective function resulted in significantly smaller temperature errors.540 C=2.59 K=31. With only the second data set active. the average or maximum composition decreased when the Maximum Likelihood Objective function was used (see Ave C7 and Max Phenol for Schemes 3 and 4). In all cases.192 K=.3 K=. 4.24 C=-22. Although we can be reasonably confident of these results.125 K=.7 K =. because they predict physically incorrect behaviour.581 K=1. while in most cases the composition errors increased slightly.489 C=.4 Scheme 5 1539 1328 .508 We will use the interaction parameters obtained using Scheme 5. 5.32 K=1.32 K=1.41 K=30.245 C=.329 C=. comparing the experimental data to the points calculated from the Property Package.83 C=14. optimize using the Activity Coefficients Objective Function. there is no liquid-liquid region. 3.480 C=.157 C=-.293 1120 Avg T Max T Ave C7 Max C7 Ave Phenol Max Phenol C=.87 K=-5.34 K=35. With both data sets active. optimize using the Maximum Likelihood Objective Function.509 K=.2 Scheme 1 1672 1580 . A liquid-liquid region is not predicted with our interaction parameters.2 K=58.713 C=2. 9 Scheme 4 1438 1172 . With only the first data set active.1 Scheme 2 1362 1001 .199 K=1. 15 of 90 2/24/99 10:21 AM .8 K=68.05 C=-27. The following table outlines the results of this analysis.528 C=. optimize using the Maximum Likelihood Objective Function.75 K=-4.460 K=.40 C=10.502 K=1.3 K=65.hyprotech. In some instances. optimize using the Objective Function which results in the smallest error. then the Property Package and/or interaction parameters would be unacceptable.

htm We can check the prediction of liquid-liquid regions from the Binary Coefficients page of the appropriate Fluid Package view (ensure that you have entered the interaction parameters as shown below): aij bij 16 of 90 2/24/99 10:21 AM .Application Example http://www.hyprotech.com/support/examples/extract/extract.

As we did in the previous section. The TXY plot (using this interaction parameter) is shown below. Toluene-Phenol Interaction Parameters Recall that only one TRC data set is available for this binary (Data Set #6814).com/support/examples/extract/extract. transfer the three components to the Selected Components group. Open the Fluid Package Manager and add a new Fluid Package: Property Package: PR Components: C7. and is concisely summarized below. Phenol Leave all other parameters at their defaults. and enter a temperature and pressure. Peng Robinson Interaction Parameters Default interaction parameters are available only for the C7-Toluene pair (0. binaries are constructed such that aij = aji.014. Toluene. Select the Ternary plot radio button.006.htm You can see the LLE ternary plot on the Binary Coeffs page of the Fluid Package view. Unlike the NRTL interaction parameters. n-Heptane-Toluene Interaction Parameters We will use the default interaction parameter aij = aji = 0.hyprotech. only one PR interaction parameter matrix pane is available.Application Example http://www. as well. Note that you may have to decrease the tolerance or step size in order to obtain adequate convergence in this order of magnitude. we will determine the Interaction Parameters based on TRC and literature experimental data. You can see the VLLE ternary plot on the Setup page of the Ternary Distillation Experiment view. no liquid-liquid region is predicted. This requires you to enter a temperature and a pressure. displaying a reasonably good fit. The procedure is essentially the same. we obtain an interaction parameter of aij = aji = 0. Here. Over a range of temperature and pressures. When we use the Bubble Temperature or Maximum Likelihood Objective Functions. 17 of 90 2/24/99 10:21 AM .006). The Activity Coefficient Objective Function should not generally be used for Equations of State as results tend to be mediocre for highly polar systems. you only enter a pressure.

hyprotech. and Kolyuchkina et al.Application Example http://www.010.03) are below: 18 of 90 2/24/99 10:21 AM . The interaction parameters predicted using the Chang data is very different from the Kolyuchkina data. we will use the data of Chang and Kolyuchkina et al. Chang predicts aij = 0.045 (Maximum Likelihood). predicts aij = 0.03. we obtain aij = 0.com/support/examples/extract/extract.htm n-Heptane-Phenol Interaction Parameters As before. The TXY plots (using an interaction parameter of 0. When we combine both data sets.

and they also indicate a liquid-liquid region. The VLLE plot at a pressure of 18 psia is shown below. Here.htm The figure on the left plots the Chang data and the figure on the right plots the Kolyuchkina data. This requires you to enter a temperature and a pressure. and we conclude that the Peng-Robinson Property Package is not acceptable for this example. You can see the ternary LLE plot on the Binary Coeffs page of the Fluid Package view. Note that using the PRSV Property Package results in a better fit. 19 of 90 2/24/99 10:21 AM .007 or less. you only enter a pressure. This can be confirmed by looking at the ternary LLE or VLLE plot. We can avoid the prediction of a liquid-liquid region by setting the n-Heptane-Phenol interaction parameter to 0. You can see the VLLE plot on the Setup page of the Ternary Distillation Experiment view. However. although a two-liquid-phase region is incorrectly predicted under certain conditions. These plots show that the dew point curve does not match the experimental data very well.hyprotech.Application Example http://www. the calculated curve still does not fit the experimental data very well.com/support/examples/extract/extract.

Parameters used in this Example For this example. We will use the NRTL Property Package. and the Interaction Parameters as defined in the previous section. For the ternary distillation experiment. so that incorrect predictions/azeotrope formation will not be a problem. we will use the NRTL Property Package with the following interaction parameters: aij bij C-6. condenser. a sidestream.com/support/examples/extract/extract. as shown in the following table: Pressure (psia) Azeotropic Composition 23 No azeotropes 24 C7=0.9923 40 C7=0.Conceptual Design allows for single-column design. reboiler and decanter. at higher pressures.9993 30 C7=0.hyprotech. The point is that we should not allow the pressure to fluctuate excessively. the column can have two feeds. NRTL does not predict any azeotropes.9825 It is important to remember that activity models generally do not extrapolate well with respect to pressure. an azeotrope between n-Heptane and Phenol is predicted.Application Example http://www. 20 of 90 2/24/99 10:21 AM . However.4 Ternary Distillation Design (NRTL) First Column HYSYS .htm Prediction of Azeotropes using NRTL At 18 psia. so we should therefore regard these results with caution.

and the recycle stream from the second column. as predicted by the NRTL Property Package (using our new interaction parameters).hyprotech. 16 and 20 psia). 50% n-Heptane on a molar basis).htm A trial-and-error type of procedure is required.Application Example http://www. There is no decanter or sidestream. The first column (extractive distillation) has two feeds to it. and select the appropriate Fluid Package from the drop down list. the process feed (50% Toluene. Open a Ternary Distillation Experiment. The Configuration/Summary page will appear as follows: 21 of 90 2/24/99 10:21 AM . as we must cycle between the two columns until the connecting streams have roughly the same compositions and flowrates. and the bottoms stream of the second column is the upper feed to the first column.com/support/examples/extract/extract. The bottoms stream of the first column feeds the second column. set a pressure of 18 psia (the average of the top and bottom pressures in the column. The program will then determine if there are any azeotropes or two-liquid regions: There are no azeotropes or liquid-liquid regions at this pressure.

For the remaining streams.hyprotech.9990 respectively.Application Example http://www. We know that the upper feed is primarily phenol.75.htm Note that we have entered the specifications for the process feed stream (Lower Feed). Before entering the specifications. This restricts our range of choice for the remaining specifications. You will see the following: 22 of 90 2/24/99 10:21 AM . Select the Bottoms radio button. Later. we do not know if this is the optimum Upper Feed/Lower Feed ratio. the Upper Feed flowrate is 1100 lbmole/hr. we will do a sensitivity analysis in order to estimate an optimum Reflux Ratio. set the Reflux Ratio to be 5. As an initial estimate. The molar flow of the process feed stream is 400 lbmole/hr.990. we will enter the specifications on the Spec Entry page. specify a Distillate C7 mole fraction of 0. With an Upper Feed/Lower Feed ratio of 2. we will use the following specifications: We have specified the C7 and Phenol mole fractions to be 1E-06 and 0.com/support/examples/extract/extract. Note that at this point. Next.

hyprotech. set the reflux ratio to be 5. the number of stages will be determined using this optimum value. all that is left is to specify a reflux ratio. This constrains the C7 mole fraction from 0 to 0. Specify the C7 fraction to be 0. You will see the following message: The optimum value for Omega (that which results in the lowest number of total stages) is automatically calculated. Set the Phenol fraction to be 0. if you simply press the Calculate button again. Select the Calculate button.846. Alternatively. you could set Omega to any value you wanted on the 2 Feed Omega page. The remaining mole fractions will be calculated based on the overall mole balance.htm It would be advantageous to maximize the phenol in the bottoms stream.0006.Application Example http://www. We will always use the optimum value in this example.0007. After you select the Calculate button. At this point. As an initial estimate.com/support/examples/extract/extract. move to the Flows / Stages page: 23 of 90 2/24/99 10:21 AM .

