Contents lists available atScienceDirect
Linear
Algebra
and
its
Applications
www.elsevier.com/locate/laa
On
a
conjecture
of
Ilmonen,
Haukkanen
and
Merikoski
concerning
the
smallest
eigenvalues
of
certain
GCD
related
matrices
Ercan Altınışıka,∗, Ali Keskina, Mehmet Yıldıza,
Murat Demirbükenb
aDepartmentofMathematics,Facultyof Sciences,GaziUniversity,06500
Teknikokullar–Ankara,Turkey
b
DepartmentofComputerScience,FacultyofEngineering,İ.DoğramacıBilkent University,06800Bilkent–Ankara,Turkey
a r t i c l e i n f o a b s t r a c t
Articlehistory:
Received14July2015 Accepted23November2015 Availableonline15December2015 SubmittedbyR.Brualdi MSC: 15A18 15A23 15B36 15B48 11B39 11C20 Keywords: GCDmatrix (0,1)-matrix Positivematrix Eigenvalue Spectralradius Fibonaccinumber
LetKnbethesetofalln× n lowertriangular(0,1)-matrices
witheachdiagonalelementequalto1,Ln={Y YT: Y ∈ Kn}
andletcnbetheminimumofthesmallesteigenvalueofY YT
as Y goes throughKn.TheIlmonen–Haukkanen–Merikoski
conjecture(theIHMconjecture)statesthatcnisequaltothe
smallesteigenvalueofY0Y0T,whereY0 ∈ Kn with(Y0)ij = 1−(−1)i+j
2 fori> j.Inthispaper,wepresentaproofofthis
conjecture.Inourproofweuseaninequalityforspectralradii ofnonnegativematrices.
© 2015ElsevierInc.All rights reserved.
* Correspondingauthor.Tel.:+903122021070.
E-mailaddresses:ealtinisik@gazi.edu.tr(E. Altınışık),akeskin1729@gmail.com(A. Keskin),
yildizm78@mynet.com(M. Yıldız),murdem91@gmail.com(M. Demirbüken).
http://dx.doi.org/10.1016/j.laa.2015.11.023
1. Introduction
Let S = {x1,x2,. . . ,xn} be a set of distinct positive integers, (xi,xj) denote the
greatest common divisor of xi and xj and let ε be apositive real number. The n× n
matrices (S) = ((xi,xj)) and (Sε) = ((xi,xj)ε) are called the GCD matrix and the
powerGCDmatrixonS,respectively.TheLCMmatrixandthepowerLCMmatrixare similarly defined. In 1876, Smith [30] proved thatif S is factor closed, then det(S) = n
k=1ϕ(xk),where ϕ isEuler’stotient.Sincethen manyresultsonthese matriceshave
been publishedintheliterature,seee.g.[5,6,10,11,13,21].
AninterestingandactiveareainthestudyofGCDtypematricesistheir eigenstruc-ture. The first results on this subject were published in the papers [32,7,23] but the paper ofHongand Loewy[16]canbe consideredasthe firstpaperon thestudyof the eigenvalues of GCD and related matrices due to the number theoretical aspect of the subject. Since their pioneeringpaper manyresults onthe subjecthave been published in the literature, see e.g. [1,3,4,14,15,17,19,24,26–28]. In that paper, Hong and Loewy investigatedtheasymptoticbehavioroftheeigenvaluesofpowerGCDmatricesbyusing sometoolsofnumbertheory.Besidetheirresultsonasymptoticbehaviorofthese matri-ces, inthesamepaperHongandLoewyintroducedaconstantcn andusedittopresent
alowerboundforthesmallesteigenvaluesofpowerGCDmatrices.LetKn bethesetof
all n× n lowertriangular (0,1)-matriceswith eachdiagonalelement equalto 1 and let
Ln={Y YT : Y ∈ Kn}.Theydefinedthenumberscn dependingonlyonn asfollows:
cn= min Z∈Ln
μ(1)n (Z) : μ(1)n (Z) is the smallest eigenvalue of Z. (1.1) Then theyproved that
λ(1)n ((Sε))≥ cn· min
1≤i≤n{Jε(xi)},
whereJεisJordan’sgeneralizationofEuler’stotientandλ(1)n ((Sε)) isthesmallest
eigen-valueofthepowerGCDmatrix(Sε),see[16, Theorem 4.2].
