Gma1 2 2013 Continut

GAZETA MATEMATIC
A
SERIA A
ANUL XXXI(CX) Nr. 12/ 2013
ANIVERS
ARI
Profesorul Ioan Tomescu la a 70-a aniversare
La 5 noiembrie 2012 domnul profesor universitar Ioan Tomescu, mem-
bru corespondent al Academiei Romane, a mplinit frumoasa v arsta de 70
de ani. La ceas aniversar mi revine onoarea si placerea de a trece n revista
principalele capitole ale unei viet i si cariere de except ie, asa cum le cunosc n
urma celor aproape patruzeci de ani de colaborare cu dansul.
Ioan Tomescu s-a nascut n orasul Ploiesti, judet ul Prahova, ntr-o fa-
milie de distinsi intelectuali.
Universitatea
In perioada 19601965 a urmat cursurile Facult at ii de Matematic a a

Universitat ii din Bucuresti, pe care a absolvit-o cu Diploma de Merit sust i-
n and lucrarea de licent a cu titlul ,,Analiza si sinteza multipolilor cu contacte.
In timpul student iei a obt inut premiul ntai si premiul al doilea la Olimpiada
Nat ional a de Matematica pentru student i, Bucuresti, 1961 si 1962.
Titlul de doctor l-a obt inut n anul 1971 cu teza ,,Metode combinatorii
n teoria automatelor nite sub conducerea academicianului Gr. C. Moisil.
In aceast a teza se propune o metoda matriciala pentru determinarea tuturor

perechilor de stari compatibile ale unei masini secvent iale Mealy si se arata ca
minimizarea unei astfel de masini poate redusa n anumite cazuri la proble-
ma determin arii partit iei cromatice de cardinal minim a mult imii varfurilor
unui graf.

In acelasi an, 1971, a obt inut Premiul pentru matematici aplicate la
First Balkan Mathematics Competition for Students and Young Researchers,
Bucuresti. Peste numai patru ani, n 1975, avea s a obt ina Premiul Gheorghe
T it eica al Academiei Romane pentru cartea ,,Introducere n combinatoric a,
care a fost tradus a n englez a si maghiara.
Cariera universitara
In cadrul catedrei de Informatica a Universitat ii din Bucuresti a parcurs

treptele carierei universitare si a fost pe rand preparator n perioada 1965
1968, asistent n perioada 19681972, lector n perioada 19721990, iar din
1990 a devenit, direct, profesor universitar.
2 Anivers ari
Sub conducerea sa, 18 matematicieni din t ara si din str ainatate si-au
sust inut tezele obt inand titlul de doctor n matematica: Eugen Mandrescu,
Virgil Domocos, Laurent iu Modan, Cristina Vertan, Hazim A. Farhan, Petri-
sor Gut a, Laura Ciupala, Akhlaq Ahmad Bhatti, Imran Javaid, Mohammad
Tariq Rahim, Mircea Adam, Syed Ahtsham Ul Haq Bokhary, Gabriela Mi-
hai (Cristea), Ruxandra Marinescu-Ghemeci (Verman), Muhammad Imran,
Salma Kanwal, Sana Javed, Ayesha Riasat.
Activitatea si pozit ii ocupate
Incepand cu anul 2005 a fost visiting professor la Abdus Salam School of

Mathematical Sciences Government College University, Lahore, Pakistan,
unde a primit premiul Best Ph. D. Advisor n anul 2009. De asemenea, a fost
visiting professor la Department of Computer Science Auckland University,
New Zealand, n anul 1995 si la Department of Applied Mathematics The
University of Tirana, Albania, n anul 1974, precum si Visiting Professor
Research Fellow la School of Computing National University of Singapore,
n anul 2002.
In perioada 19902007 a fost seful catedrei de Informatica din Univer-

sitatea din Bucuresti.
Dintre numeroasele pozit ii academice ocupate trebuie remarcate acelea
de membru n Comisia Nat ional a de atestare a titlurilor, diplomelor si grade-
lor universitare a Ministerului Educat iei Nat ionale, din anul 1996 panan anul
2006, precum si cea de secretar al Comisiei de stiint e exacte la C.N.E.A.A.,
din anul 1994 pana n anul 2005.
Domnul profesor Ioan Tomescu a fost conducatorul delegat iei Romaniei
la Olimpiada Internat ional a de Matematica n intervalele de timp 1983
1986 si 19901994, iar n perioada 19901994 a fost conducatorul delegat iei
Romaniei la Olimpiada Balcanica de Matematica.
Este referent la un numar impresionant de publicat ii si edituri: Revue
Roumaine de Mathematiques Pures et Appliquees, Bulletin Mathematique
de la Societe des Sciences Mathematiques de Roumanie, Studii si Cercetari
Matematice (Bucuresti), Analele Universitat ii Bucuresti, seria Matematica,
Gazeta Matematica seria A (Bucuresti), Journal of Graph Theory, Discrete
Mathematics, Discrete Applied Mathematics, Random Structures and Algo-
rithms, Communications in Mathematical Chemistry (MATCH), Graphs and
Combinatorics, Discussiones Mathematicae Graph Theory, Journal of Com-
binatorial Theory (A), Australasian Journal of Combinatorics, Electronic
Journal of Combinatorics, International Journal of Mathematical Sciences,
Journal of Applied Mathematics & Computing, Ars Combinatoria, Utilitas
Mathematica, Mathematical Reviews (din 1976), Zentralblatt f ur Mathe-
matik (din 1968), la editurile Academiei, Stiint ica, Didactica si Pedagogic a,
Tehnica (din 1967).
Este editor sef al revistei Bulletin Mathe matique de la Societe des Sci-
ences Mathematiques de Roumanie si membru n comitetele de redact ie ale
Profesorul Ioan Tomescu la a 70-a aniversare 3
unor reviste de prestigiu, cum ar Electronic Journal of Graph Theory and
Applications.
In anul 1987 a fost unul dintre organizatorii conferint ei Semester of

Combinatorics, Stefan Banach Mathematical Center, Warszaw, Poland.
Cursuri predate
La Facultatea de Matematic a din Universitatea din Bucuresti a pre-
dat cursurile: ,,Structuri de date, ,,Introducere n programare, ,,Algoritmi
numerici si nenumerici, ,,Programare liniara, ,,Metode numerice n infor-
matica, ,,Combinatorica si teoria grafurilor, ,,Teoria grafurilor si aplicat ii,
,,Tehnici de optimizare combinatoriala si ,,Teoria automatelor.
A mai predat de asemenea cursurile ,,Numerical and Nonnumerical Pro-
gramming Techniques si ,,Graphs and Operations Research (International
Graduate UNESCO Courses, 19781982), ,,Data Structures (Department
of Computer Science, Auckland University, New Zealand, 1995), ,,Sorting
and Searching si ,,Applications of Graph Theory to Operations Research
(Department of Applied Mathematics, The University of Tirana, Albania,
1974).
Opera
De-a lungul unei prolice cariere de cercet ator, nentrerupta p an a astazi,
domnul profesor Ioan Tomescu a obt inut importante rezultate n teoria cir-
cuitelor combinat ionale, teoria extremal a a grafurilor (n special n probleme
privind colorarea varfurilor unui graf si polinoame cromatice), hypergrafuri si
conexiunea lor cu inegalitat i de tip Bonferroni, probleme Hamiltoniene pen-
tru anumite clase de grafuri, aplicat ii ale grafurilor n chimie, combinatorica
cuvintelor, teoria metrica a grafurilor.
In ultima perioada, activitatea stiint ica a domnului profesor Ioan

Tomescu s-a desf asurat n urmatoarele direct ii: Teoria cromatica a hiper-
grafurilor; Dimensiunea metrica si dimensiunea partit ie a grafurilor; Indicele
distant a grad si indicele Randic ale unui graf; Proprietat i asimptotice ale
factorilor cuvintelor peste un alfabet dat; Teorie Ramsey pentru perechi de
grafuri particulare; Proprietat i asimptotice ale grafurilor si hipergrafurilor
relative la un diametru dat.
Este autorul a peste 155 lucr ari stiint ice publicate n reviste de specia-
litate printre cele mai bine cotate din lume, a prezentat 9 lucr ari n volume
ale unor conferint e sau colect ii, a publicat 4 cursuri si manuale universitare,
11 cart i, o monograe si 29 de alte articole multe din acestea nregistrand
numeroase citari. A sust inut 17 expuneri la conferint e si 12 conferint e la
universit at i.
In anul 2009 a fost ales membru al International Academy of Mathe-

matical Chemistry.
De asemenea, din anul 2008 este presedinte de onoare al Societat ii de
Stiint e Matematice din Rom ania.
4 Articole
In anul 2000, domnul profesor Ioan Tomescu a fost ales membru cores-
pondent al Academiei Romane.
Personalitate senina si luminoas a, nnobilata de naturalet e, modestie,
sobrietate si caracter, domnul profesor Ioan Tomescu este creatorul unei
opere matematice greu de cuprins pentru a descrisa n cateva randuri,
o opera n realizarea careia adancimea adevarului matematic descoperit si
estetica n care acesta este prezentat au dat ntotdeauna aleasa marc a a
cercetatorului.
Domnule profesor Ioan Tomescu, va urez cu aceasta ocazie sanatate
si succes n aventura cercet arii matematice cu care v-at i identicat ntreaga
viat a, devenind astfel un minunat exemplu pentru noi tot i. La Mult i Ani !
Conferent iar doctor Dragos-Radu Popescu
ARTICOLE
Variat iuni pe o problema de olimpiada
Mihai Cipu
1)
Se dedic a unui mare problemist,
Profesorul Ioan Tomescu
Abstract. After studying by elementary techniques the integers repre-
sentable as
a
2
+b
2
ab+1
, we shall consider the same problem from a dierent
viewpoint. In the second part of the paper we shall study the integers q
for which there exist integers a, b, c such that q =
a
2
+b
2
+c
2
1+ab+bc+ca
.
Keywords: generalized Pell equations; inhomogeneous ternary quadrics;
quadratic reciprocity
MSC : Primary: 11D09; Secondary: 11D72; 11Y50
1. Problema
La cea de a 29-a olimpiad a internat ional a de matematica, desfasurata
n 1988 n Australia, concurent ii au avut de rezolvat urmatoarea problema:
P6. Daca a, b si q =
a
2
+b
2
ab+1
sunt ntregi pozitivi, q este patrat perfect.
Problema, propusa din partea R.F.G. de catre Stephan Beck, are o is-
torie deosebit de interesanta, relatata cu nume si date precise de A. Engel [4].
Potrivit acestuia, niciunul dintre cei sase problemisti faimosi nsarcinat i cu
selectarea problemelor pentru olimpiada nu a reusit sa o rezolve, asa ca au
apelat la patru renumit i specialisti n teoria numerelor din Australia. Nici
acestia nu au produs o solut ie complet a n sase ore (ceea ce reprezinta un
1)
Institutul de Matematica ,,Simion Stoilow al Academiei Romane, Unitatea de cerce-
tare nr. 5, C. P. 1-764, 014700 Bucuresti, Mihai.Cipu@imar.ro
M. Cipu, Variatiuni pe o problem a de olimpiad a 5
interval de timp mult mai mare decat cel la dispozit ia concurent ilor, care
au de rezolvat trei probleme n patru ore si jumatate). Cu toate acestea,
comisia de select ie a plasat problema pe lista scurta supusa votului juriului,
notand-o cu doua stelut e, nsemnul rezervat problemelor extrem de dicile,
put in probabil a date elevilor. Totusi, dupa dezbateri ndelungate, juriul a
hotarat a o selecta ca ultima problema din concurs.
Dupa aceasta decizie curajoasa, istoria problemei intra n domeniul pu-
blic. Dintre cei 268 de participant i la OIM, 11 au gasit solut ii perfecte. Pentru
aceasta performant a, dar mai ales pentru cele ulterioare, acesti ,,recordmeni
merit a a nominalizat i. Doi dintre ei sunt din Bulgaria: Emanouil Atanassov
(devenit cercet ator n informatic a la Institutul pentru Calcul Paralel al Aca-
demiei Bulgare de Stiint e) si Zvezdelina Stankova (din acest an profesor vi-
zitator la University of California din Berkeley); alt i doi elevi au reprezen-
tat R.P. Chineza: Hongyu He (acum profesor la Louisiana State University,
Baton Rouge, S.U.A.) si Xi Chen (University of Alberta, Canada); echipa
Romaniei a avut si ea doi membri cu punctaj perfect n concurs: Nicusor
Dan (cercetator la IMAR) si Adrian Vasiu (profesor la Binghamton Uni-
versity, State University of New York); iar din U.R.S.S., Nicolai Filonov
(actualmente cercet ator la Institutul de Matematic a Steklov al Academiei
Ruse de Stiint e si conferent iar la Facultatea de Fizica a Universitat ii de
stat din St. Petersburg) si Sergei Ivanov (exist a doi omonimi: unul la St.
Petersburg, cercetator la Institutul de Matematic a Steklov, celalalt profesor
la University of Illinois din Urbana-Champaign, S.U.A.). Au mai obt inut 7
puncte la problema a sasea cate un elev din Austria, anume Wolfgang Stocher
(matematician si programator la rma SKF

Osterreich AG, Austria), Canada
Ravi Vakil (profesor preferat al student ilor de la Stanford University) si
R.P.D. Vietnam Ngo B`ao Chau (acesta, dupa studii la Ecole Normale
Superieure din Paris, a demonstrat complet ,,lema fundamentala pe care se
bazeaz a programul lui Langlands, astfel can 2010 a obt inut Premiul Fields si
un post de profesor la Universitatea din Chicago). S-a mai notat cu 6 puncte
lucrarea lui Mouaniss Belrhiti Alaoui, care a reprezentat Regatul Maroc, dar
care a studiat la Ecole Polytechnique, acum ocupandu-se, n calitate de ad-
ministrator principal de portofoliu la Agent ia de Investit ii a Emiratului Abu
Dhabi, de sporirea valorii petrodolarilor. Aproape de solut ia complet a a fost
si Julien Cassaigne (Frant a), notat cu 5, actualmente cercetator la Institutul
de Matematica de la Luminy. Cu except ia unei note de 4, obt inuta de John
Woo (S.U.A.), care n-a ment inut contactul nici cu matematica, nici cu vre-
unul din domeniile cu vizibilitate pe internet, lucrarile celorlalt i concurent i
cont ineau mai put in de jumatate din rezolvarea complet a. Sa ment ionam
ca printre cei ce au obt inut un singur punct se aa si un elev din Australia
ce tocmai mplinise 13 ani n timpul competit iei si care rezolvase perfect
cele cinci probleme precedente din concurs Terence Tao, si el laureat al
Premiului Fields (n 2006, ns a) si acum profesor la UCLA.
6 Articole
Problema a sasea din 1988 a avut multa vreme reputat ia de cea mai
dicila chestiune data la OIM, cu media punctajelor pentru cei 268 de partici-
pant i de 0,634, 11 rezolvari perfecte si 189 note de 0.

Intre timp, nsa, a
pierdut acest titlu: problema a treia din 2007 are media 0,304, ind rezolvat a
n concurs de doar doi elevi (Caili Shen din R.P. Chinez a si Mladen Radojevic
din Serbia), iar media pentru ultima problema din aceeasi edit ie este de
doar 0,152, n condit iile n care a fost rezolvat a complet de 5 din cei 520
competitori, iar 473 de lucr ari au fost notate cu 0.
2. Solutia
In aceast a sect iune este prezentat a rezolvarea problemei P6 si a catorva

altora, str ans nrudite.
Sunt posibile mai multe abordari. Dou a dintre ele invoca principiul
extremal si cel al coborarii innite. Desigur, o solut ie complet a cont ine o serie
de alte detalii, a c aror combinare nu este lipsita de ranament si atractivitate,
asa cum ne vom convinge imediat mai jos. Pornind de la astfel de solut ii,
problema init ial a poate prezentat a ntr-o multitudine de moduri, limbaje,
parafraz ari aparent dep artate de contextul init ial. De pilda, cei cu preferint e
pentru geometrie vor poate mai atrasi de un enunt de tipul Aratat i ca pentru
q numar natural, exista un punct cu ambele coordonate ntregi pozitive situat
pe cuadrica x
2
+y
2
qxy q = 0 daca si numai dac a q este patrat perfect.
Fiecare solut ie are propriile avantaje n comparat ie cu celelate, furni-
zand informat ii suplimentare despre obiectele studiate. Dupa ce n Subsec-
t iunea 2.1 vom determina toate numerele q cu proprietat ile cerute n P6,
vom reusi acelasi lucru si pentru numerele a, b.

In Subsect iunea 2.3 vom
examina modicarile induse n structura solut iilor prin renunt area la condit ia
ca ntregii s a e pozitivi.

In nalul acestei sect iuni indicam un punct de
vedere superior din care poate abordata P6 si o parte din conexiunile sale
cu chestiuni de teoria aproximarii numerelor algebrice.
2.1. Alta ipoteza, aceeasi concluzie. O extindere efectiva a problemei de
concurs a fost gasita de Shaidesh Shirali (profesor la un liceu din India) [8]:
P1. Daca a, b si c sunt ntregi strict pozitivi astfel ca
1 a
2
+b
2
abc c + 1,
atunci a
2
+b
2
abc este patrat perfect.
O problema asemanatoare (n care marginea c + 1 din inegalitate este
nlocuita prin c) a fost publicata sub numarul 1240 n revista canadiana Crux
Mathematicorum.
Autorul ncepe rezolvarea problemei P1 cu examinarea cazurilor n care
c este mic. Notam
d = a
2
+b
2
abc. (1)
Daca c = 1, atunci d 2. Se observa ca pentru d = 1 problema este
rezolvat a, iar d = 2 nu este posibil din motive de paritate. Daca c = 2, clar
d = (a b)
2
.

In continuare se consider a c 3.
Sa presupunem c a d dat de relat ia (1) nu este patrat perfect, de unde se
deduce ca d 2. Pentru c si d xate, vom considera toate perechile (u, v) de
numere naturale strict pozitive supuse restrict iei d = u
2
+v
2
cuv. Ordonam
astfel de perechi dupa valoarea luat a de suma componentelor. Observam ca
pentru indiferent care pereche (u, v) cu toate propriet at ile enunt ate avem
u ,= v (n caz contrar, egalitatea u
2
(2 c) = d nu poate avea loc ntrucat
membrul stang este strict negativ, n vreme ce membrul drept este pozi-
tiv). Pentru a face o alegere, consideram ca u > v. Este clar ca (u
1
, v), cu
u
1
= cv u, este o pereche de numere ntregi ce veric a d = u
2
1
+v
2
cu
1
v.
Condit ia ca d sa nu e patrat perfect mpreun a cu egalitatea dat a de relat iile
lui Vi`ete uu
1
= v
2
d asigura ca u
1
este nenul. Admit and ca ar strict
negativ, s-ar obt ine d = u
2
1
cu
1
v + v
2
1 + cv + v
2
> c + 1, contradict ie.
Prin urmare, u
1
1.

In acest moment conchidem ca (u
1
, v) face parte din
mult imea de perechi considerata mai sus.

In plus, din alegerea u > v rezulta
u
1
=
v
2
d
u
<
u
2
d
u
< u,
astfel ca u
1
+ v < u + v. Iterand, ajungem la concluzia ca exista un sir
innit strict descrescator de numere naturale. Contradict ia provine din pre-
supunerea c a d nu ar patrat perfect.
Exemplul a = 1, b = 2, c = 1, pentru care a
2
+ b
2
abc = 3 = c + 2,
arata ca, nlocuind valoarea c+1 cu una mai mare, nu putem obt ine concluzia
din problema P1 fara a impune ipoteze suplimentare.
2.2. Informat ii suplimentare.

