generalized disjunctive programming: a framework …...ignacio e. grossmann center for advanced...
TRANSCRIPT
![Page 1: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/1.jpg)
Ignacio E. GrossmannCenter for Advanced Process Decision-making
Carnegie Mellon University Pittsburgh, PA 15213, USA
Generalized disjunctive programming: A framework for formulation and alternative
algorithms for MINLP optimization
IMA Hot Topics Workshop:Mixed-Integer Nonlinear Optimization: Algorithmic Advances and Applications
November 17-21, 2008
![Page 2: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/2.jpg)
2
• Discrete/Continuous Optimization Nonlinear models0-1 and continuous decisions
• Optimization Models Mixed-Integer Linear Programming (MILP)Mixed-Integer Nonlinear Programming (MINLP)
Motivation
• ChallengesHow to develop “best” model?How to improve relaxation?How to solve nonconvex GDP problems to global optimality?
Alternative approach:Logic-based: Generalized Disjunctive Programming (GDP)
![Page 3: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/3.jpg)
3
1. Overview of major relaxations for nonlinearGDP and algorithms
2. Linear GDP: hierarchy of relaxations
3. Global Optimization of nonconvex GDP
Outline
Ph.D. StudentsRamesh RamanMetin TurkaySangbum LeeNick SawayaJuan Ruiz
![Page 4: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/4.jpg)
4
MINLP
f(x,y) and g(x,y) - assumed to be convex and bounded over X. f(x,y) and g(x,y) commonly linear in y
,1,0|,,|
, 0),( ..
),(min
aAyyyYbBxxxxRxxX
YyXxyx gts
yxfZ
m
ULn
≤∈=≤≤≤∈=
∈∈≤
=
• Mixed-Integer Nonlinear Programming
Objective Function
Inequality Constraints
![Page 5: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/5.jpg)
5
Algorithms Branch and Bound (BB) Ravindran and Gupta (1985), Stubbs, Mehrotra (1999), Leyffer (2001) Generalized Benders Decomposition (GBD) Geoffrion (1972)Outer-Approximation (OA) Duran and Grossmann (1986), Fletcher and Leyffer (1994) LP/NLP based Branch and Bound Quesada, Grossmann (1994)Extended Cutting Plane(ECP) Westerlund and Pettersson (1992) Codes: SBB GAMS simple B&B MINLP-BB (AMPL)Fletcher and Leyffer (1999)
Bonmin (COIN-OR) Bonami et al (2006) FilMINT Linderoth and Leyffer (2006)
DICOPT (GAMS) Viswanathan and Grossman (1990) AOA (AIMSS)
α−ECP Westerlund and Peterssson (1996) MINOPT Schweiger and Floudas (1998) BARON Sahinidis et al. (1998)
Mixed-integer Nonlinear Programming
![Page 6: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/6.jpg)
6
Generalized Disjunctive Programming
Motivation
1. Facilitate modeling of discrete/continuous optimization problems through use algebraic constraints andsymbolic expressions
2. Reduce combinatorial search effort3. Improve handling nonlinearities
![Page 7: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/7.jpg)
7
Generalized Disjunctive Programming (GDP)
( )
Ω
,0)(
0)(
)(min
1
falsetrue,YRc,Rx
trueY
K k γc
xgY
Jj
xs.t. r
xfc Z
jk
k
n
jkk
jk
jk
k
kk
∈
∈∈=
∈⎥⎥⎥
⎦
⎤
⎢⎢⎢
⎣
⎡
=
≤∈
≤
∑ +=
∨
• Raman and Grossmann (1994) (Extension Balas, 1979)
Objective Function
Common Constraints
Continuous Variables
Boolean Variables
Logic Propositions
OR operator
Disjunction
Fixed Charges
Constraints
![Page 8: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/8.jpg)
8
11 21
12 22
13 23
11 12 13
13 11 12
21 22
11 21 12 22 13 23
0, , , , , ,
U
Y YY YY YY Y YY Y YY Y
x xY Y Y Y Y Y True Fa
∨∨∨
∨ ⇒⇒ ∨
∨
≤ ≤∈
11 2 3, ,
lse
c c c
∈R
Process Network with fixed charges
1 2 3
1 2 4
6 3 5
11
3 1 2
1
. .
TMin Z c c c d xs tx x xx x x
Yx p xc
= + + +
= += +
=21
3 2
1 1
12 22
5 2 4 5 4
2 2 2
13 23
7 3 6 7 6
3 3 3
0 0
0 0
0 0
Yx x
c
Y Yx p x x xc c
Y Yx p x x xc c
γ
γ
γ
⎡ ⎤ ⎡ ⎤⎢ ⎥ ⎢ ⎥∨ = =⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥= =⎣ ⎦ ⎣ ⎦⎡ ⎤ ⎡ ⎤⎢ ⎥ ⎢ ⎥= ∨ = =⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥= =⎣ ⎦ ⎣ ⎦⎡ ⎤ ⎡ ⎤⎢ ⎥ ⎢ ⎥= ∨ = =⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥= =⎣ ⎦ ⎣ ⎦
GDP model
![Page 9: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/9.jpg)
9
Generalized Disjunctive Programming (GDP)
( )
Ω
,0)(
0)(
)(min
1
falsetrue,YRc,Rx
trueY
K k γc
xgY
Jj
xs.t. r
xfc Z
jk
k
n
jkk
jk
jk
k
kk
∈
∈∈=
∈⎥⎥⎥
⎦
⎤
⎢⎢⎢
⎣
⎡
=
≤∈
≤
∑ +=
∨
• Raman and Grossmann (1994)
Objective Function
Common Constraints
Disjunction
Fixed Charges
Continuous Variables
Boolean Variables
Logic Propositions
OR operator Constraints
Relaxation?
![Page 10: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/10.jpg)
10
Big-M MINLP (BM)
• MINLP reformulation of GDP
min ( )
. . ( ) 0 ( ) (1 ) , ,
1,
0, 0,1
k
k
jk jkk K j J
jk jk jk k
jkj J
jk
Z f x
s t r xg x M j J k K
k K
A a x
γ λ
λ
λ
λλ
∈ ∈
∈
= +
≤≤ − ∈ ∈
= ∈
≤≥ ∈
∑ ∑
∑Big-M Parameter
Logic constraints
NLP Relaxation 0 1jkλ≤ ≤
![Page 11: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/11.jpg)
11
Convex Hull Formulation
• Consider Disjunction k ∈ K( ) 0
k
jk
jkj J
jk
Y
g x
c γ∈
⎡ ⎤⎢ ⎥
≤⎢ ⎥⎢ ⎥=⎢ ⎥⎣ ⎦
∨
Theorem: Convex Hull of Disjunction k (Lee, Grossmann, 2000)Disaggregated variables ν j
λj - weights for linear combination
, 0)/(
1,0 ,1
,0
, ,|),(
kjk
jk
jkjk
jkJj
jk
kjkjk
jk
Jjjkjk
Jj
jk
Jjvg
JjUv
cvxcx
k
k
∈≤
≤<=∑
∈≤≤
∑=∑=
∈
∈∈
λλ
λλ
λ
γλ
Convex Constraints
- Generalization of Balas (1979)Stubbs and Mehrotra (1999)
=>
k
![Page 12: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/12.jpg)
12
)/(),( λλλ vgvh =
Remarks
If g(x) is a bounded convex function, is a bounded convex function Hiriart-Urruty and Lemaréchal (1993)),( λvh
1.
