Chapter 7
Linear Programming
(3) Pipage Rounding
Ding-Zhu Du
Pipage Rounding
example
Maximum Coverage
Given a family C of subsets of a set I {1,..., n} with nonnegativ e weight
function w on subsets in C and a positive integer p, find a subset X I
with | X | p to maximize the total weight of subsets in C hit by X .
ILP Formulation
m
max
w z
j
j 1
s.t.
x
i
z j , j 1,..., m,
i
p,
iS j
n
x
i 1
j
xi {0,1}
0 z j 1.
Alternative Formulation 1
m
max L( x) w j min{ 1, xi }
j 1
iS j
n
s.t.
x
i 1
i
p,
xi {0,1}
Alternative Formulation 2
m
max F ( x) w j (1 iS (1 xi ))
j
j 1
n
s.t.
x
i 1
i
p,
xi {0,1}
Relationship
F ( x) cL( x)
where c 1
1
e
Proof.
(1 xi )
iS j
1 (1 xi ) 1
k
iS j
k
(k | S j |)
xi
iS j
1 1
c xi c min( 1, xi ) for 0 xi 1.
k
iS j
iS j
iS j
k
Relationship
F ( x) cL( x)
where c 1
1
e
Proof.
(1 xi )
iS j
1 (1 xi ) 1
k
iS j
k
(k | S j |)
xi
k
iS j
1
1 1
1 1 c c min( 1, xi ) for xi 1.
k
k
iS j
iS j
k
k
z
Set g ( z ) 1 1 .
k
k
1
g (0) 0, g (1) 1 1 .
0
1
k
g ( z ) is monotone increasing and concave in [0,1].
k
1
1
So, g ( z ) z (1 1 ) z (1 ).
e
k
z
g ' ( z ) k 1
k
k 1
1 z
( ) 1
k k
z
g ' ' ( z ) (k 1)1
k
k 2
1
( ) 0.
k
k 1
0.
Relaxation
m
m
max L( x) w j min{ 1, xi }
j 1
xi p,
i 1
0 xi 1.
w z
s.t.
x
computed in time O(n 3.5 ).
j
i
z j , j 1,..., m,
i
p,
iS j
n
x
i 1
An optimal x * can be
j
j 1
iS j
n
s.t.
max
0 xi 0
0 z j 1.
Pipage Rounding
xi*
0 xi* 1
Choose 0 x *k 1 and 0 x * j 1 (i j ).
Define yi zi x *i for i k , j , and
yk x *k min( 1 x *k , x * j ), y j x * j min( 1 x *k , x * j ),
z k x *k min( 1 x * j , x *k ), z j x * j min( 1 x * j , x *k ).
y if F ( y ) F ( z )
Set x'
z otherwise.
Property
F ( x' ) F ( x*)
because F ( x( )) is convex w.r .t.
where xi ( ) x *i for i k , j , and
xk ( ) x *k , x j ( ) x * j .
y x(1 ) where 1 min( 1 xk , x j )
z x( 2 ) where 2 min( 1 x j , xk )
x* x(0)
Theorem
Maximum weight coverage problem has e /( e 1) approximat ion .
Proof
Let x be the integer solution obtained by pipage rounding. Then
L( x ) F ( x ) F ( x*) (1 1 / e) L( x*) (1 1 / e) opt.
Pipage Rounding
framework
Integer Programming
Consider a bipartite graph G (V ,U , E ).
max F ( x)
s.t.
x
e ( v )
e
max L( x)
pv for v V U
s.t.
xe {0,1} for e E.
x
e ( v )
e
pv for v V U
xe {0,1} for e E.
For integer feasible solution x,
F ( x) L( x)
Relaxation
max F ( x)
s.t.
x
e ( v )
e
max L( x)
pv for v V U
0 xe 1
for e E.
By ɛ-convexity, obtain
an integer solution
from x *, by x
pipage rounding, such
that
F ( x ) F ( x*)
x
s.t.
e ( v )
e
pv for v V U
0 xe 1
F ( x) c L( x)
for e E.