In acest moment, problema P6 este rezol-
vata. Privind cu luare aminte rat ionamentul din subsect iunea precedenta, se
constat a ca el furnizeaz a suciente informat ii suplimentare despre tripletele
de numere naturale (a, b, q) pentru a le putea determina complet.
Ideea esent ial a a solut iei date problemei P6 este construirea unui sir de
numere naturale a > a
1
> > a
k1
> a
k
= 0 astfel ca
a
2
i
+a
2
i+1
a
i
a
i+1
+ 1
= q =
a
2
1
+a
2
aa
1
+ 1
=
a
2
+b
2
ab + 1
pentru i = 1, 2, . . . , k 1.
La ncheierea procesului, atunci cand a
k
= 0, se obt ine q = a
2
k1
.
Campbell [2] interpreteaz a aceast a egalitate un pic diferit decat ar de
asteptat, anume q =
_
c.m.m.d.c.(a
k1
, a
k
)
_
2
.

Intruc at a
i+1
= qa
i
a
i1
, o
induct ie simpla ce foloseste relat ia c.m.m.d.c.(a
i1
, a
i
) = c.m.m.d.c.(a
i+1
, a
i
)
conduce la concluzia ca avem totdeauna q =
_
c.m.m.d.c.(a, b)
_
2
.
Daca numerele naturale a, b, q verica q = t
2
=
a
2
+b
2
ab+1
, rat ionamentul
din Subsect iunea 2.1 asigur a c a a si b sunt termeni succesivi ai sirului recurent
8 Articole
liniar x
k+1
= t
2
x
k
x
k1
(k 1) cu termenii init iali x
0
= 0, x
1
= t. Acest
rezultat apare n multe locuri, printre care [1] si [6].
2.3. Valori negative. Acum, ca avem o solut ie pentru problema init ial a,
putem sa consideram variat iuni ale sale. O prima tentat ie ar sa renunt am
la cerint a ca toate numerele sa e naturale. Este simplu de vazut c a o situat ie
noua apare efectiv doar c and ab < 0.

In aceste condit ii parafraz am P6 astfel:
P2. Gasit i numerele ntregi strict pozitive a, b, c pentru care c =
a
2
+b
2
ab1
.
Aceasta problema a constituit obiectul mai multor discut ii n forumuri
de pe internet si apare si n [6].
Transam rapid chestiunea pentru valori mici.
Dac a b = 1, condit ia este ca a 1 sa divida a
2
+1 = a
2
1 +2, ceea ce
are loc doar pentru a = 2 sau a = 3.

In ambele cazuri, rezulta c = 5. Daca
mina, b 2, cu inegalitatea mediilor se deduce c 3, iar din faptul ca 2
nu are divizori pozitivi de forma a
2
1 rezulta a ,= b.
Pentru b si c xate, ecuat ia x
2
bcx + b
2
+ c = 0 are, n afara de a, o
radacin a a
1
= bc a, care este ntreag a, strict pozitiva (caci aa
1
= b
2
+c) si
mai mica decat b daca a > bc a.

Intr-adev ar,
a
1
=
bc
_
b
2
(c
2
4) 4c
2
< b c < b
2
(c 2).
Ultima inegalitate este consecint a imediata pentru b 2, c 3.
Dac a a
1
2, rat ionamentul poate reluat cu a
1
n loc de a si se gaseste
un numar natural a
2
1 astfel ca a
1
> a
2
si
a
2
2
+b
2
a
2
b1
= c. Dupa un numar
nit de pasi se ajunge la relat ia
1+b
2
b1
= c, care am vazut c a nu poate avea
loc dec at pentru c = 5.
Ca si n Subsect iunea 2.2, se constata ca mult imea solut iilor proble-
mei P2 poate descris a complet: invariabil c = 5, iar a si b sunt termeni suc-
cesivi n unul din sirurile date de relat ia de recurent a liniar a y
k+1
= 5y
k
y
k1
(k 1) cu termenii init iali y
0
= 1, y
1
2, 3 (vezi, de pilda, [6]).
2.4. Un punct de vedere erudit.

In esent a, rezolvarea problemelor P6
si P2 depinde de capacitatea noastra de a rezolva n numere ntregi ecuat ia
a
2
+b
2
qab q = 0. Dupa nmult irea cu 4 si completarea patratului, devine
evident ca avem de-a face cu o ecuat ie Pell generalizat a
x
2
(q
2
4)y
2
= 4q. (2)
Pentru aceasta clasa de ecuat ii diofantice s-a pus la punct o teorie bine
articulata si algoritmi ecient i de rezolvare (prezentat i cu lux de amanunte
n [5]). Se stie, de pilda, ca toate solut iile ntregi pentru ecuat ia (2) se obt in
combin and, ntr-un mod simplu si descris precis, un num ar nit de solut ii
fundamentale ale acesteia cu solut ia minimala pentru ecuat ia Pell atasata
(n care membrul drept este 1). Solut ia minimala pentru o ecuat ie Pell de
forma X
2
DY
2
= 1 se obt ine din dezvoltarean fract ii continue a num arului
irat ional
D. Acelasi lucru este valabil pentru ecuat ia Pell generalizat a

X
2
DY
2
= C daca 1 C <
D. Aceste rezultate decurg dintr-o teorema

a lui Legendre potrivit careia daca
r
s
este o fract ie pozitiv a ireductibila si
este un numar irat ional astfel ncat [
r
s
[ <
1
2s
2
, atunci
r
s
este o convergenta
a dezvoltarii n fract ii continue pentru .
Este de asemenea binecunoscut un mod alternativ de prezentare a solu-
t iilor unei ecuat ii Pell generalizate, si anume, ca termeni ai unor siruri liniar
recurente de ordinul doi. Pana acum am folosit de fapt aceast a reprezentare.
Pentru completitudine, cit am din [7] o reformulare a teoremei din Subsect i-
unea 2.2 n contextul schit at aici:
Teorema 1. Toate tripletele de numere naturale (a, b, q) astfel ca a b si
q =
a
2
+b
2
ab+1
sunt de forma
_
ty,
tx+t
3
y
2
, t
2
_
, cu x
2
(t
4
4)y
2
= 4 si t impar, sau
_
ty, tx +
t
3
y
2
, t
2
_
, cu x
2
(4t
4
1)y
2
= 1 si t par.
Rezolvarea pentru problema P6 prezentat a anterior se bazeaz a pe prin-
cipii frecvent folosite n matematica elementara si nu invoca rezultate din teo-
ria ecuat iilor Pell.

Incercand sa rezolve direct ecuat ia (2), autorii lucrarii [7]
constat a ca abordarea ce utilizeaza dezvoltarea n fract ii continue esueaz a
din cauza lipsei unui analog pentru teorema lui Legendre, astfel ca purced
sa demonstreze o generalizare a sa, n care membrul drept sa e
2
s
2
n loc de
1
2s
2
. Aceast a modicare a ipotezei induce o schimbare dramatic a a concluziei,
aparand trei cazuri n plus fat a de ceea ce a gasit Legendre. Cele mai generale
rezultate de acest tip apar n [3] si permit, de exemplu, demonstrarea unor
rezultate de urmatoarea factura:
Teorema 2. Fie k un ntreg impar supraunitar. Daca t este un numar
natural nenul mai mic dec at 2
k
2
4 si ecuat ia x
2
(k
2
4)y
2
= 4t are
solut ii n ntregi strict pozitivi coprimi, atunci e t = 1, e t = k + 2.
3. O alt a problem a
Variat iunile pe o tema data sunt mai totdeauna posibile. De pilda, n [6]
este propus urmatorul analog al problemelor discutate anterior:
P3. Gasit i tot i ntregii a, b, c pentru care q =
a
2
+b
2
+c
2
1 +ab +bc +ca
este
ntreg.
Autorii se m arginesc sa formuleze sapte ntrebari, prima dintre ele -
ind ,,Ce propriet at i ale solut iilor problemei P3 putem descoperi examinand
solut iile gasite cu calculatorul? Probabil ca imaginea rezultata din experi-
mentele lor este deconcertanta, prea aproape de haos, mult prea complicata
10 Articole
pentru a sugera raspunsuri imediate sau conjecturi cat de cat plauzibile.
Cresterea complexitat ii era de asteptat, dat ind faptul ca pentru ecare q
nu mai avem de a face cu curbe, ci cu suprafet e.
In aceast a sect iune vom studia mult imea

Q := (a, b, c, q) Z
4
: q =
a
2
+b
2
+c
2
1 +ab +bc +ca
si brele sale
Q
q
:= (a, b, c) Z
3
: (a, b, c, q) Q (q Z).
Sa notam ca orice mult ime nevida Q
q
va cont ine, mpreun a cu un triplet
(a, b, c), opusul sau (a, b, c) = (a, b, c) si orice permutare (a, b, c) a sa.
Iata primele informat ii despre solut iile problemei P3.
Propozit ia 3. Pentru (a, b, c, q) Q, urmatoarele propriet at i sunt ndepli-
nite:
a) cel put in unul dintre numerele a, b, c este par;
b)
_
c.m.m.d.c.(a, b, c)
_
2
divide q;
c) q este multiplu de 4 daca si numai dac a c.m.m.d.c.(a, b, c) este par.
Demonstrat ie. a) Atunci cand abc este impar, numaratorul fract iei ce
da q este impar, n vreme ce numitorul s au este par, prin urmare se contrazice
cerint a ca q sa e ntreg.
b) Patratul acestui c.m.m.d.c. divide numar atorul si este coprim cu
numitorul, astfel ca fract ia nu se poate simplica cu niciun divizor al sau.
c) Rezulta din partea b) si dintr-un rat ionament simplu si scurt modulo
8.
Intruc at P3 se specializeaza la P6, din cele prezentate n Sect iunea 2

rezulta ca mult imea Q
k
2 este innita pentru orice ntreg k si elementele sale
sunt date de recurent e liniare.
Teorema 4. Pentru orice ntreg k, mult imea Q
k
2 cont ine toate tripletele
(a, b, c) constand din trei termeni cu indici consecutivi din sirul dat de relat ia
de recurent a liniara
x
n+3
= k
2
(x
n+2
+x
n+1
) x
n
(n Z)
si condit ia init iala (x
0
, x
1
, x
2
) = (0, 0, k).
In fapt, un rezultat aseman ator este valabil n condit ii mai generale.

Teorema 5. Fie q un ntreg pentru care mult imea Q
q
este nevida si e
(a
0
, b
0
, c
0
) Q
q
. Atunci Q
q
cont ine toate tripletele (a, b, c) constand din trei
termeni cu indici consecutivi din sirul dat de relat ia de recurent a liniara
y
n+3
= q(y
n+2
+y
n+1
) y
n
(n Z)
si condit ia init iala (y
0
, y
1
, y
2
) = (a
0
, b
0
, c
0
).
Aceste rezultate raspund altor ntrebari din [6]: ,,Exista relat ii de recu-
rent a ntre solut ii?, ,,Exista forma explicit a pentru solut ii?, ,,Mult imea
solut iilor este bine ordonata sau apare o structura de ordine mai compli-
cat a?. De pilda, n Q
4
se gasesc tripletele formate din termeni cu indici
consecutivi ai sirurilor
x
n
: . . . , 190, 40, 8, 2,0, 0, 2, 8, 40, 190, 912, 4368, . . . ,
x
t
n
: . . . , 720, 152, 30, 8,0, 2, 0, 8, 30, 152, 720, 3458, . . . .
(Termenul de indice nul al ecarui sir este scris cu caractere ngrosate.)
Ambele siruri sunt generate de relat ia de recurent a liniar a
a
n+3
= 4(a
n+2
+ a
n+1
) a
n
(n Z) folosind condit ia init ial a x
0
= x
1
= 0,
x
2
= 2 si respectiv x
t
0
= x
t
2
= 0, x
t
1
= 2. Ecuat ia caracteristic a relevanta pen-
tru aceast a discut ie are rad acinile 1, =
5+
21
2
, =
5
21
2
. Prin urmare,
pentru orice n Z avem
x
n
=
(
21 3)
n
(
21 + 3)
n
+ 6(1)
n
21
,
x
t
n
=
2
n1
+ 2
n1
10(1)
n
7
.
Sirul diferent a y
n
= x
n
x
t
n
are tot i termenii de indice cel put in 4 strict
pozitivi, n vreme ce tot i termenii de indice negativ sunt negativi. Din aceasta
analiza se conchide, de exemplu, ca tripletul (8, 30, 152) din Q
4
nu poate
gasit n sirul (x
n
)
nZ
.
Este clar ca descrierea pentru Q
q
indicata n Teorema 5 este inecienta,
existand multa redundant a. Ea ar putea deveni ecienta n masura n care ar
descoperit un criteriu simplu de a decide dac a doua astfel de siruri au sau
nu termeni comuni. Primii pasi n aceast a direct ie de cercetare sunt cont inut i
n rezultatul urm ator, care clarica situat ia pentru valori mici ale lui q.
Propozit ia 6. a) Q
1
este vida.
b) Q
0
= (0, 0, 0).
c) Q
1
= (a, a, a + 1) : a Z.
Demonstrat ie. a) Presupun and contrariul, se ajunge la concluzia ca trei
numere pozitive (2a +b +c)
2
, 3b
2
+ 2bc + 3c
2
, 4 au suma nula.
c) Pentru (a, b, c) Q
1
arbitrar se obt ine ecuat ia
a
2
a(b +c) +b
2
bc +c
2
1 = 0,
al carei discriminant (b +c)
2
4(b
2
bc +c
2
1) = 4 3(b c)
2
este patrat
perfect. Rezulta e b = c, e c = b 1, deci ecuat ia devine (a b)
2
= 1 si
respectiv a
2
a(2b 1) +b(b 1) = 0, avand rad acinile a = b 1 si respectiv
a = b sau a = b 1.
In continuare vom presupune [q[ 2 si vom da condit ii necesare pentru

ca un astfel de ntreg q sa e reprezentabil ca n P3.
12 Articole
Teorema 7. Daca [q[ 2 si Q
q
,= , atunci:
a) q , 2 (mod 4),
b) q + 2 nu are divizori pozitivi congruent i cu 5 sau 7 modulo 8.
Demonstrat ie. a) Conform Propozit iei 3 a), cel mult doi dintre ntregii
a, b, c sunt impari. Dac a tot i ar pari, atunci q ar multiplu de 4n virtutea
Propozit iei 3. Daca unul singur este impar, numaratorul va impar, la fel si
q.

In situat ian care a si b, sa zicem, ar impari, num aratorul ar congruent
cu 2 modulo 4, iar numitorul ar evident par, astfel ca fract ia ce da q s-ar
simplica prin 2, rezultand q impar.
b) Ecuat ia a
2
qa(b +c) +b
2
+c
2
qbc q = 0 de gradul doi n a are
radacinile ntregi. Prin urmare, discriminantul sau
(q
2
4)(b
2
+c
2
) + 2q(q + 2)bc + 4q
trebuie sa e patrat perfect. Acest patrat este congruent cu 8 modulo orice
divizor impar p al lui q +2. Altfel spus, 2 trebuie sa e rest patratic modulo
p, ceea ce se petrece doar pentru numerele pozitive p 1 sau 3 (mod 8).
In ciuda simplit at ii, aceste condit ii necesare pentru a avea Q

q
nevida
sunt foarte eciente. De pilda, dintre primele 100 de numere naturale strict
pozitive, 10 sunt patrate perfecte, 25 sunt pare, dar nu divizibile prin 4, iar
pentru alte 46 exista cel put in un divizor al lui q +2 care este congruent cu 5
sau 7 modulo 8. Astfel, pentru 81 de ntregi q, 1 q 100, am putut decide
daca este sau nu reprezentabil ca n P3.
Cu ajutorul calculatorului au fost gasite reprezentari n care intervin
ntregi 1 a b 1000 si 2000 c 4000 pentru nca alte 17 numere q
din intervalul indicat, ramanand incert statutul a doar doua valori: q = 32 si
57. Pentru simetricul fat a de origine [100, 1], situat ia se prezint a astfel:
25 de numere sunt congruente cu 2 modulo 4, 43 sunt eliminate cu ajutorul
criteriului din partea b) a teoremei precedente, 5 este reprezentabil (a se
vedea Subsect iunea 2.3), n vreme ce 1 nu este (cf. Propozit ia 6), iar pentru
alte 28 de numere reprezentarile dorite au fost gasite cu ajutorul calculatoru-
lui. Ramane cititorului curios placerea de a decide daca 20 si 84 apar sau
nu pe lista numerelor negative reprezentabile ca n problema P3.
In general, pentru orice x > 0, folosind Teoremele 4 si 7 se stabileste

daca mult imea Q
q
este sau nu vida pentru mai mult de
x+
7x
12
dintre ntregii
pozitivi q care nu depasesc x.
Criteriile din Teorema 7 permit s a raspundem si altor ntrebari referi-
toare la mult imea
M = q Z : Q
q
,= .
Avand n vedere P6, este clar ca mult imea numerelor pozitive care
admit o reprezentare de tipul
a
2
+b
2
ab+1
este nchisa la nmult ire. M nu are nsa
aceasta proprietate: de pild a, 4, ca orice patrat perfect, apart ine mult imii
M, la fel si 7 (caci (2, 3, 36) Q
7
), dar 28 , M pentru ca 30 este divizibil
cu 5.
Rezultatul urm ator ilustreaza ideea ca n contextul actual exista un
analog pentru Teorema 1.
Teorema 8.
Q
4
= (2u, 2v 4u, 4v 4u + 2w) : u, v, w Z si 9u
2
3v
2
+w
2
= 1.
Demonstrat ie. O incluziune ind usor de vericat, vom ar ata doar ca
orice triplet (a, b, c) din Q
4
are o reprezentare ca n enunt ul teoremei.
Conform Propozit iei 3, exist a ntregi u, d, e astfel ca a = 2u, b = 2d,
c = 2e si e
2
4e(d + u) + d
2
+ u
2
4du 1 = 0. Prin urmare, pentru un
ntreg convenabil w avem
4(d +u)
2
d
2
u
2
+ 4du + 1 = 3(d
2
+u
2
+ 4du) + 1 = w
2
si e = 2(d +u) +w. Similar, trebuie ca 36u
2
3(3u
2
+ 1 w
2
) sa e patrat
perfect, ceea ce nseamna ca exista unntreg v pentru care sa aiba loc relat iile
d = v 2u si 9u
2
+w
2
1 = 3v
2
.
Pentru a raspunde nca unei ntrebari formulate n [6], construim familii