0)0,( =νh for bounded g(x)
0, ( )( (0)) (0) 0 0jk jk jkif g gλ ε ε= ⇒ − = ≤
1, ((1)( ( / (1)) (0)(0) (1) ( / (1)) 0jk jk jk jk jk jkif g g gλ ν ε ν= ⇒ − = ≤
a. Exact approximation of the original constraints as ε → 0.
c. The LHS of the new constraints are convex.
b. The constraints are exact at λjk = 0 and at λjk = 1 regardless of value of ε.
2. Replace by:( / ) 0jk jk jk jkgλ ν λ ≤ 0 jk jkUν λ≤ ≤where
((1 ) )( ( / ((1 ) ))) (0)(1 ) 0jk jk jk jk jk jkg gε λ ε ν ε λ ε ε λ− + − + − − ≤
Furman, Sawaya & Grossmann (2007)
![Page 13: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/13.jpg)
13
Convex Relaxation Problem (CRP)
Property: The NLP (CRP) yields a lower bound to optimum of (GDP).
Logic constraints
, ,10 ,0 ,
, ,0)/(
, 1
, ,0
,
0)( ..
)( min
Kk JjxaA
Kk Jjg
Kk
Kk JjU
Kk x
xrts
xfZ
kjk
jk
kjk
jk
jkjk
Jjjk
kjkjk
jk
Jj
jk
Kk Jjjkjk
k
k
k
∈∈≤≤≥≤
∈∈≤
∈=∑
∈∈≤≤
∈∑=
≤
+∑ ∑=
∈
∈
∈ ∈
λλ
λλ
λ
λ
λγ
ν
ν
ν
ν
Convex HullFormulation
CRP:
Note: Hull relaxation as intersection of convex hull for each disjunction
![Page 14: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/14.jpg)
14
Strength Lower BoundsTheorem: The relaxation of (CRP) yields a lower bound that is greater than or equal to the lower bound that is obtained from the relaxation of problem (BM):
min ( )
. . ( ) 0 ( ) (1 ) ,
1,
0, 0 1
k
k
jk jkk K j J
jk jk jk k
jkj J
jk
Z f x
s t r xg x M j J k K
k K
A a x
γ λ
λ
λ
λλ
∈ ∈
∈
= +
≤≤ − ∈ ∈
= ∈
≤≥ ≤ ≤
∑ ∑
∑
RBM:
![Page 15: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/15.jpg)
15
MINLP Reformulation
, 1,0 ,0 ,
, ,0)/(
, 1
, ,0
,
0)( ..
)( min
Kk JjxaA
Kk Jjg
Kk
Kk JjU
Kk x
xrts
xfZ
kjk
jk
kjk
jk
jkjk
Jjjk
kjkjk
jk
Jj
jk
Kk Jjjkjk
k
k
k
∈∈=≥≤
∈∈≤
∈=∑
∈∈≤≤
∈∑=
≤
+∑ ∑=
∈
∈
∈ ∈
λλ
λλ
λ
λ
λγ
ν
ν
ν
ν
Specify in CRP as 0-1 variablesλ
![Page 16: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/16.jpg)
16
Logic based methods
Branch and bound(Lee & Grossmann, 2000)
DecompositionOuter-ApproximationGeneralized Benders
(Turkay & Grossmann, 1997)
Methods Generalized Disjunctive Programming
Convex-hull Big-MCutting plane
(Sawaya & Grossmann, 2004)
Reformulation MINLPOuter-ApproximationGeneralized Benders
Extended Cutting Plane
GDP
![Page 17: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/17.jpg)
17
A Branch and Bound Algorithm for GDP
• Tree Search NLP subproblem at each node
• Solve CRP of GDPlower bound
CRP
+ fix a term indisjunction
CRPCRP
+ convex hullof remaining
terms
• Branching Rule Set the largest λj as 1 Dichotomy rule
• Logic inferenceCNF unit resolution (Raman & Grosmann, 1993)
• Depth first searchWhen all the terms are fixed
upper bound • Repeat Branching until ZL > ZU.
![Page 18: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/18.jpg)
18
GDP Example
Global Optimum(3.293,1.707)
Z* = 1.172
Contour of f (x)
Local solutions
x1
x2
(0,0)
S3
S2S1
Find x ≥ 0, (x ∈ S1)∨(x ∈ S2)∨(x ∈ S3) to minimize Z = (x1 - 3)2 + (x2 - 2)2 + c
Objective Function = continuous function + fixed charge (discontinuous).
![Page 19: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/19.jpg)
19
x1
x2
Convex hull = conv(USj)
Example : convex hull
S3
S2
S1
![Page 20: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/20.jpg)
20
x1
x2
(0,0)Convex hull
optimum, ZL = 1.154xL = (3.159,1.797)
S3
S2
S1
Convex combinationof zj
Convex hull = conv(USj)
zj = vj/λj
λ1= 0.016λ2= 0.955λ3= 0.029
Local solution point
Example: CRP solution
x*
Infeasible to GDP
Weight
![Page 21: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/21.jpg)
21
Example : branch and boundFirst Node: S2
Optimal solution: ZU = 1.172
x1
x2
(0,0)
S3
S1
Optimal Solution(3.293,1.707)Z* = 1.172
S2
![Page 22: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/22.jpg)
22
Example : branch and boundSecond Node: conv(S1 U S3)Optimal solution: ZL = 3.327
x1
x2
(0,0)
S3
S2S1
Upper BoundZU = 1.172
Lower BoundZL = 3.327
![Page 23: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/23.jpg)
23
• Branching Rule: λj - the weight of disaggregated variable Fix Yj as true: fix λj as 1.
