Chapter 10 Minimization or Maximization of Functions Optimization
- Slides: 23
Chapter 10 Minimization or Maximization of Functions
Optimization Problems • Solution of equations can be formulated as an optimization problem, e. g. , density functional theory (DFT) in electronic structure, Lagrangian mechanics, etc • Minimization with constraints – operations research (linear programming, traveling salesman problem, etc), max entropy with a given energy, …
General Consideration • Use function values only, or use function values and its derivatives • Storage of O(N) or O(N 2) • With constraints or no constraints • Choice of methods
Local & Global Extremum
Bracketing and Search in 1 D c a b Bracket a minimum means that for given a < b < c, we have f(b) < f(a), and f(b) < f(c). There is a minimum in the interval (a, c).
How accurate can we locate a minimum? • Let b a minimum of function f(x), Taylor expanding around b, we have • The best we can do is when the second correction term reaches machine epsilon comparing to the function value, so
Golden Section Search x a b c • Choose x such that the ratio of intervals [a, b] to [b, c] is the same as [a, x] to [x, b]. Remove [a, x] if f[x] > f[b], or remove [b, c] if f[x] < f[b]. • The asymptotic limit of the ratio is the Golden mean
Parabolic Interpolation & Brent’s Method Brent’s method combines parabolic interpolation with Golden section search, with some complicated bookkeeping. See NR, page 404 -405 for details.
Minimization as a root-finding problem • Let’s derivative d. F(x)/dx = f(x), then minimizing (or maximizing) F is the same as finding zero of f, i. e. , f(x) = 0. • Newton method: approximate curve as a straight line.
Deriving Newton Iteration • Let the current value be xn and zero is approximately located at xn+1, using Taylor expansion • Solving for xn+1, we get
Convergence in Newton Iteration • Let x be the exact root, xi is the value in ith iteration, and εi =xi-x is the error, then • Rate of convergence: (Quadratic convergence)
Strategy in Higher Dimensions 1. Starting from a point P and a direction n, find the minimum on the line P + n, i. e. , do a 1 D minimization of y( )=f(P+ n) 2. Replace P by P + min n, choose another direction n’ and repeat step 1. The trick and variation of the algorithms are on chosen n.
Local Properties near Minimum • Let P be some point of interest which is at the origin x=0. Taylor expansion gives • Minimizing f is the same as solving the equation T for transpose of a matrix
Search along Coordinate Directions Search minimum along x direction, followed by search minimum along y direction, and so on. Such method takes a very large number of steps to converge. The curved loops represent f(x, y) = const.
Steepest Descent Search in the direction with the largest decrease, i. e. , n = - f n. T n’ = 0 n’ Constant f contour line (surface) is perpendicular to n, because df = dx f = 0. The current search direction n and next search direction are orthogonal, because for minimum we have y’( ) = df(P+ n)/d = n. T f|P+ n = 0 n
Conjugate Condition n 1 T A n 2 = 0 Make a linear coordinate transformation, such that contour is circular and (search) vectors are orthogonal
Conjugate Gradient Method 1. Start with steepest descent direction n 0 = g 0 = - f(x 0), find new minimum x 1 2. Build the next search direction n 1 from g 0 and g 1 = - f(x 1), such that n 0 A n 1 = 0 3. Repeat step 2 iteratively to find nj (a Gram-Schmidt orthogonalization process). The result is a set of N vectors (in N dimensions) ni. T A nj = 0
Conjugate Gradient Algorithm 1. Initialize n 0 = g 0 = - f(x 0), i = 0, 2. Find that minimizes f(xi+ ni), let xi+1 =xi+ ni 3. Compute new negative gradient gi+1 = - f(xi+1) (Fletcher-Reeves) 4. Compute 5. Update new search direction as ni+1 = gi+1 + ini; ++i, go to 2
The Conjugate Gradient Program
Simulated Annealing • To minimize f(x), we make random change to x by the following rule: • Set T a large value, decrease as we go • Metropolis algorithm: make local change from x to x’. If f decreases, accept the change, otherwise, accept only with a small probability r = exp[-(f(x’)-f(x))/T]. This is done by comparing r with a random number 0 < ξ < 1.
Traveling Salesman Problem Beijing Tokyo Shanghai Taipei Hong Kuala Lumpur Singapore Find shortest path that cycles through each city exactly once.
Problem set 7 1. 2. Suppose that the function is given by the quadratic form f=(1/2)x. T A x, where A is a symmetric and positive definite matrix. Find a linear transform to x so that in the new coordinate system, the function becomes f = (1/2)|y|2, y = Ux [i. e. , the contour is exactly circular or spherical]. More precisely, give a computational procedure for U. If two vectors in the new system are orthogonal, y 1 T y 2=0, what does it mean in the original system? We’ll discuss the conjugate gradient method in some more detail following the paper: http: //www. cs. cmu. edu/~quake-papers/painlessconjugate-gradient. pdf
- Standard maximization problem
- Monopsony profit maximization
- Limitations of profit maximization
- Dual simplex method maximization problem
- Profit maximization
- Simplex
- Normal profit economics
- Profit maximization and competitive supply
- Cost revenue and profit maximization guided reading
- Utility maximization problem
- Perfect substitutes utility function
- Big m method minimization with mixed constraints
- Kalkulator metode simpleks
- Module 53 featured worksheet profit maximization
- Profit maximization in financial management
- Expectation maximization tutorial
- Customer lifetime value maximization
- Meaning of financial management
- Abt model of social accounting
- The zj row in a simplex table for maximization represents
- What cj-zj represents
- Expectation maximization algorithm
- Unit 2 demand supply and consumer choice
- Special cases in simplex method pdf