Choosing Variable to generate a Node in a MILP

44 Views Asked by At

I'm trying to find or understand rigorous a criterion about choosing the variable to generate a new Node in the Branch and Bound method. Usually the methods I've seen just choose the variable nearly to a integer or the first one must to be an integer. For example, Given $$\mbox{MILP}:\hspace{0,2cm} \mbox{min}\left\lbrace c^Tx + h^Ty \mid (x,y) \in S \right\rbrace $$ Where $S=\left\lbrace (x,y) \in \mathbb{Z}^n_+ \times \mathbb{Q}^p_+ \mid Ax+Gy =b\right\rbrace $.

Let be LP$_0$ its relaxation and $(x_0,y_0)$ the optimal solution with value $z_0$.

$x_0=(x_{01},\cdots,x_{0n})$. Consider $J$ such that $j \in J$ if $x_{0j}$ $ \notin \mathbb{Z}$. Then usually they choose min$\{j \in J\}$ or min $\{|x_{0j}-\lfloor x_{0j}\rfloor|\}\cup\{|x_{0j}-\lceil x_{0j}\rceil|\}$ for $j\in J$. And then generates the nodes $N_{{0}_1}$ and $N_{{0}_2}$.

My two questions are:

1.There is not any criterion for choosing the better variable? For example, I've tried and think to choose the $x_{0j}$ such that $c_jx_{0j}$ is minimum. All times I've tried this the algorimth do less steps in the examples I've tried.

2.The same with the Node, wich I should to solve first? In many books choose the one which makes the relaxation problem optimum (I guess is better, but I would like justify it with rigor).

If you know any book or something to understand it better I would really apreciate it. Thank you very much.

1

There are 1 best solutions below

0
On BEST ANSWER

For the first question, search for variable selection strategies, including pseudocost branching, strong branching, and reliability branching.

For the second question, search for node selection strategies, including depth-first, breadth-first, best-bound, and best-estimate.