A TEXT FOR NONLINEAR PROGRAMMING Thomas W. Reiland Department of Statistics North Carolina State University Raleigh, NC 27695-8203 Email: reiland@ncsu.edu Table of Contents Chapter I: Optimality Conditions..................................................Error! Bookmark not defined. § 1.1 Differentiability.................................................................Error! Bookmark not defined. § 1.2. Unconstrained Optimization ............................................Error! Bookmark not defined. § 1.3 Equality Constrained Optimization.................................................................................... 0 § 1.3.1 Interpretation of Lagrange Multipliers......................................................................... 10 § 1.3.2 Second order Conditions – Equality Constraints ......................................................... 11 § 1.3.3 The General Case ......................................................................................................... 13 § 1.4 Inequality Constrained Optimization ...............................Error! Bookmark not defined. § 1.5 Constraint Qualifications (CQ) ........................................Error! Bookmark not defined. § 1.6 Second-order Optimality Conditions ...............................Error! Bookmark not defined. § 1.7 Constraint Qualifications and Relationships Among Constraint Qualifications ..... Error! Bookmark not defined. Chapter II: Convexity ...................................................................Error! Bookmark not defined. § 2.1 Convex Sets ......................................................................Error! Bookmark not defined. § 2.2 Convex Functions ............................................................Error! Bookmark not defined. § 2.3 Subgradients and differentiable convex functions ............Error! Bookmark not defined. 1.3 Equality Constrained Optimization page 1.3-1 § 1.3 Equality Constrained Optimization Recall that f ( x ) points in the direction from x in which the instantaneous rate of increase of f is the greatest; the length of f ( x ) is equal to that maximum rate of increase. Example 1.2 Continued: f ( x) x12 x22 f ( x) x12 x22 k provides circular contours 2x f ( x) 1 2 x2 2 2 4 At x , f 2 4 2 f (2) 4 2 Notice that f ( x) is perpendicular to any direction in which the instantaneous rate of change of f is zero; f ( x) is perpendicular to all surfaces of the form f ( x) k , a constant. Equality Constrained Problem: Max (or min) f ( x) , x E n s.t. gi ( x) bi , i 1,..., m (m n) f, gi’s have continuous 1st partial derivatives This problem can reduce to the unconstrained problem by solving the constraints for m variables in terms of the other n-m variables and substituting into the objective function. Though an important idea, equality constraints have little theoretical impact on the problem as they only reduce the dimensionality of the problem; they do not establish boundaries. Example 1-9: Min x12 x22 x32 s.t. x1 x2 x3 10 The feasible region is a plane in E 3 framed by the triangle. The solution is “constrained to the plane” but not bounded. Definition 1.3.1 A point x satisfying the constraints gi ( x) bi is a regular point of the constraints if g1 ( x ), g 2 ( x ), , g m ( x ) are linearly independent. In other words, x is a regular point if the Jacobian matrix J evaluated at x is of full rank (m). 