Solve easily to obtain
Optimal solution x *
L( x ) F ( x ) F ( x*) c L( x*) c opt
Pipage Rounding
Consider bipartite subgraph H x .
V
0 xij 1
U
Find R, either a maximal path or a cycle.
R
V
U
Find R, either a maximal path or a cycle in H x .
Decompose R into two matchings M 1 and M 2 .
Define x( ) by setting
xe if e R,
xe ( ) xe if e M 1 ,
x if e M .
2
e
Set 1 min(min
2 min(min
eM 1
eM 2
xe , min eM 2 (1 xe )),
xe , min eM 1 (1 xe )).
Then for [ 1 , 2 ], x( ) is feasible.
x( 1 ) if F ( x( 1 )) F ( x( 2 ))
x'
x( 2 ) otherwise
ɛ-convexity
For any R, F ( x( )) is convex w.r .t. .
Thus, for any [-1 , 2 ],
F ( x( )) max( F ( x(1 )), F ( x( 2 ))).
Pipage Rounding
Applications
D. Shin and S. Bagchi, Optimal Monitoring in Multi - Channel
Multi - Radio Wireless Mesh Networks, in Proc. of the
MobiHoc200 9, pp.229 - 238, 2009.
Consider a set of n normal nodes u1 ,..., un
and a set of m monitoring nodes v1 ,..., vm .
Each ui has ai normal radios ui1 ,..., uiai .
U {u11 ,..., u1a1 ,..., u1n ,..., unan }.
Each vi has ti monitoring radios.
Each monitoring radio can be tuned to a channel j 1,..., c.
Sij the set of normal radios that can be covered by any
monitoring ratio of vi tuned on channel j.
m
Si {Sij | j 1,..., c}, S i 1 Si .
Choose at most k sets from S with at most ti ones from Si
so as to maximize the number of normal radios covered by
selected sets.
n
max
x
l
l 1
s.t. xl
m
y
i , j:ul S ij
c
y
i 1 j 1
c
y
j 1
ij
ij
ij
, l {1,..., n}
k,
1, i I {1,..., m}
xl {0,1}, l {1,..., n}
yij {0,1}, i I , j J {1,..., c}.
Consider a set of n normal nodes u1 ,..., un
and a set of m monitoring nodes v1 ,..., vm .
Each ui has ai normal radios ui1 ,..., uiai .
U {u ,..., u ,..., u ,..., u }.
1
1
a1
1
1
n
an
n
U {u1 ,..., un }
Each vi has ti monitoring radios.
1
Each monitoring
radio can be tuned to a channel j 1,..., c.
Sij the set of normal radios that can be covered by any
nodes
monitoring ratio of vi tuned on channel j.
m
Si {Sij | j 1,..., c}, S i 1 Si .
Choose at most k sets from S with at most ti ones from Si
so as to maximize the number of normal radios covered by
selected sets.
nodes
y1 j
n
max L( y ) min( 1,
l 1
m
s.t.
c
y
i 1 j 1
c
y
j 1
ij
ij
y
i , j:ul S ij
ij
)
k,
j
S1
yij
Si
1, i I {1,..., m}
xl {0,1}, l {1,..., n}
yij {0,1}, i I , j J {1,..., c}.
j
n
max F ( y ) (1
l 1
m
s.t.
c
y
i 1 j 1
c
y
j 1
ij
ij
y1 j
(1 y
ij
))
i , j:ul S ij
k,
1, i I {1,..., m}
xl {0,1}, l {1,..., n}
yij {0,1}, i I , j J {1,..., c}.
yij
1
F ( y ) (1 ) L( y ).
e
F ( y ) has - convexity
because each maximal path contains only two edges.
Theorem
With pipage rounding, we can get an approximat ion
1
solution y with L( y ) (1 )opt.
e
Thanks, End
© Copyright 2026 Paperzz