20 stages would be required to achieve these bottom compositions. if we respecify the bottoms composition so that the phenol fraction is lower. If we were to specify the heptane and phenol compositions to be 0.96.844. We could specify a lower heptane fraction in the bottoms — if we define it to be 0. which means that if most of the toluene and heptane were to exit in the distillate stream of the second column.htm The total number of stages is excessively high. Also. we will require less stages in this column.hyprotech. the column profiles will not converge. Specify the heptane and phenol compositions to be 0. the heptane fraction in the bottoms coming off the first column must be small. This is not adequate.84 respectively. As well. The Heptane to Toluene ratio is now 103.0065) = 0.0015 and 0. Note that as we decrease the phenol composition.1535. below a certain point (phenol composition » 0. The toluene composition would be (1 .0. we must increase the C7 composition. The results are shown here: 24 of 90 2/24/99 10:21 AM . the best purity we could obtain would be about 0. Because we want to take relatively pure toluene off the top of the second column and relatively pure phenol off the bottom.84 . which should allow the Toluene fraction off the top of the second tower to be about 0.836).Application Example http://www.0001 for instance. the heptane composition must be even lower. 29 stages are required.0. therefore.0065 and 0.99.com/support/examples/extract/extract. and the toluene to heptane ratio about 24. Note that most of the toluene and heptane in the bottoms stream will exit in the distillate stream of the second column.

set the pressure to 18 psia. No Sidestream). Second Column As before.com/support/examples/extract/extract.hyprotech. 25 of 90 2/24/99 10:22 AM .htm We will now create a new ternary distillation experiment. transferring the bottoms specifications for the first column to the feed for the second column.Application Example http://www. and select the appropriate Fluid Package. The Feed specifications. No Decanter. Leave the settings on the Configuration/Settings page at their defaults (Single Feed. are shown here: At this point. it may take some experimentation to see what stream specifications will result in a converged column. taken from the bottoms stream off the first column. The Reflux Ratio for the second column will initially be set at 5.

0045 0.999). First Column: Second Pass The recycle stream flow (upper feed) is now 1106.0055 0.0015 Toluene 0. At this step.5000 1e-6 0.766. we must return to the first column. 6 stages are required for the second column (where the sixth “stage” is the reboiler).99 in the distillate is high (about 29). the phenol composition off the bottoms had to be adjusted to 0.0000 0.Application Example http://www.9950 0.99 0.com/support/examples/extract/extract. As well.0050 0. the minimum number of stages required to obtain a heptane composition of 0. The Flows/Stages page is 26 of 90 2/24/99 10:22 AM . We now have the following composition specifications: Component Mixed Feed Solvent Feed Distillate Bottoms C7 0.htm The following specifications work for the distillate and bottoms: Note that we should be able to obtain a higher toluene purity.5000 0. At this point. all other component fractions are calculated.8420 Component fractions in boldface are specified.1565 Phenol 0. which gives us a feed ratio of 2. we must use the Bottoms specifications obtained here for the Top Feed of the first column. using the new recycle stream specs. the feed enters on the fourth stage. Thus we will have to increase the phenol composition to compensate. If we keep the recycle compositions as they are.99 (initially we had set the phenol composition in the recycle to 0.hyprotech. In other words.3.

0003 0.9900 0. and the feed to the second column has the same composition as the bottoms stream of the first column.8420 0. the solvent feed stream to the first column has the same composition as the bottoms stream of the second column.0015 0.0097 1e-6 Toluene 0.htm shown below: Second Column: Second Pass At this point.com/support/examples/extract/extract. The specifications are shown below: Component Feed Distillate Bottoms C7 0.hyprotech.9950 The Flows/Stages page is shown below: 27 of 90 2/24/99 10:22 AM .1565 0.Application Example http://www.0050 Phenol 0.

at the expense of a higher number of stages and/or higher Reflux/Reboil ratios.com/support/examples/extract/extract. the results obtained here will not exactly match those determined in HYSYS.985/0.hyprotech.99 / 0. Using these results as a base case. one which has lower purities (0. Two configurations will be proposed. Reboil Ratio and Purities We will adjust the Reflux Ratio and Purities. there are inherent simplifications. such as the assumption of constant molal overflow. and one with higher purities (0.985). Thus. the reflux ratio and product purities are now adjusted in order to determine an optimum configuration.SteadyState. observing their effect on other column variables such as the Reboil Ratio and the Number of Stages.99) The Base Case constants and variables are tabulated below: 28 of 90 2/24/99 10:22 AM .Application Example http://www. Effect of Reflux Ratio.99/0. As well. Higher purities (0. but this is acceptable considering that this is a preliminary approximation.htm Note that the flows between the columns do not match precisely.99).

Column 1 5 Reboil Ratio.7657 0.991 0. Column 1 21 Number of Stages.9167 0.99 Reflux Ratio (Reboil Ratio) Column 1 Keeping other variables constant.5 Heptane / 0 Phenol Variable Reflux Ratio. Column 2 Distillate 0.990. Column 2 8 Number of Stages. As shown in the table below.99 Reflux Ratio.5 Toluene / 0.985 0. Heptane Fraction 0. the number of stages increases substantially as we increase the Heptane Fraction above 0.989 0. Column 1 Distillate 0.6846 3. We therefore conclude that a reflux ratio of 5 is optimum for the first column.0906 Feed Stage.995 Phenol Lower Feed Composition 0.9242 0. Column 2 1. Decreasing the reflux ratio below five causes the number of stages to increase.990 0. We will go with a Heptane fraction of 0.9188 Upper Feed Stage.994 Upper Feed Stage 7 11 13 16 20 46 Lower Feed Stage 15 19 21 24 28 54 Number of Stages 19 24 25 28 32 58 Reflux Ratio (spec.htm Constant Recycle Composition 1e-6 Toluene / 0. Column 1 13 Lower Feed Stage.990.6126 0.) 5 5 5 5 5 5 Reboil Ratio 29 of 90 0.9199 0.994. the number of stages is 58 which is much too high to be viable. Column 2 5 Reboil Ratio.9188 1.9146 2/24/99 10:22 AM . At a Heptane fraction of 0. the number of stages also increases. Column 1 25 Heptane Fraction. at the expense of some extra stages.Application Example http://www. increasing the reflux ratio above 5 gives no improvement in the number of stages required for the separation.9178 0.992 0.com/support/examples/extract/extract. Reflux Ratio 4 5 10 20 Upper Feed Stage 18 15 13 11 10 Lower Feed Stage 26 24 21 20 20 Number of Stages 30 28 25 24 25 Reboil Ratio 3 0. Although we would like a high purity.2160 Heptane Fraction Column 1 As we increase the Heptane fraction in the distillate.005 Heptane/ 0. Column 2 10 Toluene Fraction.9188 0. Column 1 0.hyprotech. the reflux ratio is adjusted.

we cannot predict toluene fractions above 0. A reflux ratio of 5 is selected as the optimum.htm Reflux Ratio (Reboil Ratio) Column 2 The number of stages in the second column is only somewhat sensitive to the reflux ratio.9088 1.985 0. Toluene Fraction 0.hyprotech. as shown below. while increasing the ratio to 10 reduces the number of stages by one.989 0.0919 1. 2. Results Using Optimized Values These are the specs for the first (high purity) column configuration: 30 of 90 2/24/99 10:22 AM .0906 Upper/Lower Feed Ratio Finally.5 2.990 Feed Stage 6 7 8 Number of Stages 8 9 10 Reflux Ratio (spec.8171 Toluene Fraction Column 1 With this configuration. the Upper/Lower Feed ratio was varied. We will keep the toluene fraction of 0.com/support/examples/extract/extract.7 3 4 Upper Feed Stage 23 13 13 11 10 Lower Feed Stage 30 22 21 20 20 Number of Stages 35 26 25 24 23 It appears that the U/L ratio that we used. and the effect on the number of stages in the first column observed: U/L Ratio 2 2.9994 3.Application Example http://www.) 5 5 5 Reboil Ratio 1.0906 1. Reflux Ratio 4 5 10 20 Feed Stage 10 8 8 8 10 9 9 Number of Stages 12 Reboil Ratio 0.99. is reasonable in this case. even though more stages are required. Any increase in the ratio does not decrease the number of stages significantly.0972 1.7.990. Decreasing the ratio to 4 is done at a cost of two extra stages.

Column 1 25 Heptane Fraction. Column 2 8 Number of Stages. 31 of 90 2/24/99 10:22 AM . Column 2 1.99 Reflux Ratio. Column 1 5 Reboil Ratio. Column 1 13 Lower Feed Stage. Note that there are feed streams at stages 13 and 21.com/support/examples/extract/extract.0906 Feed Stage. Column 2 5 Reboil Ratio.9188 Upper Feed Stage. Column 1 0. Column 1 Distillate 0.99 The temperature and liquid composition profiles for the first column are displayed below. Column 2 10 Toluene Fraction.htm Reflux Ratio. Column 1 21 Number of Stages.Application Example http://www. Column 2 Distillate 0.hyprotech.

985 / 0.5 Toluene / 0. we will start with a Reflux Ratio of 5 for both columns.75. as well. The Base Case constants and variables are tabulated below: Constant Recycle Composition 1e-6 Toluene / 0. we require a smaller phenol fraction in the Recycle (0.hyprotech.Application Example http://www.993 Phenol Lower Feed Composition 0.985) With lower purities.993).007 Heptane/ 0.5 Heptane / 0 Phenol 32 of 90 2/24/99 10:22 AM .htm The temperature and liquid/vapour composition profiles for the second column are shown below: Lower purities (0. As with the high purity case. we have set the upper/lower feed ratio to 2.com/support/examples/extract/extract.

Column 1 20 Heptane Fraction.htm Variable Reflux Ratio.985 Reflux Ratio. We therefore conclude that a reflux ratio of 5 is optimum for the first column. Reflux Ratio 3 4 5 10 20 Feed Stage 9 8 8 7 8 Number of Stages 11 10 10 9 9 Results Using Optimized Values These are the specs for the second (lower purity) column configuration: 33 of 90 2/24/99 10:22 AM .9231 1. we adjust the reflux ratio. Column 1 Distillate 0. Decreasing the reflux ratio below five causes the number of stages to increase. Column 2 10 Toluene Fraction. Column 1 5 Reboil Ratio. Column 2 8 Number of Stages.9231 Upper Feed Stage. When we increase the reflux ratio above 5.985 for both columns. Column 1 7 Lower Feed Stage. Column 1 17 Number of Stages. A reflux ratio of 4 is selected as the optimum.985 We will optimize only the reflux ratios.hyprotech.Application Example http://www. Column 2 5 Reboil Ratio. leaving the purities at 0. Column 2 1.6924 3. Reflux Ratio 4 5 10 20 Upper Feed Stage 8 7 7 6 Lower Feed Stage 19 17 17 19 Number of Stages 23 20 21 20 Reboil Ratio 0. Reflux Ratio (Reboil Ratio) Column 1 As before. Column 2 Distillate 0.7693 0.1015 Feed Stage.com/support/examples/extract/extract. as shown below.2310 Reflux Ratio (Reboil Ratio) Column 2 The number of stages in the second column is only somewhat sensitive to the reflux ratio. there is no improvement in the number of stages required for the separation. and observe the effect on the number of stages. Column 1 0.