1.3 Equality Constrained Optimization g1 ( x ) x1 J ( x ) g m ( x ) x 1 page 1.3-2 g1 ( x ) xn g m ( x ) xn Theorem 1.5 Lagrange Multipliers Let x E n be any local maximum or minimum of f ( x) s.t. gi ( x) bi , i 1, ,m , m<n. If x is a regular point of the constraints, the there exists a unique set of Lagrange multipliers , 1 m , such that f ( x ) igi (x ) 0 . m i 1 Remark: Define the Lagrangian function by F ( x, ) f ( x) i [ gi ( x) bi ] . Then the necessary conditions of Theorem 1.5 become: x F ( x , ) f ( x ) i gi ( x ) 0 F ( x , ) g ( x ) b 0 Or, putting these together yields: F ( x , ) 0, where F is an (m n) 1vector. Implicit Function Theorem Let x 0 ( x10 ,..., xn0 )T E n satisfy the following: (i) The functions gi are continuously differentiable in a neighborhood of x 0 . (ii) gi ( x 0 ) bi , i 1,..., m (iii) The m n Jacobian is nonsingular ( x 0 E n is a regular point of the constraints). Then there exists a neighborhood of xˆ 0 ( xm0 1 ,..., xn0 )T E n m such that for xˆ ( xm 1 , xm 2 ,..., xn )T in this neighborhood there exist functions i ( xˆ ), i 1, 2,..., m such that (I) i are continuously differentiable (II) xi0 i ( xˆ 0 ), i 1, 2,..., m (the i are implicit functions) (III) gi (1 ( xˆ ), 2 ( xˆ ), , m ( xˆ ), xˆ ) bi , i 1, 2, , m Example 1-10: Min f ( x) x12 x22 s.t. 3x1 4 x2 6 n=2 m=1 Solve for an implicit function: x1 6 4 x2 3 6 4 x2 2 fˆ ( x2 ) x2 3 2 Substitute into the objective function: 1.3 Equality Constrained Optimization page 1.3-3 Now the problem can be readily solved using the first derivative (unconstrained). Example 1-11: Given the constraint x12 x22 0 , n = 2, m = 1, and let x 0 (0, 0)T Condition (iii) is violated since J ( x0 ) 011 which is singular. Therefore, no exist such that x1 ( x2 ) . x1 x22 i x22 ix2 Proof of Theorem 1.5: Since x is a regular point of the constraints, the Implicit Function Theorem implies that there exists continuously differentiable 1 , 2 , , m such that (1) xi i ( xˆ ), i 1, where xˆ ( xm 1 , xm 2 , (2) ,m , xn )T , the objective function f ( x ) becomes fˆ ( xˆ ) f (1 ( xˆ ), 2 ( xˆ ), , m ( xˆ ), xˆ ) . This function has an unconstrained local maximum or minimum at xˆ E n m so (3) fˆ ( xˆ ) 0, j m 1, m 2, x j ,n Apply the chain rule for partial differentiation to (2): m fˆ f k f at x̂ for j m 1, , n . 0 x j k 1 xk x j x j f f Note: Use instead of because using (1) they are equivalent. xk k (4) By the Implicit Function Theorem, gi (1 ( xˆ ), 2 ( xˆ ), , m ( xˆ ), xˆ ) bi , i 1, , m , at x̂ and all x̂ in a neighborhood of x̂ ; thus the partial derivatives of gi with respect to the remaining components xm 1 , , xn vanish at all such points (function in constant in neighborhood): (5) gi m gi k 0, i =1, x j k 1 xk x j , m; j m 1, ,n . Since x is a regular point, define the unique numbers 1 , 2 , in (6). , m by means of the equations 1.3 Equality Constrained Optimization (6) page 1.3-4 f ( x ) m gi ( x ) i , j 1, x j x j i 1 ,m For a specific j, (5) represents m equations (one equation for each i); multiply each equation by the respective i and sum over i to obtain (7). m (7) i i 1 gi m k x j k 1 x j m gi i i 1 x j 0, j m 1, ,n . Evaluating (7) at x and subtracting from (4) gives f ( x ) m gi ( x ) k ( x ) f ( x ) m gi ( x ) i i 0, j m 1, xk x j x j x j k 1 xk i 1 i 1 m ,n . But from (6), the term in brackets is zero for every k, leaving f ( x ) m gi ( x ) i 0, j m 1, x j x j i 1 ,n We combine this with (6) to get the desired result: m f ( x ) i gi ( x ) 0 . QED i 1 ______________________________________________________________________________ Problem P : Min f ( x) , f has continuous 1st partials on some open subset containing Ω x E n Definition A vector d E n is a feasible direction at x if there exists a 0 such that x d , , 0 . Theorem If x is a local minimum point of f over Ω, then for any feasible direction d E n , f ( x )T d 0 . Proof: Suppose f ( x )T d 0 for some feasible d. From Def. of Differentiability, f ( x d ) f ( x ) f ( x )T d d ( x d ; x ) for 0 . f (x d ) f (x ) f ( x )T d d ( x d ; x ) 1.3 Equality Constrained Optimization lim f (x d ) f (x ) 0 page 1.3-5 f ( x )T d 0 f ( x d ) f ( x ) 0 for τ sufficiently small. f ( x d ) f ( x ) for τ sufficiently small. # This contradicts the fact that x is a local minimum. QED. Remark: If x int() or if E n , then f ( x )T d 0 for d E n implies f ( x ) 0 (Theorem 1.13). Example 1-12: f ( x) x 2 , [1, 2] d E1 is feasible at x 1 if d 0 x 1 is a global minimum on Ω f ( x ) f ( x ) 2 x 2 at x 1 f ( x )T d f ( x )d 2d 0 for d feasible Example 1-13: Min f ( x1 , x2 ) x12 x1 x2 x1 x2 , ( x1 , x2 )T E 2 | x1 0, x2 0 1 f has a global minimum at x1 , x2 0 2 0 2 x1 1 x2 f ( x ) 3 f ( x) x 1 1 2 3 d T d 1 E 2 is feasible at x if d 2 0 since f ( x ) d d 2 0 2 d2 ( d1 can have any value) Theorem – 2nd Order Necessary Conditions Let f have continuous second partial derivatives on an open set containing E n . If x is a local minimum of f over Ω, then for any feasible direction d E n at x we have (4a) f ( x )T d 0 (5a) if f ( x )T d 0 , then d T 2 f ( x )d 0 . Remark: If x int() or E n , then (5a) becomes (5) in Theorem 1.3: d T 2 f ( x )d 0 , d E n . Proof: (4a) already proven in previous proof. Suppose d T 2 f ( x )d 0 for some feasible direction d where f ( x )T d 0 . 1.3 Equality Constrained Optimization page 1.3-6 f ( x d ) f ( x ) f ( x )T d d T 2 f ( x ) d d f (x d ) f (x ) lim 2 d T 2 f ( x )d d f (x d ) f (x ) 0 2 2 2 (x d ; x ) (x d ; x ) d T 2 f ( x )d 0 f ( x d ) f ( x ) 0 for τ small enough f ( x d ) f ( x ) for τ small enough. # QED. Example 1-13 Continued: Min f ( x1 , x2 ) x12 x1 x2 x1 x2 , ( x1 , x2 )T E 2 | x1 0, x2 0 T 1 0 Global Minimum occurs at x 2 3 f ( x )T d d 2 (5a) applies when d 2 0 2 2 1 2 f ( x) 2 f ( x ) 1 0 d T 2 f ( x )d 2d12 2d1d 2 2d12 0 when d 2 0 . Example 1-14: Min f ( x1 , x2 ) x13 x12 x2 2 x22 , x E 2 | x1 0, x2 0 Suppose solution is in interior of Ω (ignore constraint). 3x12 2 x1 x2 0 6 f ( x) 2 0 x or x 0 9 x1 4 x2 18 12 2 f ( x ) which is not positive definite 12 4 18 12 det 2 f ( x ) I det (18 )(4 ) 144 0 12 4 2 22 72 0 22 484 288 772 28 11 11 2 2 2 2 f ( x ) is indeterminant and x is not a local minimum. 