Y2
Root NodeConvex hull of all Si
Z = 1.154λ = [0.016,0.955,0.029]
First NodeFix λ2 = 1, Z = 1.172[x1 ,x2] = [3.293,1.707]
λ = [0,1,0]
Second NodeConvex hull of S1 and S3
Z = 3.327λ = [ 0.337,0,0.623]
¬ Y2
ZU = 1.172Backtrack
ZL = 1.154Branch on Y2
ZL = 3.327 > ZU
Stop
Example: Search Tree
![Page 24: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/24.jpg)
24
Process Network with Fixed Charges• Türkay and Grossmann (1997)
Superstructure of the process
1
2
6
7
4
3
5 8
x1
x4
x6
x21
x19
x13
x14
x11
x7
x8
x12
x15
x9
x16 x17
x25x18
x10
x20
x23x22 x24x5
x3x2
A
B
: Unitj
Y1 ∨ Y2
Y6 ∨ Y7
Y4 ∨ Y5
C
D
F
E
Yi ∨ Yj
Specifications
![Page 25: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/25.jpg)
25
Minimum Cost: $ 68.01M/year
2
6
4
8
x1
x4
x19
x13
x14
x11
x12
x18
x20
x23 x24x5A
B
: Unitj
D
F
E
RawMaterial ProductsReactor Reactor
Optimal solution
x7
x6x10
x17
x25
x8
![Page 26: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/26.jpg)
26
Proposed BB MethodZL = 62.48
λ = [0.31,0.69,0.03,1.0,1,0,1]
ZU = 68.01 = Z*λ = [0,1,0,0,1.0,1,0,1]
Optimal Solution
ZU = 71.79λ = [0,1,1,1.0,1,0,1]
Feasible Solution
ZL = 75.01 > ZU
λ = [1,0,0.022,1.0,1,0,1]
ZL = 65.92λ = [0,1,0.022,1.0,1,0,1]
0
32
41
Fix λ2 = 1
Fix λ3 = 1 Fix λ3 = 0
Fix λ2 = 0
Stop
5 nodes vs. 17 nodes of Standard BB (lower bound = 15.08)
Proposed BB
0
ZL = 15.08 Big-M Std. BB
1 2
43
1413 5 6
812111615 7
10*9
Y4 = 0 Y4 = 1
Y6 = 0 Y6 = 1
Y8 = 0Y8 = 1
Y1 = 0 Y1 = 1
Y8 = 0 Y8 = 1
Y2 = 0 Y2 = 1 Y1 = 1
Y3 = 0 Y3 = 1
![Page 27: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/27.jpg)
27
Logic-based Outer ApproximationMain point: avoids solving MINLP in full space
Turkay, Grossmann (1997)
SDkiiDifalseYforc
xB
SDkDitrueYforc
xhxgts
xfcZ
kikk
i
kkiikk
ik
SDkk
∈≠∈=⎭⎬⎫
==
∈∈=⎭⎬⎫
=≤
≤
+= ∑∈
,ˆ,0
0
,ˆ0)(0)(..
)(min
ˆγ (NLPD)
x ∈ Rn, ci ∈ Rm,
NLP Subproblem:(reduced)
α+= ∑k
kcZMin
(MGDP)1,...,= 0)()()()()()(..
Lxxxgxg
xxxfxftsT
T
llll
lll
⎪⎭
⎪⎬⎫
≤−∇+
−∇+≥α
SDk
cL
xxxhxh
Y
ikk
ik
Tikik
ik
Di k
∈
⎥⎥⎥⎥⎥
⎦
⎤
⎢⎢⎢⎢⎢
⎣
⎡
=∈
≤−∇+∨∈
γl
lll 0)()()(
Ω (Y) = True α ∈ R, x ∈ Rn, c ∈ Rm, Y ∈ true, falsem
Master Problem:
Proceed as OA. Requires initialization several NLPs to cover all disjunctions
Redundant constraints are eliminated with falsevalues
Master problem solved with disjunctive branch and bound orwith MILP reformulation
![Page 28: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/28.jpg)
28
LogMIP
Part of GAMS Modeling System-Disjunctions specified with IF Then ELSE statementsDISJUNCTION D1(I,K,J);D1(I,K,J)
with (L(I,K,J)) ISIF Y(I,K,J) THEN
NOCLASH1(I,K,J);ELSE
NOCLASH2(I,K,J);ENDIF;
-Logic can be specified in symbolic form (⇒, OR, AND, NOT )or special operators (ATMOST, ATLEAST, EXACTLY)
-Linear case: MILP reformulation big-M, convex hull-Nonlinear: Logic-based OA
http://www.ceride.gov.ar/logmip/
Aldo Vecchietti, INGAR
![Page 29: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/29.jpg)
29
Carnegie Mellon
1
. .
( )
, ,
jk
k
jk
k
Tk
k K
jk jk
j Jk j k
j J
L U
jk k
k
Min Z c d x
s t Bx bY
A x a k Kc
Y k K
Y Truex x xY True False j J k K
c
γ
∈
∈
∈
= +
≥
⎡ ⎤⎢ ⎥
≥ ∈⎢ ⎥⎢ ⎥=⎣ ⎦
∨ ∈
Ω =
≤ ≤∈ ∈ ∈
∈
∑
∨
R k K∈
Linear Generalized Disjunctive ProgrammingLGDP Model
Raman R. and Grossmann I.E. (1994) (Extension Balas (1979)) (LGDP)
Objective function
Common constraints
Disjunctive constraints
Logic constraints
Boolean variables
Logical OR operator
Continuous variables
Can we obtain stronger relaxations?
![Page 30: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/30.jpg)
30
Carnegie Mellon
Disjunctive Programming
Constraint set of a DP can be expressed in two equivalentextreme forms
: ( )n i i
i QF x A x a
∈= ∈ ∨ ≥R
- Disjunctive Normal Form (DNF). A disjunction whose terms do not contain further disjunctions
0: , ( ), 1,...,j
n h h
h QF x Ax a d x d j t
∈= ∈ ≥ ∨ ≥ =R
- Conjunctive Normal Form (CNF). A conjunction whose terms do not contain further conjunctions
Disjunction: A set of constraints connected to one another through the logical OR operator ∨
Conjunction: A set of constraints connected to one anotherthrough the logical AND operator ∧
![Page 31: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/31.jpg)
31
Carnegie Mellon
1
. .
( )
, ,
jk
k
jk
k
Tk
k K
jk jk
j Jk j k
j J
L U
jk k
k
Min Z c d x
s t Bx bY
A x a k Kc
Y k K
Y Truex x xY True False j J k K
c
γ
∈
∈
∈
= +
≥
⎡ ⎤⎢ ⎥
≥ ∈⎢ ⎥⎢ ⎥=⎣ ⎦
∨ ∈
Ω =
≤ ≤∈ ∈ ∈
∈
∑
∨
R k K∈
Linear Generalized Disjunctive ProgrammingLGDP Model
(LGDP)
Objective function
Common constraints
Disjunctive constraints
Logic constraints
How to deal with Boolean and logic constraints in Disjunctive Programming?
Boolean variables
![Page 32: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/32.jpg)
32
Carnegie Mellon
Reformulating LGDP into DisjunctiveProgramming Formulation
Sawaya N.W. and Grossmann I.E. (2008)
1
. .