Column 2 10 Toluene Fraction. Column 1 0.9187 Feed Stage.985 Reflux Ratio. Column 2 0. Column 1 7 Lower Feed Stage. Column 1 17 Number of Stages.Application Example http://www.htm Reflux Ratio. Column 1 5 Reboil Ratio.985 The Stage Liquid Compositions are shown here for the first column: Also.9231 Upper Feed Stage. the Liquid composition profiles for the second column are shown: 34 of 90 2/24/99 10:22 AM . Column 2 Distillate 0. Column 1 Distillate 0.hyprotech.com/support/examples/extract/extract. Column 2 8 Number of Stages. Column 1 20 Heptane Fraction. Column 2 4 Reboil Ratio.

com/support/examples/extract/extract. the b term is the alpha term. we will construct the columns in HYSYS.Conceptual De-sign.Conceptual Design to HYSYS. We require the phenol stream to “make up” for phenol lost in the toluene and heptane product streams. the c term is the alpha term.SteadyState. Toluene.Conceptual Design Fluid Package.SteadyStateor HYSIM using the Export to HYSIM button in the HYSYS .SteadyState.hyprotech.dat and . Phenol In HYSYS. In HYSYS . Define the Fluid Package as follows: Property Package — NRTL Components — n-Heptane. Change the Interaction Parameters to match the regressed parameters obtained in Part 1 (or copy the . In the Main Environment WorkSheet.htm C-6.Conceptual Design to the Support directory).5 Building the Columns in HYSYS In this section. Note that the interaction parameters for the NRTL package can be exported from HYSYS . specify the Feed and phenol makeup streams as follows: 35 of 90 2/24/99 10:22 AM .Application Example http://www.idx files which you created in HYSYS . and obtain a steady-state solution for both column configurations.

The Feed locations remain the same.0000 Molar Flow [lbmole/hr] 400. as predicted in part 2. not 25 and 10. In the SubFlowsheet.Conceptual Design. the number of trays includes the reboiler. Reboilers and Condensers for the high purity setup: 36 of 90 2/24/99 10:22 AM .0000 High Purity Configuration In HYSYS .0000 Comp Mole Frac [Phenol] 0.SteadyState.htm Name Feed phenol makeup Vapour Frac 0. In HYSYS.5788e+07 -60190.com/support/examples/extract/extract.hyprotech.3151 Heat Flow [Btu/hr] -1.0000 220.0000 0.0000 Comp Mole Frac [Toluene] 0.0000 20.0000 1.0000 Temperature [F] 220.2000 Mass Flow [lb/hr] 38469.0000 1.1128 Liq Vol Flow [barrel/day] 3448. Therefore the number of trays in each column are 24 and 9. add the Tray Sections.1028 Comp Mole Frac [n-Heptane] 0.1605 94.5000 0.9080 6. the number of trays does not include the reboiler.5000 0.0000 Pressure [psia] 20.Application Example http://www.

htm TS-1 CONNECTIONS TRAY SECTION 24 TS-2 CONNECTIONS Number of Trays Feed (21) Number of Trays 9 Feeds (Stage) Solvent (13) Liquid Inlet Phenol Makeup (13) Liquid Inlet Reflux-2 Vapour Inlet Reflux-1 Vapour Inlet Boilup-2 Liquid Outlet Boilup-1 Liquid Outlet To Reboiler-2 Vapour Outlet To Reboiler 1 Vapour Outlet To Condenser-2 To Condenser-1 PARAMETERS Feeds (Stage) COL1 Bottoms (8) PARAMETERS Tray Section Type Standard Tray Section Type Standard TOTAL CONDENSER Condenser-1 TOTAL CONDENSER Condenser-2 CONNECTIONS CONNECTIONS Feed To Condenser-1 Feed To Condenser-2 Distillate Heptane Distillate Toluene Reflux Reflux-1 Reflux Reflux-2 Energy COL1 Cond Q Energy COL2 Cond Q PARAMETERS Pressure Drop REBOILER PARAMETERS 0 psi Reboiler-1 CONNECTIONS Pressure Drop REBOILER 0 psi Reboiler-2 CONNECTIONS Feed To Reboiler-1 Feed To Reboiler-2 Boilup Boilup-1 Boilup Boilup-2 Bottoms Product COL1 Bottoms Bottoms Product Solvent Energy Energy COL1 Reb Q PARAMETERS Pressure Drop COL2 Reb Q PARAMETERS 0 psi Pressure Drop 0 psi The PFD will appear as follows: 37 of 90 2/24/99 10:22 AM .Application Example TRAY SECTION http://www.com/support/examples/extract/extract.hyprotech.

hyprotech. bring up the Column view.htm Return to the Main Flowsheet. 38 of 90 2/24/99 10:22 AM . and enter the following specifications: Pressures Condenser-2 — 16 psia Reboiler-2 — 20 psia Condenser-1 — 16 psia Reboiler-1 — 20 psia Temperature Estimates Temperature Estimate Condenser-1 — 220 F Temperature Estimate Condenser-2 — 240 F Solving the Column Next.com/support/examples/extract/extract. we will add the specifications.Application Example http://www.

then select a specification from the list: As an example. 5) Heptane Frac = 0.hyprotech. the Reflux Ratio spec is shown here: We want to solve to the following specifications: Reflux Ratio = 5 (Reflux Ratio-1. Liquid. 5) Reflux Ratio2 = 5 (Reflux Ratio-2.99 (Heptane Frac.com/support/examples/extract/extract.99. Heptane) Toluene Frac = 0. Liquid.99 (Toluene Frac. Molar. Condenser-2. 0. Condenser-2. Condenser-1. Mole Fraction. There are several alternative methods you can use to obtain a solution. select the Add button on the Specs Page. the following two methods may work: 39 of 90 2/24/99 10:22 AM .Application Example http://www.99. Mole Fraction. 0.htm To add a specification. Toluene) It may not be possible to immediately solve to these specifications. Molar. Condenser-1.

In any case. once you have converged. the Toluene Frac spec and the Toluene Flow spec. 200 lbmole/hr) Activate the Reflux Ratio specs. Once it solves. Add the following Toluene Recovery Spec: Toluene Recovery = 0. Run the column. Once it solves. 0. 2. Re-run the column (do not reset). You may have to “approach” a spec by choosing a conservative value for the specification. Toluene. replace the Toluene Flow spec with the Heptane Frac spec. replace the Toluene Recovery spec with the Toluene Frac spec. Re-run the column. Add the following Flow Spec: Toluene Flow = 200 lbmole/hr (Toluene Flow. Toluene.Application Example http://www.htm 1.995) Activate the Reflux Ratio specs. Run (but do not Reset) the column after each change. Whether a certain set of specifications will solve depends in part on the solution history. even if you have Reset the solution. then successively approaching the actual specification.hyprotech.995 (Toluene Recovery. Molar.com/support/examples/extract/extract. you will see a view similar to the following: 40 of 90 2/24/99 10:22 AM . Molar. the Heptane Frac spec and the Toluene Recovery spec. Run the column.

This is an improvement over the specifications estimated using HYSYS .32e+07 Btu/hr Column 2 Reboiler — 2. We obtain the following Condenser and Reboiler duties: Column 1 Condenser — 1.Application Example http://www.993 (Toluene and Heptane. you may want to save the first configuration in a separate file.Conceptual Design. Reboilers and Condensers.hyprotech.Conceptual Design.64e+07 Btu/hr Column 2 Condenser — 1. Low Purity Configuration Rather than reinstalling the Tray Sections.htm Note the similarities in the temperature profile shown here with the profiles obtained using HYSYS . respectively). At this point.com/support/examples/extract/extract.04e+07 Btu/hr Note that we can further increase the distillate compositions to 0.994 and 0.72e+07 Btu/hr Column 1 Reboiler — 1. simply adjust the number of stages and feed locations as follows: 41 of 90 2/24/99 10:22 AM .

Solving the Column We want to solve to the following specifications: Reflux Ratio = 5 (Reflux Ratio-1. then bring up the column view.985 (Heptane Frac.htm Ensure that you are in the Main Flowsheet.com/support/examples/extract/extract. Condenser-1.985.985. it is not possible to meet the specifications predicted by HYSYS . 0. Toluene) In this case.985. 5) Heptane Frac = 0.Conceptual Design. The following 42 of 90 2/24/99 10:22 AM . Mole Fraction. 0. Liquid. Liquid. Simply change the Heptane and Toluene Fracs to 0.985 (Toluene Frac. Heptane) Toluene Frac = 0. Mole Fraction. therefore it is not necessary to add new specs. Molar.Application Example http://www. The pressures and temperature estimates will be defined as before: Pressures Condenser-2 — 16 psia Reboiler-2 — 20 psia Condenser-1 — 16 psia Reboiler-1 — 20 psia Temperature Estimates Temperature Estimate Condenser-1 — 220 F Temperature Estimate Condenser-2 — 240 F The types of specifications are the same as before. Condenser-2. Molar.hyprotech. Condenser-1. Condenser-2. 5) Reflux Ratio2 = 5 (Reflux Ratio-2.

the solved column using these specifications appears as follows: 43 of 90 2/24/99 10:22 AM . The first configuration is more desirable despite the high Reflux Ratio.985 Toluene Fraction 0.970 Toluene Fraction = 0.hyprotech.Application Example http://www.985 Reflux Ratio Column 1 = 5 Reflux Ratio Column 2 = 4 The first configuration maintains the purity specs.com/support/examples/extract/extract.985 Reflux Ratio Column 1 = 12 Reflux Ratio Column 2 = 4 or Heptane Fraction = 0.htm configurations for the low purity case are possible: Heptane Fraction 0. while the second maintains the reflux ratio specs.