6 x 2 x2 2 f ( x) 1 2 x1 2 x1 4 Example 1-15: Max f ( x1 , x2 ) x1 x2 s.t. g ( x1 , x2 ) x1 x2 2 F ( x, ) x1 x2 ( x1 x2 2) Set F ( x , ) 0 1.3 Equality Constrained Optimization (1) F ( x , ) x2 0 x1 (2) F ( x , ) x1 0 x2 page 1.3-7 F ( x , ) ( x1 x2 2) 0 (1) and (2) x1 x2 (3) (3) x1 x2 2 2 1 So x1 x2 1 satisfies the necessary conditions for a local maximum or minimum. f ( x ) is a linear combination of g ( x ) Theorem 1.5 gives only necessary conditions. Consider Example 1.16 below. Example 1-16: Max or min f ( x) x13 x1 x2 x 2 s.t. g ( x) x2 0 F ( x, ) x13 x1 x2 x 2 x2 x F ( x , ) 0 (1) 3 x12 x2 0 (2) 3 x1 1 0 F ( x , ) 0 x2 0 From (1), x1 0 . From (2), 1 . x (0, 0)T , 1 is the only candidate but x is neither a local maximum or local minimum. f ( x ) 0 f (ò,0) ò3 0 , f (ò, 0) ò3 0 , ò 0 Note that Max f at x1 , x2 0 Min f at x1 , x2 0 _____________________________________________________________________________ 2nd Derivation of Lagrange Multipliers 1.3 Equality Constrained Optimization page 1.3-8 Max (or min) f ( x) x E n s.t. gi ( x) bi , i 1, , m m < n ; f, gi’s have continuous 1st partial derivatives Note: gi ( x) bi , i 1, , m defines surface of dimension n-m (no gi are redundant). Associated with each point on surface S defined by the constraints is the tangent plan at that point. Definition 1.3.2 A curve on a surface S is a family of points x (t ) S continuously parameterized by t for a t b . The curve is differentiable if dx(t ) exists, and is twice dt 2 differentiable if d x(t ) 2 . A curve x(t ) is said to pass through the point x if x(t ) x for dt some t , a t b . The derivative of the curve at x is dx (t ) . dt dx ( t ) 1 dt dx(t ) dt dxn (t ) dt n1 Definition 1.3.3 Consider all differentiable curves on S passing through a point x . The tangent plane at x is the collection of the derivatives at x of all these differentiable curves. The tangent plan is a subspace of E n . Suppose a surface is definite by the constraints gi ( x) bi ; we want to obtain explicit representation for the tangent plane. In particular, we want to express the tangent plan in terms of gi . g1 ( x) i 1, m , g ( x) gi ( x) Let g ( x) ( g ( x) is the Jacobian). j 1, , n x j m n g m ( x) Define M y E n : g ( x ) y 0 is a subspace of E n . We want to investigate the conditions under which M is the tangent plane at x . Theorem 1.6 At a regular point x of the surface defined by g ( x) b the tangent plane is equal to M y E n : g ( x ) y 0 . Example 1-17: In E 2 , let g1 ( x1 , x2 ) x1 . Then x E 2 : g ( x1 , x2 ) 0 is the x2 axis. Since g1 ( x) 1 0 , every point on the x2 axis is a regular point. The tangent plan at any regular T 1.3 Equality Constrained Optimization point is the x2 axis: y E 2 page 1.3-9 :[1, 0] y 0 x2 axis. M is not a regular point when x is not a regular point. Example 1-18: Let g1 ( x1 , x2 ) x12 ; g1 ( x1 , x2 ) 2 x1 x E 2 : g ( x1 , x2 ) 0 is the surface S = x2 axis. 0 so g1 0 0 for any point in S; so no points in S are regular points. T T In this case, M E 2 , but the