innite de ntregi negativi q care admit reprezentarea ceruta n P3.
Teorema 9. Pentru orice ntreg m avem
_
2m+ 2, 2m, 2m
2
+ 2m1, 4m
4
8m
3
8m
2
4m5
_
Q,
_
m+ 1, m, m
2
+m2, m
4
2m
3
+m
2
+ 2m5
_
Q.
Demonstrat ie. Cautam ntregi a, b, c pentru care ab + bc + ca = 2.
Aceasta egalitate are loc dac a si numai daca c =
ab+2
a+b
. Alegem mai ntai a
si b astfel ca suma lor sa e 2. Din cerint a ca c sa e ntreg se deduce ca a
si b trebuie sa e simultan numere pare. Substituind, se gasesc formulele din
enunt ce descriu prima familie.
Celalalt exemplu se construieste analog, impunand (la fel de arbitrar)
ca a +b sa e 1.
Similar se pot indica familii innite de ntregi pozitivi, niciun membru
neind patrat perfect, cu reprezentarea dorita.
Teorema 10. Pentru orice ntreg m avem
(a, b, c, q) =
_
m+ 1, m, m
2
+m+ 1, m
3
+ 2m
2
+ 2m+ 1 +
m
4
+m
2
2
_
Q.
Daca m , 0 sau 7 (mod 8), q nu este patrat perfect.
Demonstrat ie. Deoarece m
4
+2m
3
+5m
2
+4m+2 = (m
2
+m+2)
2
2,
condit ia de a avea q patrat este echivalent a cu existent a unui numar natural
y astfel ca y
2
2
_
m(m+1)
2
+1
_
2
= 1. Solut iile n numere naturale ale ecuat iei
14 Articole
Pell negative Y
2
2X
2
= 1 sunt date de formula y
k
+x
k
2 = (1+
2)
2k+1
,
unde k N. Termenii sirului (x
k
)
k
sunt obt inut i din relat ia de recurent a
liniar a x
k+2
= 6x
k+1
x
k
si condit ia init ial a x
0
= 1, x
1
= 5. Folosind
aceasta informat ie, se vede imediat ca pentru orice numar natural k avem
x
k
1 (mod 4). La fel de simplu se veric a apoi ca
m(m+1)
2+1
1 (mod 4)
daca si numai daca m 0 sau 7 (mod 8).
Notam ca pentru m = 0 si m = 7 se obt ine q = 1, respectiv q = 1681 =
= 41
2
, n vreme ce pentru m = 8 avem q = 2737 = 7 17 23, iar pentru
m = 15, q = 29281 = 7 4183. Mai general, pentru m 8 sau 16 (mod 40)
rezulta q 7 (mod 10), astfel ca pentru aceste valori ale lui m cu sigurant a
q nu este patrat perfect. Concluzii similare se obt in pentru valori negative
ale parametrului m observand ca transformarea m m 1 are ca efect
permutarea valorilor pentru a si b date n teorema.
Bibliografie
[1] J.-P. Bode, H. Harborth, Positive integers
a
2
+b
2
ab+1
are squares, n Applications of Fi-
bonacci numbers, Vol. 9: Proc. Tenth Internat. Research Conf. on Fibonacci Numbers
and their Appl. (F. T. Howard, ed.), Kluwer, 2004, pp. 6367.
[2] J. Campbell, A solution to 1988 IMO question 6 (The most dicult question ever set
at an IMO), Math. Competitions, 1 (1988), 2933.
[3] A. Dujella, B. Ibrahimpasic, On Worleys theorem in Diophantine approximations,
Ann. Math. Inform., 35 (2008), 6173.
[4] A. Engel, Problem Solving Strategies, Springer-Verlag, New York, 1997.
[5] M. Jacobson, H. Williams, Solving the Pell Equation, CMS Books in Mathematics,
Springer, New York, 2009.
[6] I. Lauko, G.A. Pinter, L. Pinter, Another step further . . . On a problem of the 1988
IMO, Math. Mag., 79 (2006), 4553.
[7] F. Luca, C. F. Osgood, P. G. Walsh, Diophantine approximations and a problem from
the 1988 IMO, Rocky Mountain J. Math., 36 (2006), no. 2, 637648.
[8] S. Shirali, Innite descent but not into Hell!, Resonance, 8 (2003), 4255.
Received: August 10, 2012. Accepted: November 11, 2012
M. Merca, Partitii ntregi s i grafuri orientate aciclice 15
Partit ii ntregi si grafuri orientate aciclice
Mircea Merca
1)
Articol dedicat Prof. Dr. Ioan Tomescu
la a 70-a aniversare
Abstract. The algorithms for generating the integer partitions of a po-
sitive integer have long been invented. Nevertheless, data structures for
storing the integer partitions are not received due attention. In 2005 there
were introduced diagram structures to store integer partitions. In this
article, we present a recently obtained result in storing integer partitions:
the binary diagrams.
Keywords: integer partition, ascending composition, directed graph, tree
MSC : 15A17, 05C05, 05C20
1. Introducere
Orice numar ntreg pozitiv n poate descompus ntr-o suma de unul
sau mai multe numere ntregi pozitive
i
,
n =
1
+
2
+ +
k
.
Dac a ordinea numerelor ntregi
i
nu este importanta, aceasta reprezen-
tare se numeste partit ie a numarului ntreg n, altfel, se numeste compunere.
Cand
1

2

k
,
avem o compunere ascendenta. Numarul partit iilor lui n este dat de p(n),
funct ia lui Euler pentru partit ii [1]. Sirul
p(n)
n0
= 1, 1, 2, 3, 5, 7, 11, 15, 22, 30, 42, 56, 77, 101, 135, 176, . . .
este bine cunoscut n literatura [10, sirul A000041]. Pentru mai multe detalii
referitoare la p(n) se poate consulta [1].
Primii algoritmi pentru generarea partit iilor ntregi au fost descoperit i
de catre R. J. Boscovich n anul 1747 si K. F. Hindenburg n anul 1778, a
se vedea Dickson [3, pag. 101106].

In anul 2005, Lin [7] a propus patru
structuri de date pentru memorarea partit iilor ntregi: dou a structuri liniare
(direct a si multiplicativ a), o structura arborescenta si o structura de tip
diagrama.
Utilizarea structurilor arborescente pentru stocarea partit iilor ntregi
nu este o idee chiar atat de recenta. Fenner si Loizou [4] au introdus arborii
binari pentru reprezentarea partit iilor ntregi n 1979 si apoi s-au ocupat de
generarea partit iilor ntregi cu ajutorul metodelor de parcurgere a arborilor
binari, a se vedea Fenner si Loizou [5, 6].
1)
Department of Mathematics, University of Craiova, mircea.merca@profinfo.edu.ro
16 Articole
Ideea utiliz arii structurilor arborescente pentru stocarea partit iilor unui
numar ntreg se bazeaz a pe faptul ca doua partit ii ale aceluiasi numar pot
avea mai multe part i comune. De exemplu, compunerile ascendente
[1, 1, 1, 1, 1, 1] si [1, 1, 1, 1, 2]
au patru part i comune.

In aceast a situat ie, o secvent a de arce dintr-un arbore
poate stoca aceste p art i comune.
Lin [7] a creat structura arborescenta pentru stocarea partit iilor ntregi
conform urmatoarei reguli: rad acina arborelui este etichetat a cu (1, n) si
(x
t
, y
t
) este un descendent direct al nodului (x, y) daca si numai daca
x
t

_
x, x + 1, . . . ,
_
y
2
_
, y
_
si y
t
= y x
t
,
unde [x] este notat ia uzuala pentru partea ntreag a a lui x. Daca x
t
= y
atunci (x
t
, y
t
) = (y, 0) este un nod terminal. Este clar c a orice nod terminal
este de forma (x, 0), 0 < x n. Lin [7] a demonstrat ca [x
1
, x
2
, . . . , x
k
] este
o compunere ascendenta a lui n daca si numai daca
(x
0
, y
0
), (x
1
, y
1
), . . . , (x
k
, y
k
)
este un drum care leaga radacina (x
0
, y
0
) de nodul terminal (x
k
, y
k
). Apoi,
bazandu-se pe acest fapt, a aratat ca numarul total de noduri necesar pentru
a stoca partit iile lui n este 2p(n), iar numarul nodurilor terminale este p(n).
In continuare, vom utiliza pentru aceast a structura arborescenta denumirea

de arbore Lin.
Arborele Lin care stocheaza toate partit iile num arului 6 este prezentat
n Figura 1. Cum poate utilizat acest arbore pentru a genera toate partit iile
numarului 6? Pornind de la rad acin a, vom parcurge arborele n adancime
si, cand ajungem la un nod terminal, list am drumul parcurs. De exemplu,
drumul
(1, 6)(1, 5)(1, 4)(1, 3)(1, 2)(2, 0)
leaga radacina arborelui de nodul terminal (2, 0). Daca eliminam din acest
drum prima pereche si pastram din ecare pereche ramasa numai prima va-
loare, obt inem compunerea ascendent a [1, 1, 1, 1, 2].
In arborele Lin, descendent ii direct i ai nodului (1, n) sunt noduri de

forma (k, n k). Din regula de construire a arborilor Lin deducem ca tot i
descendent ii direct i ai nodului (k, n k), 2 k
_
n
2
, sunt descendent i
direct i ai nodului (1, nk). Aceast a observat ie i-a permis lui Lin [7] sa trans-
forme acest arborentr-un graf orientat aciclic, prin eliminarea descendent ilor
nodului (k, nk) din arbore si crearea leg aturilor corespunzatoarentre nodul
(k, n k) si descendent ii direct i ai nodului (1, n k). Structura de date ast-
fel obt inuta este denumita de catre Lin [7] diagrama partit iilor unui numar
ntreg. Vom utiliza pentru acest tip de diagrama denumirea de diagram a
Lin.

In Figura 2 este prezentata diagrama Lin obt inuta prin transformarea
arborelui Lin din Figura 1.
1,6
1,5
1,4
1,3
1,2
1,1
1,0
2,0
3,0
2,2
2,0
4,0
2,3
3,0
5,0
2,4
2,2
2,0
4,0
3,3
3,0
6,0
Figure 1. Arborele Lin pentru numarul 6
1,6
1,5
1,4
1,3
1,2
1,1
1,0
2,0
3,0
2,2 4,0
2,3 5,0
2,4 3,3 6,0
Figure 2. Diagrama Lin pentru num arul 6
Este clar ca diagrama Lin este o reprezentare concisa a arborelui Lin.
Din acest motiv vom p astra pentru nodul (1, n) denumirea de nod r adacin a,
iar pentru nodurile (k, 0), k = 1, 2, . . . , n, denumirea de noduri terminale.
De fapt, nodul r adacin a este singurul nod din diagrama Lin care are gradul
intern zero, iar nodurile terminale sunt singurele noduri din diagrama Lin
care au gradul extern zero. Este evident ca, n diagrama Lin a numarului n,
exista p(n) drumuri ntre rad acin a si cele n noduri terminale si ecare dintre
aceste drumuri reprezint a o partit ie a lui n.
Lin [7] a demonstrat urmatoarea teorema:
18 Articole
Teorema 1. Numarul nodurilor din diagrama Lin a num arului ntreg pozitiv
n este egal cu
_
n
2
4
_
+n + 1.
Pentru a avea o imagine si mai clara asupra dimensiunii diagramelor
Lin, am stabilit n [9] formula pentru num arul arcelor din diagrama Lin.
Teorema 2. Numarul arcelor din diagrama Lin a num arului ntreg pozitiv
n este egal cu
_
1
36
n
3
+
7
24
n
2
+
5
12
n + 1
_
.
2. Diagrame binare
Un arbore orientat este denit formal ca o mult ime A de unul sau mai
multe noduri, astfel ncat exista n A un nod special numit rad acina arbore-
lui si celelalte noduri din A sunt repartizate n m 0 mult imi disjuncte
A
1
, . . . , A
m
, ecare mult ime ind la randul ei un arbore orientat. Arborii
A
1
, . . . , A
m
se numesc subarborii rad acinii. Dac a ordinea relativa a subarbo-
rilor A
1
, . . . , A
m
este importanta spunem ca arborele orientat este un arbore
ordonat. Un arbore ordonat n care ecare nod are cel mult doi subarbori se
numeste arbore binar. Arborii binari n care ecare nod neterminal are exact
doi subarbori se numeste arbore binar strict. Daca ntr-un arbore orientat
stim care dintre noduri este rad acin a atunci sensurile arcelor sunt complet
determinate si nu mai este necesara reprezentarea lor.
Este bine cunoscut faptul c a orice arbore ordonat poate convertit
ntr-un arbore binar schimband legaturile dintre noduri: primul descendent
al unui nod A devine descendent stang al nodului A, iar urmatorul frate al
nodului A devine descendent drept al nodului A. Aplicand aceasta transfor-
mare asupra arborilor Lin se obt ine un arbore binar. Eliminand apoi rad acina
acestui arbore binar, obt inem un arbore binar strict care memoreaza toate
partit iile num arului ntreg n.

In Figura 3 este prezentat arborele binar strict
obt inut n urma transformarii arborelui Lin din Figura 1.
Pentru a genera compunerile ascendente, parcurgem n adancime ar-
borele binar strict pentru stocarea partit iilor. Cand ajungem la un nod ter-
minal, listam din drumul care leaga rad acina de acest nod terminal numai
nodurile care sunt succedate de descendentul stang. De exemplu,
(1, 5)(1, 4)(1, 3)(2, 2)(2, 0)
este un drum care leaga nodul rad acin a (1, 5) de nodul terminal (2, 0). Din
acest drum nodul (1, 3) este eliminat la listare deoarece este urmat de nodul
(2, 2) care este descendent drept. P astrand din ecare pereche ramasa numai
prima valoare, obt inem compunerea ascendent a [1, 1, 2, 2]. Pe de alta parte,
observam ca mai exista doua drumuri care leaga nodul rad acin a (1, 5) de alte
doua noduri terminale etichetate (2, 0):
(1, 5)(1, 4)(1, 3)(1, 2)(1, 1)(2, 0) si (1, 5)(2, 4)(2, 2)(2, 0).
1,5
1,4
1,3
1,2
1,1
1,0 2,0
3,0
2,2
2,0 4,0
2,3
3,0 5,0
2,4
2,2
2,0 4,0
3,3
3,0 6,0
Figure 3. Arborele binar strict pentru stocarea partit iilor num arului 6
Aceste drumuri reprezinta compunerile ascendente
[1, 1, 1, 1, 2] si [2, 2, 2].
Arborele binar strict pentru stocarea partit iilor numarului ntreg n
poate construit dup a urmatoarea regul a: rad acina arborelui este etichetat a
cu (1, n 1), nodul (x
s
, y
s
) este un descendent stang al nodului (x, y) daca
si numai daca
x
s
=
_
x, daca 2x y
ym, altfel
si y
s
= y x
s
, (3)
iar nodul (x
d
, y
d
) este un descendent drept al nodului (x, y) daca si numai
daca
x
d
=
_
x + 1, daca 2 +x y
x +y, altfel
si y
d
= x +y x
d
. (4)
Arborii binar strict i pentru stocarea partit iilor ntregi au fost derivat i
din arborii Lin si sunt diferit i de arborii binari introdusi de Fenner si Loizou.
Avantajele utilizarii structurilor arborescente binare sunt bine cunoscute si
n acest caz au fost concretizate n [8] prin obt inerea celui mai rapid algoritm
pentru generarea partit iilor ntregi.
In Figura 3 observ am ca subarborele stang al nodului (2, 4) este identic

cu subarborele drept al nodului (1, 3). Aceasta redundant a a informat iilor
este conrmat a de urmatoarea teorema prezentata n [9].
Teorema 3. Fie (x, y) un nod intern din arborele binar strict pentru sto-
carea partit iilor unui num ar ntreg, astfel nc at x > 1. Daca 2x y, atunci
subarborele stang al nodului (x, y) este identic cu subarborele drept al nodului
(x 1, y x + 1). Daca 2x > y, atunci subarborele st ang al nodului (x, y)
este identic cu subarborele drept al nodului (
_
y
2
, y
_
y
2
).
20 Articole
Conform acestei teoreme, n orice arbore binar strict pentru stocarea
partit iilor ntregi, subarborele stang al oricarui nod intern (x, y) cu x > 1
este identic cu subarborele drept al unui nod (x
t
, y
t
). Pentru orice nod intern
(x, y) cu x > 1, eliminam subarborele sau stang si introducem un arc de la
nodul (x, y) la descendentul drept al nodului (x
t
, y
t
). Astfel, obt inem un graf
orientat aciclic care a fost denumit n [9] diagrama binara a partit iilor ntregi.
In Figura 4 este prezentata diagrama binara obt inuta prin transformarea

arborelui binar strict din Figura 3.
Doua rezultate care caracterizeaza dimensiunea diagramelor binare au
fost prezentate n [9].
Teorema 4. Numarul nodurilor diagramei binare a partit iilor ntregului po-
zitiv n este egal cu
_
n
2
4
_
+n.
Teorema 5. Numarul arcelor diagramei binare a partit iilor ntregului pozitiv
n este egal cu
_
n
2
2
_
.
1,5
1,4
1,3
1,2
1,1
1,0 2,0
3,0
2,2
4,0
2,3
5,0
2,4
3,3
6,0
Figure 4. Diagrama binara pentru stocarea partit iilor num arului 6
Comparand Teoremele 1 si 4, nu putem spune ca diagramele binare
sunt mai eciente n stocarea partit iilor ntregi dec at diagramele Lin. Ceea ce
difera este numarul arcelor care este considerabil mai micn cazul diagramelor
binare.

In plus, nodurile diagramei binare a partit iilor ntregului pozitiv n
pot rearanjate ntr-un tablou bidimensional cu
_
n
2
+ 1 linii si n coloane,
denumit tabloul compunerilor ascendente [9].

In acest tablou, ecare nod
ocupa o celula conform urmatoarei reguli: celula indexata (i, j) este ocupata
de nodul (x, y) daca si numai daca
i =
_
_
x
2
_
+ 1, daca y = 0,
x, altfel
si j = x +y. (5)
Este clar ca exista celule n tablou care nu sunt ocupate. Utiliz and Teorema
4 deducem ca numarul celulelor neocupate este un patrat perfect,
_
n
2
2
(sirul
A007894 n [10]). De exemplu, conform (5), nodurile diagramei binare din
Figura 4 sunt rearanjate n Figura 5.
Notam cu S(i, j) si D(i, j) pozit iile succesorilor direct i ai nodului de
pe pozit ia (i, j) din tabloul compunerilor ascendente. T inand cont de (5),
deducem ca
S
i,j
=
_
_
_
(i, j i) , pentru 3i j,
__
j i
2
_
+ 1, j i
_
, pentru 2i j < 3i
si
D
i,j
= (i + 1, j), pentru 2i j.
1,0 1,1 1,2 1,3 1,4 1,5
2,0 3,0 2,2 2,3 2,4
4,0 5,0 3,3
6,0
Figure 5. Tabloul compunerilor ascendente ale numarului 6
Astfel, tabloul compunerilor ascendente pentru num arul ntreg pozitiv
n poate reprezentat concis de urmatoarea matrice
= (
i,j
)
i=1,...,[n/2]+1,j=1,...,n
denit a astfel:
i,j
=
_
_
i, daca 2i j,
j, daca i =
_
j
2
_
+ 1,
0, altfel.
In [9], aceast a matrice a fost denumita matricea compunerilor ascen-

dente. De exemplu, matricea compunerilor ascendente a numarului 6 este
_
_
_
_
1 1 1 1 1 1
0 2 3 2 2 2
0 0 0 4 5 3
0 0 0 0 0 6
_
_
_
_
.
22 Articole
3. Observatii s i concluzii
Alegerea structurilor de date pentru stocarea partit iilor ntregi este de
o important a cruciala pentru ecient a algoritmilor de generare a partit iilor.
Metoda gener arii partit iilor ntregi prin parcurgerea structurilor arborescente
utilizate pentru stocarea lor a fost introdusa de catre Fenner si Loizou [4, 5,
6].

In [8], am adaptat algoritmul general pentru parcurgerea n inordine a
arborilor binari la cazul particular al arborilor binari strict i pentru stocarea
partit iilor ntregi si am obt inut cel mai ecient algoritm pentru generarea
partit iilor ntregi [8, Algorithm 6].

