Chapter 7 Linear Programming
-
Upload
graham-mcneil -
Category
Documents
-
view
35 -
download
2
description
Transcript of Chapter 7 Linear Programming
Pipage Rounding
example
Maximum Coverage
.by hit in subsets of weight total themaximize to||with
subset a find ,integer positive a and in subsetson function
weight enonnegativ with },...,1{set a of subsets of family aGiven
XCpX
IXpCw
nIC
ILP Formulation
.10
}1,0{
,
,,...,1 , s.t.
max
1
1
j
i
n
ii
Siji
m
jjj
z
x
px
mjzx
zw
j
Alternative Formulation 1
}1,0{
, s.t.
},1min{)( max
1
1
i
n
ii
m
j Siij
x
px
xwxLj
Alternative Formulation 2
}1,0{
, s.t.
))1(1()( max
1
1
i
n
ii
m
jSi ij
x
px
xwxFj
Relationship
.10for ),1min(11
|)|(
)1(
1)1(1
11 where
)()(
jjj
j
j
j
Sii
Sii
Sii
k
Sii
j
k
Sii
Sii
xxcxck
x
Skk
x
x
ec
xcLxF
Proof.
Relationship
.1for ),1min(1
1111
|)|(
)1(
1)1(1
11 where
)()(
jj
j
j
j
Sii
Sii
k
k
Sii
j
k
Sii
Sii
xxcckk
x
Skk
x
x
ec
xcLxF
Proof.
).1
1()1
11()( So,
[0,1].in concave and increasing monotone is )(
.1
11)1( ,0)0(
.11)(Set
ez
kzzg
zg
kgg
k
zzg
k
k
k
.0)1
(1)1()(''
.01)1
(1)('
2
11
kk
zkzg
k
z
kk
zkzg
k
kk
0 1
Relaxation
.10
, s.t.
},1min{)( max
1
1
i
n
ii
m
j Siij
x
px
xwxLj
.10
00
,
,,...,1 , s.t.
max
1
1
j
i
n
ii
Siji
m
jjj
z
x
px
mjzx
zw
j
).( in time computed
becan * optimalAn 5.3nO
x
Pipage Rounding
10 * ix*ix
otherwise.
)()( if 'Set
).*,*1min(* ),*,*1min(*
),*,*1min(* ),*,*1min(*
and ,,for * Define
).( 1*0 and 1*0 Choose
z
zFyFyx
xxxzxxxz
xxxyxxxy
jkixzy
jixx
kjjjkjkk
jkjjjkkk
iii
jk
Property
.*)( ,*)(
and ,,for *)( where
.t.convex w.r is ))(( because
*)()'(
jjkk
ii
xxxx
jkixx
xF
xFxF
)0(*
),1min( where)(
),1min( where)(
22
11
xx
xxxz
xxxy
kj
jk
Theorem
.)/11(*)()/11(*)()()(
Then rounding. pipageby obtainedsolution integer thebe Let
.ionapproximat)1/( has problem coverage weight Maximum
optexLexFxFxL
x
ee
Proof
Pipage Rounding
framework
Integer Programming
.for }1,0{
for s.t.
)( max
)(
Eex
UVvpx
xF
e
vve
e
.for }1,0{
for s.t.
)( max
)(
Eex
UVvpx
xL
e
vve
e
)()(
,solution feasibleinteger For
xLxF
x
).,,(graph bipartite aConsider EUVG
Relaxation
.for 10
for s.t.
)( max
)(
Eex
UVvpx
xF
e
vve
e
.for 10
for s.t.
)( max
)(
Eex
UVvpx
xL
e
vve
e
)()( xLcxF Solve easily to obtain Optimal solution
By ɛ-convexity, obtain an integer solution from , by pipage rounding, such that
*x*x x
*)()( xFxF optcxLcxFxFxL *)(*)()()(
Pipage Rounding
.subgraph bipartiteConsider xH
V U10 ijx
cycle. aor path maximal aeither , Find R
R
V U
.in cycle aor path maximal aeither , Find xHR
otherwise )(
))(())(( if )('
feasible. is )( ],,[for Then
)).1(min,min(min
)),1(min,min(minSet
. if
, if
, if
)(
settingby )( Define
. and matchings twointo Decompose
2
211
21
2
1
2
1
21
12
21
x
xFxFxx
x
xx
xx
Mex
Mex
Rex
x
x
MMR
eMeeMe
eMeeMe
e
e
e
e
ɛ-convexity
. .t.convex w.r is ))(( ,any For xFR
))).(()),((max())((
],,[-any for Thus,
21
21
xFxFxF
Pipage Rounding
Applications
2009. 238,-pp.229 9,MobiHoc200
theof Proc.in Networks,Mesh WirelessRadio-Multi
Channel-Multiin Monitoring Optimal Bagchi, S. andShin D.
sets. selected
by covered radios normal ofnumber themaximize toas so
from ones most at with from sets most at Choose
. },,...,1|{
. channelon tuned of ratio monitoring
anyby covered becan that radios normal ofset the
.,...,1 channel a to tunedbecan radio monitoringEach
radios. monitoring has Each
}.,...,,...,,...,{
.,..., radios normal has Each
.,..., nodes monitoring ofset a and
,..., nodes normal ofset aConsider
1
11
11
1
1
1
1
ii
m
i iiji
i
ij
ii
ann
a
aiiii
m
n
StSk
SScjSS
jv
S
cj
tv
uuuuU
uuau
vvm
uun
n
i
}.,...,1{, },1,0{
},...,1{ },1,0{
},...,1{ ,1
,
},...,1{ , s.t.
max
1
1 1
:,
1
cJjIiy
nlx
mIiy
ky
nlyx
x
ij
l
c
jij
m
i
c
jij
Sujiijl
n
ll
ijl
sets. selected
by covered radios normal ofnumber themaximize toas so
from ones most at with from sets most at Choose
. },,...,1|{
. channelon tuned of ratio monitoring
anyby covered becan that radios normal ofset the
.,...,1 channel a to tunedbecan radio monitoringEach
radios. monitoring has Each
}.,...,,...,,...,{
.,..., radios normal has Each
.,..., nodes monitoring ofset a and
,..., nodes normal ofset aConsider
1
11
11
1
1
1
1
ii
m
i iiji
i
ij
ii
ann
a
aiiii
m
n
StSk
SScjSS
jv
S
cj
tv
uuuuU
uuau
vvm
uun
n
i
},...,{ 1 nuuU
nodes
1
nodes
}.,...,1{, },1,0{
},...,1{ },1,0{
},...,1{ ,1
, s.t.
),1min()( max
1
1 1
1 :,
cJjIiy
nlx
mIiy
ky
yyL
ij
l
c
jij
m
i
c
jij
n
l Sujiij
ijl
jy1
ijy
j
jiS
1S
jy1
ijy
}.,...,1{, },1,0{
},...,1{ },1,0{
},...,1{ ,1
, s.t.
))1(1()( max
1
1 1
1 :,
cJjIiy
nlx
mIiy
ky
yyF
ij
l
c
jij
m
i
c
jij
n
l Sujiij
ijl
edges. only two containspath maximaleach because
convexity - has )(
).()1
1()(
yF
yLe
yF
.)1
1()( with solution
ion approximatan get can werounding, pipageWith
opte
yLy
Theorem
Thanks, End