tangent plane is the x2 axis. Theorem 1.7 Let x be a regular point of the constraints g ( x) b and a local minimum or maximum of f subject to these constraints. Then all y E n satisfying g ( x ) y 0 must also satisfy f ( x )T y 0 . In other words, f ( x ) is perpendicular to the tangent plane. This implies that f ( x ) is a linear combination of the gradients of g at x . Proof: Let y be any vector in the tangent plane at x and let x(t ) be any differentiable curve on the constraint surface passing through x with derivative y at x . x(0) x , dx(0) y, g ( x(t )) b for a t a for some a 0 dt Since x is a regular point, the tangent plane is identical with the y ' s E n such that g ( x ) y 0 . Since x is a constrained local extremum point of f, we have the following: d f ( x(t )) t 0 0 dt n f ( x(0)) dx j (0) 0 x j dt j 1 T f ( x ) y 0 QED. Theorem 1.8 Lagrange Multipliers Let x be a local minimum or maximum point of f subject to g ( x ) b . Assume also that x is a regular point of the constraints. Then there exist E m such that m f ( x ) igi ( x ) 0 . i 1 Proof: Consider the LP ( P ) : s.t. Max z cT x Ax b , A is m n, m n 1.3 Equality Constrained Optimization page 1.3-10 Assume the objective is bounded. The partial derivatives are gi ai j . x j By Theorem 1.5, at the (finite) maximum x0 there exist i , i 1, f ( x0 ) m gi i 0, j 1, x j x j i 1 , m , such that ,n . m c j i ai j 0, j 1, ,n i 1 T zmax In matrix form, A c . The optimal value of the objective function cT x0 x0T c x0T AT bT . Since x0 is feasible, the Lagrange multipliers are just the dual variables. If x0 is optimal for Problem ( P ), there exist E n such that AT c and cT x0 bT . For Problem ( P ), we have proved the duality theorem using Lagrange Multipliers. i is the rate of change of the optimal value of f as the right hand side constraint bi changes. § 1.3.1 Interpretation of Lagrange Multipliers Consider the following problem: Max z f ( x), x E n s.t. gi ( x) b0i , i 1, , m, m n The global optimum occurs at x with associated Lagrange multipliers E m . If we substitute different bi ’s, the rest of the problem remains unchanged, though we would expect the maximum point to occur at a different point. Suppose each xj and i are continuously differentiable functions of b in some neighborhood of b0. Theorem 1.3.1 Interpretation of Lagrange Multipliers Let z f ( x ) . Take the partial derivatives of z f ( x ) with respect to b in the neighborhood of b0. Using the chain rule: (1) n z f ( x ) x j bi j 1 x j bi From g k ( x ) b0 k , we have (2) g k ( x ) n g k ( x ) x j 1 k i bi x j bi 0 k i j 1 1.3 Equality Constrained Optimization page 1.3-11 For any specific i, (2) represents m equations (one for each k); multiplying each by the respective k , k 1, , m , and summing over k gives n x m g k x j g j k k i x j k 1 j 1 x j bi j 1 bi k 1 m (3) k n Substituting the necessary conditions into (3) gives xj f ( x ) i x j j 1 bi n Hence from (1) we have the result below: i z , i 1, bi ,m The Lagrange multipliers associated with the global optimum point gives the rate of change of the optimal value of the objective function with respect to changes in