15.081J/6.251J Introduction to Mathematical Programming Lecture 2: Geometry of Linear Optimization I 1 Outline: 1. 2. 3. 4. 5. What is the central problem? Standard Form. Preliminary Geometric Insights. Geometric Concepts (Polyhedra, \Corners"). Equivalence of algebraic and geometric concepts. 2 Central Problem minimize subject to cx ai x = ai x ai x Slide 2 0 0 0 0 2 M1 2 M2 i 2 M3 j 2 N1 j 2 N2 i i bi b i b i j0 > j <0 x x 2.1 Standard Form Slide 3 cx Ax = b x0 minimize subject to 0 Characteristics Slide 1 Minimization problem Equality constraints Non-negative variables 2.2 Transformations max c x 0 ai x 0 ai x 0 > xj < 0 ; min(;c x) ai x + si = bi si 0 0 i 0 b i b , ai x ; i = 0 s i b j = xj ; xj ; j 0 xj 0 + x + x 1 ; i0 s Slide 4 2.3 Example maximize subject to x1 Slide 5 ; x2 + 21 +2 2 1 > 1 <0 2 0 x1 x x1 x x x + ;minimize ;x+1 + x1 + x2 subject to x+1 ; x1 + x2 + s1 =1 + x1 ; x1 + 2x2 ; s2 = 1 + x1 x1 x2 s1 s2 0 ; ; ; ; 3 Preliminary Insights Slide 6 minimize ; 1 ; 2 subject to 1 + 2 2 3 2 1+ 23 1 20 x x x x x x x x x2 3 - x1 - x2 = - 2 1.5 - x1 - x2 = z (1,1) - x1 - x2 = 0 3 1.5 c x1 x1 + 2x2 < 3 2x1 + x2 < 3 ;x1 + x2 1 x1 0 x2 0 There exists a unique optimal solution. There exist multiple optimal solutions in this case, the set of optimal solutions can be either bounded or unbounded. 2 Slide 7 Slide 8 x2 c = (1,0) c = (- 1,- 1) 1 c = (0,1) c = (1,1) x1 The optimal cost is ;1, and no feasible solution is optimal. The feasible set is empty. 4 Polyhedra 4.1 Denitions Slide 9 The set fx j a x = g is called a hyperplane. The set fx j a x g is called a halfspace. The intersection of many halfspaces is called a polyhedron. A polyhedron is a convex set, i.e., if x y 2 , then x + (1 ; )y 2 . 0 b 0 b P P = b3 x= a '2 a '3x b2 a3 a2 a' x < b a1 1 a 4' x = b4 =b =b a4 a5 a a5' x = b 5 (b) (a) 3 1x a' x a' x > b a' P 5 Corners 5.1 Extreme Points Slide 10 Polyhedron P = fx j Ax bg x 2 P is an extreme point of P if 6 9 y z 2 P (y = 6 x z 6= x): x = y + (1 ; )z 0 < < 1 . . . u w v P . . . z x y 5.2 Vertex Slide 11 x 2 is a vertex of if 9 c: x is the unique optimum P P minimize subject to 5.3 Basic Feasible Solution P = f( x1 x2 x3 )j x1 + x2 + cy y2 0 x3 P =1 x1 x2 x3 0g Slide 12 Slide 13 Points A,B,C : 3 constraints active Point E: 2 constraints active suppose we add 2 1 + 2 2 + 2 3 = 2. x x x 4 . w {y | c 'y =c 'w } P c . c x 'x } 'y = c {y | c x3 . A E D . . . P x2 C . B x1 Then 3 hyperplanes are tight, but constraints are not linearly independent. Intuition: a point at which inequalities are tight and corresponding equations are linearly independent. = fx 2 <n j Ax bg Slide 14 n P a1 : : : am rows of A x2P I = fi j ai x = bi g 0 Denition x is a basic feasible solution if subspace spanned by fai 2 g i is <n . 5.3.1 Degeneracy If jI j = n, then ai i 2 I are linearly independent x nondegenerate. 5 I Slide 15 If jI j > n, then there exist n linearly independent fai i 2 I g x degener- ate. A D C P B E (a) 5.3.2 Example (b) min st : : + 52 1+ 2+ x1 x x x Slide 16 ;2x3 4 2 3 6 0 x3 x1 3 2+ x1 x3 x x3 x2 x3 6 Equivalence of denitions Slide 18 Theorem: = fx j Ax bg. Let x 2 . P Slide 17 P x is a vertex , x is an extreme point , x is a BFS. 6.1 Proof 1. Vertex ) extreme point 9c : c x c y 8y 2 If x is not an extreme point 9y z 6= x: x = y + (1 ; )z . But c x c y c x c z ) c x = c y + (1 ; )c z c x contradiction 0 < 0 Slide 19 P 0 0 0 0 0 < < 0 < 0 0 2. Extreme point ) BFS Suppose x is not a BFS. 6 Slide 20 Let = f : ai x = ig. But ai do not span all of <n ) 9z 2 <n : ai z = 0 2 Let x1 = x + z x2 = x ; z ) aix1 = i aix2 = i 2 I 0 i b 0 i I 0 b 0 i i b I 62 I : aix < bi ) ai (x + z ) < bi 0 ai (x ; z) 0 0 Slide 21 i < b for small enough. ) x1 x2 2 : yet x = x1 +2 x2 ) x not an extreme point: contradiction 3. BFS ) vertex x BFS = f : ai x = i g Let i = 10 262 c = ;d A m P P P . Then c x = ;d Ax = ; iai x = ; ai x = ; i i�1 i I i I But 8x 2 P : ai x i ) x optimum min cx c x = ; P aix ; P i = c x P Slide 22 � I 0 i � b d i I i I: 0 � 0 0 0 � 0 0 d � 0 � b 2 2 b � 0 i2I 0 Why unique? i2I b 0 � 0 x2 7 P: Slide 23 Equality holds if ai x = i 2 since ai spans <n solution x = x . 0 b i I � 8 aix = i has a unique 0 b MIT OpenCourseWare http://ocw.mit.edu 6.251J / 15.081J Introduction to Mathematical Programming Fall 2009 For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.