In [9], diagramele binare pentru stocarea
partit iilor ne-au permis s a obt inem o versiune usor mbunatat it a pentru acest
algoritm.
Notam cu t
1
(n) timpul mediu de execut ie pentru [8, Algorithm 6], cu
t
2
(n) timpul mediu de execut ie pentru algoritmul Fenner-Loizou [4, 5] si cu
r(n) raportul dintre t
1
(n) si t
2
(n),
r(n) =
t
1
(n)
t
2
(n)
.
In Tabelul 1 sunt prezentate c ateva dintre valorile r(n) obt inute uti-
lizand Visual C++ 2010 Express Edition pe un calculator cu procesor Intel
Pentium Dual T3200 2.00 GHz. Pentru ecare algoritm, timpul mediu de
execut ie a fost obt inut n urma efectu arii a zece teste.

In cazul n = 130, se
poate observa c a timpul mediu de execut ie pentru [8, Algorithm 6] reprezinta
43, 5% din timpul mediu de execut ie al algoritmului Fenner-Loizou.
Tabelul 1. Rezultate experimentale pentru r(n)
n 75 90 95 110 115 130
r(n) 0,426 0,433 0,448 0,435 0,435 0,435
Pe de alta parte, analiza ecient ei algoritmilor obt inut i n [8] si [9]
pentru generarea partit iilor ntregi ne-a permis s a descoperim noi inegalit at i
care implica funct ia partit iilor p(n) (se consider a p(n) = 0 pentru n < 0).
Inegalitatea
p(n) p(n 1) p(n 2) +p(n 5) 0, n > 0,
descoperita si demonstrata n [8], este utilizata pentru a dovedi ecient a
celui mai rapid algoritm pentru generarea partit iilor ntregi. Ulterior, aceast a
inegalitate a fost prezentat a n [2] ca un caz particular al unei familii innite
de inegalit at i. Urmatoarea inegalitate
p(n) 5p(n 3) + 5p(n 5) 0, n ,= 3,
a fost descoperit a n [9] pe parcursul efectuarii testelor de ecient a a algo-
ritmilor propusi pentru generarea partit iilor ntregi. Aceast a inegalitate nu
este demonstrata si se aa n continuare n stadiul de conjectura. O varianta
mbunatat it a a acestei inegalitat i este prezentata n urmatoarea conjectura.
Conjectura 3.1. Fie n un num ar ntreg. Inegalitatea
p(n) 5p(n 3) + 5p(n 5) p(n 14) 0
este adevarata, daca si numai dac a n ,= 3.
Pentru
x
n
= p(n) 5p(n 3) + 5p(n 5) p(n 14)
obt inem
x
n
n0
= 1, 1, 2, 2, 0, 2, 1, 0, 2, 0, 2, 1, 2, 1, 4, 0, 4, 4, 5, 3, 11,
7, 15, 15, 23, 27, 44, 44, 68, 84, 113, 135, 189, 223, 298, . . .
si constat am experimental ca sirul x
n
n21
este crescator. Se poate demon-
stra acest fapt?
Autorul mult umeste domnului Dr. Mihai Cipu de la Institutul de
Matematica ,,Simion Stoilow al Academiei Romane pentru sugestiile sale
utile n ceea ce priveste aspectul si cont inutul acestui articol.
Bibliografie
[1] G. E. Andrews, The Theory of Partitions, Addison-Wesley, 1976.
[2] G. E. Andrews, M. Merca, The truncated pentagonal number theorem, J. Combin.
Theory. Ser. A, 119 (2012), 16391643.
[3] L. E. Dickson, History of the Theory of Numbers: Diophantine Analysis, AMS Chelsea
Publishing, 2002.
[4] T. I. Fenner, G. Loizou, A binary tree representation and related algorithms for gen-
erating integer partitions, Comp. J., 23 (1980), 332337.
[5] T. I. Fenner, G. Loizou, An analysis of two related loop-free algorithms for generating
integer partitions, Acta Inform., 16 (1981), 237252.
[6] T. I. Fenner, G. Loizou, Tree traversal related algorithms for generating integer parti-
tions, SIAM J. Comput., 12 (1983), 551564.
[7] R. B. Lin, Ecient data structures for storing the partitions of integers, The 22nd
Workshop on Combinatorics and Computation Theory, Taiwan, 2005.
[8] M. Merca, Fast algorithm for generating ascending compositions, J. Math. Model.
Algorithms, 11 (2012), 89104.
[9] M. Merca, Binary diagrams for storing ascending compositions, Comp. J. (2012), doi:
10.1093/comjnl/bxs111.
[10] N. J. A. Sloane, The On-Line Encyclopedia of Integer Sequences, available at
http://oeis.org.
Received: November 11, 2012. Accepted: January 15, 2013
24 Articole
Olimpiada de Matematica a student ilor din sud-estul
Europei, SEEMOUS 2013
Cornel B aetica
1)
si Gabriel Mincu
2)
Abstract. This note deals with the problems of the 7th South Eastern Eu-
ropean Mathematical Olympiad for University Students, SEEMOUS 2013,
organized in Athens, Greece, by the Hellenic Mathematical Society and the
University of Athens, between March 21 and March 25, 2013.
Keywords: Similar matrices, Jordan canonical form, Skolem-Noether
theorem, Cauchy-Schwarz inequality, eigenvalues, eigenvectors, cyclotomic
polynomials
MSC : 11C20; 15A18; 33D05; 40A30
Conform regulamentului concursului, acesta a avut o singur a proba
const and din patru probleme. Prezent am mai jos aceste probleme nsot ite
de solut ii, dintre care unele au aparut n lucrarile concurent ilor, iar altele n
juriu. Pentru solut iile ociale facem trimitere la http://www.seemous.eu.
Problema 1. Aat i funct iile continue f : [1, 8] R cu proprietatea ca
2
_
1
f
2
(t
3
)dt + 2
2
_
1
f(t
3
)dt =
2
3
8
_
1
f(t)dt
2
_
1
(t
2
1)
2
dt.
Universitatea din Patras, Grecia
Aceasta a fost considerata de juriu drept o problema usoar a.
Solut ie. Facem substitut ia t = x
3
si avem ca
8
_
1
f(t)dt = 3
2
_
1
x
2
f(x
3
)dx.
Inlocuind n relat ia init ial a obt inem

2
_
1
f
2
(t
3
)dt + 2
2
_
1
f(t
3
)dt = 2
2
_
1
t
2
f(t
3
)dt
2
_
1
(t
2
1)
2
dt.
1)
Universitatea din Bucuresti, Facultatea de Matematic a si Informatica, RO-010014
Bucuresti, Rom ania, cornel.baetica@fmi.unibuc.ro
2)
Universitatea din Bucuresti, Facultatea de Matematic a si Informatica, RO-010014
Bucuresti, Rom ania, gamin@fmi.unibuc.ro
SEEMOUS 2013 25
Trecem totul n partea stanga si relat ia se scrie astfel:
2
_
1
_
f
2
(t
3
) + 2f(t
3
) 2t
2
f(t
3
) + (t
2
1)
2
dt = 0.
Se observ a imediat ca f
2
(t
3
)+2f(t
3
)2t
2
f(t
3
)+(t
2
1)
2
= [f(t
3
)(t
2
1)]
2
.
Asadar,
2
_
1
_
f(t
3
) + 1 t
2
2
dt = 0.
Deoarece f este funct ie continu a rezulta ca f(t
3
) = t
2
1 pentru orice t [1, 2]
si de aici obt inem c a f(x) = x
2/3
1 pentru orice x [1, 8].
Problema 2. Fie M, N M
2
(C) matrice nenule cu proprietatea ca
M
2
= N
2
= 0
2
si MN + NM = I
2
. Aratat i ca exista o matrice inversabila
A M
2
(C) astfel ncat M = A
_
0 1
0 0
_
A
1
si N = A
_
0 0
1 0
_
A
1
.
Cornel Baet ica, Rom ania
Aceasta a fost considerata de juriu drept o problema de dicultate medie.
S-au gasit multe solut ii la aceasta problema, unele de catre concurent i iar
altele de catre membrii juriului. Student ii care au rezolvat problema au ales
abordari similare celor din primele doua solut ii prezentate n continuare.
Solut ia 1. Deoarece M
2
= 0
2
, singura valoare proprie a lui M este 0.
Folosind forma canonic a Jordan a unei matrice de ordin 2 obt inem ca exista o
matrice inversabila P M
2
(C) astfel ncat M = P
1
_
0 1
0 0
_
P.

Inmult ind
relat ia MN +NM = I
2
la stanga cu P si la dreapta cu P
1
rezulta ca
(PMP
1
)(PNP
1
) + (PNP
1
)(PMP
1
) = I
2
.
Notand PNP
1
=
_
a b
c d
_
si nlocuind n relat ia de mai sus obt inem:
_
0 1
0 0
__
a b
c d
_
+
_
a b
c d
__
0 1
0 0
_
= I
2
.
Deducem imediat c a c = 1 si d = a. Dar (PNP
1
)
2
= PN
2
P
1
= 0
2
, de
unde rezulta c a b = a
2
.

In consecint a, PNP
1
=
_
a a
2
1 a
_
. Din relat ia
_
1 a
0 1
__
0 0
1 0
__
1 a
0 1
_
=
_
a a
2
1 a
_
26 Articole
obt inem c a
(B
1
P)N(P
1
B) =
_
0 0
1 0
_
,
unde B =
_
1 a
0 1
_
. Acum notam A = B
1
P si nu ne mai ramane decat sa
vericam ca AMA
1
=
_
0 1
0 0
_
.
Solut ia 2. Fie E
12
=
_
0 1
0 0
_
si E
21
=
_
0 0
1 0
_
. Cautam
A M
2
(C) care sa satisfaca cerint ele problemei si, n plus, det A = 1. Fie
asadar A =
_
x y
z t
_
cu xtyz = 1. Avem ca A
1
=
_
t y
z x
_
. Atunci
AE
12
A
1
=
_
xz x
2
z
2
xz
_
si AE
21
A
1
=
_
yt y
2
t
2
yt
_
.
Pe de alta parte, o matrice X M
2
(C) satisface X
2
= 0 daca si
numai daca tr X = 0 si det X = 0. Aceasta ne arat a ca putem scrie
M =
_
a b
c a
_
cu a
2
+bc = 0 si N =
_
a
t
b
t
c
t
a
t
_
cu a
t2
+b
t
c
t
= 0.
Asadar avem de gasit x, y, z, t C cu xt yz = 1 astfel ncat
M =
_
xz x
2
z
2
xz
_
si N =
_
yt y
2
t
2
yt
_
.
Obt inem a = xz, b = x
2
, c = z
2
, a
t
= yt, b
t
= y
2
, c
t
= t
2
.
Evident putem aa pe x si z din primele trei ecuat ii, respectiv pe y si t din
ultimele trei ecuat ii. Mai trebuie s a vericam condit ia xt yz = 1. Dar din
MN + NM = I
2
rezulta c a 2aa
t
+bc
t
+b
t
c = 1, adica (xt yz)
2
= 1. Daca
xt yz = 1, atunci schimbam pe x cu x si pe z cu z, deoarece x si z
satisfac si ele primele trei ecuat ii, deci putem alege xt yz = 1.
Solut ia 3. Consideram f, g : C
2
C
2
date prin f(x) = Mx si
g(x) = Nx. Avem f
2
= g
2
= 0 si fg +gf = id
C
2.
Sa remarcam ca ker f ,= 0, altminteri din f
2
= 0 rezulta f = 0, fals.
Analog ker g ,= 0. Mai mult, ker f ker g = 0: daca f(x) = g(x) = 0 din
f(g(x))+g(f(x)) = x deducem ca x = 0.

In concluzie, C
2
= ker f ker g. Fie
acum v
1
ker f, v
1
,= 0 si v
2
ker g, v
2
,= 0. Atunci B = v
1
, v
2
formeaza o
baza n C
2
.

In aceast a baza matricea asociata transformarii liniare f este de
forma
_
0
0 0
_
, C, ,= 0, iar matricea asociata transformarii liniare
g este de forma
_
0 0
0
_
, C, ,= 0. Din fg + gf = id
C
2 deducem ca
= 1.
SEEMOUS 2013 27
Pentru a obt ine matricele dorite pentru f si g nu avem acum dec at sa
schimbam baza B n baza B
t
= v
1
, v
2
.
Solut ia 4. Ca si la solut ia precedent a, consider am f, g : C
2
C
2
date prin f(x) = Mx si g(x) = Nx. Avem f
2
= g
2
= 0 si fg + gf = id
C
2.
Compunem ultima relat ie la stanga cu fg si obt inem ca (fg)
2
= fg, deci fg
este o proiect ie a lui C
2
. Daca fg = 0, atunci gf = id
C
2, deci f si g ar
inversabile, ceea ce contrazice f
2
= 0.
Asadar fg ,= 0. Fie u Im(fg) 0 si w C
2
astfel ncat u = fg(w).
Obt inem fg(u) = (fg)
2
(w) = fg(w) = u. Fie v = g(u). Vectorul v este
nenul, altfel am avea u = f(v) = 0, fals. Mai mult, u si v nu sunt coliniari,
deoarece v = u cu C implica u=f(v) =f(u) = f(u)=f
2
(g(w))=0,
contradict ie.
Sa consideram acum baza ordonata B a lui C
2
formata din vectorii
u si v. Avem f(u) = f
2
(g(u)) = 0, f(v) = f(g(u)) = u, g(u) = v si
g(v) = g
2
(u) = 0. Asadar matricele lui f si g n raport cu B sunt
_
0 1
0 0
_
si respectiv
_
0 0
1 0
_
.
Matricea A va matricea de trecere de la baza canonica a lui C
2
la B.
Remarca 1. Aceasta problema si are originea n ncercarea de a
da o demonstrat ie elementara teoremei Skolem-Noether pentru cazul par-
ticular al C-algebrei M
2
(C).

In acest caz teorema spune ca singurele C-
automorsme ale C-algebrei M
2
(C) sunt cele interioare. Asadar, dac a este
un C-automorsm al lui M
2
(C), exista o matrice inversabila A astfel ncat
(X) = AXA
1
pentru orice X M
2
(C). Pentru a demonstra acest fapt
este sucient sa gasim o matrice inversabila A astfel ncat (E
12
) = AE
12
A
1
si (E
21
) = AE
21
A
1
. Observam nsa ca [(E
12
)]
2
= [(E
21
)]
2
= 0
2
si
(E
12
)(E
21
) +(E
21
)(E
12
) = I
2
.
Remarca 2. Solut iile 3 si 4, bazate pe considerente de spat ii vectoriale,
au avantajul ca sunt valabile peste orice corp comutativ, nu numai peste C.
Problema 3. Determinat i valoarea maxima a integralei
1
_
0
f
t
(x)
2
[f(x)[
1
x
dx
pe mult imea funct iilor f : [0, 1] R de clasa (
1
pentru care f(0) = 0 si
1
_
0
[f
t
(x)[
2
dx 1.
Pirmyrat Gurbanov, Turkmenistan
28 Articole
Aceasta problema a fost considerata de catre juriu de dicultate put in
peste medie. Rezultatele au aratat nsa ca prin prisma participant ilor aceasta
a fost cea mai dicila problema din concurs; niciun concurent nu a reusit mai
mult decat taton ari ale problemei.
Solut ie. Notand E(f) =
1
_
0
[f
t
(x)[
2
[f(x)[
1
x
dx, avem
E(f) =
1
_
0
[f
t
(x)[
2
x
_
0
f
t
(t)dt
dx
1
_
0
[f
t
(x)[
2
x
_
_
x
_
0
[f
t
(t)[dt
_
_
dx. (6)
Conform inegalitat ii Cauchy-Schwarz,
_
_
x
_
0
[f
t
(t)[dt
_
_
2
x
x
_
0
[f
t
(t)[
2
dt;
de aici si din (6) obt inem E(f)
1
_
0
(f
t
(x))
2
x
_
0
(f
t
(t))
2
dt dx; punand
g(x) =
x
_
0
(f
t
(t))
2
dt, aceasta inegalitate devine
E(f)
1
_
0
g
t
(x)
_
g(x)dx =
2
3
_
g(x)
3
1
0
2
3
.
Valoarea maxim a ceruta este
2
3
, ea ind atinsa de pilda pentru f(x) = x.
Problema 4. Fie A /
2
(Q) pentru care exista n N
astfel ncat
A
n
= I
2
. Aratat i ca A
2
= I
2
sau A
3
= I
2
.
Vasile Pop, Rom ania
Aceasta problema a fost considerata dicila de catre juriu.

In poda
acestui fapt, sapte student i au reusit sa obt ina punctajul maxim. Abord arile
lor au folosit cu precadere tehnicile din prima solut ie pe care o vom prezenta.
Solut ia 2 a aparut mai ntai n juriu; un singur concurent a rezolvat probema
folosind un rat ionament asem anator.
Solut ia 1. Notam polinomul caracteristic al lui A cu P
A
, iar valorile
proprii ale lui A cu
1
si
2
. Condit ia A
n
= I
2
ne conduce la
n
1
=
n
2
= 1.
Cum P
A
are gradul doi si coecient ii rat ionali, r ad acinile sale sunt e reale,
e complexe conjugate.
Daca
1,2
R, din
n
1
=
n
2
= 1 deducem ca n este impar si
1
=
=
2
= 1. De aici rezulta ca (A + I
2
)
2
= 0; n ind impar, putem scrie
I
2
= (A + I
2
I
2
)
n
= n(A + I
2
) I
2
.

In consecint a, A = I
2
, deci si
A
3
= I
2
.
R. Zamfir, Bounds for the real part of polynomial roots 29
Daca
1,2
sunt complexe conjugate, din
n
1
=1 deducem [
1
[ =[
2
[ =1.
Pun and
1
= cos + i sin , din
n
1
= 1 obt inem cos n + i sin n = 1,
deci cos n = 1.
Dar 2 cos cos n = cos(n + 1) + cos(n 1), iar cos k = T
k
(cos ),
unde T
k
Z[X] sunt polinoamele Cebasev de primul tip obt inute pornind
de la formula T
k
(x) = cos(narccos x) pentru 1 x 1. Se stie (sau se
probeaza cu usurint a, de exemplu prin induct ie) ca T
k
are gradul k, coe-
cientul dominant 2
k1
si termenul liber
i
k
+(i)
k
2
.
Deducem o relat ie de forma cos P(2 cos ) = 2 cos cu P Z[X]
monic de grad n. Obt inem cos = 0 sau P(2 cos ) = 2. Cum
2 cos =
1
+
2
= tr(A) Q, pentru a satisface ecuat ia P(x) + 2 = 0 el va
trebui sa e ntreg. Dar avem si [2 cos [ 2, deci cos
_
1,
1
2
, 0,
1
2
, 1
_
.
Pentru cos = 1 obt inem cazul deja studiat
1
=
2
= 1.
Daca cos =
1
2
, P
A
= X
2
+X+1, deci A
3
I
2
= (AI
2
)(A
2
+A+
I
2
) = 0, de unde A
3
= I
2
. De aici, I
2
= A
n
I
2
, A, A
2
; oricare dintre
variante contrazice nsa relat ia A
3
= I
2
.
Daca cos = 0, P
A
= X
2
+ 1, deci A
2
= I
2
.
Dac a cos =
1
2
, P
A
= X
2
X + 1, deci
A
3
+I
2
= (A +I
2
)(A
2
A+I
2
) = 0,
de unde A
3
= I
2
.
Pentru cos = 1 am ajunge la
1
=
2
= 1, deci la contradict ia
1 =
n
1
= 1.
Remarc a. Unii student i au dat solut ii care debutau similar cu solut ia
1, folosind apoi fara demonstrat ie urmatorul rezultat: ,,Daca Q si
cos Q, atunci cos
_
1,
1
2
, 0,
1
2
, 1
_
. Comisia de corectura a de-
cis acordarea punctajului maxim si pentru solut ii complete bazate pe acest
rezultat.
Solut ia 2. Notam cu
A
polinomul minimal al lui A. Din ipotez a
deducem ca
A
[ X
n
+ 1 [ X
2n
1. Prin urmare,
A
este un produs de
polinoame ciclotomice care nu are rad acini multiple. Cum nsa grad
A
2,
factorii sai au si ei aceeasi proprietate.