the right hand side of the constraints. § 1.3.2 Second order Conditions – Equality Constraints Consider the problem posed at the beginning of §1.3.1 except now the functions are twice differentiable. Theorem 1.9 2nd Order Necessary Conditions Suppose x is a local maximum of f subject to gi ( x) bi , i 1, , m, m n , and that x is a regular point of the constraints. There exist a unique E m such that m f ( x ) i gi ( x ) 0 i 1 If we denote by M the tangent plane M y E n : g ( x ) y 0 , then m 2x F ( x , ) 2 f ( x ) i 2 gi ( x ) i 1 2 x is negative semidefinite on M. y F ( x ) y 0, y M T Theorem 1.10 2nd Order Sufficient Conditions 1.3 Equality Constrained Optimization page 1.3-12 Suppose there exists a point x satisfying gi ( x ) bi , i 1, , m, and a E m such m m that f ( x ) i gi ( x ) 0 . Suppose also that yT 2 f ( x ) i 2 gi ( x ) y 0 for all i 1 i 1 n y M y E : g ( x ) y 0 , y 0 , Then x is a strict local maximum of f subject to gi ( x ) bi , i 1, ,m . Example 1-17: Max x1 x2 x2 x3 x1 x3 s.t. x1 x2 x3 3 f ( x ) g ( x ) 0 (1) x2 x3 0 x1 x3 0 (2) x1 x2 0 (3) (1) x2 x3 (1) and (2) x1 x2 0 (1) and (3) x1 x3 0 x1 x2 x3 1, 2 0 1 1 f ( x) 1 0 1 1 1 0 2 g ( x) 0 2 0 1 1 F ( x , ) 1 0 1 1 1 0 2 x 0 M y E 3 : g ( x ) y 0 y E 3 : y1 y2 y3 so yT 2x F ( x , ) y y1 ( y2 y3 ) y2 ( y1 y3 ) y3 ( y1 y2 ) ( y12 y22 y32 ) 0 for y 0 so x is at least a local maximum Example 1-18: Max or min f ( x) x13 x1 x2 x2 s.t. g ( x) x2 0 candidate: x 0 0 , 1 T 1.3 Equality Constrained Optimization page 1.3-13 3x 2 x2 f ( x) 1 x1 1 0 g ( x) 1 6 x 1 2 f ( x) 1 1 0 0 1 2 F ( x , ) 1 0 0 0 2 g ( x) 0 0 y T 2 F ( x , ) y 0 M y E 2 : g ( x ) y 0 y E 2 : y 2 0 § 1.3.3 The General Case The necessary conditions can be generalized further for cases when the Jacobian is not assumed to be at full rank. Consider the problem: Max or min f ( x) x En s.t. gi ( x) bi , i 1, , m, m n f, gi’s have continuous 1st partial derivatives Define the Augmented Jacobian to have m+1 rows and n columns: g1 g1 x xn 1 J J f g m g m f x xn 1 f f x1 xn ( m 1)n Assume x is a local maximum or minimum and the augmented Jacobian matrix is not full rank at x . r ( J f ) m 1 at x . When r ( J f ) m 1 at x then x cannot be a relative maximum or minimum. Consider the equations: () 0f ( x ) 1g1 ( x ) mg m ( x ) 0 Which can be rewritten as: 1.3 Equality Constrained Optimization page 1.3-14 f ( x0 ) g1 ( x0 ) g m ( x0 ) 0 1 m 0, j 1,, n x j x j x j 1 2 m 0 J f 0 (Linear Combination of the rows of Jf) Since r ( J f ) m 1 at x , there exist nontrivial solutions to ( ) . If r ( J f ) m 1 , then m+1 rows are linearly independent and the only way a linear combination of the rows is zero is if all coefficients are zero. Case (i) r ( J f ) m at x We can select any m equations (i.e. columns of Jf ) containing a nonsingular submatrix of order m whose elements are in J, and solve uniquely for 1 ,, m in terms of 0 (m equations with m + 1 unknowns). The value of 0 can be designed arbitrarily; it cannot be zero, however, if we want a nontrivial solution. For convenience, set 0 1 . Case (ii) r ( J f ) m at x , but r ( J ) m 1 Select m equations from ( ); the matrix of whose coefficients contains a nonsingular matrix of order m. There does not exist a solution to this set of equations which does not have 0 0 . If 0 0 , there exist 1 ,, m , not all zero, which satisfy ( ). The λi are not uniquely determined, but instead, the solutions span a one dimensional subspace of E m . 