( )
, ,
jk
k
jk
k
Tk
k K
jk jk
j Jk j k
j J
L U
jk k
k
Min Z c d x
s t Bx bY
A x a k Kc
Y k K
Y Truex x xY True False j J k K
c
γ
∈
∈
∈
= +
≥
⎡ ⎤⎢ ⎥
≥ ∈⎢ ⎥⎢ ⎥=⎣ ⎦
∨ ∈
Ω =
≤ ≤∈ ∈ ∈
∈
∑
∨
R k K∈
LGDP
1
. . 1
1
0 1 ,
jk
k
k
Tk
k K
jk jk
j Jk j k
jkj J
L U
jk k
k
Min Z c d x
s t Bx b
A x a k Kc
k K
H hx x x
j J k K
c
λ
γ
λ
λ
λ
∈
∈
∈
= +
≥
=⎡ ⎤⎢ ⎥
≥ ∈⎢ ⎥⎢ ⎥=⎣ ⎦
= ∈
≥
≤ ≤≤ ≤ ∈ ∈
∈
∑
∨
∑
R k K ∈
LDP => Integrality λ guaranteed
Proposition. LGDP and LDP have equivalent solutions.
![Page 33: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/33.jpg)
33
Carnegie Mellon
Equivalent Forms in DP Through Basic Steps
tt TF S
∈= ∩
,t T∈ , a polyhedron, .t
t i i ti QS P P i Q
∈= ∪ ∈
Thus the RF is:
where for
There are many forms between CNF and DNF that are equivalent
Regular Form (RF): form represented by intersection of unions of polyhedra
Proposition 1 (Theorem 2.1 in Balas (1979)). Let F be a disjunctive set in RF. Then F
can be brought to DNF by | | 1T − recursive applications of the following basic steps,
which preserve regularity:
For some , , ,r s T r s∈ ≠ bring r sS S∩ to DNF, by replacing it with:
( ).rs
rs i ti Qt Q
S P P∈∈
= ∪ ∩
![Page 34: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/34.jpg)
34
Carnegie Mellon
Illustrative Example: Basic Steps1 2 3F S S S= ∩ ∩
1 11 21( )S P P= ∪ 2 12 22( )S P P= ∪ 3 13 23( )S P P= ∪
12 11 12 11 22 21 12 21 22( ) ( ) ( ) ( )S P P P P P P P P= ∩ ∪ ∩ ∪ ∩ ∪ ∩
11 12 13 11 22 13 21 12 13 21 22 13123
11 12 23 11 22 23 21 12 23 21 22 23
( ) ( ) ( ) ( )( ) ( ) ( ) ( )P P P P P P P P P P P P
SP P P P P P P P P P P P
∩ ∩ ∪ ∩ ∩ ∪ ∩ ∩ ∪ ∩ ∩⎛ ⎞= ⎜ ⎟∪ ∩ ∩ ∪ ∩ ∩ ∪ ∩ ∩ ∪ ∩ ∩⎝ ⎠
Then F can be brought to DNF through 2 basic steps.
which is its equivalent DNF
1 2 3F S S S= ∩ ∩We can then rewrite
12 3as F S S= ∩
1 2 11 21 12 22( ) ( )S S P P P P∩ = ∪ ∩ ∪Apply Basic Step to:
12 3 11 12 11 22 21 12 21 22 13 23(( ) ( ) ( ) ( )) ( )S S P P P P P P P P P P∩ = ∩ ∪ ∩ ∪ ∩ ∪ ∩ ∩ ∪
Apply Basic Step to:
12 3F S S= ∩ 123as F S=We can then rewrite
![Page 35: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/35.jpg)
35
Carnegie Mellon
Equivalent Forms for GDP
1
. .
( )
, ,
jk
k
jk
k
Tk
k K
jk jk
j Jk j k
j J
L U
jk k
k
M in Z c d x
s t B x bY
A x a k Kc
Y k K
Y T ru ex x xY T ru e F a lse j J k K
c
γ
∈
∈
∈
= +
≥
⎡ ⎤⎢ ⎥
≥ ∈⎢ ⎥⎢ ⎥=⎣ ⎦
∨ ∈
Ω =
≤ ≤∈ ∈ ∈
∈
∑
∨
R k K∈ 1
. . 1
1
0 1 ,
jk
k
k
Tk
k K
jk jk
j Jk j k
jkj J
L U
jk k
k
M in Z c d x
s t B x b
A x a k Kc
k K
H hx x x
j J k K
c
λ
γ
λ
λ
λ
∈
∈
∈
= +
≥
=⎡ ⎤⎢ ⎥
≥ ∈⎢ ⎥⎢ ⎥=⎣ ⎦
= ∈
≥
≤ ≤≤ ≤ ∈ ∈
∈
∑
∨
∑
R k K ∈
LGDP LDP
| | | |
0: ( , , ) : ( )k
k K
k
n J Ki i jk jk
i T k K j JF z x c b z b A z aλ ∈
+ +
∈ ∈ ∈
∑⎧ ⎫= = ∈ ∩ ≥ ∩ ∪ ≥⎨ ⎬
⎩ ⎭R
| | | |
0 ˆˆ ˆ: ( , , ) : ( ) ( )
kk K
k n
n J Ki i jk jk mn mn
j J m Ji T k K n KF z x c b z b A z a A z aλ ∈
+ +
∈ ∈∈ ∈ ∈
∑⎧ ⎫= = ∈ ∩ ≥ ∩ ∪ ≥ ∩ ∪ ≥⎨ ⎬
⎩ ⎭R
All possible equivalent forms for GDP, obtained through any number of basic steps,are represented by:
LDP’
![Page 36: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/36.jpg)
36
Carnegie Mellon
Proposition 2 (Theorem 3.3 combined with Corollary 3.5 in Balas (1979)). Let
0, : , n i ii ii Q
F P P x A x a i Q∈
= ∪ = ∈ ≥ ∈R , where Q is an arbitrary set and each
0( , )i iA a is an ( 1)im n× + matrix such that every iP is a bounded non-empty polyhedron.
Furthermore, let ( )Qζ be the set of all those nx ∈R such that there exist vectors1( , ) , i n
iv y i Q+∈ ∈R , satisfying
0
0
0 0
1
i
i Q
i i ii
i
ii Q
x v
A v a y i Qy i Q
y i Q
∈
∈
− =
− ≥ ∈
≥ ∈
= ∈
∑
∑
Then ( ).cl conv F Qζ=
Proposition 3 (Corollary 3.7 in Balas (1979).
Let ( ) : ( ) : 0,1, I iQ x Q y i Qζ ζ= ∈ ∈ ∈ .