Intruc at (a) 2 numai pentru
a 1, 2, 3, 4, 6 ( desemnand indicatorul lui Euler),
A
X 1, (X 1)(X + 1), X
2
+X + 1, X + 1, X
2
+ 1, X
2
X + 1.
Primele doua situat ii contrazicnsa
A
[ X
n
+1, iar a treia ne conduce la
A
3
= I
2
, deci I
2
= A
n
I
2
, A, A
2
, aceste relat ii neputand nsa simultan
adevarate. Ramane deci ca
A
X +1, X
2
+1, X
2
X +1; primele doua
variante conduc la A
3
= I
2
, iar cea de-a treia la A
2
= I
2
.
Received: March 28, 2013; Accepted: April 16, 2013
30 Articole
Bounds for the real part of polynomial roots and
applications to irreducibility
Ric a Zamfir
1)
Abstract. In this article we give new bounds for the real part of poly-
nomial roots by using the Frobenius generalized companion matrix and
Bombieris norm.
Keywords: Bounds, polynomial roots, companion matrix.
MSC : 12D10
1. Introduction
Let
f
n
(x) = x
n
a
1
x
n1
a
2
x
n2
a
n1
x a
n
(1)
be a polynomial with complex coecients.
We suppose that a
n
,= 0 and there are complex numbers b
1
, b
2
, . . . , b
n1
,
c
1
, c
2
, . . . , c
n
such that
a
1
= c
1
, a
2
= c
2
b
1
, a
3
= c
3
b
1
b
2
, . . . , a
n
= c
n
b
1
b
2
. . . b
n1
. (2)
If we consider the Frobenius generalized companion matrix
A =
_
_
_
_
0 b
n1
0 . . . 0
0 0 b
n2
. . . 0
. . . . . . . . . . . . . . . . . . . . .
c
n
c
n1
c
n2
. . . c
1
_
_
_
_
M
n
(C), (3)
then we have the equality (see [3, pp. 43])
f
n
(x) = det (xI
n
A) (4)
which shows that the roots of the polynomial f
n
are the eigenvalues of matrix
A.
Using the classical Frobenius matrix, Kittaneh [2, Theorem 1] proved
the following
Theorem 1. If f(x) = x
n
+a
n
x
n1
+ +a
2
x+a
1
is a monic polynomial
of degree n 2 with complex coecients then for every root z of f we have
the inequalities
Re(z
j
) , (5)
1)
Professor, Bucuresti, rzamfir62@gmail.com
where
=
1
2
_
_
Re (a
n
)
_
(Re (a
n
))
2
+
n1
j=1
[a
j
[
2
_
_
+ cos
n
n + 1
and
=
1
2
_
_
Re (a
n
) +
_
(Re (a
n
))
2
+
n1
j=1
[a
j
[
2
_
_
+ cos
n
n + 1
.
In what follows we give new bounds using the generalized Frobenius
companion matrix. Afterwards we give a bound for the real part of polyno-
mial roots using the Bombieris norm.
2. Main Results
If A is the generalized Frobenius companion matrix, then
A = S +iT, (6)
where S =
1
2
(A +A
) and T =
1
2i
(AA
) are hermitian matrices (here A
is the adjoint of A, i.e., the conjugate transpose of A).

For a hermitian matrix X, the eigenvalues
1
(X),
2
(X), . . . ,
n
(X) are
real and we arrange them so that
1
(X)
2
(X)
n
(X).
In the following we will use the next two results from [2].
Lemma 1. If A M
n
(C) and A = S + iT with S, T M
n
(R), then
for every j 1, 2, . . . , n we have
1
(S) Re(
j
(A))
n
(S). (7)
Lemma 2. If B, C M
n
(C) are hermitian matrices, then for every
j 1, 2, . . . , n we have
j
(B) +
1
(C)
j
(B +C)
j
(B) +
n
(C). (8)
In particular,
1
(B) +
1
(C)
1
(B +C) (9)
and
n
(B +C)
n
(B) +
n
(C). (10)
If in (2) we choose b
1
= b
2
= . . . = b
n1
= b > 0 we can prove the next
result.
Theorem 2. For every root z
j
of
f
n
(x) = x
n
a
1
x
n1
a
2
x
n2
a
n1
x a
n
32 Articole
we have
Re(z
j
) , (11)
where
=
1
2
_
_
Re(c
1
)
_
(Re(c
1
))
2
+
n
j=2
[c
j
[
2
_
_
+b cos
n
n + 1
and
=
1
2
_
_
Re(c
1
) +
_
(Re(c
1
))
2
+
n
j=2
[c
j
[
2
_
_
+b cos
n
n + 1
.
Proof. We have
S =
1
2
(A+A
) =
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
0
1
2
b 0 . . . 0
1
2
c
n
1
2
b 0
1
2
b . . . 0
1
2
c
n1
0
1
2
b 0 . . . 0
1
2
c
n2
0 0
1
2
b . . . 0
1
2
c
n3
. . . . . . . . . . . . . . . . . .
0 0 0 . . . 0
1
2
(b +c
2
)
1
2
c
n
1
2
c
n1
1
2
c
n2
. . .
1
2
(b +c
2
) Re(c
1
)
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
.
We write
S = P +Q, (12)
where
P =
_
0 x
x Re(c
1
)
_
M
n
(C), (13)
Q =
_
_
_
_
_
_
_
_
_
_
_
_
_
_
0
1
2
b 0 . . . 0 0
1
2
b 0
1
2
b . . . 0 0
0
1
2
b 0 . . . 0 0
. . . . . . . . . . . . . . . . . .
0 0 0 . . . 0
1
2
b
0 0 0 . . .
1
2
b 0
_
_
_
_
_
_
_
_
_
_
_
_
_
_
M
n
(R), (14)
and
x =
_
_
_
_
_
_
_
_
_
1
2
c
n
1
2
c
n1
.
.
.
1
2
c
2
_
_
_
_
_
_
_
_
_
C
n1
. (15)
Similarly to [2, pp. 662] we obtain the eigenvalues of P:
1
(P) =
1
2
_
_
Re(c
1
)
_
(Re(c
1
))
2
+
n
j=2
[c
j
[
2
_
_
, (16)
2
(P) =
3
(P) = . . . =
n1
(P) = 0, (17)
n
(P) =
1
2
_
_
Re(c
1
) +
_
(Re(c
1
))
2
+
n
j=2
[c
j
[
2
_
_
. (18)
Indeed, we observe that if P
n
() = det(P I
n
) then
P
n
() =
0 . . . 0
1
2
c
n
0 . . . 0
1
2
c
n1
. . . . . . . . . . . . . . .
0 0 . . .
1
2
c
2
1
2
c
n
1
2
c
n1
. . .
1
2
c
2
Re(c
1
)
. (19)
Expanding the determinant along the rst row we get
P
n
() = P
n1
() (1)
n
1
4
[c
n
[
2
n2
and inductively we arrive at
P
n
() = (1)
n
n2
(
2
Re(c
1
)
1
4
n
j=2
[c
j
[
2
).
The eigenvalues of Q are
j
(Q) = b cos
(n j + 1)
n + 1
, j = 1, . . . , n. (20)
34 Articole
In order to prove this it is enough to consider the tridiagonal matrix
R =
_
_
_
_
_
_
_
_
_
_
0 1 0 0 . . . 0 0
1 0 1 0 . . . 0 0
0 1 0 1 . . . 0 0
0 0 1 0 . . . 0 0
. . . . . . . . . . . . . . . . . . . . .
0 1 0 0 . . . 0 1
0 0 0 0 . . . 1 0
_
_
_
_
_
_
_
_
_
_
. (21)
Once again the characteristic polynomial R
n
() = det(R I
n
) can
be computed by an iterative process: R
n
() = R
n1
() R
n2
(), with
R
0
() = 1 and R
1
() = .
By Gershgorin Circle Theorem, all roots
k
of R
n
() satisfy [
k
[ 2.
The characteristic equation of the recurrence R
n
() = R
n1
()R
n2
()
is x
2
+x +1 = 0 with =
2
4 0. Then the equation has two complex
conjugate roots x
1,2
=
i
4
2
2
. Putting cos =

2
and sin =
4
2
2
,
the general solution of the recurrence is R
n
() = c
1
cos n+c
2
sin n, where
the coecients c
1
and c
2
can be determined from the starting conditions as
follows: c
1
= 1 and c
1
cos +c
2
sin = . We nd c
2
=

4
2
=
1
tan
, and
therefore R
n
() = cos n +
sin n
tan
.
In order to nd the eigenvalues of R we have to solve the equation
R
n
() = 0. If R
n
() = 0, then cos n +
sinn
tan
= 0 sin cos n +
+cos sin n = 0 sin(n + 1) = 0 (n + 1) = k =
k
n+1
,
k = 1, . . . , n. Solving for the equation tan =
4
2
we get
2
= 4 cos
2
and we may take = 2 cos , which gives us

k
= 2 cos
k
n+1
, k = 1, . . . , n.
Now this leads us immediately to (20).
If we apply Lemma 1 and Lemma 2 we get
1
(P) +
1
(Q)
1
(S) Re(
j
(A))
n
(S)
n
(P) +
n
(Q) (22)
We have
1
(P)+
1
(Q)=
1
2
_
_
Re(c
1
)
_
(Re(c
1
))
2
+
n
j=2
[c
j
[
2
_
_
+b cos
n
n + 1
, (23)
n
(P)+
n
(Q)=
1
2
_
_
Re(c
1
) +
_
(Re(c
1
))
2
+
n
j=2
[c
j
[
2
_
_
+b cos
n
n + 1
, (24)
Re(
j
(A)) = Re(z
j
), (25)
and the conclusion follows from (23), (24), (25), and (22).
Remark 1. If we choose b = 1, from (11) we get Kittanehs theorem.
In the next theorem we give a bound for the real part of polynomial
roots by using Bombieris norm. We remind that if
f(x) = a
n
x
n
+a
n1
x
n1
+ +a
1
x +a
0
C[x]
and p 1, then the Bombieris norm of order p is
[f]
p
=
_
_
n
j=0
[a
j
[
p
_
n
j
_
p1
_
_
1/p
. (26)
If f, g C[X], deg(f) = n and deg(g) = m, we have the Bombieri
inequality (see [1])
[fg]
2

1
_
n +m
n
_
[f]
2
[g]
2
. (27)
Theorem 3. Let f R[X] be a monic polynomial which has a root
C R. We have the inequality
[Re()[
_
n
2
_
[f]
2
1. (28)
Proof. Because C R and f R[X], is also a root of f. We can
write
f(x) = (x )(x )g(x)
and apply Bombieris inequality. Thus we obtain
[f(x)]
2
2

1
_
n
2
_ [(x ) (x )]
2
2
[g(x)]
2
2
and therefore
[f(x)]
2
2

1
_
n
2
_ [(x ) (x )]
2
2
(29)
because [g]
2
1 (g is monic). We have the identity
(x ) (x ) = x
2
2Re()x +[[
2
,
so we obtain
[(x ) (x )]
2
2
= 1 + 2Re()
2
+[[
4
1 + 2Re()
2
+ Re()
4
=
_
1 + Re()
2
_
2
.
36 Note Matematice
Using (29) we get
1 + Re()
2
_
n
2
_
[f]
2
,
which leads to the conclusion.
References
[1] B. Beauzamy, E. Bombieri, P. Eno, H. L. Montgomery, Products of polynomials in
many variables, Journal of Number Theory, 36 (1990), 219245.
[2] F. Kittaneh, Bounds and a majorization for the real parts of the zeros of polynomials,
Proc. Amer. Math. Soc., 135 (2007), 659664.
[3] H. Linden, Bounds for the zeros of polynomials from eigenvalues and singular values
of some companion matrices, Linear Algebra and its Applications, 271 (1998), 4182.
[4] M. Mignotte, D. Stefanescu, Polynomials. An Algorithmic Approach, Springer, 1999.
Received: April 11, 2011; Accepted in revised form: February 14, 2013
NOTE MATEMATICE
Coecient i binomiali si margini ale radacinilor unui polinom
Doru Stef anescu
1)
Profesorului Ioan Tomescu cu ocazia
celei de a 70-a aniversari.
Abstract. Elementary properties of binomial coecients are used to lo-
calize positive roots of univariate polynomials with real coecients. These
results are thereafter employed to compute bounds for the modulus of com-
plex roots for polynomials over the complex eld.
Keywords: Polynomial roots, real roots, localization of roots
MSC : Primary: 12D10; Secondary: 30C15
Introducere
Calcularea radacinilor reale ale unui polinom ntr-o variabila ce are
coecient ii numere reale se realizeaz a parcurgand mai multe etape. Prima
dintre acestea consta n gasirea unor intervale cat mai mici care sa cont ina
aceste r adacini. Pentru a atinge acest scop este necesar a gasirea unor margini
ale radacinilor. Pentru aceasta, un procedeu este acela de a calcula margini
pentru modulele radacinilor complexe ale polinoamelor, vezi [1], [2]. Exisa
nsa si metode specice radacinilor reale.
1)
Universitatea din Bucuresti, Facultatea de Fizica, Bucuresti, Romania,
doru.stefanescu@fizica.unibuc.ro
D. Stef anescu, Margini pentru r ad acinile polinoamelor 37
In aceast a nota vom arata cum pot folosite rezultate privind coe-
cient ii binomiali, vezi [3], la gasirea unor intervale care s a cont ina radacina
pozitiva a unui polinom care are o singura schimbare de semn. Procedeul
permite gasirea unui interval care sa cont ina si r adacinile pozitive ale derivatei
polinomului.
Astfel de rezultate sunt utile si la calcularea marginilor modulelor ra-
d acinilor complexe ale unui polinom cu coecient ii complecsi. Se stie ca o
astfel de margine este chiar unica rad acin a pozitiva polinomului asociat prin
metoda lui Cauchy, vezi [2].
Margini pentru r ad acini pozitive
O clasa importanta de polinoame este aceea a celor care au coecient ii
reali iar sirul coecient ilor are o singura schimbare de semn. De exemplu,
daca se considera polinomul Q(X) = X
n
+a
1
X
n1
+ +a
n
C[X] , atunci,
dupa o teorema lui Cauchy [2], unica rad acin a pozitiva a polinomului
F(X) = X
n
[a
1
[X
n1
[a
n
[ R[X]
este o margine superioara a modulelor rad acinilor complexe ale polinomului
considerat Q.
Pentru nceput vom calcula o margine superioar a a radacinii pozitive a
unui polinom de o forma speciala.
Lema 1. Fie ntregii d > e 0 si polinomul
P(X) = X
d
+a
d1
X
d1
+ +a
e+1
X
e+1
X
e
+b
e1
X
e1
+ +b
1
X+b
0
,
unde coecient ii a
i
, si b
j
sunt pozitivi. Numarul
1/(de)
este o margine
superioara a radacinilor pozitive ale polinoamelor P
(i)
, pentru tot i i N.
Demonstrat ie. S a observam ca daca >
1/(de)
, avem
de
> ,
asadar
de
> 0. Deci
P() =
e
(
de
) +a
d1
d1
+ +a
e+1
e+1
+b
e1
e1
+ +b
0
> 0.
Pe de alta parte, pentru polinomul
P
t
(X) = dX
d1
+ +(e +1)a
e+1
X
e
eX
e1
+(e 1)b
e1
X
e2
+ +b
1
observam ca avem
d
d1
e
e1
= d
e1
(
de
e
d
) > d
e1
(
de
) > 0.
Rezulta ca avem P
t
() > 0.
Propozit ia 2. Fie ntregii d > e 0 si polinomul
R(X) =
d
k=e+1
_
k
e
_
a
k
X
ke
, unde a
d
= 1, > 0 si tot i a
i
0.
38 Note Matematice
Notam M = maxa
d
, a
d1
, . . . , a
e+1
. Unica radacina pozitiv a a polino-
mului R satisface inegalitatea
>
_
1 +
M
_
1/d
1.
Demonstrat ie. Sa observam ca polinomul R are coecient ii reali si o
singura schimbare de semn, asadar regula lui Descartes a semnelor ne asigura
ca avem o singura radacin a reala pozitiva.
S a observam ca, punand a
e
= 1, avem
d
k=e+1
_
k
e
_
a
k
X
ke
=
d
k=e
_
k
e
_
a
k
X
ke
1 =
=
de
j=0
_
e +j
e
_
a
j+e
X
j
1 .
Pe de alta parte
_
e +j
j
_

_
d
j
_
pentru tot i j = 0, . . . , d e,
de unde obt inem
d
k=e
_
k
e
_
a
k
ke
=
de
j=0
_
e +j
e
_
a
j+e
de
j=0
_
d
j
_
a
j+e
j
M
de
j=0
_
d
j
_
j
.
Intruc at
de
j=0
_
d
j
_
d1
j=0
_
d
j
_
j
= (1 +)
d
d
,
rezulta
0 = R() = 1 +
d
k=e
_
k
e
_
a
k
ke
1 +M
_
(1 +)
d
d
_
< 1 +M(1 +)
d
,
de unde concluzia.
Corolarul 3. Fie ntregii d > e 0 si numerele reale > 0, a
d
= 1, a
i
0,
cu e < i < d. R adacina pozitiv a a polinomului
R(X) =
d
k=e+1
_
k
e
_
a
k
X
ke
se a a n intervalul
_
_
1+
M
_
1/d
1,
1/(de)
_
.
Proposed problems 39
Demonstrat ie. Cu notat iile din Lema 1 si Propozit ia 2, este sucient sa
observam ca R = P
(e)
.
Invitam cititorul sa gaseasca particularizari cat mai interesante pentru
rezultatele indicate aici.
Nota. Autorul mult umeste referentului anonim si editorilor pentru
sugestiile pertinente care au dus la mbunatat irea manuscrisului init ial.
Bbliografie
[1] M. Mignotte, Computer Algebra O introducere n algebra computat ionala, Editura
Universitat ii din Bucuresti, 2000.
[2] D. Stefanescu, Margini pentru r ad acinile polinoamelor cu coecient i complecsi, Gazeta
Matematic a seria A, 26 (2008), 287294.
[3] I. Tomescu, Introducere n combinatoric a, Editura Tehnic a, Bucuresti, 1972.
Received: March 11, 2013; Accepted: April 16, 2013
PROBLEMS
Authors should submit proposed problems to gmaproblems@rms.unibuc.ro.
Files should be in PDF or DVI format. Once a problem is accepted and considered
for publication, the author will be asked to submit the TeX le also. The referee
process will usually take between several weeks and two months. Solutions may also
be submitted to the same e-mail address. For this issue, solutions should arrive
before 15th of November 2013.
PROPOSED PROBLEMS
379. For any prime p we denote by [ [
p
: Q R the p-adic norm given
by [0[ = 0 and [p
t
a/b[
p
= 1/p
t
if a, b Z, p ab. The p-adic eld Q
p
is the
completion of (Q, [ [
p
).
Prove that for every prime p and every a
1
, . . . , a
n
Q
p
we have
max
i,j
min
(k,l),=(i,j)
[(a
i
a
j
) (a
k
a
l
)[
p
1/4
n1
min
i,=j
[a
i
a
j
[
p
.
Proposed by Alexandru Zaharescu, University of Illinois at Urb-
ana-Champaign, USA.
380. Let p be a prime and let A M
n
(Z) be a matrix such that p [ tr A
k
for
all integers k 1.
(i) Prove that if n < p then there is some integer m 1 such that
A
m
pM
n
(Z).
40 Problems
(ii) Prove that if n = p then A
p
(det A)I
p
pM
p
(Z).
Proposed by Vlad Matei, student, University of Wisconsin,
Madison, USA.
381. Prove or disprove: for any ring A there exists a map f : A Z(A)
such that f(1) = 1 and f(a + b) = f(a) + f(b) for all a, b A. Here Z(A)
denotes the center of A, Z(A) = r A : ra = ar, a A.
Proposed by Filip-Andrei Chindea, student, University of
Bucharest, Romania.
382. Let a
1
, b
1
, . . . , a
s
, b
s
Z
2
and let f : Z
2s
2
Z
2
,
f(X
1
, Y
1
, . . . , X
s
, Y
s
) =
s
i=1
(a
i
X
2
i
+X
i
Y
i
+b
i
Y
2
i
).
Determine [f
1
(0)[.
Proposed by Constantin-Nicolae Beli, Simion Stoilow Institute of
Mathematics of the Romanian Academy, Bucharest, Romania.
383. Let C = R
2
R
>0
be the the parameter space of all plane circles. Let
n > 0 and denote by H
k
the subset of C
n
parameterizing the congurations
of n plane circles whose interiors contain a xed point Q, such that any two
circles intersect and no k circles pass through the same point. Show that H
4
is path connected.
Proposed by Marius Cavachi, Ovidius University of Constant a,
Romania.
384. Let M M
2
(C) be an invertible matrix and let k > 1 be an integer.
a) Show that if M is not scalar, then for any two matrices A, B M
2
(C)
with A
k
= B
k
= M there exists E M
2
(C) such that B = AE = EA,
E
k
= I
2
, and [tr(E)[ 2.
b) Is the converse of a) true?
Proposed by Cornel B aetica, University of Bucharest, Romania
385. Let x (0, ) and f(x) =
1
tan x

