Constraint integer programming [Elektronische Ressource] / vorgelegt von Tobias Achterberg

De
GrötscJulitInBerlintegeragProgrammingnat.vDr.orgelegtverton,Dipl.-Math.henDipl.-Inf.DTgenehmigteobiaster:AThUnivterbDr.ergBixbBerlinersitderUSAFwissenscakultäthe:IBerlinIConstrainMathematikDissertationundhProf.haftenDr.derMartinThelecechehenersitätUnivProf.ersitätRobBerlinE.zuryErlangungUnivdesyakHouston,ThenderGradesDoktorAusspracder12.2007haften200783Dr.rer.JuliaFürhenacscassungEinDiesevArbaufeitenstelltteeinenvineinetegriertenerhaltenAnsatzausdassConstrFaintbPrProgramoutzen.grvammingw(CP)onenundkGemiscExphzwt-GanzzahligereinProgrammierungutoma-(ScMixefeh-dinInteFgerenePrdiesesogebgrConstrainamminger-,undMIPdas)t-Ganzzahligenvorliegendeor,hreibungdenewwirtenConstraufaintaufwInteeigerAspPrArboChip-Design-VgrThemaamming(CIPm)Ennennen.enenSoAndernfallswaufwohlGerätesMovdellierungs-Talswird,hwurfLösungsteceist.hnikConstrainenerdenbveiderdieFGesamelderungenieÿenAnsatzinT-TdenErfolgneuendieintetegriertenGemiscAnsatzbein,DieumeitdieheunKundvhenVStärkHinenihrenderGesambveidenpraktiscGebieteten.zubkalgorithmiscomeingegangen.binieren.
Publié le : lundi 1 janvier 2007
Lecture(s) : 19
Source : OPUS.KOBV.DE/TUBERLIN/VOLLTEXTE/2007/1611/PDF/ACHTERBERG_TOBIAS.PDF
Nombre de pages : 418
Voir plus Voir moins

hel
Constrain
2007
t
Univ
In
h
teger
Prof.
Programming
der
v

orgelegt
Dr.
v
he
on
E.
Dipl.-Math.
Houston,
Dipl.-Inf.
Aussprac
T
83
obias
Dissertation
A
Prof.

Martin
h
ec
terb
ersit?t
erg
Rob
Berlin
y
der
y
F
T
akult?t

I
12.
I
2007

nat.
Mathematik
genehmigte
und


ter:
haften
Dr.
der

T
Gr?tsc
ec
T


hen
Univ
Univ
Berlin
ersit?t
Dr.
Berlin
ert
zur
Bixb
Erlangung

des
ersit
ak
,

USA
hen
ag
Grades
wissensc
Doktor
hen
der
he:

Juli
haften
Berlin

D
Dr.
rer.Julia
F?rAlgorithmen
Zusammenf
hgebiets
assung
hiedene
Diese

Arb
on
eit
die
stellt
tspric
einen

in
erw
tegrierten
ausf?hrlic
Ansatz
anhand
aus
h
Constr
der
aint
ehlfunktionen
Pr
ei
o
mo
gr
T
amming
ten,
(CP)
erden.
und
ten
Gemisc
auf
h
erimen
t-Ganzzahliger
eite
Programmierung

(
tion
Mixe
haltung
d
lerhaftes
Inte
dem
ger
eld
Pr
Sc
o
Pro-
gr
en
amming
ts
,

MIP
b
)
ezien
v
Programmen
or,
Arb
den
dieser
wir
ertet
Constr
in
aint
das
Inte
endigen
ger
wird
Pr
ekte
o
eit
gr
erik
amming
innerhalb
(CIP
onic
)
?ssen
nennen.
t
So
Sp
w
w?rde
ohl
eisen,
Mo
f?hren
dellierungs-
erw
als
eilproblem

hafts-V
h
ob
L?sungstec
eine
hnik
Wir
en
t
b
k
eider
on
F
Struktur
elder

ie?en
auf
in
mit
den
ec
neuen
unserer
in
f?r
tegrierten
L?sen
Ansatz
h
ein,
en?tigt
um
v
die
liefert
un
Besc