Material and Utility costs that are used in this simulation are shown below: Economic Cost Index 1996 to 1990 1.Application Example http://www.07 44 of 90 Tax Rate 28% Interest minus Inflation 7% Working Days/Year 300 2/24/99 10:22 AM . This section is divided into the following parts: Raw Data — Data which is used in the calculation of capital costs. Net Present Worth — Economic calculation taking into account the Capital Cost and Gross Income. used to obtain the net present worth.SteadyState to calculate the economics of the process.72e+07 Btu/hr Column 1 Reboiler — 3. the Condenser and Reboiler duties are much higher. but are sufficient to provide a preliminary estimate.htm Although less stages are required for this configuration. incurred at the end of each year.92e+07 Btu/hr C-6.6 Optimization Economics: Background A Spreadsheet will now be set up in HYSYS. and as they are formula-based. can be used in the optimization calculations. This will be confirmed in the next section. The benefit of these methods is that they are easy to implement.hyprotech. Nomenclature and Constants — A list of the nomenclature and constants used in the various expressions in this section. incurred at the end of each year. Raw Data Some of the Economic. The methods used here to calculate capital costs. expenses. expenses and revenue are relatively simple. revenue and net present worth.com/support/examples/extract/extract. Capital Cost — Initial equipment and related costs associated with the construction of the process. and it is unlikely that the reduced capital cost will compensate for the increased utility cost.54e+07 Btu/hr Column 2 Condenser — 1. namely Toluene and Heptane. Annual Expenses — Expenses associated with the operation of the plant. We obtain the following Condenser and Reboiler duties: Column 1 Condenser — 3.34e+07 Btu/hr Column 2 Reboiler — 1. Revenue — Income obtained from the sale of the process products. incurred at time zero.

Column 1 Number of Trays.hyprotech. and will be imported into the Spreadsheet. Column 2 Capital Cost A percentage of delivered-equipment cost method is used to determine the total capital investment. the equipment (column tray sections. That is. Product and Utility Costs Toluene ($/gal) 0. such as piping. The Reboiler and Condenser expressions are regressed 45 of 90 2/24/99 10:22 AM .58 Phenol ($/lb) 0.25 Natural Gas ($/1E6 Btu) 3. all additional costs.Application Example http://www.76 Heptane ($ /gal) 0.20 In addition to these.74 Feedstock ($/gal) 0. Equipment Cost Equipment Cost Reference First Column Condenser 1-373 Second Column Condenser 1-373 First Column Reboiler 1-373 Second Column Reboiler 1-373 First Column Tray Section 1-712 Second Column Tray Section 1-712 All expressions here are derived from a graph or table. the following variables are required from the Steady-State solution.41 Water ($/1000 gal) 0.htm Raw Material. reboilers and condensers) are sized and priced.com/support/examples/extract/extract. construction and so on are calculated as a percentage of the equipment cost. First Condenser Duty Second Condenser Duty First Reboiler Duty Second Reboiler Duty Phenol Mass Flow Toluene Mass Flow Heptane Mass Flow Feed Flow TS 1 Liquid Mass Flow TS 2 Liquid Mass Flow Feed Standard Density Toluene Standard Density Heptane Standard Density Number of Trays.

Direct. Contracting is estimated to be 5% of the sum of all Equipment. Direct and Indirect Costs.10 1 (174) New Building 0. Item Factor Reference Installation 0. Direct and Indirect Costs.40 Instrumentation 0. Service and Land.30 Contracting and Contingency These are applied based on the Equipment. Electrical. New Building.10 1 (182) The total Fixed Capital Investment (FCI) is the sum of all Equipment.07.htm linearly. Indirect. Piping.33 1 (182) Construction 1 (182) 0. while the Tray Section expression assumes a linear relationship on a log-log scale. 1 (183) Piping 0. Direct costs include Installation. multiplied by the cost index factor of 1. Instrumentation.18 1 (171) 1 (172). Direct and Indirect Costs The costs of each item below is estimated as the Equipment cost multiplied by the respective Factor for that item. Working Capital The working capital is estimated to be 15% of the Fixed Capital Investment.20 1 (175) Yard 0. Item Factor Reference Contracting 0.05 1 (182) Contingency 0.com/support/examples/extract/extract.Application Example http://www. 46 of 90 2/24/99 10:22 AM . Contracting and Contingency costs.hyprotech.60 1 (173) Electrical 0.70 1 (182) Land 0.10 1 (182) Service 0. The sum of the costs of these six items is the Equipment cost (based on 1990 prices).06 1 (182) Engineering / Supervision 0. Yard. and Contingency (unforeseen events) is estimated as 10% of the sum of these costs. Annual Expenses The following table lists the expenses which are considered in the economic analysis of this plant. Indirect costs include Engineering /Supervision and Construction.

Application Example Expense http://www. and multiplied by a cost index factor (1.hyprotech.07) to account for 1990 to 1996 inflation.com/support/examples/extract/extract. Cost of Phenol 1 (816) Cost of Feedstock 1 (197) Labour* 1 (198) Supervision and Clerical Labour Cost x 15% 1 (202) Maintenance and Repairs FCI x 6% 1 (203) Operating Supplies Maintenance x 15% 1 (204) Lab Charges Labour Cost x 15% 1 (204) Condenser 1 Cooling Water 1 (815) Condenser 2 Cooling Water 1 (815) Reboiler 1 Natural Gas 1 (815) Reboiler 2 Natural Gas 1 (815) Depreciation FCI x 10% 1 (205) Local Taxes FCI x 2% 1 (205) Insurance FCI x 1% 1 (205) Plant Overhead (Labour Cost + Supervision and Clerical Cost + Maintenance and Repairs Cost) x 60% 1 (205) Administrative Labour Cost x 20% 1 (206) Distribution Gross Income x 4% 1 (207) Research and Development Gross Income x 4% 1 (207) * Linearly Regressed from Graph The individual expenses are totalled.htm Annual Cost Reference 1 (197). Revenue 47 of 90 2/24/99 10:22 AM .

It is assumed that there is no salvage value. The Income after tax is the Annual Operating Income multiplied by one minus the tax rate.htm Revenue Annual Cost Reference Toluene 1 (816) Heptane 1 (816) The total gross revenue is the sum of the amount obtained from selling the products. from years one to five. Nomenclature and Constants used in Economic Analysis 48 of 90 2/24/99 10:22 AM . The revenue and expenses are applied at the end of each year.hyprotech. multiplied by the cost index factor (1. The Net Present Worth of the Annual Cash Income is determined using the following formula: This expression simplifies to: The Total Net Present Worth is the total capital investment (negative cash flow) plus the Net Present Worth of the Annual Cash Income.07). This expenditure is the total cash flow for year zero. Calculation of Net Present Worth The following points outline the simplified calculation for net present worth: The total capital investment is the Fixed Capital Investment plus the Working Capital. It is assumed that the life of the process is five years.com/support/examples/extract/extract.Application Example http://www. the Annual Cash Income is exactly the same for years one to five. which was earlier discounted as an annual expense. The Annual Operating Income is the Annual Income minus the Annual Costs. The Annual Cash Income is the Income after tax plus the Depreciation Expense.

STD = Standard Density of Heptane (lb/ft3) rT.Application Example http://www.STD = Standard Density of Toluene (lb/ft3) CF = Cost of Feedstock ($/gal) CGAS = Cost of natural gas ($/1E6 Btu) CH2O = Cost of water ($/1000 gal) CHEP = Cost of Heptane ($/gal) CPH = Cost of Phenol ($/lb) CpH2O = Heat Capacity of water (1 Btu/lb F) CTOL = Cost of Toluene ($/gal) DTRAY1 = Diameter of First Tray Section (ft) DTRAY2 = Diameter of Second Tray Section (ft) FFOL = Feed Flow (lb/hr) FPH = Phenol Flow (lb/hr) FTOL = Toluene Flow (lb/hr) FTRAY1.STD = Standard Density of Feedstock (lb/ft3) rH.hyprotech. interest minus inflation) N = Number of working days/yr NTRAY1 = Number of Trays in First Tray Section NTRAY2 = Number of Trays in Second Tray Section n = Life of project (y) QCOND1 = Duty. Column 1 Reboiler (Btu/hr) QREB2 = Duty. from Stage 20 (lb/h) FTRAY2.com/support/examples/extract/extract. Column 2 Reboiler (Btu/hr) xH = Mole Fraction of Heptane in Heptane Product xT = Mole Fraction of Toluene in Toluene Product 49 of 90 2/24/99 10:22 AM . = Liquid Mass Flowrate for First Tray Section.htm Nomenclature rF. = Liquid Mass Flowrate for Second Tray Section. Stage 1 (lb/h) i = Annual rate of interest (in this case. Column 1 Condenser (Btu/hr) QCOND2 = Duty. Column 2 Condenser (Btu/hr) QREB1 = Duty.