1
x
. Prove that f
(n)
(x) < 0 for
n = 0, 1, . . ..
George Stoica, Department of Mathematical Sciences, University
of New Brunswick, Canada.
386. Let n 1 be an integer. Find the minimum of
f() =
in/2
(i) +
i>n/2
1
(i),
taken over all permutations S
n
. Determine an explicit value of that
realizes this minimum.
Proposed by Filip-Andrei Chindea, student, University of
Bucharest, Romania.
387. Let a > 0 and let (a
n
)
n0
be the sequence dened by a
0
= 0 and
a
n+1
=

a +a
n
for all n 1. Prove that the set of all n such that a
n
Q
is nite.
Romania.
388. Let a, b, c (0, 1) be real numbers. Prove the following inequality:
cyc
1
1 a
4
+
cyc
1
1 a
2
bc

sym
1
1 a
3
b
.
Proposed by Cezar Lupu, University of Pittsburgh, USA, and
Stefan Sp ataru, International Computer High School of Bucharest,
Romania.
389. Let T be the real vector space of the continuous functions f : [0, 1] R.
We consider on T the distance d(f, g) =
1
_
0
[f(x) g(x)[dx.
a) Show that the intersection of any ane line directed by a nowhere
zero function with any sphere consists of at most two points.
b) Does this property hold for every ane line in T?
Proposed by Gabriel Mincu, University of Bucharest, Romania.
390. Let a
0
, . . . , a
n
C be pairwise dierent. Solve the linear system of
equations
n
j=0
a
k
j
z
j
=
_
0, if k = 0, . . . , n 1,
1, if k = n.
391. Let x
n
be the sequence dened by x
1
= 1, x
n+1
= p
xn
, where p
n
is the
nth prime number. Determine the asymptotic behavior of
n
x
n
, i.e., nd a
function f such that
n
x
n
f(n) as n .
392. We consider on R
n
the following norm:
x = (x
1
, x
2
, . . . , x
n
) |x| =
n
k=1
[x
k
[.
42 Problems
a) Prove that if n = 2 then the norm | | has the property
(M) For all z, z
t
, d R
2
, |z| > |z d| and |z
t
| > |z
t
d| imply
|z +z
t
| > |z +z
t
d|.
b) Does the norm | | havee the property (M) if n = 3?
Proposed by Gheorghit a Zb aganu, University of Bucharest,
Romania.
SOLUTIONS
351. Let (a
n
)
n1
be a sequence of positive integers and let >
1
2
such that
n1
a
n
= . Prove that for any k there is an integer that can be represented
in at least k ways as a sum of two elements of the sequence.
Romania.
Solution by the author. For any n N let p(n) = [i : a
i
< n[. We
note that there is m N such that p(2
m+1
) p(2
m
) > 2k 2
m/2
, otherwise
we would have
n1
1
a
n
=
m0
2
m
a
i
<2
m+1
1
a
m0
2k 2
m/2
2
m
= 2k
m0
2
m(
1
2
)
< .
For such m we take x = 2
m
and denote by b
1
, . . . , b
l
the terms of our
sequence belonging to the interval [x, 2x). Then there are
l(l1)
2
sums b
i
+b
j
with i < j and they take values in the interval [x+(x+1), (2x2)+(2x1)] =
= [2x + 1, 4x 3]. Thus there are at most 2x 3 < 2x possible values for
these sums.
Since l = p(2x) p(x) > 2k
x we get
l(l1)
2
>
2k
x(2k
x1)
2
> 2kx, for
k 2. It follows that there is at least an S [2x + 1, 4x 3] which is the
value of b
i
+ b
j
for at least k pairs (i, j) with i < j. Thus S can be written
in at least k ways as a sum of two elements in the sequence (a
n
)
n1
.
352. Let K be a eld and let m, n, k be positive integers. Find necessary
and sucient conditions the integers a, b, c should satisfy such that there
exist some matrices A M
m,n
(K) and B M
n,k
(K) with rank(A) = a,
rank(B) = b and rank(AB) = c.
Solution by the author. Let A M
m,n
(K) and B M
n,k
(K) with
rank(A) = a, rank(B) = b and rank(AB) = c. We have a minm, n,
b minn, k and c mina, b and, by Sylvesters theorem, c a +b n.
Hence we have the following three necessary conditions:
(1) 0 a minm, n.
(2) 0 b minn, k.
(3) max0, a +b n c mina, b.
We prove that these conditions are also sucient.
Let e
t
i,j
[ 1 i m, 1 j n be the canonical basis of M
m,n
(K),
where e
t
i,j
has 1 on position (i, j) and 0 everywhere else.
Similarly we consider the canonical bases e
tt
i,j
[ 1 i n, 1 j k
and e
i,j
[ 1 i m, 1 j k for M
n,k
(K) and M
m,k
(K), respectively.
We have
e
t
i,j
e
tt
q,r
=
_
e
i,r
, if j = r,
0, otherwise.
We dene A =
a
i=1
e
t
i,i
and B =
c
i=1
e
tt
i,i
+
bc
i=1
e
tt
a+i,c+i
. (Here we make
the convention that a sum of the type
a
i=a+1
x
i
is 0. Since a, c, b c 0, the
three sums from the denitions of A and B are dened.) First note that A
and B are well dened. The terms of the sum giving A are dened because
a minm, n, which is simply (1). The rst sum from B is dened because
c minn, k, which follows from (2) and (3) (we have c b minn, k).
For the second sum of B we note that its terms are of the form e
tt
q,r
with
a+1 q a+b c and c +1 r b. But e
tt
q,r
is dened only for 1 q n
and 1 r m. We have a, c 0, so 1 a + 1, c + 1, so we still need
a + b c n and c k. But the rst condition follows from (3) (we have
a +b n c) and the second from (2) and (3) (we have c b k).
By (3) we have a c. Hence AB =
c
i=1
e
i,i
. (See the formula for e
t
i,j
e
tt
q,r
.)
The a a matrix made by the rst a rows and the rst a columns of
A is I
a
. All larger minors of A are zero, so that rank(A) = a. By a similar
argument rank(AB) = c. For B we note that the b b matrix made of the
rows 1 to c and a +1 to a +b c and the rst b columns of B coincides with
I
b
. All larger minors of B are zero, so rank(B) = b.
353. Let f : [1, 1] R be a continuous and dierentiable function in 0.
Denote
I(h) =
h
_
h
f(x)dx, h [0, 1].
Show that
lim
n
1
n
2
n
k=1
(k)k[I(1/k)[ =
6
2
[f(0)[.
44 Problems
(Here denotes the Eulers totient function.)
Proposed by Cezar Lupu, University of Pittsburgh, USA, and
C alin Popescu, Simion Stoilow Institute of Mathematics of the
Romanian Academy, Bucharest, Romania.
Solution by the authors. Let us consider the function : (0, 1] R
dened by
(h) =
I(h) 2f(0)h
h
2
,
which can be extended continuously in 0 because we have
lim
h0
(h) = lim
h0
(I(h) 2f(0)h)
t
2h
= lim
h0
f(h) +f(h) 2f(0)
2h
=
1
2
lim
h0
_
f(h) f(0)
h

f(h) f(0)
h
_
=
1
2
(f
t
(0) f
t
(0)) = 0.
It follows that is bounded on (0, 1]. Let M = sup[(h)[ : 0 < h 1.
It follows that
[I(h) 2f(0)h[ Mh
2
, h (0, 1].
For h = 0 the above inequality is obvious. Since
[[I(h)[ 2f(0)h[ [I(h) 2f(0)h[ Mh
2
, 0 < h 1,
it follows that
2[f(0)[h Mh
2
[I(h)[ 2[f(0[h +Mh
2
, 0 < h 1.
Thus, we obtain
2[f(0)[(k) M
(k)
k
(k)kI
_
1
k
_
2[f(0)[(k) +M
(k)
k
,
which is equivalent to
2[f(0)[
n
k=1
(k) M
n
k=1
(k)
k

n
k=1
(k)kI
_
1
k
_
2[f(0)[
n
k=1
(k) +M
n
k=1
(k)
k
.
Let us consider the sequence (a
n
)
n1
dened by
a
n
=
1
n
2
n
k=1
(k)k
I
_
1
k
_
.
We obtain that
2[f(0)[
1
n
2
n
k=1
(k)
M
n
2
n
k=1
(k)
k
a
n
2[f(0)[
1
n
2
n
k=1
(k) +
M
n
2
n
k=1
(k)
k
.
In what follows we prove the following
Lemma. The following estimations hold true:
i) lim
n
1
n
2
n
k=1
(k) =
3
2
;
ii) lim
n
1
n
n
k=1
(k)
k
=
6
2
.
Proof. i) Let (n) be the Mobius function dened by
(n) =
_
(1)
(n)
= (1)
(n)
, if (n) = (n),
0, if (n) < (n),
where (n) is the number of distinct primes dividing the number n and (n)
is the number of prime factors of n, counted with multiplicities. Firstly, we
prove by induction the following identity:
n
k=1
(k)
k
=
n
k=1
(k)
k
_
n
k
_
. (30)
The base case is n = 1 and we see that the claim holds:
(1)
1
= 1 =
(1)
1
1| .
For the induction step we need to prove that
(n + 1)
n + 1
=
n
k=1
(k)
k
__
n + 1
k
_
_
n
k
_
_
+
(n + 1)
n + 1
.
The key observation is that
_
n + 1
k
_
_
n
k
_
=
_
1, if k[(n + 1),
0, otherwise,
so that the sum is
k|(n+1)
k<n+1
(k)
k
+
(n + 1)
n + 1
=
k[(n+1)
(k)
k
.
46 Problems
Now the fact that
k[(n+1)
(k)
k
=
(n+1)
n+1
is a basic totient identity. To
see that it holds, let p
v
1
1
p
v
2
2
. . . p
vq
q
be the prime factorization of n + 1. Then
(n + 1)
n + 1
=
q
l=1
_
1
1
p
l
_
=
k[(n+1)
(k)
k
,
by denition of (k). This concludes the proof of (30).
Now, we prove the identity
n
k=1
(k) =
1
2
_
1 +
n
k=1
(k)
_
n
k
_
2
_
. (31)
The base case is n = 1 and we have (1) = 1 =
1
2
_
1 +(1)
_
1
1
_
2
_
which is true. The induction step requires us to show that
(n + 1) =
1
2
n
k=1
(k)
_
_
n + 1
k
_
2
_
n
k
_
2
_
+
1
2
(n + 1)
_
n + 1
n + 1
_
2
.
Next observe that
_
n + 1
k
_
2
_
n
k
_
2
=
_
2
n+1
k
1, if k[(n + 1)
0, otherwise.
Therefore, the right side of the desired equality is
1
2
k|(n+1)
k<n+1
(k)
_
2
n + 1
k
1
_
+
1
2
(n + 1) =
1
2
k[n+1
(k)
_
2
n + 1
k
1
_
,
that is
(n + 1)
k[n+1
(k)
k

1
2
k[n+1
(k).
The rst of these two terms is precisely (n +1) as we saw earlier, and
the second is zero, by a basic property of the Mobius function (using the
same factorization of n + 1 as above, we have
k[n+1
(k) =
q
l=1
(1 1) = 0.)
This concludes the proof of (31).
Now, we are ready to prove that
1
n
2
n
k=1
(k) =
3
2
+O
_
log n
n
_
. (32)
Using the elementary inequalities x 1 < x| x, we get the upper
bound
n
k=1
(k)
_
n
k
_
2
=
n
k=1
(k)=1
_
n
k
_
2
k=1
(k)=1
_
n
k
_
2
k=1
(k)=1
n
2
k
2

n
k=1
(k)=1
_
n
2
k
2
2
n
k
+ 1
_
,
that is
n
2
n
k=1
(k)
k
2
+
n
k=1
(k)=1
_
2
n
k
1
_
< n
2
n
k=1
(k)
k
2
+ 2nH
n
k=1
(k)=1
1.
Similarly, we have the lower bound
n
k=1
(k)=1
_
n
k
_
2
k=1
(k)=1
_
n
k
_
2
k=1
(k)=1
_
n
2
k
2
2
n
k
+ 1
_
k=1
(k)=1
n
2
k
2
,
that is
n
2
n
k=1
(k)
k
2

n
k=1
(k)=1
_
2
n
k
1
_
> n
2
n
k=1
(k)
k
2
2nH
n
+
n
k=1
(k)=1
1,
where H
n
= 1 +
1
2
+ +
1
n
is the harmonic sequence.
Now using the asymptotic expansion of the harmonic sequence H
n
, we
have
2nH
n
O(nlog n),
n
k=1
(k)=1
1 O(n), and
n
k=1
(k)=1
1 O(n).
The term
n
k=1
(k)
k
2 is O(1) by comparison with (2), where (s) is the
Riemann zeta function
(s) =
n=1
1
n
s
, 1(s) > 1.
So far we have shown that
1
n
2
n
k=1
(k) =
1
2
n
k=1
(k)
k
2
+O
_
log n
n
_
.
It remains to evaluate
n
k=1
(k)
k
2
48 Problems
asymptotically, which we have seen converges.
The Euler product for the Riemann zeta function is
(s) =
p
_
1
1
p
s
_
1
for 1(s) > 1.
Now it follows immediately from the denition of the M obius function that
1
(s)
=
p
_
1
1
p
s
_
=
n1
(n)
n
s
.
This means that
1
2
n
k=1
(k)
k
2
=
1
2
1
(2)
+O
_
1
n
_
,
where the integral
_
n+1
1
t
2
dt was used to estimate
k>n
(k)
k
2
. But
1
2

1
(2)
=
3
2
and (32) is established.
ii) The above techniques together with the identity
n
k=1
(k)
k
=
n
k=1
(k)
k
_
n
k
_
also yield a proof that
1
n
n
k=1
(k)
k
=
6
2
+O
_
log n
n
_
.
Reasoning as before, we obtain the upper bound
n
k=1
(k)
k
_
n
k
_
n
n
k=1
(k)
k
2
+
n
k=1
(k)=1
1
k
and the lower bound
n
k=1
(k)
k
_
n
k
_
n
n
k=1
(k)
k
2

n
k=1
(k)=1
1
k
.
The proof of the Lemma is now nished.
Returning to our problem, we showed the double inequality
2[f(0)[
1
n
2
n
k=1
(k)
M
n
2
n
k=1
(k)
k
a
n
2[f(0)[
1
n
2
n
k=1
(k) +
M
n
2
n
k=1
(k)
k
.
By applying the above lemma and the squeeze theorem, we obtain the desired
conclusion.
354. For x > 1, dene the function f(x) =
+
_
1
e
it
x
dt. Prove that there exists
L C
such that
lim
x
xf(x) = L.
Proposed by Moubinool Omarjee, Jean Lur cat High School, Paris,
France.
Solution by the author. With the change of variable u = t
x
and an
integration by parts we get
f(x) =
1
x
_
1
e
iu
u
1
x
1
du =
ie
i
x

i
x
_
1
x
1
_

_
1
e
iu
u
1
x
2
du.
By the Lebesgue dominated convergence theorem we have
lim
x
_
1
e
iu
u
1
x
2
du =
_
1
e
iu
u
2
du =: C,
so f(x) =
1
x
(ie
i
iC +o(1)), which implies that
lim
x
xf(x) = L := i(e
i
C).
To prove that L ,= 0 we write L = ie
i
_
1
_
1 e
i(u1)
_
u
2
du and we note
that the real part of the integral is
_
1
(1 cos(u 1))u
2
du > 0.
Note that, again by an integration by parts, we have a simpler formula,
L =
_
1
e
iu
u
1
du.
355. Let p be an odd prime number and
_
0;

