--~
LABORATOiRE
LYSE ET
MODÉLiSA-TiON
DE SYSTEMES POUR
AIDE À LA DÉCISION
.
CNRS
SUR LA CONVERGENCE D'UN SYSTÈME
DIFFÉREN-TIEL DE PREMIER ORDRE, VECTORIEL, ORDINAIRE,
LINÉAIRE NON-HOMOGÈNE ET NON-AUTONOME
jlTÉ DE RECHERCHE
ASSOCIÉE
ESA
7024
.
Joël Biot
CERMSEM,UniversitéParis1
106-112, boulevardde l'Hôpital
75647 ParisCedex 13, France
Joel.Blot@univ-parisl.fr
UNiVERSITE
PARIS DAUPHINE
PLACE DU \1' DE
LATTRE DE
CEDEX
16
Vangelis Paschos
LAMSADE, Université Paris-Dauphine
Place du Maréchal De Lattre de Tassigny
75775 Paris Cedex 16, France
paschos@lamsade.dauphine.fr
TASS GNY
F-75775
PARIS
.
TELEPHONE
33 1)(01)
44 05 44 34
TÉLECOPIE
\ 44 05 40 91
.
NOTE DE RECHERCHE (RESEARCH PAPER) n° 27
septembre 2000
t!amsade
aauph nG.fr
\NES
!amsadedauph
nefr
.
ordinaire, linéaire non-homogene et non-autonome
On the onvergen e of a ve torial rst-order linear non-homogeneous
non-autonomous ordinary dierential equation
Joe lBlot
CERMSEM, UniversitéParis I
106-112boulevard de l'Hpital 75647 Paris Cedex 13,Fran e
Joel.Blotuniv-paris1.fr
VangelisTh. Pas hos
LAMSADE, Université Paris-Dauphine
Pla e duMaré hal deLattre de Tassigny 75775 Paris Cedex16,Fran e pas hoslamsade.dauphine.fr
Abstra t
We present a detailed study of the asymptoti al behaviourof the solution of a ve to-rialrst-orderlinearnon-homogeneousnon-autonomousordinarydierentialequation. This systemdes ribes,in termsof ontinuous mathemati s,the behaviour of thenatural greedy algorithmforthe minimumhittingset problemandit isdened in ourpaperAverage ase analysis ofgreedyalgorithms for optimisationproblems onsetsystems.
Résumé
Nousprésentonsuneétudedétailléedu omportementasymptotiquedelasolutiond'un système diérentiel de premier ordre, ve toriel, ordinaire, linéaire non-homogene et non-autonome.Cesystèmemodéliseentermesdemathématiques ontinuesle omportementde l'algorithme glouton pourle problème dutransversalminimum d'un hypergrapheet il est dénidansnotrearti leAverage aseanalysisofgreedyalgorithmsforoptimisationproblems onsetsystems.
In[2 ℄wehaveprovedthatthebehaviourofthenaturalgreedyalgorithmforminimumhittingset problem(or, equivalently, for minimum set overingproblem) an be des ribed bythe following
systemof equations: Em j = (j+1)m j+1 S jm j S; 1jh 1; Em h = 1 hm h S:
Then, usingstandard results on erningthe approximationofsmall stepsMarkov hains ([3℄), the solution of the above system is well asymptoti ally approximated by the solution ~y of the
following systemofdierential equations(where we useNewton'snotation for dy=d)
(S ;h ) 8 < : _ y ;h;j () = (j+1) h(k 1) R ;h hk y ;h;j+1 () j h(k 1) R ;h hk y ;h;j (); 1jh 1 _ y ;h;h () = 1 h h(k 1) R ;h hk y ;h;h ()
in the sensethat, sup
jy ;h;j
() (1=n)m
j
()j!0asn!1, inprobabilityfor ea h1j h ifonly the initial onditions y
;h;j
(0), 1j tend to xed limits (where nis the size of the instan eof theminimum hittingsetproblem; forthe rest ofthe notations, the interestedreader
;h ; y I ;;1 (0) = e + 1 X j=+1 e j (j 1)! y I ;;j (0) = e j j! ; 1j:
Observe that for ea h j, the orresponding RHS of the above equalities is equal to the almost sure limit ofm j (0)=n asn!1. Also,R ; = P j=1 jy I ;;j . Wedenote by ; the rst time at
whi h the last oordinate be omes0:
;
=minf>0:y ;;
()=0g
and set, for1j 1,
y I ; 1;j = y ;;j ( ; ) R ; 1 = 1 X j=1 jy I ; 1;j :
Then, weturnto thesystem(S ; 1
) ofdimension( 1)( 1)whi h weintegrate with the initial onditions: =0 andy
; 1;j (0)=y I ; 1;j , 1j 1. We dene ; 1 by ; 1 =minf >0:y ; 1; 1 ()=0g
and we repeat indu tively this pro edure for ea h h 1, taking as initial onditions for the
system(S
;h
)the nal onditionsfor (S ;h+1 ). We dene = X h=1 ;h :
In [2℄ we have given the main steps and intermediate results that prove the onvergen e of
. Sin e this proof is not only usefull in des ribing the asymptoti behaviour of the hitting set
algorithm but, also, interesting by itself, in this manus ript we present in details the study of the asymptoti behaviourof the sequen e(
)
2IN .
[1℄ N. Bourbaki,Fon tions d'une Variable Réelle(Hermann, Paris, 1976).
[2℄ J. Blot, W. Fernandez de la Vega, V. Th. Pas hos and R. Saad, Average ase analysis of greedy algorithms for optimisation problems on set systems,Theoreti al Computer
S i-en e,147 (1995), 267298.
[3℄ T. G. Kurtz, Solutions of ordinary dierential equations as limits of pure jump Markov
pro esses,J. Appl.Prob.7 (1970) 4958.