it is a good idea to set it up as a template file.2”) 1/120 hr (Residence time on tray . Create a new template and enter Fluid Package data. and add a subflowsheet.4 lb/ft3 (Density of water) 1000 No units (Cost = $/1000 gal) 1E6 No units (Cost = $/1E6 Btu) 0.25 No units (Portion of formula for area) 1/6 ft (Height of weir .219 Constant in Labour Cost Expression 2.1/2 minute) Setting up the Spreadsheet Note that the Spreadsheet we are constructing contains some information which is not used in this example.htm Constant Used in Expression Unit 24 hours/day 7. but which may be of generic use. Save the template. Simulation Data Economic Data Capital Cost Data Expense and Revenue Data Capital Cost Calculation Expense and Revenue Calculation Net Present Worth Calculation 3.Application Example http://www. 2. 5. but which may be of generic use. using the previously created file as a template. Retrieve the process case.844 Constant in Labour Cost Expression 70 F (DT Cooling Water) 62. These are the steps: 1.481 gal/ft3 2000 lb/ton 0.com/support/examples/extract/extract. Import data links into Spreadsheet. but which may be of generic use. 50 of 90 2/24/99 10:22 AM . Note that the Spreadsheet we are constructing contains some information which is not used in this example. Add a Spreadsheet and enter the following information: Note that the Spreadsheet we are constructing contains some information which is not used in this example. then insert it as subflowsheet in the appropriate case.hyprotech. If you want to use this Spreadsheet as a template for other processes. 4.

we will then import the appropriate variables into this Spreadsheet. Select NRTL with components n-Heptane.htm Creating a New Template From the file menu. Enter the headings as shown below. Adding the Spreadsheet Simulation Data Column A lists the headings.com/support/examples/extract/extract. 51 of 90 2/24/99 10:22 AM . When we load this template into the case.Application Example http://www. Toluene and Phenol. it is necessary to define a Fluid Package.hyprotech. Enter the Main Environment. As with creating a Case. select New Template. while column B will contain the data imported from the case file. or the appropriate formula.

Economic and Annual Data The Economic. HYSYS .5 B23 +((b26)/(5*3.SteadyState.hyprotech.615))/c17*b17 B18 +b2 B19 +b3 B20 +b4 B21 +b5 B22 +((b25)/(5*3.com/support/examples/extract/extract.Dynamic Design does not accept the import.1415*b7))^. Annual Expense and Revenue Data is shown below: 52 of 90 2/24/99 10:22 AM . and it is necessary to enter the formula as shown in cell B16.B23 (excluding B17) are: B16 +((b15*24)/(b17*5.5 Although the Feed Standard Density could also be imported in HYSYS.htm The formulae for cells B16 .1415*h16))^.Application Example http://www.

htm Enter all data exactly as shown. and we must use the following formulae: 53 of 90 2/24/99 10:22 AM . Capital Cost Data The Capital Cost Data is set up in columns G and H. respectively. There is also some additional Simulation Data in this area (H16 H19).Dynamic Design does not accept the import. HYSYS . Although the Standard Densities could also be imported in HYSYS.com/support/examples/extract/extract. There are no formulae on this page.Application Example http://www.SteadyState.hyprotech. The Toluene Density and Heptane Density will be imported into cells H16 and H17.

02 and 0.hyprotech.B26 have already been completed. 54 of 90 2/24/99 10:22 AM .958*@ln(b22*12)+4.44)*b14 D24 +d18+d19+d20+d21+d22+d23 D25 +h2+h3+h4+h5+h6+h7+h8+h9+h10+h11+h12 D26 +d24*(1+d25) D27 +h12+h13 D28 +d26*(1+d27) D29 +d2*d28 D30 +d29*h14/100 Expense Calculation The Total Expenses and Adjusted Expense (incorporating Cost Index Factor) are displayed in cells B50 and D50. Note that cells A18 .44)*b13 D23 @exp(.do not enter the values 3.htm H18 +((b11*24)/(h16*5.615))/i6*h16 H19 +((b12*24)/(h17*5.com/support/examples/extract/extract.615))/i7*h17 Capital Cost Calculation The Capital Cost Calculation is performed in cells C18 . respectively. All of the cells in column D shown here are formulae .Application Example http://www.75*b18/9000+9300 D19 +12.958*@ln(b23*12)+4.5/5300+15000 D22 @exp(.5/5300+15000 D21 +b21*19.15! The formulae are listed below: D18 +12.D30.75*b19/9000+9300 D20 +b20*19.

hyprotech.Application Example http://www. These are the formulae used: 55 of 90 2/24/99 10:22 AM .com/support/examples/extract/extract.htm The Expenses are listed in column B. rows 32-49.

Application Example http://www.hyprotech. the purities of the Toluene and Heptane in the respective distillates will be imported to cells D33 and D34.htm B32 +b10*f2*24*d5 B33 +b15*f3*24*7.219*@ln(b11*24/2000)+2.481*f9/(70*62. 56 of 90 2/24/99 10:22 AM .48*b11*d33^1/h18 E34 +d16*24*d5*7. E33 +d15*24*d5*7.4*1000) B40 +b3*24*d5*7.481*f9/(70*62.F50. as shown below.48*b12*d34^1/h19 F33 +d2*e33 F34 +d2*e34 F35 +f33+f34 Calculation of Net Worth This is determined in cells E37 .com/support/examples/extract/extract.481*d5/b16 B34 +(@exp(.4*1000) B41 +b4*24*d5*f10/(1e6) B42 +b5*24*d5*f10/(1e6) B43 +d28*d8/100 B44 +d28*d9/100 B45 +d28*d10/100 B46 +d13*(b34+b35+b36)/100 B47 +f13*b34/100 B48 +f14*(e33+e34)/100 B49 +f15*(e33+e34)/100 B50 +b32+b33+b34+b35+b36+b37+b38+b39+b40+b41+b42+b43+b44+b45+b46+b47+b48+b49 D50 +b50*d2 Revenue Calculation The calculation of Revenue is shown below.844))*f4*d5 B35 +f5*b34/100 B36 +f6*d29/100 B37 +f7*b36/100 B38 +f8*b34/100 B39 +b2*24*d5*7.

select the Read an Existing Template button when prompted to select the source for the sub-flowsheet.g. Importing Variables into Spreadsheet First.com/support/examples/extract/extract.Application Example http://www. Save it (e. add the subflowsheet.TPL). 57 of 90 2/24/99 10:22 AM . The formulae are listed below: F37 +d29 F38 +d30 F39 +f37+f38 F41 +f35 F42 +d50 F43 +f41-f42 F44 +(1-d3)*f43 F45 +b43+f44 F48 +f39 F49 +((1-(1/(1+d4)^f47))/d4)*f45 F50 +f49-f48 The template is now complete.ECONANAL. and load the HYSYS case. .htm Note that cell F47 contains the constant 5.hyprotech. indicating the life of the project.

htm Select the template from the list (in this case. we have called the template ECONANAL.com/support/examples/extract/extract. They are all listed below: 58 of 90 2/24/99 10:22 AM .TPL.hyprotech. There are a large number of variables to import into the Spreadsheet. and saved it in the c:\hysys\template directory).Application Example http://www. HYSYS will create a new subflowsheet with the Spreadsheet you set up when you created the template.

and choose the variable using the Variable Navigator (For more information. ensure that you are in the Main Flowsheet and that your column is solved. 2. then your Spreadsheet is complete. select the Add Import button. When you move to the Spreadsheet page. Note that you may also drag variables into the Spreadsheet. type or select from the drop down list the Spreadsheet cell to be connected to that variable. 59 of 90 2/24/99 10:22 AM .Navigation). see HYSYS Reference. Make sure that no cells read <empty>. If Cell F50 (Net Present Worth) has calculated. Choose Import Variable. Importing Variables from the Spreadsheet Page (Browsing) You may also import a variable by positioning the cursor in an empty field of the Spreadsheet and clicking the right mouse button.hyprotech. Importing From the Connections Page To add an import.com/support/examples/extract/extract. Chapter 4 .htm There are two ways to import the variables to the Spreadsheet: 1. Once you have imported all the variables. Chapter 4 . that variable will appear in the cell you specified. In the Cell column. You will see the menu shown to the right.Navigation) select the flowsheet variable you wish to import to the Spreadsheet. and using the Variable Navigator (see HYSYS Reference.Application Example http://www.

84 Million. Comparison of Configurations “Heptane Purity” refers to the Heptane composition in the first column distillate stream (Heptane).htm For the high purity configuration (RR1 = 5. Also.72 Million. change the first column stage on which the tray liquid molar flow is being measured to 16 (Cell B5).99. Recall that we could improve the Toluene Purity and Heptane Purity to 0.99 The Net Present Worth is $3.41 Million. the Net Present Worth is $3. Optimization We will use the following procedure in determining the optimum location of the feed streams: 1. Set the location of the feed stream. “Toluene Purity” refers to the Toluene composition in the second column distillate stream (Toluene). The Net Present Worth increases to $4. Even though we could improve this figure.985 Toluene Purity = 0.985 In the Spreadsheet.994 and 0. The Net Present Worth is $1. set the number of stages in cells B13 and B14 to 19 and 9. All further analysis will consider only the first configuration. we have the High Purity Configuration with the following specifications: Column 1 Solvent Stage = 13 Column 1 Feed Stage = 21 Column 1 Number of Stages = 24 Column 2 Feed Stage = 8 Column 2 Number of Stages = 9 Reflux Ratio 1 = 5 Reflux Ratio 2 = 5 Heptane Purity = 0. First.99 Toluene Purity = 0.99).84 Million.hyprotech.993 respectively.Application Example http://www.com/support/examples/extract/extract. RR2 = 5. 60 of 90 2/24/99 10:22 AM . Toluene Purity = 0. it is safe to say that the first configuration (high purity) is economically superior. respectively. For the Low Purity Configuration specify the following: Column 1 Solvent Stage = 7 Column 1 Feed Stage = 17 Column 1 Number of Stages = 19 Column 2 Feed Stage = 8 Column 2 Number of Stages = 9 Reflux Ratio 1 = 12 Reflux Ratio 2 = 4 Heptane Purity = 0. Heptane Purity = 0.

