Professional Documents
Culture Documents
flL tLl
LAS rsl
LineorAlgebroCommunity
Servingthe Internotionol
of ILASElections
Results (Richard
A. Brualdi) ...... .2
SomeObservations on 'A Genealogyof the Spottiswoode (R.WilliamFarebrother
Family'' & GeorgeP H. Styan)........ ..2
MagicSquares,Melancholy andTheMoore-Penrose lnverse(Dietrich
Trenkler&GoETrenkler) ,. .. .. ..... .3
PaulDanielsandSpacesof MagicSquares(SanjoZlobec) .. .. ... .. 10
MagicSquares on theWebandtheDeterminants of 4 x 4 MagicSquares(George P H. Styan) ........... 10
TheNamingof Parts(R.WilliamFarebrother) . ..... . 11
TheElectronicJoumalof LinearAlgebn (ELA):
Volumes1-4& 5-7 (James R. Weaver) . . . . 12
New&Forthcoming BooksonLinearAlgebn&RelatedTopics(George P.H. Styan). .......13
Challengesin MatrixTheory: 1OthILASConference(FrankUhlig) 17
Resultsof I LASElections Third, the ancestryof our first John (1510-1585), superin-
tendent of the Church of Scotland, is not known. Howevet it
seemscertain that he was not a son of the William Spottiswood
The ballots for this year's ILAS elections have now been of Spottiswoode(actually Archibald) who died at Flodden Field
counted. In accordancewith the election rules in the ILAS By- in 1513. This William should therefore be deletedfrom the head
Laws, on March 1,2002, of the genealogical table and from the opening paragraphsof the
article [3, p. 19].
Danny HenssKowlTz becomesILAS President (3-year term),
Fourth, the entail of 1863 was drawn up by our sixth John
Ravi Bnpar and Miki NBUvIANN (1780-1866) to protect his wife Helen Wauchopeand his wid-
become members of the ILAS Board (3-year terms). owed daughterAlicia Ann, seeBarry & Hall [1, pp. xxii-xxiii],
with the added proviso that those following were required to
We congratulate these winners but we also give our sincere ap- 'Spottiswoode'.
take the name of
preciation to all of those who showed their dedication to ILAS Fifth, the undifferenced arms of the Spottiswoodes of Spot-
by agreeing to stand for election. tiswoode illustrated in [2, p. 19] are those of the lairds of Spot-
We all look forward to working with Danny, Ravi, and Miki tiswoode including our fifth John (1741-1805) and our sixth
in service to our linear algebra community. John (1780-1866). As the son of a younger brother of our sixth
I want to personally thank the members of the nomination JohnoWilliam (1825-1883) was obliged to matriculate a dis-
committee [Roger Horn (Chair), Avi Berman, Tom Laffey, Hans tinct set of anns with the Office of the Lord Lyon in Edinburgh.
Schneider, and Bryan Shaderl for their efforts on behalf of These arms are illustrated in a special form in an armorial panel
ILAS. I also want to thank Roger Horn, Raphi Loewy, md Judi in the SpeechRoom of Harrow School. In this panel, the arms
MacDonald for counting the ballots in a timely manner. of William Spottiswoode(1825-1883) ateimpaled (set side-by-
side) with those of the Royal Society of London. The arms of
Richard A. B nUeLDI, IIAS P residenf.' brualdi@math.wisc.edu the Royal Society appearin the more prestigiousposition on the
D ept. of M athematic s, lJ nivers ity of Wisc ons in-M adis on WI 53 706 - 13 I 8, USA left of the shield as seen(or on the right as worn), and William's
own on the right (or on the left as worn).
Finally, wo note, with Barry & Hall [1, PP. 34] that, how-
ever spelt, the nrrme of Spottiswoode should be pronounced
'spotswood' as the letters 'is' representthe old Scottish geni-
'his' in English. Mackenzie
tive caseusually representedby 14,
pp. 108, 3l2l notes two contemporary documentsreferring to
SomeObseruationson 'Godis holy word' and to 'JesusChrist his birth, passion, resur-
'A Genealogyof the SpottiswoodeFamily" rection'.