Then ( ) .I Q Fζ =
Converting LDP to MIP reformulations
=> MIP representation
iν disaggregated variables
=> Convex Hull
![Page 37: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/37.jpg)
37
Carnegie Mellon
Family of MIP Reformulations For GDP| | | |
0 ˆˆ ˆ: ( , , ) : ( ) ( )
kk K
k n
n J Ki i jk jk mn mn
j J m Ji T k K n KF z x c b z b A z a A z aλ ∈
+ +
∈ ∈∈ ∈ ∈
∑⎧ ⎫= = ∈ ∩ ≥ ∩ ∪ ≥ ∩ ∪ ≥⎨ ⎬
⎩ ⎭R LDP’
MIP’
General template for any MILP reformulation
1
1
1
2 2
0
0
2
. .
ˆ ( , ) ,
ˆ
ˆ
k
n n n
n
n
Tj k jk
k K j J
i iB
i iH
L Ui i i X
mnjk jk S H
m J
mn
m J
i
Min Z y d x
s tb x b i I
h y h i I
x x x i I
y u j k L K I n N
x v n N
b v
γ∈ ∈
∈
∈
= +
≥ ∈
≥ ∈
≤ ≤ ∈
= ∈ ∪ ∪ ∈
= ∈
∑∑
∑∑
2
2
2
0
0
ˆ , ,
ˆ ˆ , ,
ˆ ˆ , ,
ˆ ˆ ( , ) , ,
n
n
k
n
mn imn B n
mnjk mn S n
j J
i mn imn H n
mnjk mn mn n
b y i I m J n N
u y k K m J n N
h u h y i I m J n N
u y j k M m J n N
∈
≥ ∈ ∈ ∈
= ∈ ∈ ∈
≥ ∈ ∈ ∈
= ∈ ∈ ∈
∑
3
ˆ ˆ ( , ) , ,ˆ ˆ ˆ ,
ˆ ˆ0 ( , ) , ,
ˆ 1
ˆ , ,
1
n
n
jk
n jk
k
jk mn jkmn mn n
L mn Umn mn n
mnjk mn n
mnm J
mn jk jk km Q
jkj J
A v a y j k M m J n N
x y v x y m J n N
u y j k L m J n N
y n N
y y n N j J k K
y
∈
∈
∈
≥ ∈ ∈ ∈
≤ ≤ ∈ ∈
≤ ≤ ∈ ∈ ∈
= ∈
= ∈ ∈ ∈
=
∑∑
∑
ˆ 0 , 0,1 ,
mn n
jk k
k K
y m J n Ny j J k K
∈
≥ ∈ ∈∈ ∈ ∈
![Page 38: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/38.jpg)
38
Carnegie Mellon
Particular case: Convex Hull Reformulation of LGDP
. .
,
,
1
0,1
k
k
k
Tj k jk
k K j J
jk
j J
jk jk jkjk k
L jk Ujk jk k
jkj J
jk
Min Z y d x
s t Bx b
x v k K
A v a y j J k K
x y v x y j J k K
y k K
Hy hy
γ∈ ∈
∈
∈
= +
≥
= ∈
≥ ∈ ∈
≤ ≤ ∈ ∈
= ∈
≥∈
∑∑
∑
∑
,kj J k K∈ ∈
Raman and Grossmann I.E. (1994) (CH)
Disaggregated variables
While this MILP formulation has stronger relaxation than big-M, it is not strongest!!
![Page 39: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/39.jpg)
39
Carnegie Mellon
Proposition 4. For | | | 1i T K∈ + − let iGDPF be a sequence of regular forms of the disjunctive set:
| | | |
0 ˆˆ ˆ: ( , , ) : ( ) ( ) ,
kk K
k n
n J Ki i jk jk mn mn
j J m Ji T k K n KF z x c b z b A z a A z aλ ∈
+ +
∈ ∈∈ ∈ ∈
∑⎧ ⎫= = ∈ ∩ ≥ ∩ ∪ ≥ ∩ ∪ ≥⎨ ⎬
⎩ ⎭R such that
i) 0GDPF corresponds to the disjunctive form:
| | | |
0: ( , , ) : ( ) ;k
k K
k
n J Ki i jk jk
i T k K j JF z x c b z b A z aλ ∈
+ +
∈ ∈ ∈
∑⎧ ⎫= = ∈ ∩ ≥ ∩ ∪ ≥⎨ ⎬
⎩ ⎭R
ii)| | | | 1
:T KGDP tF F
+ −= is in DNF;
iii) for 1, , ,i t= … iGDPF is obtained from
1iGDPF−
by a basic step.
Then,
0 1 | | | | 1 | | | | 1 .
T K T KGDP GDP GDP GDP th rel F h rel F h rel F clconv F clconv F+ − + −
− ⊇ − ⊇ ⊇ − = = (true convex hull)
A Hierarchy of Relaxations for GDP
![Page 40: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/40.jpg)
40
Carnegie Mellon
Illustrative Example: Hierarchy of Relaxations
1 2
1 2
1 1
2 2
0.5 01 0
0 10 1 0 1
x xx xx x
x x
− + ≥− − + ≥
= =⎡ ⎤ ⎡ ⎤∨⎢ ⎥ ⎢ ⎥≤ ≤ ≤ ≤⎣ ⎦ ⎣ ⎦
1 2 1 2
1 2 1 2
1 1
2 2
0.5 0 0.5 01 0 1 0
0 10 1 0 1
x x x xx x x x
x xx x
− + ≥ − + ≥⎡ ⎤ ⎡ ⎤⎢ ⎥ ⎢ ⎥− − + ≥ − − + ≥⎢ ⎥ ⎢ ⎥∨⎢ ⎥ ⎢ ⎥= =⎢ ⎥ ⎢ ⎥
≤ ≤ ≤ ≤⎣ ⎦ ⎣ ⎦
Application of 2 Basic Steps
Convex Hull of disjunction
Convex Hull of disjunction
1x
2x
TighterRelaxation!
LP Relaxation
![Page 41: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/41.jpg)
41
Carnegie Mellon
y
xL = ?
W
(0,0)
Set of small rectangles
ij
ji
j
(xi,yi)
j
Numerical Example:Strip-packing problem
Problem statement: Hifi (1998)Given a set of small rectangles with width Hi and length Li.Large rectangular strip of fixed width W and unknown length L.Objective is to fit small rectangles onto strip without overlap and rotation while minimizing length L of the strip.