omp

b
hen
er-
St?rk
arian
en
blic
der
Einu?
b
t-L?sungsv
eiden
on
Gebiete
hen
zu
Dab
k
esonders
om
hen
binieren.
Der
Als
der
w
efasst
eite-
der
ren
die
Beitrag
tiges
stellen
F
wir
Ele
der
A
wissensc
Chip-Hersteller

herstellen,
hen
he
Gemeinsc
einer
haft
gegeb
die
ation
Soft
t.
w
Chip
are
erhalten
SCIP
zu
zur
des
V
ann,
erf?gung,
Chip
die
wird.
ein
tiges
F
diesem
ramew
das
ork
ations-Problem,
f?r
gepr?ft
Constrain
gegeb
t
t
In
h
teger
aufw
Programming
wie
darstellt
als
und
teger

w
h
und
T
Reihe
ec
ezisc
hnik
die
en
einzelnen
des
der
SA
ausn
T-L?sens
hn
b
v
einhaltet.
unseren
SCIP
bisher
ist
SA
im
en

den
Co
de.
de
die
f?r
das
ak
te
a-
v
demisc
Gemisc
he
t-Ganzzahligen
und
b

w
h
Die
t-k
orliegende
ommerzielle
eit
Zw
eine
ec
he
k
hreibung
e
K
frei
onen

und
h.
ew
Unser
v
Ansatz
sc
des
V
Constrain
ten
t
Hin
In
k
teger
ihren
Programming
auf
ist
Gesam
eine
erhalten
V
v
erallgemeinerung
aufw
v
praktisc
on
Exp
MIP,
ten.
die
ei

b
h
auf
die
algorithmisc
V
Asp
erw
eingegangen.
endung
zw
b
Hauptteil
eliebiger
Arb
Constrain
b
ts

erlaubt,
mit
solange
Chip-Design-V

ation,
h
ein
diese
h
durc
Thema
h
des
lineare
ac
Bedingungen
der


k
Design
en
utoma-
lassen
darstellt.
falls
m
alle

ganzzahligen
dass
V
logisc
ariablen
En
auf
wurf
feste
Sc
W
der
erte
enen
eingestellt
ezik
sind.
en
Die
h
Constrain
Andernfalls
ts
der
w
feh-
erden
V
v
aufw
on
dass
einer
F
b
innerhalb
eliebigen
Ger?tes
K
k
om-
in
bination
der
aus
v
CP-
endet
und
Ein
MIP
h
-T
T
ec
in
hnik
F
en
ist
b

ehandelt.
erik
Dies
b
b
dem
einhaltet
wird,
insb
der
esondere
ene
die
haltkreisen
Domain
wurf
Pr

op
te
agation
haft
,
eist.
die
zeigen,
Relaxierung
dieses
der
blem
Constrain
Constrain
ts
In
durc
Program
h
delliert
lineare
erden

ann
h
geb
ungen,
eine
so
v
wie
problemsp
die
hen
V
an,
erst?rkung
die
der
der
Relaxierung
Constrain
durc
und
h
Gesam
dynamisc
haltung
h
utzen.
generierte
estrec
Sc
ungen
hnitteb
Industrie-Beispielen
enen.
er-
Die
hen
derzeitige
Ansatz
V
den
ersion
v
v
endeten
on
T-T
SCIP
hnik
en
und
th?lt
elegen
alle
Erfolg
K
Metho
omp
v
onenpresen
Abstra
imp

is
This
of
thesis
and
in
of
tro
ecied

h
the
building
no
to
v
to
el
vior
paradigm

of
mo

as
onstr
of
aint
dieren
inte
er.
ger
p
pr
with
o
design
gr
of
amming
w
(CIP

),
v
whic
v
h
w
in
and
tegrates
of


onstr
metho
aint
these
pr

o
ts
gr
of
amming

(CP
the
)
solv
and
programming,
mixe
ation
d
t
inte
ufacturers
ger
that
pr
to
o
Otherwise,
gr
an
amming

(MIP)
emplo
mo
of
deling
pr
and
whic
solving


.
hniques.
problem
It
t
is
n
supplemen
that
ted
ts
b
set
y

the
t
soft

w
vii
are
ts
SCIP
extensiv
,
to
whic
v
h
the
is

a
MIP
solv
e
er
the
and

framew
v
ork
of
for
In

in
t
thesis
in
design
teger
whic
programming
imp
that
of
also
Chip
features
v
SA
e
T

solving


sp
hniques.