Application Example http://www. After the column solves.06 and 0. add the Solvent Rate specification as a Primary Variable: Set low and high bounds of 0. Set up the Optimizer to Maximize the Net Worth by adjusting the Solvent Rate specification: On the Variables Page of the Optimizer.com/support/examples/extract/extract. replace the First Column Reflux Ratio specification with the following spec: Solvent Rate = Current Value 4.99 3.htm 2.hyprotech. Solve the column to the following specifications: First Column Reflux Ratio = 5 Second Column Reflux Ratio = 5 Toluene Purity = 0.99 Heptane Purity = 0.150: 61 of 90 2/24/99 10:22 AM .

Application Example http://www. and reduce the Maximum Change/Iteration to 0. On the Functions page of the Optimizer.htm Import the Net Worth from the Case Spreadsheet into Cell A1 of the Optimizer Spreadsheet.com/support/examples/extract/extract. set the tolerance to 1e06.hyprotech.1000: 62 of 90 2/24/99 10:22 AM . specify the Objective Function Cell as A1 and select the Maximize radio button: On the Parameters page. select the Mixed scheme.

2.776 5 0. It is a stable specification to adjust.99 677 $4. If.641 5 0. Next. Select the Start button.86 M 10 3. we will adjust the location of the Mixed Feed to the first column: 63 of 90 2/24/99 10:22 AM .com/support/examples/extract/extract.hyprotech. allowing HYSYS to adjust the Solvent stream in order to Maximize the Net Worth of the process. adjust the location of the Solvent feed to the first column. for example.99 0. Stage RR1 RR2 Toluene Heptane Solvent Flow Net Worth 13 3.99 0. There is a point somewhere in the middle of the range of Solvent Rates where the Net Present Worth is maximized. the Net Present Worth would be maximized close to the minimum Reflux Ratio for which the column solves.69 M 12 3. Similar logic applies for the Toluene and Heptane Fractions.99 0. we will feed the solvent recycle on stage 11. Location of Feed Stream We adjust the Solvent Rate for two reasons: 1. that is. The results are tabulated below.99 807 $4. making that method inherently unstable.Application Example http://www.819 5 0. the Column will solve over an extensive range of Solvent Rates. First.99 0.939 5 0.83 M From this point on.htm 5. using the Optimizer to maximize the Net Worth for each feed configuration.99 880 $4.99 777 $4. we were to adjust a Reflux Ratio.79 M 11 3.

with the following results: Case Description Reflux Reflux Heptane Toluene Solvent Rate Capital Ratio 1 Ratio 2 Purity Purity (lbmole/hr) Investment Annual Income Net Present Worth Base Case I 5 5 0.99 2203 3.73 M 7 3.98 M 1. the process feed to stage 20 of the first column.81 M 3.71 M 2. 5 0.99 741 19 3. we adjust the location of the Feed to the second column: Stage RR1 RR2 Toluene Heptane Solvent Flow Net Worth 8 3.347 0.Application Example http://www.99 1401 3.429 5 0. 5 5 0.99 777 20 3.370 5 0.89 M 4.77 M Maximize Net Present Worth by adjusting the Solvent 5 Rate and allowing the Reflux Ratio for Column 2 to vary. Finally.60 M Maximize the Heptane Purity.350 Rate and allowing the Reflux Ratio for Column 1 to vary.350 5 0.hyprotech.99 713 2. 1. 5 0.996 2120 3.414 5 0.99 648 2.99 764 From this point on.htm Stage RR1 RR2 Toluene Heptane Solvent Flow 21 3.641 5 0.82 M 4.99 0.75 M 3.99 0.58 M 1.60 M 1.99 698 $4. which in turn maximizes the Net Present Worth.99 0. we will feed the solvent recycle to stage 11 of the first column.03 M 5.99 716 $4.99 0. and the first column bottoms to stage 7 of the second column.99 0.60 M Maximize Net Present Worth by adjusting the Solvent 3.99 0.99 0.761 M For the rest of the optimization.438 5 0.86 M Maximize the 64 of 90 2/24/99 10:22 AM . Optimization of Purities and Reflux Ratios Several variations of the reflux ratios and purities are now tested.39 M 1.99 0. which 5 in turn maximizes the Net Present Worth.08 M Maximize the Toluene Purity.994 0.765 M 6 3.99 0. we will feed the process feed to stage 20 (minimum Reflux Ratio).99 713 $4.99 0.com/support/examples/extract/extract.

570 0.89 M 4. Toluene Fraction Low Bound — 0.985 High Bound — 0. We therefore have to be cautious when we select the primary variable ranges.038 0.12 Primary Variable 2: Source — T-100 Variable — T-100. 1.70 M The last results (bottom row of table) are the best up to this point.996 782 2. Spec Value.htm Maximize the Heptane Purity.996 (set at the maximum) Primary Variable 4: Source — T-100 Variable — T-100. Heptane Fraction Low Bound — 0.hyprotech. The danger with this approach is that we cannot simply input the maximum purities as the high limit and the minimum reflux ratios as the low limit. There would be many combinations in this range which would not solve.07 M 5.com/support/examples/extract/extract.57 M Maximize the Toluene purity.994 0.19 M 1.99 1068 3. The search for the optimum result has gone to the point where we have to include more primary variables and allow HYSYS to find the appropriate solution. 4.Application Example http://www.985 High Bound — 0. and/or provide a small value for the Maximum Change/Iteration. then maximize the Net Present Worth by 5 adjusting the Solvent Flow and allowing Reflux Ratio 2 to vary. Spec Value. then maximize the Net Present worth by 5 adjusting the Solvent Flow and allowing Reflux Ratio 2 to vary. Spec Value.99 0.78 M 2. Reflux Ratio 1 65 of 90 2/24/99 10:22 AM . The Optimizer is set up as follows: Primary Variable 1: Source — T-100 Variable — T-100. Solvent Rate Low Bound — 0.994 Primary Variable 3: Source — T-100 Variable — T-100. Spec Value. due to the fact that we are pushing the limits on the column feasibility.08 High Bound — 0.

65 M Heptane Purity 0. The Optimizer Variables page is shown below.205 Solvent Rate 757 lbmole/hr Reflux Ratio 2 1.Application Example http://www. In the next section. rather than a maximum imposed by a boundary constraint.93 M It is important to note that although this appears to be the optimum steady-state solution.htm Low Bound — 2.00 It is important to ensure that the current (starting) values of these variables are within the bounds.09 M Toluene Purity 0.hyprotech. The Reflux Ratio for the second column will be allowed to vary while we attempt to find the maximum Net Worth. after an Optimum is found: Note that none of the Actual Values are at the Boundary limits.50 High Bound — 5.6 Heptane Fraction — 0.com/support/examples/extract/extract.994 2/24/99 10:22 AM .593 Capital Investment $2.989 Annual Income $2.994 Net Present Worth $5. The results are tabulated below: Reflux Ratio 1 4. as it means that a true maximum has been found. we will study the dynamics of the process.989 Toluene Fraction — 0. This is significant. it does not mean that this configuration is controllable in dynamics. The column configuration is summarized below: First Column Second Column Number of Stages — 24 Number of Stages — 9 Process Feed Stage — 20 Feed Stage (from Column 1 Bottoms) — 7 Solvent Recycle Stage — 11 66 of 90 Reflux Ratio — 4.2 Reflux Ratio — 1.

Column 2 (1-11) and Column 1 (12-37) Component Summary. The Results are shown here: Temperature Profile.hyprotech.htm Note that the temperature profile for the second column is shown first.Application Example http://www.com/support/examples/extract/extract. Column 2 67 of 90 2/24/99 10:22 AM .

hyprotech.htm Column Worksheet 68 of 90 2/24/99 10:22 AM .Application Example http://www.com/support/examples/extract/extract.

com/support/examples/extract/extract.99 Toluene Fraction — 0. Nevertheless. the results are 69 of 90 2/24/99 10:22 AM . and the Heptane fraction was relaxed to 0. the same specifications could not be met. The Reflux Ratio for Column 1 was relaxed to 10.hyprotech.htm Component Summary. an attempt was made to reproduce the process as set up in the original HYSYS Reference Manual (page 463).97. Annual Income = $1.44 M However. As in the example.Application Example http://www.8 Reflux Ratio — 12 Heptane Fraction — 0. Column 1 Column Worksheet As a point of interest.985 Solvent Rate = 1145 lbmole/hr Capital Investment = $4. the Peng Robinson Property Package was used (earlier shown to be unacceptable) with the following results: First Column Second Column Number of Stages — 20 Number of Stages — 10 Process Feed Stage — 13 Feed Stage (from Column 1 Bottoms) — 10 Solvent Recycle Stage — 6 Reflux Ratio — 3.04 M. This is clearly an unacceptable option. but the best possible using the same configuration. when the NRTL Property Package with the updated interaction parameters is used.34 M Net Present Worth = $1.

com/support/examples/extract/extract. we require that Partial Condensers be used.99 Solvent Rate = 337 lbmole/hr Capital Investment = $4.69 M Net Present Worth = ($1.65 M) C-6.97 Toluene Fraction — 0.hyprotech.7 Dynamics In Dynamics.htm shown below: First Column Second Column Number of Stages — 20 Number of Stages — 10 Process Feed Stage — 13 Feed Stage (from Column 1 Bottoms) — 10 Solvent Recycle Stage — 6 Reflux Ratio — 10 Reflux Ratio — 12 Heptane Fraction — 0. If you installed your Condensers as Total Condensers. change them to Partial Condensers as shown below: 70 of 90 2/24/99 10:22 AM . Annual Income = $0.50 M.Application Example http://www.

