Successive shortest paths: Difference between revisions

From Algowiki
Jump to navigation Jump to search
No edit summary
Line 1: Line 1:
== Abstract view ==
== Abstract view ==


'''Invariant:'''
'''Definition:'''
# The capacity constraints are fulfilled, that is, <math>0\leq f(a)\leq u(a)</math> for all <math>a\in A</math>.
# The '''balance discrepancy''' of a node <math>v\in V</math> is defined as
# The '''balance discrepancy''' of each node <math>v\in V</math> is '''underestimating''', that is,
# The balance discrepancy of a node <math>v\in V</math> is '''underestimating''' if:
## If <math>b(v)>0</math>, then <math>\sum_{w:(v,w)\in A}f(v,w)-\sum_{w:(w,v)\in A}f(w,v)\leq b(v)</math>.
## If <math>b(v)>0</math>, then <math>\sum_{w:(v,w)\in A}f(v,w)-\sum_{w:(w,v)\in A}f(w,v)\leq b(v)</math>.
## If <math>b(v)<0</math>, then <math>\sum_{w:(v,w)\in A}f(v,w)-\sum_{w:(w,v)\in A}f(w,v)\geq b(v)</math>.
## If <math>b(v)<0</math>, then <math>\sum_{w:(v,w)\in A}f(v,w)-\sum_{w:(w,v)\in A}f(w,v)\geq b(v)</math>.
## If <math>b(v)=0</math>, then <math>\sum_{w:(v,w)\in A}f(v,w)-\sum_{w:(w,v)\in A}f(w,v)=b(v)</math>.
## If <math>b(v)=0</math>, then <math>\sum_{w:(v,w)\in A}f(v,w)-\sum_{w:(w,v)\in A}f(w,v)=b(v)</math>.


'''Invariant:'''
# The capacity constraints are fulfilled, that is, <math>0\leq f(a)\leq u(a)</math> for all <math>a\in A</math>.
# There is no negative cycle in the [[Basic flow definitions#Residual network|residual network]] of <math>f</math>.
# The balance discrepancy of ever node is underestimating.


'''Variant:'''
'''Variant:'''
Line 20: Line 25:
'''Proof:'''
'''Proof:'''
Obvious.
Obvious.
== Induction step ==
# If there
#

Revision as of 08:58, 23 October 2014

Abstract view

Definition:

  1. The balance discrepancy of a node [math]\displaystyle{ v\in V }[/math] is defined as
  2. The balance discrepancy of a node [math]\displaystyle{ v\in V }[/math] is underestimating if:
    1. If [math]\displaystyle{ b(v)\gt 0 }[/math], then [math]\displaystyle{ \sum_{w:(v,w)\in A}f(v,w)-\sum_{w:(w,v)\in A}f(w,v)\leq b(v) }[/math].
    2. If [math]\displaystyle{ b(v)\lt 0 }[/math], then [math]\displaystyle{ \sum_{w:(v,w)\in A}f(v,w)-\sum_{w:(w,v)\in A}f(w,v)\geq b(v) }[/math].
    3. If [math]\displaystyle{ b(v)=0 }[/math], then [math]\displaystyle{ \sum_{w:(v,w)\in A}f(v,w)-\sum_{w:(w,v)\in A}f(w,v)=b(v) }[/math].


Invariant:

  1. The capacity constraints are fulfilled, that is, [math]\displaystyle{ 0\leq f(a)\leq u(a) }[/math] for all [math]\displaystyle{ a\in A }[/math].
  2. There is no negative cycle in the residual network of [math]\displaystyle{ f }[/math].
  3. The balance discrepancy of ever node is underestimating.

Variant: The total balance discrepancy strictly decreases, that is, the value

[math]\displaystyle{ \sum_{v\in V}\left|\sum_{w:(v,w)\in A}f(v,w)-\sum_{w:(w,v)\in A}f(w,v)-b(v)\right| }[/math].

Induction basis

Abstract view: Start with the zero flow.

Proof: Obvious.

Induction step

  1. If there