References
We are indebted to Mrs Dorothy Spottiswoode Dor6 of "The
Spottiswoode Family History Society" (Nottingham, England),
a
for several observations on our recent articles on the genealogy tll T. Barry & D. Hall (1997). Spottiswoode: Ltfe and Labour on
of William Spottiswoode t2l and the Spottiswoode family [3]. Berwickshire Estate, 1753-1793. Infroduction by Douglas Hall
and Commentary by Tom Barry. Ttrckwell Press,East Linton, East
First, we should note that our genealogical table [3, P. 19]
Lothian, Scotland.
is not intended to represent a full genealogy of the Spottis-
woode family, but only a table that is suffrcient to trace the links t2l R. W. Farebrother (1999). A genealogy of William Spottiswoode
1825-1883. Inwge, 23, 34.
between three principal members of the family: John (1565- the
1639),the Archbishop of St Andrews; Alexander (1676-1740), t3l R. W. Farebrother & G. P. H. Styan (2000). A genealogy of
SpottiswoodeFamily: 1510-1900 . Image, 25, 19-21'
the Lieutenant-Governorof the Colony of Virginia; and William and
(L525-1883), the author of the first book ever published on de- l4l A. M. Mackenzie (1936). The Scotland of Queen Mary
the ReligiousWars, 1513-1638. Alexander Maclehose, London.
terminants (1851) and President of the Royal Society (1878-
lReprinted by Oliver & Boyd, Edinburgh, 1957, and Greenwood
1883). For a far more detailed, if still incomplete, genealogy Press,Westport, Connecticut, 1976.1
of the Spottiswoode family, we again refer to the tables in the
frontal matter of the book by Barry & Hall U, PP. xii-xxiiil. .ec.man.ac.uk
R. William FaReSROTHER: Msrbsrf@fsl
We understand that thesetables may be considerably extended I1 Castle Road, Bayston Hill, Shrewsbury,England SY3ONF
using information in the records Mrs Dor6 has on file.
George P. H. SrvnN: gphs@videotron.ca
Second, wo should note that the sevenJohns in our geneal- Dept. of Mathematics and Statisfdcs,McGill University
ogy [3, p. 19] were numberedfor the convenienceof our readers, Montrdal (Qudbec), Canada H3A 2K6
and that thesenumbershave no other significance. In particular,
they do not indicate the passageof the lairdship of Spottiswoode.
IMAGE27: October2001 page3
lnverse
Melancholyand TheMoore-Penrose
MagicSquares,
DietrichTrenklerand Cotz Trenkler
We find that
According to Ball & Coxeter (1987), this array is sometimesto
27L -L7r -205 169 be seenon the deck of large passengerships, for scoring in such
\ girmesas shuffleboard.
-103 67 101 -1
-6e -35 | The introduction of magic squares into Europe may be at-
33 135
I tributed to the Greek mathematician Emanuel Moschopulus,
-L37 237 2A3 -239 /
who lived in Constantinople in the fifteenth century (Balt &
Surprisingly enough, the patterns of numbers in p# adding to Coxeter, 1987).
1134 are the same as those of D (Fig. 3). The matrix D for Methods for consfirrcting magic squaresare described exten-
Diirer's magic squile and its Moore-Penrose inverse p+ have sively in the book by Moran (L982). The easiestmethod, also de-
52 patternsin common, and there are 34 patterns occurring only scribedthere, is applicableto any squareof odd order n. This so-
tn D and 34 patternsto be seenonly in l)* . called pyramid method starts with the desired,but empty, square
The magic squareof type n x n in its classical form contains which is then augmentedby additional boxes so that one obtains
the integers 1 ,2 , . . . , ?t2suchthat every row, column andthe two the "pyramid pattern", which is considered here for n = 3 and
main diagonals add to n (n' + L) 12. In early history suchsquares n=5:
were devised in Arabia, China, Greece, India and Japan (Hi-
rayama & Abe 1983), where they also served as charms against
all kinds of ailments. One of these squares,also known as "Loh-
Shu" or "scroll of the river Loh", is attributedl to the Great Em-
peror Yu, c. 2000 B.C. According to a Chinese legend,the Great
Emperor Yu once was standing by the Yellow River when a tor-
toise rose up from the water, delighting the Emperor with the
following urrrayon its back:
4 I 2
3 5 7
Ftc. 5: Empty box. Ftc. 6: Pyramid pattern.
8 1 6
9 5 I
2 7 6 11 4 t 7 1 0 23
24 l2 5 18 6
Flc. 7: Filled-uppyramid. Frc. 8: 3 x 3 magicsquare.
7 25 13 1 r9
The final magic square is obtained by reflecting the outside
numbers 1, 3,7 and 9 to the empty positionsoppositethe origi- 20 8 2l 14 2
nal location. How this method works for n = 5 is illustrated in
Figs. 9 and 10. 3 16 9 22 15
16 22
1 48 3 1 50 33 r 6 63 1 8
11 17 23
30 5 1 46 3 62 t9 T4 35
6 t2 18 24
47 2 49 32 l5 34 T7 64
1 7 t3 T9 25
52 29 4 45 20 6r 36 13
2 8 T4 20
5 44 25 56 9 40 21 60
3 9 15
28 53 8 4l 24 57 t2 37
4 10
43 6 55 26 39 1 0 59 22
5
54 27 42 7 58 23 38 11
w\)s
t\ L 4 n
cluding the famous method of the French writer Simon de La
(
'lN 'r\Xr/ N L 7
/
Loubdre (1642-1729) presentedin his Relation du Royaume de , n/ (
Siam publishedin t693.