SCIP

is
sho
freely
b
a
ma
v
in
ailable
it
in
ed.

t

hip
de
is
for
erty
academic
oblem
and
is

whether
purp
a
oses.
ert
Our
e

w
t
b
in
as
teger
teger
programming
vide

b
h

is
the
a
individual
generalization
the
of
whole.
MIP
extensiv
that

allo
CIP
ws
the
for
SA
the
and

the
of
metho
arbitrary
of

ingredien
ts,
and
as
t
long
e
as
results
they

turn
t
in
arian
to
for
linear
individual

blo
ts
ks
on
a
the
solv

W
tin
fo
uous
on
v
algorithms
ariables
their
after
on
all
o
in
erall
teger

v
the
ariables
er.
ha
addition
v
mixed
e
teger
b
the
een
deals
xed.

The
veric

,
ts,
h
ma
an
y
ortan
they

b
electronic
e
automation.
linear
man
or
ha
more
e

mak
are
sure
treated
the
b
design
y
a
an

y
the


bination
the
of
hip.
CP
the
and
hip
MIP
ould

w
hniques:
erroneous
the
eha
propagation
that
of
y
the
failures
domains
the
b
where
y
is

y
t
An
sp
ortan

subproblem
algorithms,

the
design
generation

of
the
a
op
linear

relaxation
pr
and
,
its
h
solving
to
b
erify
y
a
LP
satises
metho
sp
ds,
prop
and
y
the
W
strengthening
sho
of
ho
the
this
LP

b
e
y
deled


plane
in
separation.
program
The
pro

a
t
um
v
er
ersion
problem-sp
of
algorithms
SCIP
exploit


with
the
all

of
and
the


a

Another
onen
of
ts
e
to
b
solv
hmarks
e
our
mixed

in
to
teger

programs.
state-of-the-art
In
T
the
dology
thesis,
do
w
ts
e


our
v
d.
er
most
A
station

lots
wledgements
w
W
lot
orking
w
at
and
the

Zuse
with
Institute
the
Berlin
I
w
to
as
t
a
almost
great

exp
to
erience
momen
for
d
me,
e
and
dieren
thanks
to
to
o
the
little
v
e
ery
as
exible
y
p
do
eople
not
in
king
the
on
administrations
again
and
Infineon
managemen
p
t

lev
During
els
as
of
hips
ZIB
all
and
industry
m

y
and
new
the
emplo

y
T
er

ILOG,
v
this
his
exp
him
erience
at

Russians.
tin
e
ues.
hel
It
he
is
lik
a
y
pleasure
pap
to
m
b
y
e
to
surrounded

b
the
y
Gr?tsc
lots
W
of
b

)

at
ev

en

though
teger
some
o-y
of

them
,
ha

v
its
e
thank
the
Ha
nast
p
y
y
habit
of
to
at

b
in
the
to
within
m
in
y
w

o
(without
a
b
Minsk
eing
OneSpin
formally
Berlin
in
the
vited!)
to
with
the
the
German
only
I
purp
sub
ose
ed
of
the
w
y
asting
that
m
ha
y
thank
time
Martin
b
his
y
the
asking
me
strange
er
questions
F
ab
w
out
I
SCIP
a
and
Instead,
Cplex
sitting
.

And
whole
eating
m
a
in
piece
a
of
h
sp
the
onsored
And

o
e
to
from
his
time
group
to
B?ttner
time
h
while
the
discussing
OneSpin
m
in
uc
pro
h
p
more
solving
in
design
teresting
an


than
t
optimization
metho
or
t
mathematical
p
programming
the
is
h
alw
V
a
learned
ys
out
a
of
v
ho
aluable


eried.
Thank
Brinkmann
y
supp
ou,
him
Marc!

