Linear Programming Learning Notes (3) Degeneracy
All the resources come from Linear Programming: Foundations and Extensions by Professor Robert J. Vanderbei.
Explore the link below for further information:
LP Book Resources
Part 1
Basic Theory: The Simplex Method and Duality
Chapter 3 Degeneracy
Definition
We say a dictionary is degenerate if bi vanishes for some i∈B .
Problem arise when a degenerate dictionary produces degenerate pivots. A pivot is a degenerate pivot if one of the ratios in the calculation of the leaving variable is +∞ .
Example
ζ=3−0.5x1+2x2−1.5w1−−−−−−−−−−−−−−−−−−−−−−
x3=1−0.5x1−0.5w1
w2=x1−x2+w1
As b2 is missing, this is a degenerate dictionary. Picking x2 as the entering variable, x2 cannot increase and ζ remains unchanged. This is a degenerate pivot. The values of the variables haven’t changed throug this pivot, they remains:
(x1,x2,x3,w1,w2)=(0,0,1,0,0)
It is usual typical that it can “break away” from the degeneracy. The real danger is that the simplex method will make a sequence of degenerate pivots and eventually return to a dictionary that has appeared before, which is called cycling .
It has been shown that if a problem has an optimal solution but cycles off-optimum, then the problem must involve dictionaries with at leat four(un-slack) variables and two constraints. If the simplex method cycles, then all the pivots within the cycle must be degenerate.
Also, there is a Theorem3.1 stating that:
If the simplex method fails to terminate, then it must cycle.
The problem of cycling is not as that common. Also, computer implementations of the simplex method written with floating point numbers can avoid this, because a zero will generally show up not as an exact zero but rather as a very small number. Therfore, the dictionary appears to be slightly off from actually being degenerate.
To solve the problem of cyclings, we can use a method similar to what the floating point numbers do, which is The Perturbation/Lexicographic Method.
The Pertubation/Lexicographic Method
We can write down a series of fixed perturbation for each constraint as:
0<ϵ