D \.,
\ I
v\ ry
I \
v' \r 6 4
I
96 tl 89 6B A R E P o
S A T o R
BB 69 9 l t6
squareof Capesffano.
Flc. l6: 5 x 5 alphabeticalmagic
6l B6 IB 99
Four of the five words are Latin. AREPO is the exception,
though it recalls the word ARATRO, the ablative of the Latin
t9 98 66 BI word for plough. The symmetry of the array becomes evident
when reading ROIAS and OPERA backwards to get AREPO
magicsquare.
Ftc. 13: Reversible and SATOR. TENET reads the silme from either end, i.e., it is
palindromic.
Is it possible to consffuct magic squares containing prime
numbers only? Yes! The famous British puzzle expert Henry 2. Semi-Magic Squares
ErnestDudeney (1857-1930) found the following anay in 1900
where 1 is regarded as a prime number (Devendran, L990, p.
25): The matrix A denoting a semi-magic squile may be character-
rzedby the properties
7 6r 43 A e @ )- , r ( n ) - A , e ( ' ) ,
t(A' ) - t(A)
FIG. 15: Magicsquarewith consecutiveprimes.
Booth & Booth (1955) showedthat the inverse of the matrix for
a nonsingular semi-magic squile A is semi-magic, again with /k+l+m*n -l-m, -k-t\
the magic constant I I t@). In Schmidt & Trenkler (2001) this tl
^r-l -k-t t k (2)
result is extendedto singular semi-magic squares.Indeed, they |
\ -n1 -n m n /
showed that it A is semi-magic, then 4+ is semi-magic with
magic constant s(A)+. Here ,4+ denotesthe M-inverse of A, We note that the magic constant s = t(A) since 16t(s)
and r(-4)+ - 0 if s(/) - 0 and r(A)+ = Ilt@), otherwise. 0 = I{' e(3). From this we get ((t l3)r @)e1s)r)I/ =
Consider now e(3)r)
lr ( G l3)€(3)
ry+.
Mn - - ar@)"(")'
{ AIA ,* € R} If p - ln - km + 0, then the MP-inverseof ,A/is
and ry+=frx \
- 0}. / -k+l-m*n -k+l+2m,-2n 2k-21 -m*n \
I{n- tAlA € Rn and r(A) I
| 2k+l- m-2n 2k+t+2m*4n -4k-2t-m-2nl
It is clear that Mn and lf" are both ideals of Rn As stated in I
\-f -iltzm*n
^ .
- kt - 2 1n t - 4 1m - 2 n^ At
2 k +, 4. 1
r ! A
+2m*n /
I
Weiner (1955), the algebra .R' is the direct sum of thesetwo ide-
als. Furthermore,the matrices A;i
For p - 0 the MP-inverse of /f is also calculated in GroB et al.
2, . . . ) n, where the e,; are the members of the canonical basis
(1999), along with rank, eigenvaluesand eigenvectorsof A (for
of IR', generateI{n Since the A;i are linearly independent,the
arbitrary p).
algebr" lVnhasdimension (n - L)2.
The maffix B for Beverly's 8 x 8 semi-magic squareB (Fig.
11) has rank 5 and MP-inverse
-42457041. -8977065
22454t27 42704331 -3t5t7297 5342929 11114383 28384337
22454727 42663095 -42a5704t -8935829 L2954143 29a94209 -33357057 3833057
4085361 16299017 29641905 -36701931 8743313 -22A38705 249a3953 2435797
29641905 -36660695 4085361 76257781 23t44L93 925919 10583073 -21328833
3+=
6928663040 2576815 43322a71 -22979729 -9595605 -1.64a20I7 5736257 -3920897 27991009
-22979729 -9636841 2576815 43364707 -20aLL37 29500881 -LA32L777 42263a5
49519277 -36000919 -15791951 15598005 -6291 967 -23232033 40019233 2829119
-15791951 15639241 495t92L7 -36042155 3aL79473 t319247 -4452207 -21722767
We observethat for A € Rn and a +0 we have All rows and columns of g+ add to I 1260.Its four 4 x 4 corner
squaresare semi-magicagain with magic constant I1520.
( A + o " @ ) r { " ) ' ) *= A * + 1 e @ ) e ( n, Y The knight-moves sfrrrcture has disappeared, with two
where 6 = exceptions.The largest entry 495I92LT 16928663040occurs
twice, and the distance between the positions of these two en-
tries is a knight's move. The sirmeis true for true for the smallest
1
aF if s(-4) - 0
entry -4285704116928663040which again occurs twice and
the distancebetween the positions of these two entries is again
_ M1 if r(A) + 0 andr(,4) * an - 0
a knight's move.