W
of
ork
made
on

this
from
thesis
other
actually
e
b
I
egan
test
already
hmark
in

2000
early
with
e
the
duration
w

ork
p
on
to
m
the
y
Y

No
science
He
master's
t
thesis.
her
This
is
w
orking
as
tions
ab

out
is
applying
exible
neural
of
net
he
w
in
orks
with
to
aairs
learn
try
go
.
o
met
d
k

at
hing
a
strategies
he
for
the
MIP
A
solv
he
ers.
ugged

gang
h
though
on
is

Eastern
h
v
a
to

m
is
advisor
only
Gr?tsc
p
for
ossible

if
for
y
freedom
ou
oered
ha
to
v
whatev
e
I
the
ed.

or

t
de
o
of
ears,
a
did
state-of-the-art
publish
MIP
single
solv
er!
er.
I
F
as
ortunately
in
,
y
the

former
the
ZIB
da
mem
on
b
y
er
de
Alexander
order
Martin
get
made
basis
his
whic
solv
I
er

SIP
actual
a
h.
v
then,
ailable

to

me,
thanks

Martin
h
hel:
that

I
the
w
of
as
olfram
reliev
at
ed

from
later
in
ecame
v
spin-o
en
y
ting
Solutions
the
resulted
wheel
the
a



time.
the
With

the
t:
help
the
of
hip
Thorsten
v
K
problem,
o
ideal

to
h,
kle
I

learned
in
a
programming
lot
ds.
b
the
y
w
analyzing
ear
his
erio
algorithms.
of
Since
pro
I
whic
studied
w
b

oth,
alse-XT
mathematics
I
and
a

ab
science,
the
I
design
alw

a
and
ys
w
lo

ok
b
ed
v
for
I
a
Raik

for
that
his

ort.
bines
ving
the
as
t

w

o
erson
elds.
the
In
partner
2002,
m
it
pro


to
t
m
most
y
the
mind
pro
that
w
the
had
in
ZIB:
tegration
obtained
of
of
in
and
teger

programming
data,
and
I

ed
t
data
programming
enough
w
b
ould
useful
b
the
e
of
a
pro
p
Another

teresting

erson
in
learned
this
kno
regard.
within
Unfortunately
pro
,
is

ak
h
v
an
vik
in
v.
tegration
is
w
brillian
as
SA
w

a
from
y
who
b
no
ey
w
ond
for
the
Solu-
scop
in
e
h.
of
the
SIP
administration
,
a

more
h
than
that
one
I
Ba
had
aria,
to

start
Berlin
from
order
scratc
deal
h
all
at

the
regarding
end
en
of
to
2002
y
with
W
a
rst
quic
when
kly

gro
ed
wing
up

a
de
w
that
y
I
after

arriv
SCIP
with
in
train
order
Berlin-Ostbahnhof.
to
t
emphasize
Ostbahnhof,
its
w
relation
m
to
b
SIP
a
.
of
A
They
t
t
this
he
p
a
oin
ypical
t,
ix
Iit
x
v
A


impro
wledgements
allo
Europ
Miek
ean

who
the

Orlo
a
ganze
lot
ic
of
h!

pro
in
yd
order
to
to
the
buy
,
a
are

hmark
in
vier
German
henenden
y
b
.
h
F
SCIP
ortunately
I
,
are
b
Markus
y
All
sho
Most
wing
(and
some
a
of
ecame
his
liv
pap
Eisen
ers
h
he
that

spare

ic
vince
freigehalten
them
T
that
T
he
e.
is
dieser
only
die
a
v
p
en
o
Berlin,
or
10
re-
this

grateful
her
y
who
Heinz,
do
Pfetsc
es
TU
not
Zonghao
ha
help
v
thesis
e
w
an
h,
y
around
money
thank
.
her
After
but
telling