2
such that cos =

1
p
.
Prove that for any n N
, n > 1, there is no m N
such that cos (n) =

1
m
.
Proposed by Vlad Matei, University of Cambridge, UK.
Solution by the author. We argue by contradiction. Assume indeed that
cos (n) =
1
m
.
By Moivres formula we know that
cos (n) =
(1 +i
_
p
2
1)
n
+ (1 i
_
p
2
1)
n
2p
n
=
]n/2|
i=0
(1)
i
(p
2
1)
i
_
n
2i
_
p
n
.
50 Problems
It would follow, since p is prime, that there exists a positive integer l
such that l n and m = p
l
, thus
]n/2|
i=0
(1)
i
(p
2
1)
i
_
n
2i
_
= p
nl
. (33)
We have that
]n/2|
i=0
(1)
i
(p
2
1)
i
_
n
2i
_
]n/2|
i=0
_
n
2i
_
(mod p), thus
]n/2|
i=0
(1)
i
(p
2
1)
i
_
n
2i
_
2
n1
(mod p).
If l < n this would lead to p [ 2 so p = 2, in contradiction with the hypothesis.
Thus l = n. From this we can rewrite (33) as
]n/2|1
i=0
(1)
i
(p
2
1)
i
_
n
2i + 2
_
= 0. (34)
In what follows v
2
(x) denotes the exponent of 2 in the natural number
x, also known as the 2-adic valuation.
Let v
2
(p
2
1) = a and v
2
(n(n 1)) = b. We note that a 3 since p
is odd. Our aim is to prove that v
2
_
(p
2
1)
i
_
n
2i + 2
__
b for i 1, since
the rst term in our sum is
_
n
2
_
and it has 2-adic valuation b 1.
We have that
v
2
_
(p
2
1)
i
_
n
2i + 2
__
= v
2
((p
2
1)
i
) +v
2
__
n
2i + 2
__
=
= ai +v
2
(n(n1)(n2) (n2i 1))v
2
((2i +2)!) ai +bv
2
((2i +2)!).
Using Legendres formula we know that v
2
((2i+2)!) = 2i+2s
2
(2i+2),
where s
2
(x) is the sum of digits in the base two expansion of x, thus
v
2
((2i + 2)!) 2i + 1. Putting it all together, from our rst observation
a 3 and the fact that i 1 it results that
v
2
_
_
p
2
1
_
i
_
n
2i + 2
__
ai +b 2i 1 b + 3i 2i 1 = b +i 1 b.
Thus we have obtained our desired inequality and this gives an imme-
diate contradiction in the equality (34), by looking at the 2-adic valuations
of the rst term
_
n
2
_
and of the others
_
p
2
1
_
i
_
n
2i+2
_
, for i 1.
356. Let b
n
n0
be a sequence of positive real numbers. The following
statements are equivalent:
i)
n=0
[b
r
n+1
b
r
n
[
bn
< for all r R;
ii)
n=0
[b
n+1
bn[
bn
< ;
iii)
n=0
[b
n+1
b
n
[ < and limb
n
> 0;
iv)
n=0
[b
n+1
bn[
b
n+1
< ;
v)
n=0
[b
r
n+1
b
r
n
[
b
n+1
< for all r R.
Proposed by Alexandru Kristaly, Babe s-Bolyai University,
Cluj-Napoca, Romania, and Gheorghe Moro sanu, Central European
University, Budapest, Hungary.
Solution by the authors. First, we prove
n=0
[b
n+1
b
n
[
b
n
< m, M > 0 : m b
n
M, n 0. (35)
Let a
n
=
b
n+1
bn
bn
. From the hypothesis it follows that
n=0
[a
n
[ < , and
b
n+1
b
n
= 1 +a
n
, n 0.
Note that in particular lim
n
a
n
= 0. Therefore, we may assume without
any loss of generality that [a
n
[ < 1 for all n 0. It follows from the above
relation that
b
n
b
0
=
n1
k=0
(1 +a
k
). (36)
From (36) we obtain that
b
n
b
0
n1
k=0
(1+[a
k
[)
k=0
(1+[a
k
[)
k=0
exp([a
k
[) = exp(
k=0
[a
k
[) =: M
0
< ,
thus
b
n
b
0
M
0
=: M.
Using again (36) we obtain that
b
n
b
0
n1
k=0
(1 [a
k
[)
k=0
(1 [a
k
[).
52 Problems
The latter term is positive. We assume by contradiction that it is 0.
Consequently,
k=0
ln(1 [a
k
[) = .
Since lim
x0
ln(1[x[)
[x[
= 1 and lim
k
a
k
= 0, the above series has the same nature
as
n=0
[a
n
[. Contradiction. Thus,
k=0
(1 [a
k
[) = m
0
> 0, i.e.,
b
n
b
0
m
0
=: m.
ii) i) If r = 0, we have nothing to prove. If r ,= 0, by the mean value
theorem we obtain that b
r
n+1
b
r
n
= rc
r1
n
(b
n+1
b
n
), where m c
n
M.
This proves the claim.
In fact, i) and ii) are equivalent because ii) is obtained by specializing
r = 1 in i). It is also obvious that ii) iii). We have just to notice that
n=0
[b
n+1
b
n
[ < implies that b
n
is convergent. The converse implication,
iii) ii), is trivial.
The equivalence of the last three statements follows by similar argu-
ments.
357. Describe the functions : R R with (0) = 0 such that the set of
functions + y : y R is a semigroup with respect to the operation ,
the composition of functions. Prove that such a semigroup is a monoid if and
only if is the identity map.
Proposed by Dan Schwarz, Bucharest and Marcel T ena, Saint Sava
National College, Bucharest, Romania.
Solution by the authors. On the set of functions T = f : f : R R let
dene the relation f
1
f
2
if f
2
f
1
is a constant function. One immediately
checks is an equivalence relation. Then the class

f of any function f
contains exactly one (canonical) representative with (0) = 0.
Assume now that is a subsemigroup of (T, ). Let (+y) = +y
t
.
Computed at x = 0 it yields (y) = y
t
, thus (+y) = +(y) for all y.
Computed at y = 0 it yields = , thus also ( +y) = + y for all
y, i.e., is a neutral element to the right.
Conversely, for a function satisfying (0) = 0 and ((x) + y) =
= (x) +(y) for all x, y, we have
(( +a) ( +b))(x) = ( +a) (( +b)(x))
= ( +a)((x) +b) = ( + ((b) +a))(x),
hence is a semigroup.
2
Consider now K
= k R : (k) = k. From = follows

K
= Im. We have 0 K
and (k
1
+k
2
) = ((k
1
)+k
2
) = (k
1
)+(k
2
) =
= k
1
+ k
2
. Also, 0 = ((k) + (k)) = k + (k), hence (k) = k. It
follows that (K
, +) (R, +), an additive subgroup of R.

The improper cases are
K
= 0, corresponding to = 0;
K
= R , corresponding to = id
R
.
Let y be a coset of R/K
; we have (k+y) = k+(y) for k K
. Then
the function is dened by id on K
, and (arbitrarily) on representatives y of

the cosets y, prolonged to the whole coset by (k+y) = k+(y). Conversely,
any such function satises (0) = 0 and ((x) +y) = (x) +(y) for all
x, y. The function (x) = x| is such an example, for K
= Z, with ,= id
R
(obviously, the semigroup will have no neutral element).
But if + y : y R is a monoid, then = = (where is
the neutral element), hence is the neutral element of the monoid. Then
+(y) = ( +y) = +y for all y, hence = id
R
. On the other hand,
it is clear that for = id
R
the corresponding subset is a monoid (in fact a
group).
358. Prove that for any coloring of the lattice points of the plane with a
nite number n 1 of colors and for any triangle ABC having angles with
rational tangents there is a triangle with lattice vertices of the same color
which is similar to ABC.
Proposed by Benjamin Bogo sel, West University of Timi soara,
Romania.
Solution by the author. Let tan A =
a
b
=
abcd
b
2
cd
and tan B =
c
d
=
abcd
abd
2
,
with a, b, c, d N
, gcd(a, b) = gcd(c, d) = 1. We consider the lattice points

D, E, F of coordinates (0, 0), (b
2
cd + abd
2
, 0) and (b
2
cd, abcd), respectively.
By construction the triangle DEF is similar to ABC as tan A = tan D,
tan B = tan E.
From now on we will focus on points in the set
/ = X [

DX = p
DE +q
DF, p, q N.
Given k points P
1
, . . . , P
k
/ in an arithmetic progression on a line parallel
to Ox, let P be a point with P
1
P|DF and P
k
P|EF so that the triangle
P
1
P
k
P is similar to ABC. We denote by (P
1
, P
k
, k) the set of the vertices
of the (k 1)
2
congruent triangles similar to ABC in which the triangle
P
1
P
k
P can be divided. Then (P
1
, P
k
, k) /. We denote by (n) the
smallest number of points P
1
, . . . , P
(n)
/ in an arithmetic progression on
2
In fact this induces on R the associative operation ab = a+(b), with neutral element
to the right e = 0. To check, (a b) c = a (b c) = a +(b) +(c), a 0 = a +(0) = a.
54 Problems
a line parallel to Ox such that, regardless of the coloring, in (P
1
, P
(n)
, (n))
there are three points of the same color that are the vertices of a triangle
similar to ABC. (Assuming that such number exists.) We will prove by
induction on n that (n) is dened.
When n = 1 there is nothing to prove, as all points have the same color,
so (1) = 2.
For the induction step we use the Van der Waerdens theorem, which
states that for any r, k 1 there is a number N such that if the elements
of an arithmetic progression of length N are colored with r colors then we
can extract an arithmetic progression of length k with elements of the same
color. The smallest N with this property is the Van der Waerden number,
denoted W(r, k).
Assume that (n) is dened. We will show that (n+1) is also dened
and (n+1) W(n+1, (n) +1). We consider a coloring with n+1 colors.
Let P
1
, . . . , P
W(n+1,(n)+1)
/ be some points in an arithmetic progression
on a line parallel to Ox.
We want to prove that (P
1
, P
W(n+1,(n)+1)
, W(n+1, (n)+1)) contains
three points of the same color that are the vertices of a triangle similar
to ABC. By the Van der Waerdens theorem there are Q
1
, . . . , Q
(n)+1

P
1
, . . . , P
W(n+1,(n)+1)
in an arithmetic progression of the same color.
Obviously
(Q
1
, Q
(n)+1
, (n) + 1) (P
1
, P
W(n+1,(n)+1)
, W(n + 1, (n) + 1)).
We have
(Q
1
, Q
(n)+1
, (n) + 1) = Q
1
, . . . , Q
(n)+1
(R
1
, R
(n)
, (n)),
where R
1
, . . . , R
(n)
are the points of (Q
1
, Q
(n)+1
, (n) + 1) from the rst
line parallel to [P
1
, P
(n)+1
]. If (R
1
, R
(n)
, (n)) contains some point R of
the same color as the Q
i
s then R and two points Q
i
, Q
j
are the vertices of
a triangle which is similar to ABC and they have the same color, so we are
done. Hence we may assume that all points of (R
1
, R
(n)
, (n)) are colored
in the remaining n colors. By the induction hypothesis there are three points
of the same color in (R
1
, R
(n)
, (n)) that are the vertices of a triangle
similar to ABC and again we are done.
359. Determine how many permutations of the 81 squares of the Sudoku
grid have the property that for any solution of the Sudoku game, if we apply
the permutation to the 81 squares we obtain another solution of the Sudoku
game.
Solution by the author. Denote I := 1, . . . , 9. A completion of
the Sudoku grid with numbers from 1 to 9 can be regarded as a function
f : I I I. We denote by A the set of such completions that are solutions
of the Sudoku game.
For any set X we denote by
X
its group of permutations. In particular,
S
n
=
1,...,n
.
If
II
and f : I I I then, after applying to the 81 squares
of the grid, for any s I I the value on the square (s) will be f(s). Hence
the value on a square s will be f(
1
(s)). Thus the new completion of the
grid will be the function f
1
. Hence we want to determine [G[, where
G =
II
[ f
1
A f A. Note that if , G then for any
f A we have f
1
A and so f ()
1
= (f
1
)
1
G, whence
A. Hence G is a subsemigroup of
II
. But
II
is nite so G is a
subgroup.
For any i I we denote by r
i
and c
i
the i-th row and column, respec-
tively, r
i
= i I, c
i
= I i.
We have I = I
1
I
2
I
3
, where I
k
= 3k 2, 3k 1, 3k. Then
the grid can be divided into 3 blocks of 3 consecutive rows, R
1
, R
2
, R
3
, where
R
k
= r
3k2
r
3k1
r
3k
= I
k
I. Similarly we have the blocks of 3 consecutive
columns C
1
, C
2
, C
3
, C
k
= I I
k
. For 1 k, l 3 we denote by S
k,l
the nine
3 3 squares of the grid, S
k,l
= I
k
I
l
= R
k
C
l
. We have
A = f : I I I [ f(r
i
) = f(c
i
) = I i I, f(S
k,l
) = I 1 k, l 3.
For any i I we have i I
(i)
, where (i) =
_
i
3
_
. Then the row r
i
is a
part of the block R
(i)
and the column c
i
is a part of the block C
(i)
. Also a
square (i, j) I I belongs to the 3 3 square S
(i),(j)
.
If
I
then we denote by
r
,
c

II
the corresponding per-
mutation of the rows and columns, respectively. Namely,
r
is given by
(i, j) ((i), j) and
c
by (i, j) (i, (j)). Then
r
(r
i
) = r
(i)
,
r
(c
i
) = c
i
and
c
(r
i
) = r
i
,
c
(c
i
) = c
(i)
i I.
If ,
I
and =
r
c
then (r
i
) = r
(i)
and (c
i
) = c
(i)
, so
sends rows to rows and columns to columns. Also ((i, j)) = ((i), (j)).
In particular, this implies that is uniquely determined by and , i.e.,
if =
r
c
=
tr
tc
then
t
= ,
t
= . Conversely, if
II
sends
rows to rows and columns to columns, i.e., if there are ,
I
such that
(r
i
) = r
(i)
and (c
i
) = c
(i)
, then for any (i, j) I I we have (i, j) =
r
i
c
j
so ((i, j)) = (r
i
) (c
j
) = r
(i)
c
(i)
= ((i), (j)). Thus
is given by (i, j) ((i), (j)) and so =
r
c
. In conclusion, the mapping
(, )
r
c
is a bijection between
I

I
and the permutations of I I
sending rows to rows and columns to columns.
For any S
3
we denote H
=
I
[ (I
k
) = I
(k)
and
H =
S
3
H
. Note that H
, which implies HH H, so H
is a subgroup of
I
. Note that H
,= for any S
3
. Indeed, since for any
k, l the mapping i 3(l k) + i is a bijection from I
k
to I
l
, the mapping
56 Problems
: I I, given by i 3((k) k) + i when i I

k
, is a bijection with
(I
k
) = I
(k)
, so it belongs to H
.
We have a morphism : H S
3
given by if H
. Then
ker = H
1
and is surjective since H
,= S
3
. It follows that
[H[ = [S
3
[ [H
1
[. But [S
3
[ = 6 and H
1
=
I
[ (I
k
) = I
k
k is the
internal product of its subgroups
I
1
,
I
2
,
I
3
, each of them having 3! = 6
elements, so [H
1
[ = 6
3
. Hence [H[ = 6
4
.
Let H
, H
for some , S
3
and let =
r
c
. Then for
any i I we have (r
i
) = r
(i)
and (c
i
) = c
(i)
. Also since is given
by (i, j) ((i), (j)) we have (S
k,l
) = (I
k
I
l
) = (I
k
) (I
l
) =
= I
(k)
I
(l)
= S
(k),(l)
k, l 1, 2, 3. It follows that for any f A
we have f (r
i
) = f(r
(i)
) = I, f (c
i
) = f(c
(i)
) = I and f (S
k,l
) =
= f(S
(k),(l)
) = I. Thus f A for any f A, so
1
G, whence
G.
In conclusion, G contains H
r
H
c
=
r
c
[ , H, which is the
image of H H under the isomorphism (, )
r
c
dened above. We
have [H
r
H
c
[ = [H H[ = 6
8
. G also contains the reexion in the diagonal
, given by (i, j) (j, i). Indeed, if f A we have f (r
i
) = f(c
i
) = I and
f (c
i
) = f(r
i
) = I i I and f (S
k,l
) = f(S
l,k
) = I, k, l 1, 2, 3,
so f A. It follows that G H
r
H
c
H
r
H
c
. We claim that
G = H
r
H
c
H
r
H
c
, which will imply that [G[ = 2 6
8
= 3, 359, 232.
Lemma. For any (i, j) I I we denote by A
i,j
the union of the row,
the column and the 3 3 square containing it, A
i,j
= r
i
c
j
S
(i),(j)
.
If (i, j), (r, s) I I, (i, j) ,= (r, s), then the following are equivalent :
(1) f(i, j) ,= f(r, s), f A.
(2) (r, s) A
i,j
.
Proof. The implication (1)(2) follows from the fact that any f A is
injective on every row, column and 3 3 square. For the reverse implication
we assume that (r, s) / A
i,j
and we prove that there is some f A with
f(i, j) = f(r, s). Since (r, s) / r
i
and (r, s) / c
j
we have r ,= i and s ,= j.
Also (i, j) and (r, s) belong to dierent 33 squares, say S
k,l
and S
m,n
, with
(k, l) ,= (m, n).
Let f
0
A be arbitrary. In the square S
m,n
there is some (p, q) with
f
0
(p, q) = f
0
(i, j). Since (i, j) and (p, q) belong to dierent 3 3 squares
they are dierent so, by the (1)(2) implication, (p, q) / A
i,j
. Arguing in
the same way as for (r, s), we get p ,= i, q ,= j.
We consider the transpositions = (r, p), = (s, q) and we dene
=
r
c
. Since (r, s), (p, q) S
m,n
we have r, p I
m
, s, q I
n
. This
implies that
Im
H,
In
H, so H
r
H
c
G. Therefore
f := f
0
A.
We have (r) = p, (s) = q and, since i ,= r, p, j ,= s, q, we have
(i) = i, (j) = j.
Therefore (i, j) = (i, j) and (r, s) = (p, q), so f(r, s) = f
0
(p, q) =
= f
0
(i, j) = f(i, j), as claimed.
We denote M = X I I [ [X[ = 9, f(X) = I, f A.
Corollary. M is the set of all rows, columns and 3 3 squares.
Proof. Obviously, by the denition of A, M contains all rows, columns
and 3 3 squares.
Conversely, assume that X M. Then [X[ = [I[ = 9, so the condi-
tion that f(X) = I means that f
[X
is bijective. Hence if (i, j), (r, s) X,
(i, j) ,= (r, s), then for any f A we have f(i, j) ,= f(r, s), so, by the Lemma,
(r, s) A
i,j
. Hence X A
i,j
.
Let (i, j) X. If i I
k
, j I
l
then X A
i,j
= r
i
c
j
S
k,l
. If X = S
k,l
then we are done. So we may assume that X (r
i
S
k,l
) or X (c
j
S
k,l
) is
nonempty. We consider the rst case, so let (i, h) X (r
i
S
k,l
). We have
h I
m
with m ,= l since otherwise (i, h) I
k
I
l
= S
k,l
.
Since (i, j), (i, h) X, we have X A
i,j
A
i,h
. But c
j
S
k,l
C
l
, so
r
i
A
i,j
r
i
C
l
. Similarly, r
i
A
i,h
r
i
C
m
, so r
i
A
i,j
A
i,h

(r
i
C
l
) (r
i
C
m
) = r
i
. (We have C
l
C
m
= .)
Thus X A
i,j
A
i,h
= r
i
, whence X = r
i
. (We have [X[ = [r
i
[ = 9.)
Similarly, if X (c
j
S
k,l
) ,= we get X = c
j
and we are done.
Let now G. We prove that H
r
H
c
H
r
H
c
to get our claimed
result, G = H
r
H
c
H
r
H
c
. If X M then for any f A we have
f A, so f((X)) = f (X) = I. It follows that (X) M. Hence
will permutate the elements of M. We have M = M
1
M
2
, where
M
1
= r
i
, c
i
[ i I and M
2
= S
k,l
[ 1 k, l 3. Note that [r
i
c
j
[ = 1
i, j I and if X = S
k,l
then for any Y M, Y ,= X, we have either
[X Y [ = 3, when Y = r
i
with i I
k
or Y = c
j
with j I
l
, or [X Y [ = 0
otherwise. Thus M
1
= X M [ Y M, [X Y [ = 1.
If X M
1
then there is Y M with [X Y [ = 1. It follows that
[(X) (Y )[ = [(X Y )[ = 1, so (X) M
1
. Hence sends M
1
to M
1
and M
2
to M
2
, i.e., it sends rows and columns to rows and columns and 33
squares to 3 3 squares.
Note that if X Y ,= then (X) (Y ) ,= , so for any X M
1
sends Y M
1
[ X Y ,= to Y M
1
[ (X) Y ,= . Take
X = r
1
. Then (r
1
) = r
h
or c
h
for some h I. In the rst case will send
Y M
1
[ r
1
Y ,= = c
i
[ i I to Y M
1
[ r
h
Y ,= = c
i
[ i I.
Hence sends columns to columns and rows to rows. In the second case
sends c
i
[ i I to Y M
1
[ c
h
Y ,= = r
i
[ i I. Hence sends
columns to rows and rows to columns. We consider the two cases separately.
If sends rows to rows and columns to columns then =
r
c
for some
,
I
. We have (r
i
) = r
(i)
and (c
i
) = c
(i)
, i I. Let k 1, 2, 3.
Then sends S
k,1
to another 3 3 square, say S
m,n
. Then for any i I
k
we
have r
i
S
k,1
,= and therefore r
(i)
S
m,n
= (r
i
) (S
k,1
) ,= . It follows
58 Problems
that (i) I
m
. Hence (I
k
) I
m
so, since is bijective, (I
k
) = I
m
. It
follows that there is some S
3
with (I
k
) = I
(k)
k 1, 2, 3. Hence
H. By a similar reasoning H and so H
r
H
c
.
If sends rows to columns and columns to rows then so does . Thus
G will send rows to rows and colums to columns. It follows that
H
r
H
c
, so H
r
H
c
.
Remark. Recall that H
1
is the kernel of the surjective homomorphism
: H S
3
and H
=
1
() S
3
. Hence H
are the classes of H/H