-@ 1 The eigenvaluesof B are 0, 0, 0 ,260 plus the four roots of
if r(,4) + 0 andr(,4) * an + 0
the polynomial
As Diirer's magic squiue D is also semi-magic, then by this re-
14 - 22n3- 3784r' + 1 rSTgzr+ 1Tzog2.
sult we obtain
These four roots are all real, two of them negative.
( D + 6 e @ r) ( n ) ' ) * - D * * ^ / e @r)( n ) ' Finally we should mention Benjamin Franklin (1706-1790),
for given d € IR and for some j. Thus the rich pattern of D+ is the U.S. statesman,scientist and writer. Well-known for the in-
inheritedby (D + 6e0')rtn)')*. vention of the lightning conductor, he is less renowned for his
semi-magicsquares,but seeMarder (1940) and Murphy & Walz
GroB et al. (1999) studied semi-magic squiles of type 3 x 3
(2001). The matrix F for Franklin's 16 x 16 square(Andrews,
and showed that the associatedmatrices have the form
1960, p. 97) has rank 3 and magic constant2056; the row and
column sumsof the MP-inverse F* ire, of course, 1/2056. A
A* (tlile(s)e(3)'-ltr, (1) comprehensivereport on Franklin's semi-magic squarescan be
where found in the recent article by Pasles(2001).
pageI fMAGE 27: October2OO1
3. Magic Squares for the 4 x 4 magic squaregiven by GroB (2000) has rank 3. The
MP-inverse is
The matrix M for a magic squ:ue of type n x n with magic -3 -3
constant s is characterize.dby the following four equations I
I -c) o
(iii)
P7r,(n)
n(M)
=
=s
se(')
I -3 -3
which obviously is not magic.
(iv)t(RM) =s, In general, the matrix for 4 x 4 magrc squareshas the form
where tr(.) denotestrace and R - (roi) is the n x n rcflection a*b f+t, e*g-h c*d
m a t r i x w i t h e n t r i ersi , n - i + 1 - I a n d r 6 i = 0 i f i + n - i +
1; i,j c+e d+g a* f b
be written as
M - ( t / 3 ) e ( 3 )e ( 3 ) '+ l / ,
d+f b+e a+ g
\s -c-f -g
We now consider the singular case. c
-1 1 1 -1
wherec1 = -4rf 4cg af7 2f2 2g',dt
29'te!=2c2-2f' a n d b 1- c 1
-1 -1
f g + 92 + "(f + g). Hence S+ is also super-magic.
-1 -1 It might be interesting to identify all casesof magic squares
of order 4 such that their MP-inverse is magic again. This is
-3 postponed for future research.
fMAGE 27: October2001 page 9
p*
4. Most Perfect Pandiagonal Magic Squares -L
+ b+ e c*d*e a*c b+d W. S. Andrews (1960). Magic Squaresand Cubes: With Clnpters by
/" OtherWriters. SecondEdition, Revisedand Enlarged. Dover, New
b a*b+d+e c*e
p_ f "+c*d York.
u* d+e a*c*e c*d a*b
I W. W. R. Ball & H. S. M. Coxeter (1987). Mathematical Recreations
\C' a*b+d b+e a*c*d*e and Essays. 13th Edition. Dover, New York, and University of
Toronto Press.
with magic constant s A. D. Booth & K. H. V. Booth (1955). On magic squares.The Mathe-
eigenvaluesof P are given by matical Gazette,39, 132-133.
S. L. Campbell & C. D. Meyer, Jr. (1979). Generalized Inverses of
)1-0,,\z-sl Linear Transformations. Piunan, London. [Unabridged corrected
reprint edition, Dover, New York, 1991.1
. \ 3= - 2 f f i , \ + = -2ffi- -A3. N. Chater & W. J. Chater (1949). On the determinants of pan-magic
squaresof even order. The MathematicalGazette,33,94-98.
Thus P here is always singular; when ae * cd T. Devendran (1990). Das Beste aus dem mathematischen Kabinea.
eigenvaluesare always purely imaginary. DeutscheVerlags-Anstalt, Stuttgan.
Chater & Chater (1949) proved that MPPM-squffes of even A. Domenicano & I. Hargittai (2000). Alphabetic magic square in a
order are singular in general, in accordancewith our result. As medieval church. The M athematical Intelli g encer, 22 (I), 52-53 .
seen above in our Section 3, the MP-inverse of a magic square J. L. Fults (1974). Magic Squares. Open Court, Chicago.
need not be magic. MPPM-squares of order 4 behave differ- M. Gardner (1996). The magic of 3 x 3. Quantum: The Student Maga-
ently! Indeed, the MP-inverse of an MPPM-squareis an MPPM- zine of Math and Science, January/February1996,24-26.
squareagain. J. GroB (2000). Personal communication.
