next up previous contents index
Next: Alternative: Reducing IP to Up: Optimal Partitioning Previous: Example

Subsections

Correctness

  

In this section, we show that IP-opt is correct by proving the following

Theorem 28822

The algorithm IP-opt delivers an optimal partition for an instance of IP if there is a partition that satisfies (5.7).  

Proof:

The proof is by induction over the loop variable i for which

 
  (18)
is the invariant of the for-loop in IP-opt where

and

Using the facts we can conclude that c(qn) is the number of overlaps in an optimal partition as specified by IP and that an optimal partition is given by Pn.

If there is no partition that satisfies constraint (5.7) then there must be some qi such that for all j<i. This causes the set J to be empty and thus the algorithm to report this fact and then to stop. In the remainder of the proof we assume that there is an optimal partition and consequently that J is non-empty for all .

Base case:
i=1
The algorithm provides

with . Thus which causes the sum in (5.11) to evaluate to 0 which matches with (*). Thus the assumption holds.
Hypothesis:
Assume that (5.11) holds for all   (**).

Inductive Step:
i=x+1
Let be the minimising point with j < i = x+1 as guaranteed by the algorithm. Then

Therefore (5.11) holds for i=x+1.

  


next up previous contents index
Next: Alternative: Reducing IP to Up: Optimal Partitioning Previous: Example

Thomas Zurek