In2008,Ilmonen,HaukkanenandMerikoski[19]studiedeigenvaluesofmeetandjoin matriceswhichareabstractgeneralizationsofGCDandLCMmatrices,respectively,and they generalizedthe aboveresultconcerning the numberscn for positivedefinite meet
matrices definedonlocally finite meetsemilattices.They alsoobtainedasimilar result for joinmatrices andhence,inparticular,forLCM matrices.Inthesamepaper,inthe light of their MATLAB calculations for n = 2,3,. . . ,7, they presented an interesting conjecture abouttheconstants cn.
Conjecture 1.1 (The IHM conjecture). (See [19,Conjecture 7.1].)Let Y0 = (y0ij)∈ Kn
yij0 = 1− (−1)
i+j
2 (1.2)
ifi> j.Thencn isequaltothesmallesteigenvalue ofY0Y0T.
Wehave furthernumericalevidence of the IHM conjecture as follows. Recently,the first, the second and the fourth author of the paper have investigated that the IHM conjectureholdsforn= 7 and n= 8 withthehelpofaMATLABcode.OurMATLAB code running on a computer1 has verifiedthe truth of the IHM conjecture for n = 7
in 23 minutes and for n = 8 in 3.5 days. Since |L8| = 228 = 268,435,456 and |L9| =
236= 68,719,476,736,itwouldtakeabout3 yearstoverifytheIHMconjectureforn= 9
withthehelpofourMATLABcode.Toovercomethisdifficultyabouttime,wewritea differentcodeinCprogramminglanguage.WeuseNewton’sidentities(see[20])toobtain thecharacteristicpolynomialofamatrixZ inLnandwecalculatethesmallesteigenvalue
of Z by using Newton’s method (see [31]) to shorten therunning time. Indeed, our C coderunningonthesamecomputerhas verifiedthe truthof theIHM conjecturein30 minutesforn= 8 andin7daysforn= 9.Thus,wehaveconcludedthatConjecture 1.1 holdsfor n= 8 and n= 9. Thisinvestigationhasbeen presented bythefirstauthor of thispaperin[2].
Afterobtainingenoughnumericalevidence thatthe IHMconjecture canbe true,we getthemotivation to findoutaproofof it.The strategyoftheproof isas follows. We provethatfor anymatrixY inKn, |Y−1|≤ |Y0−1|, wherethe matrixY0 given by(1.2)
and|Y−1| istheelement-wiseabsolutevalueofthematrixY−1.Secondly,weshowthat
|Z−1| ≤ |(Y
0Y0T)−1| for all Z ∈ Ln. Then, by using an inequalityfor spectral radii of
nonnegativematrices, weobtainaproofoftheIHM conjecture.Weconcludethepaper with a conjecture on the uniqueness of the matrix Y0 and a discussion about further
studiesontheconstantcn.
2. PropertiesofmatricesinKn
Firstwepresentasimplefactaboutaparticularnilpotent(0,1)-matrixwhichweuse inthe courseof our proofs. Here we give theproof of this fact thoughone canfind in theliterature.
Lemma 2.1. Let Y ∈ Kn and N := Y − I, where I is the n× n identity matrix. We
denoteby(Nk)ij theij-entryof thepositiveintegerk-th powerofN . Thenwehavethe
followingproperties.
i) (Nk)
ij = 0 wheneveri− j < k,
ii) Y−1 = I− N + N2− · · · + (−1)n−1Nn−1.
1
Proof. Since N is a strictly lower triangular (0,1)-matrix, the proof of the first claim follows fromthematrixmultiplication.Fortheproofofthesecond claim,consider
In− (−N)n = (I + N )(I− N + N2− · · · + (−1)n−1Nn−1). Since Nn = 0 andY = I + N ,wehave
Y−1= I− N + N2− · · · + (−1)n−1Nn−1. 2
Now we investigate the inverse of any matrixY in Kn. In the following lemma we
obtainarecurrencerelationfortheentriesoftheinverseof Y .
Lemma2.2.LetY ∈ Kn andN := Y − I.Denote N = (nij) andY−1 = (aij).Thenwe
have thefollowingrecurrencerelationforakl:
akl= ⎧ ⎪ ⎨ ⎪ ⎩ 0 if k < l, 1 if k = l, − k−1i=l nkiail if k > l.