L e t a * b + c * d + e + 0 , ( b - c ) 2 + 1 2 + } a n d a 2 + d 2+ 0 , J. GroB, G. Trenkler & S.-O. Troschke(1999).On properties of 3 x 3
semi-magic squares. International Journal of Mathematical Edu-
then
cation in ScienceandTechnology,30,861-865.
page 10 IMAGE 27: October2001
H. D. Heinz and J. R. Hendricks (2000). Magic Square Lexicon: Illus- PaulDanielsand Spacesof Magic Squares
trated. H. D. Heinz, Surrey, B. C.
J. R. Hendricks (1999). Inlaid Magic Squares and Cubes. J. R. Hen- I did not know that the MP-inverse of a magic squire may not be
dricks, Victoria, B. C. a magic square.In my undergraduatelectures on magic squaresI
A. Hirayama & G. Abe (1983). Researches in Magic Squarer (in sometimes mention Paul Daniels's magic square-Paul Daniels
Japanese).Osaka Kyoikutosho Co. Ltd., Osaka. (b. 1938) is a well-known British magician and TV personal-
G. James & K. James (2000). Reversible magic square. The American ity who I often saw on CanadianTV in the 1970s.I copied the
Mathematical Monthly, lO7, 821.
matrix for Daniels's magic squareonce from TV. It is:
N. A. Khan (1957). Characteristic roots of semi-magic squarematices.
The American M athematical M onthly, 64, 26I-263 . 24 11 22 17
E. Lanners (1977). Illusions. Holt, Rinehart & Winston, New York.
C. C. Marder (1940). The Magic Squares of Benjamin Franklin. The 2L 18 23 12
Brick Row Book Shop, New York.
F. Mayoral (1996). Semi-magic squaresand their orthogonal comple- 15 20 13 26
ments.The Mathematical Gazette,80, 308-313.
H. Minc (1988). Nonnegative Matrices. Wiley, New York. 14 25 16 19
J. S. Montague & R. J. Plemmons (1973). Doubly stochastic magic
The MP-inverse of this magic squareIS a magic square!
equations./sra el Joumal of Mathematics, 15, 216-229.
One of my favourite exercisesfor linear algebra undergrad-
J. Moran (1982). The Wonders of Magic Squares.With a Foreword by
Martin Gardner. Vintage Books, New York. uatesis to replace a given set of, say, three 3 x 3 magic squares'
by an orthogonal set of magic squaresusing Gram-Schmidt and
F. Murphy & R. Walz (2OOl). Ben Franklin and the Magic Squares.
Random House, New York. the usual inner product fr(At B). So, magic squareshave their
place in linear algebra coursesif only for illustratiolls.
K. Ollerenshaw& D. S. Br6e (1998). Most-Perfect Pandiagonal Magic
Squares: Their Construction and Enumeration. The Institute of I wonder what is known about spacesof magic squaresand
Mathematics and its Applications, Southend-on-Sea. their dimensions. For exirmple, the dimension of 4 x 4 magic
P. C. Pasles(2001). The lost squaresof Dr. Franklin: Ben Franklin's squaresis 8, but I wonder if the dimension is known for 5 x 5
missing squaresand the secret of the magic chcle. The American magic squares,etc. I think that this might still be an open ques-
MathematicalMonthly, 108, 489-51 1. tion, see T. J. Fletcher (1972), Linear Algebra: Thtough its Ap-
C. R. Pickover (2002). The Zen of Magic Squares, Circles, and Stars: plications,Van Nostrand Reinhold, London.
An Exhibition of Surprising Structures AcrossDimensions. Prince-
ton University Press. Sanjo Zl-oge c: zlobec@math.mcgi Il.ca
K. Schmidt & G. Trenkler (2001). The Moore-Penrose inverse of a Dept. of Mathematics and Statistics, McGilI University
semi-magic square is semi-magic. International Journal of Mathe' Montrdal (Qudbec), CanadaH3A 2K6
matical Education in Scienceand Technology,3z, rn press.
G. E. Sharpe & G. P. H. Styan (1965). Circuit duality and the general
network inverse.IEEE Transactionson Circuit Theory,I2,22-27 .
p. H. Styan & G. E. Subak-Sharpe (1997). Inequalities and equali-
Magic Squareson the Web and the
G.
ties associatedwith the Campbetl-Youla generahzedinverse of the Determinantsot4 x 4MagicSquares
indefinite admittance of resistive networks. Linear Algebra and its
Applications, 249, 349-37 O- possibly my favourite Web site for bibliographic research is
G. Trenkler (1 gg4). Singular magic squares. International Journal of
AMICUS Web from the National Library of Canada:
Mathematical Education in Science and Technology,25, 595-597 .