![Page 42: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/42.jpg)
42
Carnegie Mellon
. . i i
Min ltst lt x L≥ +
1 2 3 4
, ,
ij ij ij ij
i i j j j i i i j j j i
i i i
i N
Y Y Y Yi j N i j
x L x x L x y H y y H y
x UB L
∀ ∈
⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤∨ ∨ ∨ ∀ ∈ <⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥
+ ≤ + ≤ − ≥ − ≥⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎣ ⎦ ⎣ ⎦ ⎣ ⎦ ⎣ ⎦≤ −
i i
i NH y W i N
∀ ∈
≤ ≤ ∀ ∈1 1 2 3 4
, , , , , , , , , i i ij ij ij ijlt x y Y Y Y Y True False i j N i j+∈ ∈ ∀ ∈ <R
GDP/DP Model forStrip-packing problem
Objective functionMinimize length
Disjunctive constraintsNo overlap between rectangles
Bounds on variables
. . i i
Min ltst lt x L≥ +
1 2 3 4
1 2 3 4
1 1 1 1 , ,
1 , ,
ij ij ij ij
i i j j j i i i j j j i
ij ij ij ij
i
i N
i j N i jx L x x L x y H y y H y
i j N i j
x U
λ λ λ λ
λ λ λ λ
∀ ∈
⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤= = = =∨ ∨ ∨ ∀ ∈ <⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥
+ ≤ + ≤ − ≥ − ≥⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎣ ⎦ ⎣ ⎦ ⎣ ⎦ ⎣ ⎦+ + + = ∀ ∈ <
≤
i i
i i
B L i NH y W
− ∀ ∈
≤ ≤1 1 2 3 4
, , , 0 , , , 1 , , i i ij ij ij ij
i N
lt x y i j N i jλ λ λ λ+
∀ ∈
∈ ≤ ≤ ∀ ∈ <R
![Page 43: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/43.jpg)
43
Carnegie Mellon
25 Rectangle Problem Optimal solution= 31
Original CH1,112 0-1 variables4,940 cont vars7,526 constraintsLP relaxation = 9
Strengthened1,112 0-1 variables5,783 cont vars8,232 constraintsLP relaxation = 27!
=>
31 Rectangle Problem Optimal solution= 38
Original CH2,256 0-1 variables9,716 cont vars14,911 constraintsLP relaxation = 10.64
Strengthened2,256 0-1 variables11,452 cont vars15,624 constraintsLP relaxation = 33!
=>
![Page 44: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/44.jpg)
44
Cutting Planes for LinearGeneralized Disjunctive Programming
Min Z = + hTx Objective Function
s.t. Bx ≤ b Common Constraints
Ω(Y) = True Logic Constraints
x∈ Rn, Yjk∈ True, False, ck∈ R
j∈ Jk , k∈ K
=
≤
jkk
jkjk
jk
c
axA
Y
γ
∨∈ kJj
∑∈ Kk
kc
k ∈ K
GDP Model:
OR Operator
Boolean Variables
Disjunctive Constraints
Sawaya, Grossmann (2004)
![Page 45: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/45.jpg)
45
Motivation for Cutting Plane Method
Trade-off: Big-M fewer vars/weaker relaxation vs Convex-Hull tighter relaxation/more vars
Big-MRelaxed Feasible Region
x2
x1Convex Hull
Relaxed Projected Feasible Region
Strengthened Big-MRelaxed Feasible Region
Cutting Plane(x - xSEP)T(xSEP - xR
BM) ≥ 0
xSEP
xRBM
![Page 46: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/46.jpg)
46
Global Optimization Algorithms
• Most algorithms are based on spatial branch and bound method (Horst & Tuy, 1996)
•Nonconvex NLP/MINLPαBB (Adjiman, Androulakis & Floudas, 1997; 2000)
BARON (Branch and Reduce) (Ryoo & Sahinidis, 1995, Tawarmalani and Sahinidis (2002))
OA for nonconvex MINLP (Kesavan et al., 2004)
Branch and Contract (Zamora & Grossmann, 1999)
•Nonconvex GDPTwo-level Branch and Bound (Lee & Grossmann, 2001)
![Page 47: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/47.jpg)
47
Spatial Branch and Bound to
obtain the Global Optimum Guaranteed to converge to global optimum given a certain tolerance between
lower and upper bounds
![Page 48: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/48.jpg)
48
Obj
ectiv
e
Multiple minima
Lower bound
LB
LB
LBUB = Upper bound
LB < UB
LB > UB LB < UB
Global optimum search Branch and bound tree
![Page 49: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/49.jpg)
49
Nonconvex GDP
( )
Ω
,0)(
0)(
)(min
1
falsetrue,YRc,Rx
trueY
K k γc
xgY
Jj
xs.t. r
xfc Z
jk
k
n
jkk
jk
jk
k
kk
∈
∈∈=
∈⎥⎥⎥
⎦
⎤
⎢⎢⎢
⎣
⎡
=
≤∈
≤
∑ +=
∨
Objective Function
Common Constraints
Disjunctions
Logic Propositions
OR operator
f, g and r: nonconvex
![Page 50: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/50.jpg)
50
• Introducing convex underestimators
Convex Underestimator GDP (R)
( )
convexgandrf
falsetrue,YRc,Rx
trueY
K k γc
xg
Y
J
xrs.t.
xfc Z
jk
k
n
jkk
jk
jk
k
kk
j
:,
Ω
,0)(
0)(
)(min
1
∈
∈∈=
∈⎥⎥⎥
⎦
⎤
⎢⎢⎢
⎣
⎡
=
≤
≤
∑ +=
∈∨
Convex underestimatorsBilinear: LinearMcCormick (1976), Al-Khayyal (1992)
Linear fractional: Convex nonlinearQuesada and Grossmann (1995)
Concave separable: Linear secant
Problem (R) yields a valid lower bound to Problem (GDP)
![Page 51: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/51.jpg)
51
Convex envelopesConcave function
xba
Secant g(x)
[ ( ) ( )]( ) ( ) ( )f b f ag x f a x ab a
−= + −
−
f(x)
![Page 52: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/52.jpg)
52
Bilinearw = xy
L U L Ux x x y y y≤ ≤ ≤ ≤
L L L L
U U U U
L U L U
U L U L
w x y y x x y
w x y y x x y
w x y y x x y
w x y y x x y
≥ + −
≥ + −
≤ + −
≤ + −
McCormick convex envelopes
For other convex envelopes/underestimators see:Tawarmalani, M. and N. V. Sahinidis, Convexification and Global Optimization in Continuous and Mixed-Integer Nonlinear Programming: Theory, Algorithms, Software, and Applications, Vol. 65, Nonconvex Optimization And Its Applications series, Kluwer Academic Publishers, Dordrecht, 2002
![Page 53: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/53.jpg)
53
Basic Ideas Global Optimization GDP
1. Branch and bound enumeration on disjunctionsof convex GDP (R)
Yj ¬Yj
Disjunctive B&B
Spatial B&B
Feasible discrete
2. When feasible discrete solution foundswitch to spatial branch and bound (NLP subproblem)
![Page 54: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/54.jpg)
54
S S
ABC
Mixing CrystallizationReaction Drying
1 2 3 4
ABC
S
Unit 1 Unit 2 Unit 3 Unit 4 Unit 5Cast Iron
w/ AgitatorStainless Steel
w/ AgitatorCast IronJacketed
Stainless SteelJacketed w/ Agitator
TrayDryer
More than 100 alternatives: each requires nonlinear optimization
Equipment
Tasks
Synthesis Multiproduct Batch Plant(Birewar & Grossmann, 1990)
![Page 55: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/55.jpg)
55
∑ ≤
=≥
==∑=
==≥
∑+∑=
=
∈
=
p
j
N
iLii
Piii
PTt
itjij
Piti
T
t
jj
M
jj
EQ
j
HTn
NiQBn
MjNiptyptTtNiSBVts
CSCN
1
1
,...,1
,...,1;,...,1,...,1;,...,1..