Proof. When we multiply both sides ofthe equality inLemma 2.1(ii) from the left by
−N,wehave
−NY−1 =−N + N2− · · · + (−1)n−1Nn−1+ (−1)nNn.
Since Nn = 0,onecaneasily obtain
I− NY−1= Y−1. (2.1)
ByLemma 2.1,itisclearthatakl= 0 ifk < l andakl = 1 ifk = l.Now,from(2.1),we
have akl=− n i=1 nkiail
forallk > l and hence,byLemma 2.1,weobtain
akl=− k−1 i=l nkiail forallk > l. 2
Inthefollowing theorem,wefindanupper boundforeachaij interms ofFibonacci
Theorem 2.1. LetY ∈ Kn and denoteY−1 = (aij). Then, for 1≤ j < i ≤ n, we have
|aij| ≤ fi−j,wherefi−j isthe(i− j)-th Fibonaccinumber.
Proof. LetN = (nij) beasinLemma 2.2.Letj = 1,2,. . . ,n−1.Weprovebyinduction
ont= 1,2,. . . ,n−j that|aj+t,j| ≤ ftforallY ∈ Kn.ByLemma 2.2,wehave|aj+1,j| =
nj+1,j,where nj+1,j canbe0 or1.Thus,|aj+1,j| ≤ 1= f1forallY ∈ Kn.Now,assume
that for each t = 1,2,. . . ,k− 1 we have |aj+t,j| ≤ ft for all Y ∈ Kn. We provethat
|aj+k,j| ≤ fk for allY ∈ Kn. ByLemma 2.2, we haveaj+k,j =− j+ki=j−1nj+k,iaij for
allY ∈ Kn.
Case 1. Assume nj+k,j+k−1 = 0. Then |aj+k,j| = j+ki=j−2nj+k,iaij. Also, by
Lemma 2.2, aj+k−1,j = − j+k−2i=j nj+k−1,iaij. Since both of nj+k,i and nj+k−1,i for
each i = j,j + 1,. . . ,j + k− 2 can arbitrarily be 0 or 1, it is clear that aj+k,j and
aj+k−1,j gothrough thesamevalues,as Y goes throughthe setKn. Therefore, bythe
inductionhypothesis,weobtain|aj+k,j| ≤ fk−1≤ fk forallY ∈ Kn.
Case 2.Assumenj+k,j+k−1 = 1.
Subcase i.Assumenj+k−1,j+k−2= 0.ByLemma 2.2,wehave
|aj+k,j| ≤ j+k−2 i=j nj+k,iaij +|aj+k−1,j| .
Also, by Lemma 2.2, it is clear that aj+k−1,j = − j+k−2i=j nj+k−1,iaij. Since both of
nj+k,iandnj+k−1,i foreachi= j,j + 1,. . . ,j + k− 2 canarbitrarilybe0 or1,itisclear
that− j+k−2i=j nj+k,iaij and aj+k−1,j go through the samevalues, as Y goes through
the set Kn. Thus, by the inductionhypothesis, we obtain j+ki=j−2nj+k,iaij ≤ fk−1.
Beside this, by ourassumption in Subcase i, |aj+k−1,j| = j+ki=j−3nj+k−1,iaij. Since
both of nj+k−1,i and nj+k−2,i for each i = j,j + 1,. . . ,j + k− 3 can arbitrarily be 0 or 1,itisobviousthat j+k−3i=j nj+k−1,iaij andaj+k−2,j gothroughthesamevalues,as
Y goesthrough the set Kn. Bytheinduction hypothesis, we obtain|aj+k−1,j| ≤ fk−2
forallY ∈ Kn.Thus,|aj+k,j| ≤ fk−1+ fk−2= fk forallY ∈ Kn.
Subcase ii.Assumenj+k−1,j+k−2= 1. ByLemma 2.2,wehave
|aj+k,j| ≤ j+k−3 i=j nj+k,iaij + j+k−1 i=j+k−2 nj+k,iaij .
Since j+ki=j−3nj+k,iaij andaj+k−2,j gothroughthesamevalues,asY goesthroughthe
setKn,bytheinductionhypothesis,wehave i=jj+k−3nj+k,iaij ≤ fk−2forallY ∈ Kn.