G. Trenkler & H. Knautz (1991). Diirer, Moore-Penrose, Draztn .... / aaweb/ami Iogi ne.htm?
http:lI arnic us.nIc-bnc .ca:BO
Prarcisder Mathematik, 33, 183-1 84.
W. p. Wardlaw (1992). Nonsingular magic matrices: Problem 83444. For'"Title Keyword" = Magrc Squares,I found 39 records.
The American Mathematical Monthly,9&,437 -
For Andrews (1960) I discovered that the "other writers" are
D. Watson (1966). Matrices with equal row-sums and column-sums.
Paul Cants, L. S. Frierson and C. A. Browne, Jr., and that the
The Mathematical Gazette, 50, 391-392.
book includes an introduction by Paul Carus. The 1960 Dover
L. M. Weiner (1955). The algebra of semi-magic squares. The Ameri-
reprint edition is an "IJnabridged and unaltered publication of
can Mathematical M onthly, 62, 237 -239 - 'l'9L7".
the secondedition published by Open Court, Chicago, in
The first edition was published in 1908.
Dietrich TReNf IER : dtre n k ler@nts6.oec.u n i -osn abrueck.de 'oThefirst of a seriesof four papersdescrib-
(Jniversity of Osnabriick, D-49060 Osnabriick, Germany
Marder (1940) is
Dept. of Economics,
ing the technique of Leonhard Euler applied to the Lahireian
Gdtz TneNf LER : tren k ler@statistik. u n i-dortm u nd.de
method of forming magic squaresof all sizes, under the general
Dept. of Statistics, (Jniversity of Dortntund, D-44227 Dortrnund, Germany
'The intrinsic harmony of number' ". Also included is "The
title
IMAGE27: October2001 page 11
simple method of raising any 8 x 8 squareto the 16 x 16 size I also found 9 books on "magic squares" in "MathSciNet:
with the aid of an auxiliary squarepreserving all the qualities of Mathematical Raiews on the Web" @American Mathematical
the original" and "Three entirely new methods of producing the Society
bent diagonals of Benjamin Franklin". http://www.ams.o rglmathsc i neVsearch
Harvey Heinz of Surrey,B. C., Canada,maintains a wonder-
including
fuI2s-page Web site on magic squaresat
J. Sesiano,(1996).Un TraftAMAditval sur les CarrdsMagiques(in
http://www.geocities.com/CapeCanaveral/Launchpad/ 4057/ French).Annotatededition and Eanslationof an anonymousAra-
@ Harvey D. Heinz. This Web site links directly to an exten- bic text describingvariousmodesof constnrction.PressesPoly-
sive annotatedbibliography of more than 160 items on magic techniques Lausanne,
et UniversitairesRomandes, ISBN 2-88074-
squares(and magic stars), Franklin squares,and order-S pandi- 310-9,MR 98e:01007.
agonal magic squares.Here I found that in Andrews (1960) "Ttle The MathPagesWeb site
essayswhich comprise this volume appearedfirst in an Amer-
com/
http://mathpages.
ican journal called The Monisl between 1905 and 1916". Ball
'oTheDeter-
& Coxeter (1987) was "Originally published in 1892, H. S. M. has severalinteresting articles on magic squares.In
Coxeter brought it up to date with the 1938 publication of the minants of 4 x 4 Magic Squares"it is observedthat "Up to sign,
11th edition, the lzth edition in 1974 and edition 13 in 1987. there are only 12 distinct determinantsfor 4 x 4 magic squares
Chapter 7 is on magic squares". (using the elements 0 to 15)". The numbers of squareshaving
Furthermore Heinz & Hendricks (2000) has 228 pp., I7l each of thesedeterminants are listed there.
captioned illustrations and tables, 239 terms defined, 2 appen- Moreover, the author then wonders "if all the magic squues
dices of bibliographies, ISBN 0-9687985-0-0.It is available (of a given order) with the same determinant constitute a group
postagepaid (in North America) for C$32.00 or US$25.00 per under some set of 'simple' transformations.It's certainly true
copy from H. D. Hernz, 15450 92A Avenue, Surrey, B. C., for the setsof 32, since these can be shownto constitute a group
CanadaV3R 98 1; harveyheinz@home.com under rotation, reflection, etc. But what about the larger sets?