COSTmin
Synthesis Multiproduct Batch Plant
Tt
jjptyptpty
VVY
Jj
itj
T
ititj
T
tj
tj
t
∈
⎥⎥⎥⎥⎥
⎦
⎤
⎢⎢⎢⎢⎢
⎣
⎡
≠=
=
≥
∈∨
',0'
Objective function
Process time
Disjunction forTask Assignments
Nonconvexfunctions
Sizing
Demand
Horizon time
Nonconvex GDP Model
![Page 56: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/56.jpg)
56
,,,,;,,,,,,,,0
)()()()()(
,,,
34241434
34241434241424
34241434241434241414
34241404
34241404
4553424144
33322122111
falsetrueWYCYYEXptyNptTBnVVCYYYW
YYYYYYWYYYYYYYYYW
YYYWWWWW
YYEXYYYYEXYYEXYYYEXYYEX
ljcjtjtjitj
EQ
jijLiii
T
tjj ∈≤
∧∧⇔
∧∧¬∨¬∧∧⇔
∧¬∧¬∨¬∧∧¬∨¬∧¬∧⇔
¬∧¬∧¬⇔
∨∨∨
⇔∨∨⇔
⇔∨⇔⇔
Jj
CSVST
BBYS
VSTCSVST
NEQBSVSTNEQBSVST
BBYS
Jj
Tpt
NVC
YEX
ptTN
YC
ptTN
YC
ptTN
YC
ptTN
YC
VVVVC
YEX
j
j
ijij
j
jj
j
jijijj
jijijj
ijij
j
Li
ij
EQ
j
j
j
j
ijLi
EQ
j
j
ijLi
EQ
j
j
ijLi
EQ
j
j
ijLi
EQ
j
j
U
jj
L
j
jjjj
j
∈
⎥⎥⎥⎥⎥
⎦
⎤
⎢⎢⎢⎢⎢
⎣
⎡
=
=
=
¬
∨
⎥⎥⎥⎥⎥⎥⎥⎥
⎦
⎤
⎢⎢⎢⎢⎢⎢⎢⎢
⎣
⎡
+=
≤≤
≥
≥
≤−≤−
∈
⎥⎥⎥⎥⎥⎥⎥⎥
⎦
⎤
⎢⎢⎢⎢⎢⎢⎢⎢
⎣
⎡
≥
=
=
=
=
¬
∨
⎥⎥⎥⎥⎥⎥⎥⎥
⎦
⎤
⎢⎢⎢⎢⎢⎢⎢⎢
⎣
⎡
⎥⎥⎥
⎦
⎤
⎢⎢⎢
⎣
⎡
≥
=∨⎥⎥⎥
⎦
⎤
⎢⎢⎢
⎣
⎡
≥
=∨⎥⎥⎥
⎦
⎤
⎢⎢⎢
⎣
⎡
≥
=∨⎥⎥⎥
⎦
⎤
⎢⎢⎢
⎣
⎡
≥
=
≤≤
+=
00
80500010000100
''
000
00
44
33
221
'
5.0
''
'
4321
6.0
φφ
αγ
Disjunction for Equipment
Disjunction forStorage Tank
Logic Propositions
GDP model (continued)
![Page 57: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/57.jpg)
57
Proposed Algorithm for Nonconvex GDP
Nonconvex MINLPStep 0
ZU
OA (Viswanathan and Grossmann, 1990)
BoundContraction
New Bound
Step 1 (Zamora and Grossmann, 1999)
BB with Y’sUpdate ZL
Spatial BB
When solution is Integral
Update ZU
Stop whenZL ≥ ZU
Add Integer Cut
Fixed Y’s
Step 2
Step 3
(Lee and Grossmann, 2000)
(Quesada and Grossmann, 1995)
![Page 58: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/58.jpg)
58
Upper Bound Solution
• Use 4 Stages (6 units) without Storage Tank
j = 1 j = 4 j = 5
Mixing Reaction
ABC
V1 = 4,842 L V2 = 2,881 L V4 = 2,469 L V5 = 8,071 L
DryingCrystallization
ABC
j = 2
2
8
4
9
A 243 batches, 4.5hrs
2
4
3
12
B 260 batches, 6hrs
7
4
9
3
C 372 batches, 9hrs 6000 hrs
1093 hrs 1562 hrs 3345 hrs
Cost = $ 277,928 (by GAMS/DICOPT++)
![Page 59: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/59.jpg)
59
Optimal Solution: Multiproduct Batch Plant
S
Mixing Reaction Storage Tank
ABC
j = 2 j = 3 j = 5V2 = 4,309 L VST2 = 4,800 L V3 = 3,600 L V5 = 11,753 L
DryingCrystallization
ABC
Global optimal cost = $ 264,887 (5% improvement)3 Stages + 1 storage tank (5 units) (43 nodes, 48 sec)
9 12 3167 batches, 9hrs
184 batches, 12 hrs
255 batches, 9hrs
10
4
A 250 batches, 5hrs
6
3
B 293 batches, 3hrs
11
9
C 418 batches, 5.5hrs 6000 hrs
Storage 1503 hrs 2202 hrs 2295 hrs
![Page 60: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/60.jpg)
60
Carnegie Mellon
∑∈
+=Kk
kcxfZ )(
Global Optimization of Bilinear Generalized Disjunctive Programs
eTrue,FalsR,Y,cRx ikkn ∈∈∈
Ω(Y)= True
i Dk , k K ∈∈
Disjunctions
Logic Propositions
Objective Function
0)( ≤xg
kDi∈∨
⎥⎥⎥
⎦
⎤
⎢⎢⎢
⎣
⎡
=≤
ikk
ik
ik
cxrY
γ0)(
Min
s.t. Global Constraints
Bilinearities
Bilinearities may lead to multiple local minima → Global Optimization techniquesare required
Relaxation of Bilinear terms using McCormick envelopes leads to a LGDP → Improved relaxations for Linear GDP has recently been obtained (Sawaya & Grossmann, 2007)
k K ∈
Juan Ruiz
![Page 61: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/61.jpg)
61
Carnegie Mellon
Guidelines for applying basic steps in Bilinear GDP
• Replace bilinear terms in GDP by McCormick convex envelopes (LGDP)• Apply basic steps between those disjunctions with at least one variable in
common. The more variables in common two disjunctions have the more the tightening can be expected
• If bilinearities are outside the disjunctions apply basic steps by introducing them in the disjunctions previous to the relaxation.If bilinearities are inside the disjunctions a smaller tightening effect is expected.