1 In the figure below, g1 ( x0 ) 3g 2 ( x0 ) . 1 2 3 Example 1-19: Let f ( x ) denote components of f ( x ) chosen in m equations from ( ). 0f 1g1 mgm 0 m 0 0 f ( x ) i gi i 1 0 1.3 Equality Constrained Optimization page 1.3-15 r ( J f ) m if r ( J ) m 1 Example 1-20: There exist i , i 1,, m , not all zero such that 1g1 mg m 0 r ( J ) m 1 , so one gradient, say g1 , is a linear combination of the others. m g1 i gi i 1 m 1 i gi 2g 2 mg m 0 i 2 (1 2 2 )g2 (13 3 )g3 (1 m m )gm 0 Since r ( J ) m 1 , all coefficients must equal zero. 1 2 2 1 1 3 3 m m Each equation for 1 is a hyperplane (of dimension m 1 ) in m-space; the intersection of m 1 hyperplanes is a 1-dimensional space. Case (iii) r ( J f ) r ( J ) r m at x Select r equations from ( ) (r columns from Jf ). There exists the matrix of coefficients that contains a nonsingular submatrix of order r whose elements are in J. Then we can solve for r of the λ’s in terms of 0 and the other m r λ’s. 0 is arbitrary; for convenience set 0 1 . m r of the λ’s are also arbitrary. 1.3 Equality Constrained Optimization page 1.3-16 Case (iv) r ( J f ) r m, r ( J f ) r ( J ) at x Select r equations from ( ) the matrix of whose coefficients contains a nonsingular submatrix of order r. As in case (ii), there are no solutions which do not have 0 0 . If we set 0 0 , there exist i , i 1,, m , not all zero, which satisfy ( ). The i are not unique but span a m r 1 dimensional subspace of E m . In the figure below, m r 1 2 . To specify i : g1 ( x ) 2 g3 ( x ) 3 g1 ( x ) 2g 2 ( x ) 3 i gi ( x ) 0 i 1 g1 ( x )(1 1 2 3 3 ) 0 2 2 3 1 2 3 1 2 2 Theorem 1.11 Let f and gi have continuous partial derivatives. If x is a local maximum or minimum of f ( x) subject to gi ( x) bi , i 1,, m , then there exist m 1 real numbers 0 , 1 ,, m , not all zero such that: m 0f ( x ) igi (x ) 0 i 1 Example 1-21: Min f ( x) x1 x E3 s.t. g1 ( x) (1 x1 )3 x2 0 g 2 ( x) x2 0 x (1, 0, x3 )T has minimizing point x (1,0,0)T . 1.3 Equality Constrained Optimization 1 f ( x ) 0 0 page 1.3-17 0 g1 ( x ) 1 0 0 g 2 ( x ) 1 0 Looking from above: We have case (ii). r ( J f ) 2 , r ( J ) 1 2 0f ( x ) igi (x ) 0 i 1 We can see pictorially that 0 0, 1 2 1 0 0 0 1 1 2 1 0 0 0 0 0 0 1 2 0 0 1 Here the objective function f plays no role in the solution. Any f such that f ( x ) 0 k will work, where k E1 . For example, f ( x) ln( x1 ) x23 x32 e x3 would also give (1, 0, x3 )T as a candidate point. Theorem 1.12 If r ( J f ) at x0 is m 1 , then f ( x) does not take on a local maximum or minimum at x0 . Proof: Assume r ( J f ) m 1 at x0 . In particular, let the first m 1 columns of Jf form an nonsingular submatrix of order m 1 at x0 . Then the Implicit Function Theorem implies that for the m 1 equations: 1.3 Equality Constrained Optimization page 1.3-18 gi ( x) bi , i 1,, m, f ( x) z 0 In n 1 variables x1 ,, xn , z, there exists an ε-neighborhood of xm0 2 , , xn0 , z0 , where z0 f ( x0 ) , in which we can solve explicitly for x1 ,, xm 1 to yield the following: xi i ( xm 2 ,, xn , z ), i 1,, m 1 Note that z is one of the variables that can be assigned an arbitrary value in i . Consequently, if we set xm 2 xm0 2 , , xn xn0 , then z z z0 , there exists x such that f ( x) z . Since the functions i are continuous (from the Implicit Function Theorem) it follows that if we select any δ-neighborhood such that xm 2 xm0 2 , , xn xn0 with the property that for some of these points, f ( x) z0 , while for others f ( x) z0 . QED. Result from the Duality Theorem: If the problem is unbounded, no ’s will exist. If the problem is infeasible, there will be an infinite number of ’s. Example 1-22: Max x1 x2 s.t. x1 x2 1 1 1 0 1 2 0 2 0 No solution for Problem is Unbounded. Example 1-23: Max 0T x (this is trivial with equality constraints) s.t. x1 x2 x3 1 x1 x2 x3 0 1 2 0 1 2 0 1 2 1 2 0 There are an infinite number of solutions. This example is like case (iii). r ( J f ) r ( J ) r m 1.3 Equality Constrained Optimization page 1.3-19 Here m = 2, r = 1. 1 1 1 J f 1 1 1 Can solve for r ’s in terms of the other m r ’s and 0 . 0 0 0 Example 1-24: Examples of cases (i) – (iv) Min x12 x22 ( x3 1) 2 s.t. x1 x2 x3 1 J x2 x1 1 , r ( J ) 1 full rank 2 x1 1 x2 0 (1) 2 x2 1 x1 0 (2) 2( x3 1) 1 0 (3) 1 2(x3 1) (1) 2 x1 2 x2 ( x3 1) 0 (2) 2 x2 2 x1 ( x3 1) 0 x1 x2 ( x3 1) x2 x1 ( x3 1) (5) x1 x2 ( x2 x1 )( x3 1) (i) if x2 x1 , x1 , x2 0 x3 1 1 x3 0, 1 2 (1) and (2) 2 x1 2 x2 0 (4) x1 x2 1 This results in the unpleasant x12 1 . (ii) x2 x1 , x1 0 (5) x3 0 1 2 x3 (4) x1 0 x3 1 1 0 Also not good. (iii) x1 x2 (a) x1 x2 0 (4) x3 1 1 0 (b) x1 x2 0 (4) x3 1 x12 Case (ii) r ( J f ) m at x0 , r ( J ) m 1 at x0 1.3 Equality Constrained Optimization page 1.3-20 0 0; i ’s not unique Case (iii) r ( J f ) r ( J ) r m 2 0 1 1 2 1 1 2 Jf 0 5 1 4 2 x1 0 0 2( x4 1) 0 1 r ( J f ) r ( J ) 2 at x0 1 1 Can solve for r = 2 of the λ’s in terms of λ0 and the other m – r = 3 – 2 λ’s. For convenience, let λ0 = 1. 0 1 Solving you get x 1 1 1 2 1 2 2 3 1 Case (iv) r ( J f ) r m and r ( J f ) r ( J ) 0 0 , λi’s not unique but span m – r + 1 dimensional subspace of E m . Example 1-25: Min x12 ( x4 1)2 x52 s.t. x1 2 x2 x4 2 x5 4 2 x1 x2 x3 x42 x5 4 5 x2 x3 x42 x5 10 x1 1 x2 1 x3 1 x42 1 x5 2 2 2 2 2 n 5, m 4, x0 0 1 1 2 0 T 2 1 2 1 5 J f 0 1 1 2 0 0 0 1 1 4 1 6 1 2 2 0 2 2 1 3 1 2 0 r ( J f ) 3 r, r ( J ) 2 At x0, 2g1 g2 g3 and 0g1 1 g 2 g 4 2 1.3 Equality Constrained Optimization () f ( x0 ) 5 g ( x ) 0 i i 0 0, j 1,, 5 choose j 1,3, 4 x j x j i 1 (1) 1 22 4 0 (2) 21 2 53 1 4 0 2 1 2 3 4 0 2 20 1 42 63 24 0 (3) (4) (5) page 1.3-21 21 2 33 1 4 0 2 We can choose r = 3 equations from () (i.e. choose three columns from Jf ) such that the 5 3 matrix of coefficients has a 3 3 nonsingular submatrix. Chose columns 1, 3, 4 (i.e. choose equations (1), (3), and (4). There exist i , i 1,, 4 , not all zero, which satisfy equations (1) – (5). The λi’s are not unique but span an m r 1 4 3 1 2 dimensional subspace of E m E 4 . (3) 2 3 1 4 (4) 20 0 2 (1) 1 22 4 23 3 , 4 arbitrary