1
.
Since
, we have H
H
1
. So any H writes as =
0
t
,
where
0
=
for some S
3
and
t
H
1
. But H
1
is the internal direct
product of
I
1
,
I
2
,
I
3
, so
t
=
1
3
with
k

I
k
.
It follows that
r
=
r
0
r
1
r
2
r
3
. Since
0
permutates by translation
the sets I
1
, I
2
, I
3
,
r
0
permutates by translation the blocks R
1
, R
2
, R
3
. For
k = 1, 2, 3
k

I
k
, so
r
k
will permutate the rows of the block R
k
= I
k
I.
We proceed similarly with
c
when H. In conclusion, an element of G
has the form
=
s
r
0
r
1
r
2
r
3
c
0
c
1
c
2
c
3
,
where s 0, 1, is the reection in the diagonal of the grid,
r
0
permutates
the blocks R
1
, R
2
, R
3
,
r
1
,
r
1
,
r
3
permutate the rows within R
1
, R
2
, R
3
,
c
0
permutates the blocks C
1
, C
2
, C
3
, and
c
1
,
c
2
,
c
3
permutate the columns within
the blocks C
1
, C
2
, C
3
.
360. Let M
n
(C) be the ring of square matrices of size n and A M
n
(C).
The adjugate (classical adjoint) adj(A) of A is dened as follows: the (i, j)-
minor M
ij
of A is the determinant of the (n 1) (n 1) matrix that
results from deleting row i and column j of A, and the i, j cofactor of A
as C
ij
= (1)
i+j
M
ij
. The adjugate of A is the transpose of the cofactor
matrix C
ij
of A.
Show that if for all positive integers k we have det((adj(A))
k
+I
n
) = 1,
then (adj(A))
2
= O
n
.
Romania, and Cezar Lupu, University of Pittsburgh, USA.
Solution by the authors. First of all, we prove the following.
Lemma 1. If A M
n
(C) is a matrix such that the adjugate adj(A) is
nilpotent, then (adj(A))
2
= O
n
.
Proof. From the hypothesis, it follows that det(adj(A)) = 0. If A would
be invertible, from the identity Aadj(A) = det(A)I
n
(which follows from
the Laplace development for the determinant) it would result that adj(A) is
invertible as well. This contradiction shows that one has det(A) = 0. Thus,
rank(A) n1. If rank(A) n2 then adj(A) = O
n
and the conclusion is
obvious. If rank(A) = n 1, from Aadj(A) = O
n
and Sylvesters inequality,
we deduce
0 = rank(Aadj(A)) rank(A) + rank(adj(A)) n = rank(adj(A)) 1.
Since rank(adj(A)) = 1, it follows that there exist M M
1,n
(C) and
N M
n,1
(C) such that adj(A) = NM and MN = C. As adj(A) is
nilpotent, there exists a positive integer k such that (adj(A))
k
= O
n
. On the
other hand, by iteration, we deduce that
O
n
= (adj(A))
k
= N (MN)
k1
M =
k1
NM =
k1
adj(A).
Since adj(A) ,= O
n
, we have = 0 and the conclusion follows immediately
from (adj(A))
2
= N (MN) M = adj(A).
Lemma 2. If X is a square matrix of size n with complex entries such
that for any positive integer k we have det(X
k
+I
n
) = 1, then X
n
= O
n
.
Proof. Indeed, let
1
,
2
, . . . ,
n
be the eigenvalues of the matrix X. It
is easy to see that 1 +
k
1
, 1 +
k
2
, . . . , 1 +
k
n
are the eigenvalues of X
k
+ I
n
.
By the hypothesis we have
(1 +
k
1
)(1 +
k
2
) (1 +
k
n
) = 1.
By expanding this writes as
x
k
1
+ +x
k
m
= 0,
where x
1
, . . . , x
m
are the products
i
1

it
with 1 t n and 1 i
1
<
. . . < i
t
n. (We have m = 2
n
1.)
We denote by S
k
the elementary symmetric polynomials dened by
S
k
=
1i
1
<...<i
k
m
x
i
1
x
i
k
and P
k
= x
k
1
+ +x
k
m
. Since P
1
= . . . = P
m
= 0
(see above), from the well-known Newtons identities
kS
k
=
k
i=1
(1)
i1
S
ki
P
i
valid for all integers k 1 we readily get S
1
= = S
m
= 0. It follows
that x
1
= = x
m
= 0. This in turn implies
1
= =
n
= 0. By
Hamilton-Cayley theorem we deduce that X
n
= O
n
.
Applying Lemma 2 for X = adj(A), we obtain that (adj(A))
n
= O
n
.
By Lemma 1 this implies (adj(A))
2
= O
n
.
361. 88% of the surface of a sphere is colored in red. Prove that there is a
cube inscribed in the sphere with all vertices red.
Solution by the author. For 1 i 8 we denote by X
i
the event that
the vertex A
i
of a random cube with vertices A
1
, . . . , A
8
inscribed in the
60 Problems
sphere is not red. By hypothesis
Prob(X
i
) =
12
100
for i = 1, . . . , 8,
where Prob(X
i
) is the probability of the event X
i
.
By Booles inequality we have
Prob
_
8
_
i=1
X
i
_
i=1
Prob(X
i
) =
96
100
,
so the probability that at least a vertex of the cube is not red is
96
100
. Hence
the probability that all vertices are red is
4
100
> 0. Thus there are cubes
with all vertices red.
362. Given a function f : X X, we will make the notations
f
0
(X) := X, f
n
(X) := f(f
n1
(X)) for n 1, f
(X) :=
n0
f
n
(X).
(i) Prove that f(f
(X)) f
(X).
(ii) Prove that for X = R and f a continuous mapping, f
(R) is R, a
half-line, a bounded segment, a singleton, or the empty set .
Moreover, let it now be given that f(f
(R)) = f
(R).
(iii) Prove that if f
(R) is bounded, then it is a closed interval (inclu-

sively degenerate cases a singleton, or the empty set ); also give
examples for each of these cases.
(iv) Give an example for f
(R) being an open half-line.

Proposed by Dan Schwarz, Bucharest, Romania.
Solution by the author. (i) We have f
(X) f
n
(X) f
n1
(X) X
for all n 1. We also have f(f
(X)) f
(X), since
f(f
(X)) = f
_

n0
f
n
(X)
_

n0
f(f
n
(X)) =
n0
f
n+1
(X) = f
(X).
(ii) f being continuous, f
n
(R) is a connex set for all n, hence f
(R) is
a connex set, thus belonging to the given list.
(iii) An example for f
(R) = is f(x) = x
2
+ 1; for f
(R) = C is
f(x) = C (constant); while for f
(R) = [a, b], a < b, is f(x) = a for x a,

f(x) = b for b x and f(x) = x for a x b.
Assume that f
(R) = (a, b), a < b. Let y

n
= a + (b a)/2n (a, b) for
n 1, hence lim
n
y
n
= a. Then there exists x
n
(a, b) for which y
n
= f(x
n
).
Since the sequence (x
n
)
n1
is bounded, it will contain a convergent sub-
sequence (x
kn
)
n1
; let its limit be . But then a = lim
n
y
kn
= lim
n
f(x
kn
) =
= f(), since f is continuous.
Will it be (a, b), it would follow a = f() (a, b), absurd. So
, (a, b), but as x
n
(a, b), it will follow [a, b], hence a, b.
But f(a) = a is absurd, since then a f
(R) = (a, b), thus f(b) = a.

A similar argumentation leads to f(a) = b, thus f(f(a)) = a, and again
a f
(R) = (a, b), absurd.

Analogously it is shown that f
(R) = [a, b) or f
(R) = (a, b], a < b,

leads to a contradiction.
(iv) Let f : R R be given by f(x) = 1 for x 1 and
ef(x) = (1 x)x
(1)
x
+xx
(1)
x
for x 1.
(For n N
, f(n) = n for n even, f(n) =

1
n
for n odd, while for x 1, f(x)
is dened as the broken line connecting these points on the graph of f f
is piecewise linear). It is then clear that f
(R) = f(R) = (0, ).

Remarks. Assume that f
(R) = [a, b], a < b. Consider the obviously

closed set f
1
(a) = x [ f(x) = a. Were it f
1
(a) [a, b] ,= , then
a f([a, b]). Were it f
1
(a) [a, b] = , then there exists some > 0
such that f
1
(a) [a , b + ] = , but this contradicts the fact that
a = sup
n1
inf f
n
(R). Similar considerations lead to b f([a, b]). It follows
[a, b] f([a, b]) [a, b], therefore f(f
(R)) = f
(R).
Thus, when f
(R) is bounded, then f(f
(R)) = f
(R) if and only if

f
(R) is a closed interval (inclusively the degenerate cases a singleton, or

the empty set ), which strengthens point (iii).
363. For a given sequence (x
n
)
n1
of real numbers and n
0
a xed positive
integer, consider the following conditions:
(C
1
): n
2
(x
n+1
x
n
) (2n + 1)x
n
has the same sign for all n n
0
;
(C
2
): x
n+m
x
n
+x
m
for all n, m ,= n
0
;
(C
3
):
n=1
n
2
x
n
< ;
(C
4
): lim
n
x
n
n
= 0.
Prove that:
(a) none of (C
1
), (C
2
), (C
3
) implies (C
4
);
(b) (C
4
) follows from (C
3
) and either (C
1
) or (C
2
);
(c) the converse of (b) is false.
Proposed by Arpad Benyi, Western Washington University,
Bellingham, WA, and Kasso Okoudjou, University of Maryland, College
Park, Washington DC, WA, USA.
Solution by the authors. Let us start by noting that the condition (C
1
)
about the sequence (n
2
(x
n+1
x
n
) (2n +1)x
n
)
nn
0
having a constant sign
is equivalent to
(C
1
) : (n
2
x
n
)
nn
0
is monotone (increasing or decreasing).
For example,
n
2
(x
n+1
x
n
) (2n + 1)x
n
0, n n
0
62 Problems
is equivalent to
n
2
x
n+1
(n + 1)
2
x
n
0
x
n+1
(n + 1)
2

x
n
n
2
, n n
0
.
For the remainder of our solution, we will work with statement

(C
1
) instead
of (C
1
).
(a) We want to show that neither one of the conditions (C
1
), (C
2
), (C
3
)
implies (C
4
).
To see that (C
1
) , (C
4
), let x
n
= 2
n
. Clearly, the sequence
xn
n
2
=
2
n
n
2
is
increasing for n 3. To make it decreasing, pick x
n
= 2
n
. In both cases,
xn
n
=
2
n
n
as n .
For (C
2
) , (C
4
), let x
n
= n
2
. Clearly, x
n+m
= (n + m)
2
n
2
m
2
= x
n
+x
m
, but
xn
n
= n .
Finally, for (C
3
) , (C
4
), let x
n
= (1)
n
n. We have
n
2
x
n
=
=
(1)
n
n
1
which is well known to converge as an alternating series,
while n
1
x
n
= (1)
n
diverges.
(b) We want to show that either of the combinations (C
1
) (C
3
) or
(C
2
) (C
3
) implies (C
4
). We start with the rst combination.
(b1) We know that (n
2
x
n
) is a monotone (increasing or decreasing)
sequence. If we combine this with the convergence of the series given in (C
3
),
and use Pringsheims theorem, we conclude that lim
n
n n
2
x
n
= 0, and we
are done.
We recall here Pringsheims theorem: If
n=1
a
n
< and (a
n
) is mono-
tone (increasing or decreasing), then lim
n
na
n
= 0; see, for example, W.L.
Ferrar, A Text-Book of Convergence, Oxford, 1938.
(b2) We note rst that, given (C
2
), the sequence (n
1
x
n
)
n1
is conver-
gent on the extended line R, that is lim
n
n
1
x
n
exists (and is either nite or
). We sketch the proof of this standard fact here.
Let l = inf
n1
n
1
x
n
. If l > , then for all > 0, there exists m 1
such that m
1
x
m
< l +. For all n > m, we can write n = qm+r, for some
0 r m1. Therefore,
l n
1
x
n
= (qm +r)
1
x
qm+r
(qm+r)
1
(qx
m
+x
r
)
= m
1
x
m
qm
qm+r
+n
1
x
r
(l +)
qm
qm+r
+n
1
x
r
,
where in the second inequality we used the condition x
n+m
x
n
+x
m
for all
n, m. From here we immediately conclude that
l liminf n
1
x
n
limsupn
1
x
n
l +,
which proves that lim
n
n
1
x
n
= l. A similar argument works if l = .
We turn now our attention to the other condition of the hypothesis.
Using Kroneckers lemma we see that the convergence of the series
n=1
1
n
(n
1
x
n
) implies lim
n
1
n
n
k=1
k
1
x
k
= 0. Since (n
1
x
n
) is convergent on
the extended line, we can use the Cesaro-Stolz lemma to conclude that
0 = lim
n
1
n
n
k=1
k
1
x
k
= lim
n
n
1
x
n
,
and we are done.
We recall Kroneckers lemma: If (u
n
)
n1
is a monotone increasing, un-
bounded sequence of positive numbers, and (v
n
)
n1
is a sequence of real
numbers such that the series
n=1
vn
un
is convergent, then lim
n
1
un
n
k=1
v
k
= 0.
(c) To see that (C
4
) , ((C
1
) (C
2
)) (C
3
), select, for example,
x
n
=
n
lnn
, n 2. In this case, n
1
x
n
=
1
ln n
0, so (C
4
) holds, but
(C
3
) clearly fails since
n
2
x
n
=
1
nln n
= .
Incidentally, for the sequence considered, both (C
1
) and (C
2
) are satised.
The simple modication x
n
=
(1)
n
n
lnn
, n 2, provides an example for which
neither one of (C
1
) or (C
2
) holds, while (C
3
) does.
364. Let (x
n
)
n1
be a sequence of real numbers such that
limsup
n
(1 x
n
) log n < .
Show that if the series of positive reals
n1
a
n
converges then the series
n1
a
xn
n
also converges.
Proposed by Cristian Ghiu, Politehnica University of Bucharest,
Romania.
Solution by the author. From limsup
n
(1 x
n
) log n < we get that
(1 x
n
) log n M, so 1 x
n

M
log n
, n 2 for some M > 0 large enough.
Since
n1
a
n
is convergent, we have a
n
0, so there is some N
1
N
such that a
n
< 1, n N
1
.
We take an integer N
2
> e
2M
. Then
M
log n
<
1
2
, n N
2
. In particular
for n N
2
we have 1 x
n

M
log n
<
1
2
, so that x
n
>
1
2
.
Let n
0
:= maxN
1
, N
2
.
If n n
0
then a
n
(0; 1) and x
n
>
1
2
. There are two cases:
Case I: x
n
< 1. We use the inequality
A
t
B
1t
tA+ (1 t)B, t [0; 1], A, B (0; ), (37)
64 Problems
which follows immediately from the concavity of the logarithm function.
Since x
n
< 1 and 1 x
n

M
log n
, we have
1
1xn
log n
M
.
From (37) and the inequality above we get
a
xn
n

1
e
2M
= a
xn
n

_
_
1
e
2M
_ 1
1xn
_
1xn
x
n
a
n
+ (1 x
n
)
_
1
e
2M
_ 1
1xn
x
n
a
n
+ (1 x
n
)
_
1
e
2M
_log n
M
= x
n
a
n
+ (1 x
n
)
1
e
2 log n
=
= x
n
a
n
+ (1 x
n
)
1
n
2
a
n
+
1
n
2
.
It follows that a
xn
n
e
2M
a
n
+
e
2M
n
2
.
Case II: x
n
1. Now it is clear that we have a
xn
n
a
n
e
2M
a
n
+
e
2M
n
2
.
So we have proved that for any n n
0
it holds
a
xn
n
e
2M
a
n
+
e
2M
n
2
.
Since the series
n1
a
n
and
n1
1
n
2
are convergent, this implies that
nn
0
a
xn
n
is also convergent, and so is
n1
a
xn
n
.

Gma1 2 2013 Continut

Încărcat de

Informații document

Titlu original

Drepturi de autor

Formate disponibile

Partajați acest document

Partajați sau inserați document

Opțiuni de partajare

Vi se pare util acest document?

Este necorespunzător acest conținut?

Drepturi de autor:

Formate disponibile

Gma1 2 2013 Continut

Încărcat de

Drepturi de autor:

Formate disponibile

GAZETA MATEMATIC

In perioada 19601965 a urmat cursurile Facult at ii de Matematic a a

In aceast a teza se propune o metoda matriciala pentru determinarea tuturor

In cadrul catedrei de Informatica a Universitat ii din Bucuresti a parcurs

Incepand cu anul 2005 a fost visiting professor la Abdus Salam School of

In perioada 19902007 a fost seful catedrei de Informatica din Univer-

In anul 1987 a fost unul dintre organizatorii conferint ei Semester of

In ultima perioada, activitatea stiint ica a domnului profesor Ioan

In anul 2009 a fost ales membru al International Academy of Mathe-

In aceast a sect iune este prezentat a rezolvarea problemei P6 si a catorva

D. Acelasi lucru este valabil pentru ecuat ia Pell generalizat a

D. Aceste rezultate decurg dintr-o teorema

In aceast a sect iune vom studia mult imea

Intruc at P3 se specializeaza la P6, din cele prezentate n Sect iunea 2

In fapt, un rezultat aseman ator este valabil n condit ii mai generale.

In continuare vom presupune [q[ 2 si vom da condit ii necesare pentru

In ciuda simplit at ii, aceste condit ii necesare pentru a avea Q

In general, pentru orice x > 0, folosind Teoremele 4 si 7 se stabileste

Pentru a raspunde nca unei ntrebari formulate n [6], construim familii

In continuare, vom utiliza pentru aceast a structura arborescenta denumirea

In arborele Lin, descendent ii direct i ai nodului (1, n) sunt noduri de

In Figura 3 observ am ca subarborele stang al nodului (2, 4) este identic

In Figura 4 este prezentata diagrama binara obt inuta prin transformarea

In [9], aceast a matrice a fost denumita matricea compunerilor ascen-

Inlocuind n relat ia init ial a obt inem

) are hermitian matrices (here A

is the adjoint of A, i.e., the conjugate transpose of A).

and we may take = 2 cos , which gives us

such that cos =

such that cos (n) =

= k R : (k) = k. From = follows

, +) (R, +), an additive subgroup of R.

; we have (k+y) = k+(y) for k K

, and (arbitrarily) on representatives y of

, gcd(a, b) = gcd(c, d) = 1. We consider the lattice points

: I I, given by i 3((k) k) + i when i I

are the classes of H/H

(R) is bounded, then it is a closed interval (inclu-

(R) being an open half-line.

(R) = [a, b], a < b, is f(x) = a for x a,

(R) = (a, b), a < b. Let y

(R) = (a, b), thus f(b) = a.

(R) = (a, b), absurd.

(R) = (a, b], a < b,

, f(n) = n for n even, f(n) =

(R) = f(R) = (0, ).

(R) = [a, b], a < b. Consider the obviously

(R) is bounded, then f(f

(R) if and only if

(R) is a closed interval (inclusively the degenerate cases a singleton, or

S-ar putea să vă placă și