From the Heinz Web site I was led to the JapaneseWeb site Interestingly, the unique (up to sign and 32-group transfor-
- mations) squarewith determinant 6720is the only squarewhose
http://www.pse.che.toho ku.ac.jpI msuz uki/MagicSquare.htmI
determinantis divisible by any prime other than 2,3, or 5.
run by Matsumi Suzuki. Here I found a photograph taken at It's also interesting to experimentwith various matrix opera-
Gaudi's cathedral in Barcelona of a magic squile with magic tions on two magic squiuesA , B. For example,the squaresgiven
constant 33 "which is popularly acknowledged as the age at by ABA-L have the samerow and column sums as,4 and B.
which JesusChrist died". The cornmutatorAB - B A also gives some interesting results,
The most helpful Web site @ Glenn Fleishman as doesreplacing each element of a squareA with its cofactor."
http://isbn.nu/cgi-bin/design-books
allows one to searchfor books and compare prices. If the ISBN GeorgeP. H. SrvaN: gphs@videotron.ca
Dept. of Mathematics and Statistics, McGill University
(International Standard Book Number) is known then one can
Montrial (Quibec), Canada H3A 2K6
go directly to the price comparisons.For Pickover (2001), visit-
ing isbn.nu/0-691-0704t-s showedpricesranging from US$17.00-
$23.88 and availability from 1-5 days to 6-15 days.; visit The Namingof Parts
-o7o4i-sfor the table of contents. The ISBN may
isbn.nu/toc/0-6e1
be entered with or without hyphens. For "magic squares"in the
title I found 13 books including Some recent research suggeststhat the following terminology
Shen,C. T. (1989).GeneralSolutionsfor EvenOrderMagic Squares. may be useful:
Victory Press,Monterey,CA. If the real maffix B - -A is idempotent,i.e., 82 - B, then
A is "skew idempotent": A2 = -,4.
The "kntralblatt MATFI" Web site @ European Mathemati-
cal Society, FIZ Karlsruhe & Springer-Verlag If the imaginary matrix C
then A is "skew tripotent": AB - -4.
htt p://www.em is.deIZMATH/
If the imaginary matrix C
has a "Demo mode with a limit of ma:r. 3 allowed answers for then A is "skew involutory": A2 - - I .
unregisteredusers". I found 9 books on "magic squares"includ- Moreover "involutory" not "involutary".
ing Readersmay have other suggestions.
Yu. V. Chebrakov(1995).Magic Squares:Number Theory,AIge-
bra, CombinatorialAnalysis(in Russian).Populyarnye Lektsii po R. Wiltam FaREBRoTHER
: Msrbsrf@fs1
.ec.man.ac.uk
Matematike.Sankt-Peterburgskij Gosudarstvennyj Tekhnicheskij 11 Castle Road, Bayston Hill, Shrewsbury, England SY3ONF
Universiret,ISBN 5-7422-0015 -3,Zbl 0879.050
I 3.
IMAGE 27: October2OO1
?fiiilYi::,ffi!,!":Xf:ty,LyE"rra(EI-A)
vo,1-4"'.ffi
The pricesbelow includesurface-mailshipping
Total Price
Name
Shipping Address
pleasesenda copy of this form with eithera checkor moneyorder (US currencyonly) to:
ProfessorJamesR. Weaver
of Mathematicsand Statistics,UniversityofWestFlorida
Depa.rtment
110fl0UniversityParh,vay,Pensacola,FL 32514' USA
H. Fassbender (2000). Symplectic Methods for the Symplectic Eigen- E. Kaszkurewicz & A. Bhaya (2000). Matrix Diagonal Stability in Sys-
problem. Kluwer/Plenum, New York: xvi + 269 pp.nISBN 0-306- tems and Computation. Birktrfluser, Boston: xiv + 267 PP., ISBN
46478-0. 0-8176-4088-6.
G. Fischer (2000). Lineare Algebra: Eine Einfiihrung fr Studien- N. M. Katz & P. Sarnak(1999). Random Motrices, FrobeniusEigenval-
anfiinger (in German). I2th revised edition. Vieweg Studium: ues, and Monodromy. American Mathematical Society Colloquium
Grundkurs Mathematik. Vieweg, Wiesbaden: x + 384 pp., ISBN Publications 45. American Mathematical Society, Providence, R[:
3-528-87217-glpbk. xi + 4I9 pp., ISBN 0-8218-1017-0.
