A family of valid inequalities for a general integer linear programming
problem may be derived on the basis of the following simple algebraic
considerations.
Let P be a polyhedron defined by a system of constraints written in standard
form, that is to say having the expression Ax = b, x non negative and
assume that the matrix A has m rows.
For each vector u with m components we may consider the linear combination
of the rows of the system Ax = b with coefficients given by the components
of u. In so doing we obtain the equality 1) which is trivially satisfied
by any element in P. Let's take a new step and round off all the coefficients
of the left hand side of equality 1). After this operation, each coefficient
of the left hand side is not increased.
So, since by hypothesis the elements of P have non negative components,
inequality 2) is satisfied by each element in P.
Moreover,note that each coefficient of the left hand side of inequality
two is integer and thus, if we evaluate it in an integer vector x we obtain
an integer number. This implies that, if we perform a new step and round
off also the right hand side of inequality 2), we obtain a new inequality,
the number 3, that is surely satisfied by each integer vector in P and
thus by each element in P(S). In general, inequality 3) may be violated
by fractional elements of P.
An inequality obtained in this way is called a Chvátal Inequality.
Evidently, we may construct an inequality of this type for each element
u in Rm.
|
|