Inadditiontothis, sincenj+k−1,j+k−2= 1 we obtain j+k−1 i=j+k−2 nj+k,iaij = (nj+k,j+k−2− 1)aj+k−2,j− j+k−3 i=j nj+k−1,iaij.
Here each(1− nj+k,j+k−2),nj+k−1,j,. . . ,nj+k−1,j+k−3 canarbitrarily be 0 or1.Thus, j+k−1
i=j+k−2nj+k,iaij andaj+k−1,j gothroughthesamevalues,asY goesthroughtheset
Knandhence,bytheinductionhypothesis, j+k−1i=j+k−2nj+k,iaij ≤ fk−1forallY ∈ Kn.
Therefore, weobtain|aj+k,j| ≤ fk−2+ fk−1 = fk forallY ∈ Kn.
Theprincipleofinductioncompletestheproof. 2
LetA= (aij),B = (bij)∈ Mn(R),thatis,thesetofalln× n realmatrices.Wewrite
A ≥ 0 ifallaij ≥ 0.Also,we write A≥ B ifA− B ≥ 0.In additionto this, wedefine
|A| = (|aij|),thatis,|A| istheelement-wiseabsolutevalueofA.Thelargest eigenvalue
of A inmodulusisdenotedbyρ(A) andcalledthespectral radiusofA.Nowwefix the notation fortherest ofthepaper.
Theorem 2.2. Let Y0 = (y0ij)∈ Kn be as in (1.2) andlet Z0 := Y0Y0T. ForallZ ∈ Ln,
we have|Z−1|≤ |Z0−1|.
Proof. Firstly we obtaintheinverseofY0.LetN0= Y0− I andN0= (mij).Then
mij =
0 if i≤ j,
1−(−1)i+j
2 otherwise.
WeclaimthattheinverseofY0 isthen× n matrix(cij),where
cij= ⎧ ⎪ ⎨ ⎪ ⎩ 0 if i < j, 1 if i = j, (−1)i−jfi−j if i > j.
Since Y0∈ Kn byLemma 2.2, itisclearthatcij = 0 ifi< j andcij = 1 ifi= j.Also,
byLemma 2.2,wehave cij =− i−1 k=j mikckj
fori> j.Nowweprovethatcij = (−1)i−jfi−j wheneveri> j byinductionont= i−j.
Fort= 1,
cj+1,j =−mj+1,j=−1 = −f1.
Assumethatcj+t,j= (−1)tftforallt= 1,2,. . . ,k− 1.Recall that
cj+k,j=− j+k−1
s=j
Bytheinductionhypothesis, ifk iseventhenwe have cj+k,j = k/2 s=1 f2s−1= fk
andifk isoddthen
cj+k,j =−1 −
(k−1)/2
s=1
f2s=−fk.
Here thelast equalities follow from the well-knownFibonacciidentities, see [22, Theo-rem 5.2andCorollary 5.1].Thus,cj+k,j= (−1)kfk.
Secondly,wecalculatetheinverseofZ0.SinceZ0= Y0Y0T andY0−1= (cij),wehave
(Z0−1)ii = n k=1 c2ki= 1 + n k=i+1 fk−i2 (2.2)
foralli= 1,2,. . . ,n.Nowlet1≤ i< j≤ n.Then
(Z0−1)ij= n t=1 ctictj = cji+ n t=j+1 ctictj = (−1)j−ifj−i+ n t=j+1 (−1)−i−jft−ift−j = (−1)j−i(fj−i+ n t=j+1 ft−ift−j). (2.3)
SinceZ0−1 issymmetric, forall1≤ j < i≤ n,
(Z0−1)ij = (−1)i−j(fi−j+ n
t=i+1
ft−ift−j). (2.4)
Nowweprovetheclaimofthetheorem.ForeachZ ∈ Ln,there existsamatrixY in
Kn such thatZ = Y YT. Let Y−1 = (aij). Then, by Lemma 2.2 and Theorem 2.1, we
(Z−1) ii= n k=1 a2ki = n k=1 |aki|2 = n k=i |aki|2 ≤ 1 + n k=i+1 fk−i2 =(Z0−1)ii
foralli= 1,2,. . . ,n. Letj > i. ByLemma 2.2andTheorem 2.1,wehave (Z−1) ij= n t=1 atiatj ≤ n t=1 |ati||atj| =|aji| + n t=j+1 |ati||atj| ≤ fj−i+ n t=j+1 ft−ift−j =(Z0−1)ij.
