23.06.2013 Views

Implementation of Cutting Plane Separators for Mixed Integer ... - ZIB

Implementation of Cutting Plane Separators for Mixed Integer ... - ZIB

Implementation of Cutting Plane Separators for Mixed Integer ... - ZIB

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

3.3 Algorithmic Aspects 25<br />

resulting single mixed integer constraint and by the bounds imposed on the integer<br />

variables is than relaxed to obtain the mixed knapsack relaxation X MK <strong>of</strong> X, as<br />

explained in Section 3.2.<br />

We want to comprehend the strategies suggested by Marchand and Wolsey [39,<br />

42] <strong>for</strong> the selection <strong>of</strong> the substitutions, and we also want to develop new strategies.<br />

Furthermore, in [39, 42], a real variable is either bounded by a simple upper bound<br />

or by a variable upper bound. The same hold <strong>for</strong> the lower bounds: a real variable<br />

is either bounded by a simple lower bound or by a variable lower bound. Thus, we<br />

have to extend the bound substitution heuristic given in [39, 42] to the more general<br />

situation considered here.<br />

There<strong>for</strong>e, let the MIR inequality <strong>for</strong> X MK be given in the <strong>for</strong>m<br />

<br />

j∈N<br />

Ff ′ (α<br />

α<br />

0<br />

′<br />

j)xj + <br />

j∈M<br />

¯Ff ′ (γ<br />

α<br />

0<br />

′<br />

j)¯yj ≤ ⌊α ′<br />

0⌋. (3.10)<br />

Note that <strong>for</strong> d ≥ 0 and 0 ≤ α < 1, ¯ Fα(d) = 0. Furthermore, let k ∈ M be the index<br />

<strong>of</strong> a real variable with γk not equal to zero in (3.9), uk < ∞, and ũk < ∞.<br />

To understand the effect <strong>of</strong> the different bound substitutions per<strong>for</strong>mable <strong>for</strong> yk,<br />

we analyze the effect on the MIR inequality (3.10) when we use one <strong>of</strong> the different<br />

bounds imposed on yk <strong>for</strong> the substitution <strong>of</strong> yk in comparison to the case where no<br />

bound substitution is per<strong>for</strong>med.<br />

Using a simple bound <strong>for</strong> the substitution <strong>of</strong> yk has a different effect than using a<br />

variable bound. On the one hand, among other changes, using a simple bound may<br />

change the value <strong>of</strong> α ′<br />

0 and there<strong>for</strong>e, may change the value <strong>of</strong> fα ′ , which would lead<br />

0<br />

to different values <strong>of</strong> the coefficients <strong>of</strong> all variables in the MIR inequality (3.10).<br />

On the other hand, among other changes, using a variable bound does not change<br />

the value <strong>of</strong> α ′<br />

0<br />

, whereas the value <strong>of</strong> α′<br />

k <strong>for</strong> the integer variable xk involved in<br />

the variable bound used may change, which would lead to a different value <strong>of</strong> the<br />

coefficient <strong>of</strong> xk in the MIR inequality (3.10).<br />

In addition, using a lower bound <strong>for</strong> the substitution <strong>of</strong> yk has a different effect<br />

than using an upper bound. On the one hand, using a lower bound does not change<br />

the value <strong>of</strong> γ ′<br />

k . On the other hand, using an upper bound changes the value <strong>of</strong><br />

γ ′<br />

k , i.e., if γk ≥ 0 in (3.9), γ ′<br />

k becomes negative when using an upper bound <strong>for</strong> the<br />

substitution <strong>of</strong> yk, and the other way around. Thus, the decision <strong>of</strong> using a lower<br />

bound or an upper bound influences the value <strong>of</strong> γ ′<br />

k and there<strong>for</strong>e, also the coefficient<br />

<strong>of</strong> ¯yk in the MIR inequality (3.10). If γk is nonnegative in (3.9), we assume that<br />

in practice using a lower bound per<strong>for</strong>ms better than using an upper bound. If γk<br />

is negative in (3.9), we can not decide in advance whether using a lower bound or<br />

using an upper bound leads to a better per<strong>for</strong>mance.<br />

We suggest to use a two step procedure <strong>for</strong> deciding which bound is used <strong>for</strong> the<br />

substitution <strong>for</strong> each real variable. In the first step, we decide whether a simple or<br />

variable bound is used, i.e., we select a lower bound lbj (simple or variable bound)<br />

and an upper bound ubj (simple or variable). And, in the second step we decide<br />

whether a lower or upper bound is used, i.e., we decide which <strong>of</strong> the two bounds<br />

selected in the first step we will actually use <strong>for</strong> the substitution. From our analysis<br />

the question arises, which <strong>of</strong> the following three criteria <strong>for</strong> the first step does lead<br />

to the best per<strong>for</strong>mance <strong>of</strong> the separation algorithm in practice.

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!