• A smaller increase in the size of the formulation is expected when basic steps are applied between improper disjunctions and proper disjunctions.
![Page 62: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/62.jpg)
62
Carnegie Mellon
Step 3: Branch and Bound Procedure (Lee & Grossmann, 2001)
Step 1: GDP reformulation (Apply basic steps following the rules presented)
Intersecting disjunctionsContracting
Bounds
Spatial B&B
Step 2: Bound Contraction (Zamora & Grossmann, 1999)
Methodology
![Page 63: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/63.jpg)
63
Carnegie Mellon
Process superstructure
Min Z = ∑∈PUk
kCP
∑∈
=kMi
ji
jk ff j∀ MUk ∈
∑∈
=kSi
jk
ji ff j∀ SUk ∈
∑∈
=kSi
ki 1ζ SUk ∈
jk
ki
ji ff ζ= j∀ kSi ∈ SUk ∈
jk
ji ff ,0 ≤ kji ,,∀
10 ≤≤ kiζ kj,∀
kDh ∈∀ PUk ∈∀, falsetrueYP hk ∈
kCP≤0
s.t.
⎥⎥⎥⎥⎥
⎦
⎤
⎢⎢⎢⎢⎢
⎣
⎡
∂=
∈=∀∈∈=
∨ ∑∈
kikk
kj
jik
kkj
ijh
kj
i
hk
Dh
FCP
OPUifFjIPUiOPUiff
YP
k ,,',,'β
PUk ∈
k∀
Generalized Disjunctive Program
A/B/C
G/H/I
S4
S5
S6
M1
M2
M3
S1
S2
S3
M4D/E/F
Optimal structure
Z* = 1.214
A S4
S5
M1
M2
S1
S2
S3
M4D
N of cont. vars. : 114N of disc. vars. : 9N of bilinear terms: 36
Case Study I: Water treatment network design
![Page 64: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/64.jpg)
64
Carnegie Mellon
⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥
⎦
⎤
⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢
⎣
⎡
=
=
∈∈∀=
∈∀=
≤
∑∑
∑ ∑∑ ∑
∈
∈
∈ ∈
∈ ∈
jj
Kk
kj
Iiijw
kjjkw
Kk Iiijwjkw
Ii Wwijw
loj
CP
KkWwff
Wwff
ffYP
γ
ζ
ζ
1
,,
,
∑ ∑ ∑∑ ∑ ∑ ∑ ∑∈ ∈ ∈ ∈ ∈ ∈ ∈ ∈
−++Jj Ii Jj Ii Ww Kk Jj Ww
jkwkijwijij fdfcCSTCP
∑ ∑∑ ∑∈ ∈∈ ∈
=Kk Ww
jkwIi Ww
ijw ff Jj ∈∀
0=−∑ ∑∈ ∈
kJj Ww
jkw Sf Kk ∈∀
∑∈
=Ww
ijwiwijw ff'
'λ WwJjIi ∈∀∈∀∈∀ ,,
∑ ∑ ∑∈ ∈ ∈
=−Jj Jj Ww
jkwkwjkw fZf'
' 0 WwKk ∈∀∈∀ ,
Min Z =
s.t.
∨
⎥⎥⎥⎥
⎦
⎤
⎢⎢⎢⎢
⎣
⎡
=
≤ ∑ ∑∈ ∈
ii
Jj Wwijw
loi
CST
ffYST
α ⎥⎥⎥
⎦
⎤
⎢⎢⎢
⎣
⎡
==
¬
00
i
ijw
i
CSTf
YSTIi ∈∀
∨
⎥⎥⎥⎥⎥⎥
⎦
⎤
⎢⎢⎢⎢⎢⎢
⎣
⎡
=∈∈∀=
∈∈∀=¬
0,,0
,,0
j
jkw
ijw
j
CPWwKkf
WwIifYP
Jj ∈∀
upijwjkw
kj fff ≤≤≤≤ ,0;10 ζ
,,;,0 falsetrueYPYSTCPCST jiji ∈≤
Generalized Disjunctive ProgramProcess superstructure
S1
S2
S3
S4
S5
P1
P2
P3
P4
1
2
3
Stream i Pool j Product k
Optimal structure
Z* = -4.640
S1
S2
S5
P1
P3
1
2
3
Stream i Pool j Product k
N of cont. vars. : 76N of disc. vars. : 9N of bilinear terms: 24
Case Study II: Pooling network design
![Page 65: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/65.jpg)
65
Carnegie Mellon
51% 51% 204 399 Nodes
99.7%Bound contraction
24.90%499.86400.66Initial Lower BoundExample 1
Relative Improvement
Global OptimizationTechnique using
proposed relaxation
Global OptimizationTechnique using Lee &Grossmann relaxation
9%9%683 748 Nodes
8%Bound contraction
0.90%-5468-5515Initial Lower BoundExample 2
Relative Improvement
Global OptimizationTechnique using
proposed relaxation
Global OptimizationTechnique using Lee &Grossmann relaxation
Performance
![Page 66: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/66.jpg)
66
Carnegie Mellon
Conclusions
Unified Linear GDP with Disjunctive Programming- Developed DP equivalent formulation for GDP- Developed a family of MIP reformulations for GDP- Developed a hierarchy of relaxations for GDP- Numerical results have shown great improvement in lower bound
for strip packing problem
Nonconvex GDPs- Spatial branch and bound methods can be developed- Tighter lower bounds can be obtained in bilinear problems by
applying basic steps
GDP modeling framework- Provides a logic-based framework for discrete-continuous optimization- big-M and convex hull alternative formulations different relaxations- Solution methods: reformulation, branch and bound, decomposition
![Page 67: Generalized disjunctive programming: A framework …...Ignacio E. Grossmann Center for Advanced Process Decision-making Carnegie Mellon University Pittsburgh, PA 15213, USA Generalized](https://reader030.vdocuments.mx/reader030/viewer/2022040916/5e8edf2095766c2f7a27fbae/html5/thumbnails/67.jpg)
67
Carnegie Mellon
Open Cyberinfrastructure for Mixed-integer Nonlinear Programming: Collaboration and Deployment via Virtual Environments
CMU: Grossmann, Biegler, Belotti, Cornuejols, Margot, Ruiz, SahinidisIBM: Lee, Wächter
(a) Create a library of optimization problems in different application areas in which one or several alternative models are presented with their derivation. In addition, each model has one or several instances that can serve to test various algorithms.
(b) Provide a mechanism for researchers and users to contribute towards the creation of the library ofoptimization problems.
(c) Provide a forum of discussion for algorithm developers and application users where alternativeformulations can be discussed as well as performance and comparison of algorithms.
(d) Provide information on MINLP tutorials and bibliography to disseminate this information.
General Goals
Major emphasisCollect optimization problems in which alternative model formulations aredocumented with corresponding computational results(engineering, finance, operations management, biology)