Interior Point Method
“Justice will not be served until those who are unaffected are as outraged as those who are. – Benjamin Franklin”
Summary:
Starts within a feasible region, using a slack variable that is always >= 0 to convert all inequality constraints to equality constraints.
Intuition:
Punish the proximity to the constraint instead of the violation of it.
Steps:
-
Introduce slack variables in inequality conditions, convert them to equality condition.
-
Convert objective function to Lagrangian form, with additional term of -ve log of slack variable.
-
To satisfy KKT conditions of the new objective function, we obtained as system of linear equations.
-
Rearrange the system of linear equations into symmetric linear system of equations, by only including the non-slack variables. Solving it explictly.
-
Use the solution from step 4 above to find the search direction of slack variable.
-
Use line search methods to find a step size that optimizes the merit function.
-
Optimize iteratively by repeating the steps above.
Reference:
https://www.youtube.com/watch?v=zm4mfr-QT1E