Tray Sections In the economic analysis. Setting the Dynamic Property Model Parameters — The proper choice of these parameters will ensure numeric stability and accurate extrapolation.com/support/examples/extract/extract.Application Example http://www. Condensers. We do not want any vapour flow off the condensers. we estimated the diameter of the first tray section as follows: 71 of 90 2/24/99 10:22 AM . for both columns’ Reflux. there are several important steps: Sizing the Vessels — The Tray Sections. Condenser Duty and Reboiler Duty. It is also imperative that the Cooling Volume and Tower Volume (set in the Condenser) are accurate. Once we have completed these steps. typically to span twice the steady-state value. introducing various upsets to the system to ensure that our control system can adequately handle them. so the specifications will be as shown: Overview Before we can run the process dynamically. Adding the Controls — We require at least ten controllers.htm Call the Vapour streams Vapour 1 and Vapour 2. The control scheme (selection of Process Variable) and Tuning are very important in ensuring a stable control configuration. Setting up the Strip Charts — We will track key variables while we run the simulation. Sizing the Valves — All of the valves must be sized. Bottoms.hyprotech. Note that we already did some sizing calculations in the Steady-State portion of this simulation. and Reboilers must be appropriately sized based on their respective liquid flowrates. Distillate. We will now have to provide two additional columns specifications. Sizing the Vessels It is important to correctly size the vessels in order to ensure a reasonable dynamic response. we can run the process dynamically.

hyprotech. we have: The diameter is calculated to be 6. Using the Steady-State values.16 ft). the diameter is calculated to be 13.htm where FTRAY1 is the liquid volume flowrate on stage 20 of the first tray section. 1/120th of an hour or half a minute. Condensers The volume of the condensers are calculated as follows: For the first condenser: For the second condenser: 72 of 90 2/24/99 10:22 AM . we have: When you enter this value on the Dynamics page of the Tray Section.com/support/examples/extract/extract.Application Example http://www. For the second tray section.23 ft. The volume of the first tray section is: The factor (1/120) is the residence time.24 ft (assuming a weir height of 0.

hyprotech.Application Example http://www.htm The tower volume and cooling volumes are estimated as follows: For the first column: For the second column: Reboilers The volumes of the reboilers are calculated as follows: For the first reboiler: For the second reboiler: Adding the Controls and Sizing the Valves Various approaches could be taken in the development of the control scheme.com/support/examples/extract/extract. The control scheme which we will be using is outlined here: 73 of 90 2/24/99 10:22 AM .

As shown later.com/support/examples/extract/extract. We must ensure that the controllers are tuned such that any adverse interaction is minimized. As noted in other examples. and it is therefore important to closely control the condenser pressure. The pressure of the condenser determines the pressure profile of the column. The Controller parameters are displayed below: 74 of 90 2/24/99 10:22 AM . set up and dynamically tested.htm Control Scheme One benefit of using HYSYS to develop a control scheme is that several different schemes could be considered. therefore. we will have a Pressure Controller maintaining the Partial Condenser pressure by manipulating the Condenser duty. However. for instance. Therefore. that you did not want dual-point temperature control on the two distillation columns but instead wanted ratio controllers to manipulate the reboiler duties. Note that fairly conservative tuning parameters have been chosen for the controllers. the comparison and fine-tuning of different control schemes is beyond the scope of this paper. therefore no effort is made to fine-tune the parameters. it would be fairly straightforward to set it up. the dynamic response is reasonable. the control scheme as shown in this figure will be used.Application Example http://www. if you decided. Condenser Duty Controllers For each column.hyprotech. the tray temperature (PV of the Reboiler Duty Controllers) and condenser pressure are interacting variables.

Application Example http://www. so that we can see where the change in temperature is the greatest: 75 of 90 2/24/99 10:22 AM .com/support/examples/extract/extract. the utility fluid flowrate for each condenser duty stream will be calculated and displayed. which ultimately implies composition control. temperature control is achieved. we want to control the temperature of the tray where the temperature sensitivity is the highest. Generally. Reboiler Duty Controllers By manipulating the reboiler duty. set the Minimum and Maximum Flow to 0 and 10000 lbmole/hr.0000 psia TUNING 16.htm CONTROLLER Cond 1 Pressure CONTROLLER Cond 2 Pressure CONNECTIONS CONNECTIONS PV Object Condenser-1 PV Object Condenser-2 PV Vessel Pressure PV Vessel Pressure OP Object Condenser-1 Duty OP Object Condenser-2 Duty Control Valve Control Valve Duty Source From Utility Fluid Duty Source From Utility Fluid Min Flow 0 lbmole/hr Min Flow 0 lbmole/hr Max Flow 10000 lbmole/hr Max Flow 10000 lbmole/hr PARAMETERS PARAMETERS PV Min & Max 10 & 20 psia PV Min & Max 10 & 20 psia Action Direct Action Direct Controller Mode Auto Controller Mode Auto SP SP 16.hyprotech.8 Kp 0.8 Ti 15 Ti 15 Td <empty> Td <empty> For the Utility Fluid. Note that when you enter Dynamic Mode. To determine which tray has the highest sensitivity to temperature.0000 psia TUNING Kp 0. we will do a steady-state sensitivity analysis which varies the reboiler duty by a small amount.

76 of 90 2/24/99 10:22 AM .hyprotech. the greatest change in temperature in the first column occurs on Stage 18.htm The Case Study tool can be used to produce these plots. we will use the Stage 8 Temperature as the PV. As is apparent from the graphs.com/support/examples/extract/extract. We will use the Stage 18 Temperature as the Process Variable for the first column. For the second column.Application Example http://www. it is not a recommended practice to have a feed stage as the process variable for a controller. Although Stage 7 has a large % Change (roughly equal and opposite to Stage 8).

com/support/examples/extract/extract.8 Kp 0.Application Example http://www.hyprotech.0e+07 Btu/hr Max Available 2.8 Ti 15 Ti 15 Td <empty> Td <empty> Column 1 Material Stream Controllers The parameters for the Material Stream Controllers in the first column are displayed below: 77 of 90 2/24/99 10:22 AM .0e+07 Btu/hr PARAMETERS PARAMETERS PV Min & Max 200 & 300 F PV Min & Max 300 & 400 F Action Reverse Action Reverse Controller Mode Auto Controller Mode Auto SP SP 241. OP Object Reboiler-1 Duty OP Object Reboiler-2 Duty Control Valve Control Valve Duty Source Direct Q Duty Source Direct Q Min Available 0 Btu/hr Min Available 0 Btu/hr Max Available 2.9 F TUNING 347. PV Stage 8 Temp.htm CONTROLLER TS-1 Stage 18 CONTROLLER TS-2 Stage 8 CONNECTIONS CONNECTIONS PV Object TS-1 PV Object TS-2 PV Stage 18 Temp.6 TUNING Kp 0.

We want the flowrate of this stream to vary with changes to the Feed flowrate and composition.8 Ti 20 Ti <empty> Ti <empty> Td <empty> Td <empty> Td <empty> For the Reflux Stream. PV Liquid Level PV Liquid Level OP Object Reflux-1 OP Object Heptane OP Object Col-1 Bottoms Control Valve Control Valve Control Valve Flow Type Molar Flow Flow Type Molar Flow Flow Type Molar Flow Min Flow 0 lbmole/hr Min Flow 0 lbmole/hr Min Flow 0 lbmole/hr Max Flow 1600 lbmole/hr Max Flow 400 lbmole/hr Max Flow 2000 lbmole/hr PARAMETERS PARAMETERS PARAMETERS PV Min & Max 200 & 300 F PV Min & Max 40 & 60% PV Min & Max 40 & 60% Action Direct Action Direct Action Direct Controller Mode Auto Controller Mode Auto Controller Mode Auto SP SP SP 221. This Stage is especially sensitive to variations in the feed flowrate.8 Kp 1. Set the Control Valve range from 0 to 1600 lbmole/hr. Level CONNECTIONS CONNECTIONS CONNECTIONS PV Object TS-1 PV Object Condenser-1 PV Object Reboiler-1 PV Stage 10 Temp.com/support/examples/extract/extract.4 Kp 1. The Minimum and Maximum Flow are set at 0 and 400 lbmole/hr. Column 2 Material Stream Controllers The parameters for the Material Stream Controllers in the second column are displayed below: 78 of 90 2/24/99 10:22 AM .hyprotech. The bottoms stream also has Level control. we use the temperature for Stage 10 (TS-1) as the Process Variable. The Minimum and Maximum Flow are set at 0 and 2000 lbmole/hr.7 F TUNING 50% TUNING 50% TUNING Kp 0.htm CONTROLLER Reflux 1 TS-1(10) CONTROLLER Heptane (Level)) CONTROLLER COL1 Bott. the first Reboiler’s setpoint is a 50% Liquid Level.Application Example http://www. The Heptane stream will be set on Level control. so that the first Condenser is 50% full.