Finally,sinceZ−1 andZ0−1 aresymmetric,|Z−1|≤ |Z0−1| forallZ∈ Ln. 2
3. ProofofConjecture 1.1
Thefollowing lemmaiscrucialintheproofofConjecture 1.1.
Lemma 3.1. (See [18, Theorem 8.1.18].) Let A,B ∈ Mn(R). If |A| ≤ B, then ρ(A) ≤
ρ(|A|)≤ ρ(B).
Proofoftheconjecture. LetZ0 beas inTheorem 2.2.Firstweprovethatthematrices Z0−1 and|Z0−1| have thesame characteristic polynomial.Bythedefinition ofthe trace of asquarematrix,itisclearthat
tr((Z0−1)k) = n i1,...,ik=1 (Z0−1)i1i2. . . (Z −1 0 )ik−1ik(Z −1 0 )iki1
foreachk = 1,2,. . . ,n.Also,fromformulae(2.2)–(2.4)intheproofofTheorem 2.2,one caneasily showthatsgn(Z0−1)ij= (−1)i−j forall1≤ i,j≤ n.Thus,we have
sgn((Z0−1)i1i2. . . (Z
−1
0 )ik−1ik(Z −1
0 )iki1) = 1
and hence tr(|Z0−1|k) = tr((Z0−1)k). By Newton’s identities [20], we obtain that Z0−1
and |Z0−1| have the same characteristic polynomial. Thus, ρ(|Z0−1|) = ρ(Z0−1). From
Theorem 2.2and Lemma 3.1,now weobtain
ρ(Z−1)≤ ρ(|Z−1|) ≤ ρ(|Z0−1|) = ρ(Z0−1),
forall Z inLn. Sinceall Z inLn are positivedefinite,thesmallesteigenvalue ofZ0 is
lessthanor equalto thesmallesteigenvalueofZ forallZ inLn. 2
Finally, inour investigation [2], we cannot find any matrixY other than Y0 in Kn
suchthat cn is equalto the smallesteigenvalue ofY YT foreach n= 2,3,. . . ,9.After
thisobservationwecanpresent thefollowing conjecture.
Conjecture3.1.Letn≥ 2.ThereisauniquematrixY inKn suchthatcn isequaltothe
smallesteigenvalue ofY YT.In otherwords,ifthesmallest eigenvalueof Y YT isequal tocn thenY = Y0,whereY0 isdefined by(1.2).
4. Lowerboundsforcn
Intheliteraturetherearenotsomanyresultsonestimatingthevalueofcn.Recently,
Mattila[24] haspresented alower bound forcn.Indeed, heproved thatcn isbounded
belowby( 6
n4+2n3+2n2+n)
n−1
2 .Thenheshowedthatthislowerboundcanbereplacedwith
( 48
n4+56n2+48n)
n−1
2 when n is even, and ( 48
n4+50n2+48n−51)
n−1
2 when n is odd. Recently,
beside Mattila’s results, Altınışık and Büyükköse [4] have obtaineda lower bound for thesmallest eigenvaluetn of then× n matrix EnTEn, where the ij-entry of En is 1if
j|i and0 otherwise,i.e.,tn≥
n nk=1μ2(k)−1.Indeed,thisboundcanbeusedinstead
oflower bounds includingcn for thesmallesteigenvaluesof GCDand related matrices
definedonS ={1,2,. . . ,n} intheliterature,see[14,16,19,24,27].Afterabovestudieson estimatingthevalueofcn,wenaturallyraise thefollowing problem.
Problem4.1.Canoneimprovethelowerboundsmentionedaboveforcn?
In the review process, Professor Jorma K. Merikoski proposed a solution to Prob-lem 4.1,whichispresented inAppendix A.
Acknowledgements
TheauthorswouldliketothankProfessorIlkerInam,ProfessorPenttiHaukkanenand two anonymous reviewers for their valuable suggestions inimproving the presentation
of ourmanuscript.Wealsowouldlike tothanktoProfessorJormaK.Merikoskiforhis solutionto Problem 4.1.
