Constrained NLP
Created on: 2024-12-08
Tags: #constrainedNLP
Course: Optimization
Definition
Non-linear programming problems with constraints are solved using different methods based on the type of constraints.
Lagrange Multiplier (Single Equality Constraint)
- The method works for a NLP with a single equality constraint.
- The necessary conditions of optimality can be obtained using a Lagrange function given by
\(L(X,\lambda) = f(X)\, +\, \lambda\,h(X)\) - Partial differentiation with each variable, we get \(\nabla_{x}\,L(X^*, \lambda) = \nabla f(X^*)\,+\, \lambda \nabla h\,(X^*) = 0\)
- Partial differentiation with Lagrange multiplier ($\lambda$), we get
Solving the above equations, we get the value of $X^*$ and $\lambda$ which optimize the objective function.
Kuhn-Tucker Conditions
For a general constrained NLP,
Minimize $f(X)$ $h_{j}(X) = 0$, where $j=1,2,\dots,m$ $g_{k}(X)\le0$, where $k=1,2,\dots,p$
Here, we have $m$ equality constraints and $p$ inequality constraints.
\[L(X,\lambda,\mu) = f(X) + \sum_{j=1}^m \lambda_{j}h_{j}(X) + \sum_{j=1}^p \mu_{j}g_{j}(X)\]Kuhn-Tucker necessary conditions of optimality states that At local minimum point ($X^$) $h_{j}(X^) = 0$ $g_{k}(X^) \le 0$ $\mu_{k}\,g_{k}(X^) = 0$ $\mu_{k} \ge 0$ $\nabla_{X}\,\, L(X^*,\lambda) = 0$
Kuhn-Tucker sufficient condition of optimality
Use Hancock’s equation (without considering inequality constraints), \(J=\left[ \begin{array}{l l l} {{\nabla_{x x}^{2}L(X^{*},\lambda)}} & {{\nabla_{x\lambda}^{2}L(X^{*},\lambda)}} \\ {{\nabla_{\lambda x}^{2}L(X^{*},\lambda)}} & {{\nabla_{\lambda\lambda}^{2}L(X^{*},\lambda)}} \\ \end{array}\right]=\left[ \begin{array}{l l} {{\nabla_{x x}^{2}L(X^{*},\lambda)}} & {{\nabla_{x}\,h_{j}(X^{*})}} \\ {{\nabla_{x}\,h_{j}^{T}(X^{*})}} & {{0}} \end{array}\right]\)
Linked from
- No other published notes link here yet.