but the “Calculated” rate of the Solvent will be the Set Point for this control.8 Ti 15 Td <empty> B3: Calculated Solvent Flow Type Molar Flow Spreadsheet Cell Min Flow 0 lbmole/hr Control Valve Max Flow 400 lbmole/hr Flow Type Molar Flow Min Flow 0 lbmole/hr 1500 lbmole/hr PV Min & Max 40 & 60% Max Flow Action Direct PARAMETERS Controller Mode Auto SP PV Min & Max 0 & 1500 lbmole/hr Action Reverse Cascaded SP 50% TUNING TUNING Molar Flow PV PARAMETERS PARAMETERS SP CONNECTIONS Control Valve Control Valve Solvent Flow Kp 1. Set the Control Valve range from 0 to 800 lbmole/hr.hyprotech. the bottoms stream (Solvent) has a cascaded set point. we use the temperature for Stage 2 (TS-2) as the Process Variable for the second Reflux control.8 Ti 15 Td <empty> Similar to the first Reflux control. This Stage is especially sensitive to variations in the feed flowrate. Note that we must select Spreadsheet Cell B3 when setting up the Cascaded Control. Finally.Application Example http://www. The Calculated Solvent rate is simply the Column 1 Bottoms Flowrate minus the Toluene (Distillate) flowrate. Create a new Spreadsheet in the Main Flowsheet and set it up as follows: 79 of 90 2/24/99 10:22 AM . We want the flowrate of this stream to vary with changes to the Feed flowrate and composition.9 F Liquid Level OP Object Solvent OP Object Toluene Cascaded SP Source SPRDSHT-1 Kp 0. The Flowrate is chosen as the Process Variable. The Toluene stream will be set on Level control.8 Controller Mode Ti <empty> TUNING Td <empty> Kp 0.com/support/examples/extract/extract. so that the second Condenser is 50% full. The Minimum and Maximum Flow are set at 0 and 400 lbmole/hr. OP Object Reflux-2 Flow Type Molar Flow Min Flow 0 lbmole/hr Max Flow 800 lbmole/hr CONTROLLER Toluene (Level)) PV Object CONNECTIONS PV PV Object Condenser-2 Solvent PV Min & Max 200 & 300 F Action Direct Controller Mode Auto 239.htm CONTROLLER Reflux 2 CONTROLLER TS-2(2) CONNECTIONS PV Object TS-2 PV Stage 2 Temp.

On the Parameters page.Application Example http://www. Enter the formula +B1-B2 into cell B3. which will be set up as follows: 80 of 90 2/24/99 10:22 AM .hyprotech. cell B3 will always equal cell B4. However.com/support/examples/extract/extract. Import the Toluene Molar Flow into cell B2. these cells will not necessarily be the same. in Dynamics. you may wish to enter a Variable Name for cell B3 so that it will be recognizable when you set up your controller.htm Import the Column 1 Bottoms Molar Flow into cell B1. In Steady-State. Import the Solvent Molar Flow into cell B4.

flowrates and compositions may also be adjusted from the WorkSheet. so they are not crucial to the simulation. you may put Manual controllers on the Feed and Phenol streams. Setting up the Strip Charts Enter the following variables in the DataBook: 81 of 90 2/24/99 10:22 AM .com/support/examples/extract/extract. However.hyprotech.Application Example http://www.htm Feed Stream Controllers If you wish.

hyprotech. In Dynamics.Application Example http://www. The second Strip Chart contains the following variables: 82 of 90 2/24/99 10:22 AM . All that is required is to replace the formula in the cell which calculates the Adjusted FCI with the actual figure in that cell.htm We will be setting up two Strip Charts.com/support/examples/extract/extract. it will be useful to see the effect of certain variables on the bottom line. and the “Net Worth”. you need to ensure that the initial capital cost does not fluctuate when there are changes in process variables. the Solvent Molar Flow rate. each having four variables. Change cell D29 to the figure that is currently being displayed (2.3042e+06). Although the concept of an instantaneous Net Worth is of no practical use. In the Net Worth Analysis. The first Strip Chart will plot the Heptane and Toluene Molar Fractions. so that the FCI will not change as the simulation progresses. certain variables such as the column diameter were dependent on key flowrates.

htm The temperatures of the stages which are used as the Process Variables in the Reboiler Duty controllers are plotted. along with the Condenser Pressures.com/support/examples/extract/extract. this causes the integration to proceed at a much slower rate. the default parameters are sufficient: If you were concerned that you were not achieving proper accuracy over a range of temperatures and pressures. in order to avoid a large “bump” as soon as you start the integrator.hyprotech. In this case. However. Enthalpies or Entropies. switching models does not seem to be justified. and in this case. whichever is appropriate for that control). then “on” again (to Auto or Cascade control. ensure that the starting point of each controller is correct. you might want to use the Property Package Method or Local Model in calculating the K-values. Setting the Dynamic Property Model Parameters It is always important to ensure that appropriate parameters are used for the Dynamic Property Model.Application Example http://www. and before running the integrator. The control FacePlates appear as follows: 83 of 90 2/24/99 10:22 AM . This can be achieved by resetting each controller by turning it off. Dynamic Simulation After switching to dynamics.

com/support/examples/extract/extract.hyprotech. the process variables will line out: 84 of 90 2/24/99 10:22 AM . and we will instead be making changes from the WorkSheet. After a period of time. however.Application Example http://www. these are turned off. Run the integrator.htm Note that we have included a Feed Flow and Phenol Flow controller.

htm Now introduce a feed composition upset as shown below: 85 of 90 2/24/99 10:22 AM .Application Example http://www.hyprotech.com/support/examples/extract/extract.

hyprotech.htm As shown in the Strip Charts. As well. 86 of 90 2/24/99 10:22 AM .Application Example http://www. The purities line out at different values.com/support/examples/extract/extract. since we have changed the composition of the feedstock. the pressures and temperatures shift somewhat from their Set Points but eventually return. the Solvent Molar Flow lines out at a higher value. which is expected.

Change the Molar Flow of stream Feed from 400 to 360 lbmole/hr.Application Example http://www.com/support/examples/extract/extract. we will introduce a Feed Molar Flow upset.hyprotech.htm Next. The Strip Charts are shown here: 87 of 90 2/24/99 10:22 AM .

htm At this point.Conceptual Design was used to estimate interaction parameters for the NRTL and Peng-Robinson Property Packages. Therefore. we can safely conclude that our control scheme is reasonable.com/support/examples/extract/extract. the lined-out value is valuable.Application Example http://www. Without this assurance. applying the new interaction parameters regressed from experimental data.Conceptual Design was used to obtain low-purity and high-purity column configurations. A good fit was obtained for the NRTL property package. drastically increasing the overall Net Worth. HYSYS . This step was 88 of 90 2/24/99 10:22 AM . one would probably end up designing either an inefficient or an impossible column configuration. this is because the cost of the Feed decreases suddenly. Both Equation of State models incorrectly predicted liquid-liquid behaviour.Conceptual Design was crucial to this simulation. in that we could be confident that the predicted VLE would closely match experimental behaviour. but not for the Peng-Robinson and PRSV Property Packages. Note that the Net Worth spikes as soon as we add the upset. We also may be able to achieve better control with a different scheme.8 Summary and Conclusions The use of HYSYS . the Net Worth has actually decreased. C-6. This is an example where this instantaneous Net Worth function is certainly not realistic. However.hyprotech. It is interesting to note that even though the purities on the output stream increased. NRTL was used for this simulation. there is no doubt that the scheme could be refined further. However. HYSYS .

..65 Million Capital Investment. An Economic Analysis Spreadsheet was set up in HYSYS. The system responded reasonably to these upsets. Fourth Edition. Y. valves were sized.Conceptual Design (i.93 Million.99 Toluene) required more stages than the low purity configuration (0. American Institute of Chemical Engineers. J. Chang. even higher purities were possible than what was predicted using HYSYS . 1983. also. as it gave a fundamental understanding of the separation process.. Sixth Edition.. 1957. Perry's Chemical Engineer's Handbook. 0. controllers were added. 5. but not exactly the same. and any differences could be attributed to the fact that an approximate solution was obtained in HYSYS . Extractive Distillation (Seader. tuning parameters were defined. which calculated the Present Net Worth by incorporating the Fixed Capital Cost. 13-53. Klaus D. and Dynamic Model parameters were checked. 4.. it was possible to obtain a Net Worth of $5. and Black.985 Heptane.. Peters.9 Bibliography 1.SteadyState. ed. and one of the Reflux Ratios had to be increased in order to obtain a solution.994).C. 0. The high purity configuration (0.hyprotech. additional assumptions were made.Conceptual Design and Steady State. For the high purity configuration. Plant Design and Economics for Chemical Engineers. The results were very similar between HYSYS . Finally. Perhaps most importantly. and Timmerhaus. "Toluene Recovery by Extractive Distillation".Conceptual Design (0.Conceptual Design and HYSYS.com/support/examples/extract/extract. Based on the preliminary economic data. 89 of 90 2/24/99 10:22 AM .99 Heptane. Annual Revenues and Economic and Plant Data. McGraw-Hill. The feed locations for both columns. 1.985 Toluene). 2. McGraw-Hill.SteadyState was used to build the two column configurations. C. from the definition of property package interaction parameters to the dynamic system response were carried out entirely using HYSYS . The process was run dynamically.). McGraw-Hill. 0. The high purity configuration was shown to be superior (in terms of the Present Net Worth) to the low purity configuration. such as constant molal overflow).a solution in which the passed streams between the two columns were similar. For the low purity configuration.e. Green. allowing us to see the process limitations and perimeters. 3. Don W. with a $2. James H. The vessels were sized.Dynamics. 1991. Cost and Optimization Engineering.993. Acta Focalia Sinica 2.. strip charts were set up. indicating that the control scheme was satisfactory. C-6. . 1946. Feed composition and feed flow upsets were individually introduced. the solvent feed location. the reflux ratios and product purities were all varied in an effort to maximize the Present Net Worth. Max S. and key variables were observed to ensure that the control system was adequate. et al. indicating that this is an economically viable process. but the Reflux Ratios were roughly the same. Annual Expenses. the specifications could not be met. Dunn. the setup of this process. HYSYS.htm important.SteadyState and dynamics. The Optimizer was used to further refine the high purity configuration.L. although it is acknowledged that further improvements are certainly possible. 1984. the process was set up in HYSYS.. Jelen.D.Application Example http://www. Frederic C.