me
the
this
y
story
y
,
Marik
he
k,
explained
W
to
orking
me
to
the
to
k
m
ey
ab

Du
in
den
SA
esondere
T
ic
solving
h
during
ommen
the
die
remaining
erbrac
10
edaure
min
und
utes
zu
in
e.
the
Leb
sub
sc
w
jetzt
a
einem
y
h
train:




y
analysis
I
and
thankful
the
F
two
v
watche
all
d
of
liter
whic
als
Berthold,
scheme
K
for
and
fast
from
Bo
edler
olean
and

e
strain
from
t
y
propagation.
me
I
e
w
a
as
all,
v
lik
ery
Marc

actually
and
ted
started
%
with
I
the
Calla
in
(ILOG)
tegration
to
of
and
the
she
two
k
watche
mak
d
time.
liter
has
als
with
scheme
deciencies.
in
thank
to

the
Stefan
set
Neumann,

Christian
v
Sc
ering
and

der,
t
ones
handler

a
iden
few
and
da
ed
ys
end
later.

The
b
generalization
Am
of
dank

Dir,
analysis
mir
to
?b
mixed
k
in
es
teger
den
programming
ertragen,
follo
fast
w
erst
ed
h
after
Hause
three
und
w

eeks.
o
As
ZIB

t
analysis
h
turned
sehr,
out

to
Kinder
b
k
e
h
a
ommen
k
esondere
ey

ingredien
v
t

for
an
solving
ergezogen.
the
erde

er
hip
F
v
knden.

e
problem
obias
with
terb

y
t
of
in
b
teger
another
programming,
%.
I
am
am
ery
v
for
ery

thankful
urthermore,
to
am
Y
ery
ak
to
o
the
v
of-readers
for
m
p
thesis,
oin
h
ting
Timo
me
Stefan
in
Thorsten
to
o
this
h,

Marc
I
h
though
ZIB,
t
W
after
from
ha
Kaiserslautern,
ving
Llo
implemen
Clark
ted
and
more
Gu
than
ILOG.
250
of
000
ou
lines
ed
of
to
C
v

the
de
quite
for
bit.
SCIP
of
and
I
the
ould

e
hip
thank
v
Pfetsc

who
solv
read
er,

the
on)
writing
75
of
of
the
thesis.
thesis
also
w
Kathleen
ould
w
b
y
e
for
a
willingness
piece
review
of
grammar

punctuation,
e.
unfortunately
What
b
a

mistak
and
e!
not
The
e
time
in
passed
Therefore,
b
reader
y
to
,
e
and
m
suddenly
language
I
Finally

I
m
m
y
ZIB
self-imp
Andreas
osed
bl?tter,
dead-
Heinz,
line
a
(to
Sebastian
b
wski,
e
Raac
nished
Thomas
b
hlec
efore
te,
m
Steen
y
ei-
30th
who
birthda
the
y)
w
without
on
ha
hines
ving
are
ev

en
mine,
started
who
to
w
write
me
m
sp
y
their
thesis.
CPU
Of
on

y
SCIP

impro
runs.
v
allermeisten
ed
er

e
during
h
that
Julia.
time,
hast
whic
die
h
Zeit
is
er
also

the
en

und
tribution
insb
of
in
m
letzten
y
Monaten
great
dass
studen
h
ts
jeden
Kati
ag
W

olter

and
t
Timo
h
Berthold.
gek
A
bin
dditionally
zum
,
eil
the
h
public
W
visibilit

y
am
of
v
SCIP
h

hab
dramatically
Ic
,
b
thanks
es
to
dass
Hans
h
Mittelmann's
h
b
die

in
hmarking
Zeit
w
aum
ebsite.

Nev
t
ertheless,
ek
it
hab
needed
Insb
Thorsten
sind
K
ersten
o
hs

ensmonate
h
on
to
e

h
vince
hnell
me
mir
that
or?b
starting
Ab
to
w
write
ic
things
ab
do
zu
wn
geregelten
is
amilienleb
more

imp
Ic
ortan
lieb
t

than
T
to
A
impro
h
v
erg
e
Ma
the
2007
p

Soyez le premier à déposer un commentaire !

17/1000 caractères maximum.