Appendix A. AsolutiontoProblem 4.1(byJorma K.Merikoski) Recall thatthematrixZ0−1= (ζij) satisfies
ζii= 1 + n k=i+1 fk−i2 , |ζij| = f|j−i|+ n k=j+1 fk−ifk−j, i= j.
Alsorecallthatthespectralradius ofanonnegative squarematrixislessthan orequal to themaximalrowsum,andequalityholds ifand onlyifthematrixisirreducible and allrowsumsareequal(e.g., [9,Chapter 2,Theorem (2.35)]).Now
c−1n = ρ(Z0−1) = ρ(|Z0−1|) < max i n j=1 |ζij| = n j=1 |ζ1j| =|ζ11| + n j=2 |ζ1j| = 1 + n k=2 fk−12 + n j=2 fj−1+ n k=j+1 fk−1fk−j = 1 + n−1 k=1 fk+ n j=1 n k=j+1 fk−1fk−j = 1 + S1+ S2.
It iswell-known[12,Eq. (22)]that
m k=1 fk = fm+2− 1; (A.1) so S1= fn+1− 1.
Moreworkmustbe donewith S2. Firstwehave
S2= (f12+ f22+· · · + fn−12 ) + (f2f1+ f3f2+· · · + fn−1fn−2) +· · ·
= f1(f1+· · · + fn−1) + f2(f2+· · · + fn−1) +· · · + fj(fj+· · · + fn−1) +· · · + fn−2(fn−2+ fn−1) + fn2−1 = n−1 j=1 fj n−1 k=j fk. By(A.1), n−1 k=j fk = n−1 k=1 fk− j−1 k=1 fk= (fn+1− 1) − (fj+1− 1) = fn+1− fj+1, whichimplies S2= n−1 j=1 fj(fn+1− fj+1) = fn+1 n−1 j=1 fj− n−1 j=1 fjfj+1= T1− T2.
ItremainstosimplifyT1 and T2.By(A.1),
T1= fn+1(fn+1− 1).
Itiswell-known[29,SequenceA064831]that
m j=1fjfj+1= fm+12 − 1 ifn iseven, m j=1fjfj+1= fm+12 ifn isodd. Hence T2= fn2ifn iseven, T2= fn2− 1 ifn isodd. Inotherwords, T2= fn2− ηn, where ηn= 1− (−1)n 2 . Now c−1n < 1 + S1+ T1− T2 = 1 + (fn+1− 1) + fn+1(fn+1− 1) − fn2+ ηn
= fn+12 − fn2+ ηn
= (fn+1− fn)(fn+1+ fn) + ηn
= fn−1fn+2+ ηn.
It iswell-known[8,Eq. (2)]that
fm−rfm+s− fmfm+s−r= (−1)m−r−1frfs. Hence fn−1fn+2= fnfn+1+ (−1)n, and so c−1n < fnfn+1+ (−1)n+ ηn = fnfn+1+ θn, where θn= 1 + (−1)n 2 . Therefore cn> (fnfn+1+ θn)−1 = γn,
whichseemstobemuchbetterthanMattila’s[25]bounds.Forexample,c6= 0.0148 (in
three digits). Thebound γ6 = 0.00952 is rather good, while Mattila’s betterbound is
0.0000205. References
[1]E. Altınışık, On inverses of GCD matrices associated with multiplicative functions and a proof of the Hong–Loewy conjecture, Linear Algebra Appl. 430 (2009) 1313–1327.
[2]E. Altınışık, On a conjecture on the smallest eigenvalues of some special positive definite matrices, in: 3rd International Conference on Applied Mathematics & Approximation Theory – AMAT 2015, 28–31 May 2015, Ankara, Turkey, 2015.
[3]E. Altınışık, Ş. Büyükköse, A proof of a conjecture on monotonic behavior of the largest eigenvalue of a number-theoretic matrix, in: 12th International Conference of Numerical Analysis and Applied Mathematics, Rhodes, Greece, 2014.
[4]E. Altınışık, Ş. Büyükköse, A proof of a conjecture on monotonic behavior of the smallest and the largest eigenvalue of a number-theoretic matrix, Linear Algebra Appl. 471 (2015) 141–149. [5]E. Altınışık, B.E. Sagan, N. Tu˜glu, GCD matrices, posets, and nonintersecting paths, Linear
Mul-tilinear Algebra 53 (2) (2005) 75–84.
[6]E. Altınışık, N. Tu˜glu, P. Haukkanen, A note on bounds for norms of the reciprocal LCM matrix, Math. Inequal. Appl. 7 (4) (2004) 491–496.
[7]F. Balatoni, On the eigenvalues of the matrix of the Smith determinant, Mat. Lapok 20 (1969) 397–403 (in Hungarian).
[8]A.T. Benjamin, N.T. Cameron, J.J. Quinn, Fibonacci determinants – a combinatorial approach, Fibonacci Quart. 45 (2007) 39–55.
[9]A. Berman, R.J. Plemmons, Nonnegative Matrices in the Mathematical Sciences, SIAM, 1994. [10]S. Beslin, S. Ligh, Greatest common divisor matrices, Linear Algebra Appl. 118 (1989) 69–76. [11]K. Bourque, S. Ligh, On GCD and LCM matrices, Linear Algebra Appl. 174 (1992) 65–74. [12] P. Chandra, E.W. Weisstein, Fibonacci number, from MathWorld – a Wolfram web resource,
http://mathworld.wolfram.com/FibonacciNumber.html.
[13]P. Haukkanen, J. Wang, J. Sillanpää, On Smith’s determinant, Linear Algebra Appl. 258 (1997) 251–269.
[14]S. Hong, Asymptotic behavior of largest eigenvalue of matrices associated with completely even functions (mod r), Asian-Eur. J. Math. 1 (2008) 225–235.
[15]S. Hong, K.S. Enoch Lee, Asymptotic behavior of eigenvalues of reciprocal power LCM matrices, Glasg. Math. J. 50 (2008) 163–174.
[16]S. Hong, R. Loewy, Asymptotic behavior of eigenvalues of greatest common divisor matrices, Glasg. Math. J. 46 (2004) 303–308.
[17]S. Hong, R. Loewy, Asymptotic behavior of the smallest eigenvalue of matrices associated with completely even functions (mod r), Int. J. Number Theory 7 (2011) 1681–1704.
[18]R. Horn, C.R. Johnson, Matrix Analysis, second ed., Cambridge University Press, Cambridge, London, 2013.
[19]P. Ilmonen, P. Haukkanen, J.K. Merikoski, On eigenvalues of meet and join matrices associated with incidence functions, Linear Algebra Appl. 429 (2008) 859–874.
[20]D. Kalman, A matrix proof of Newton’s identities, Math. Mag. 73 (4) (2000) 859–874.
[21]I. Korkee, P. Haukkanen, On meet and join matrices associated with incidence functions, Linear Algebra Appl. 372 (2003) 127–153.
[22]T. Koshy, Fibonacci and Lucas Numbers with Applications, John Wiley, New York, 2001.
[23]P. Lindqvist, K. Seip, Note on some greatest common divisor matrices, Acta Arith. 84 (2) (1998) 149–154.
[24]M. Mattila, On the eigenvalues of combined meet and join matrices, Linear Algebra Appl. 466 (2015) 1–20.
[25]M. Mattila, On the eigenvalues of combined meet and join matrices, Linear Algebra Appl. 466 (2015) 1–20.
[26]M. Mattila, P. Haukkanen, On the eigenvalues of certain number-theoretic matrices, in: International Conference in Number Theory and Applications, 2012.
[27]M. Mattila, P. Haukkanen, On the eigenvalues of certain number-theoretic matrices, East-West J. Math. 14 (2) (2012) 121–130.
[28]M. Mattila, P. Haukkanen, On the positive definiteness and eigenvalues of meet and join matrices, Discrete Math. 326 (2014) 9–19.
[29] N.J.A.Sloane,Theon-lineencyclopediaofintegersequences,http://oeis.org/.
[30]H.J.S. Smith, On the value of a certain arithmetical determinant, Proc. Lond. Math. Soc. (1) 7 (1876) 208–212.
[31]E. Süli, D. Mayers, An Introduction to Numerical Analysis, Cambridge University Press, Cambridge, London, 2003.