J. N. Franklin (2000). Matrix Theory. Reprint edition. Dover, New K. Kleinfeld & M. Kleinfeld (2001). Understanding Linear Algebra
York: xii + 292pp., ISBN O-48641179-6lpbk. using Merlas. Prentice Hall, Upper Saddle River, NJ: vii + 210
S. A. Fulling, with M. N. Sinyakov & S. V. Tischchenko (2000). Lin- pp., ISBN 0- 13-060945-5.
earity and the Mathematics of Several Variables. World Scientific, B. Kolman & D. R. Hill (2000). Elementary Linear Algebra.Tth edi-
Singapore:x + 445 pp., ISBN 981-02-4196-8lhbk. tion. Prentice Hall, Upper Saddle River, NJ: xv + 568 PP., ISBN
P. Gabriel (200l). Matrices, G6omdtrie, Algibre Lindaire (in French). 0-13-085199-X.
Translated by G. Arnaudibs, Jean-Marie Arnaudibs & J.-D. Eiden B. Kolman & D. R. Hill (2001). Introductory Linear Algebrawith Ap-
from the German. Nouvelle Bibliothbque Mathdmatique 6. Cassini, plications. 7th edition. PrrenticeHall, Upper Saddle River, NJ: xvi
Paris: 7lO pp., ISBN 2-84225-018-4/hbk. + 655 pp., ISBN 0-13-018265-6.
M. Golubitsky & M. Dellnitz (1999). Linear Algebra and Dffirential A. I. Kosffikin (2000). Introduction to Algebra, Part 2: Linear Alge-
Equations using MnrlAB. Brooks/Cole, Pacific Grove, CA: xiv + bra Textbook (in Russian). Fiziko-Matematicheskaya Literatura.,
704 pp., ISBN 0-534-35425-4. Moscow : 367 pp., ISBN 5-922L-0018-1.
G. Gramlich & W. Werner (2000). Numerische Mathematik mit Mat- R. Larson & B. H. Edwards, with the assistanceof D. C. Falvo (2000).
Iab: Eine Einfiihrung for Naturwissensclw{tler und Ingenieure (rn Elementary Linear Algebra.4th edition. Houghton Miffiin, Boston:
German). Includes CD-ROM. dpunkt Verlag, Heidelberg: xv + 458 xiv + 435 pp., ISBN 0-395-96717-1.
pp., ISBN 3-932588-55-)VPbk. G. Latouche & V. Ramaswami (1999). Intrcduction to Matrix Analytic
M. D. Greenberg (2O01). Differential Equations & Linear Algebra. Methods in Stochastic Modeling. ASA-SIAM Series on Statistics
Prentice Hall, Upper Saddle River, NJ: 611 PP., ISBN 0-13- and Applied Probability, SIAM, Philadelphia: xiv + 334 pp., ISBN
011118-X. 0-89871425-7.
A. K. Gupta & D. K. Nagar (2000). Matrix Variate Distributions. Chap- D. C. Lay (2000). Linear Algebra and its Applications. 2nd updated
man & HaIUCRC Monographs and Surveys in Rre and Applied edition. Addison-Wesley,Readitg, MA: xxxiv + 590 PP.,ISBN 0-
Mathematics lO4. CRC Press,Boca Raton, FL, 367 pp., ISBN 1- 20r-64845-8.
58488-046-5. S. J. Leon (2W2). Linear Algebra with Applications.6th edition. Pren-
D. J. Hartfiel (2001). Matrix Theory and Applications with MATIA,B. tice Hall, Upper Saddle River, NJ: xv + 5M pp., ISBN 0-13-
CRC Press,Boca Raton,FL: ix + 371 pp., ISBN 1-58488-108-9. 033781-1.
D. A. Harville (2001). Matrix Algebra: Exercises and Solutions. S. Lipschutz (2001). Sclnum's Outline of Theory and Problems of Lin'
Springer,New York: 295 pp., ISBN 0-387-95318-3/pbk. ear Algebra. 3rd edition. McGraw-Hill, New York: viii + 424 pp.'
ISBN 0-07 - 1362W-2lPbk.
R. D. H. Heijmans, D. S. G. Pollock & A. Satorra,eds. (2000). Inno-
vations in Multivariate Statistical Analysis.'A Festschrift for Heinz B. Liu & H.-J. Lai (2O00).Matrices in Cambinatorics and GraphThe-
Neduecker. Advanced Snrdies in Theoretical & Applied Economet- ory. Network Theory and Applications 3. Kluwer, Dordrecht, vii +
rics 36. Kluwer, Dordrecht xiii +295 pp., ISBN 0-7923-8636-1. 310 pp., ISBN O-7923-6469-4lhbk.
E. A. Herman (2001). Linear Algebra: Modules for Interactive J. R. Magnus & H. Neudecker (1999). Matrix Calculus with Applica-
Learning using MAPLE 6. The Linear Algebra Modules Project tions in Statisticsand Econometrics. Revised reprint of the 1988
(LAMP). Addison-Wesley,Boston: xviii + 496 PP., ISBN 0-201- original. Wiley Series in Probabitity and Statistics. V/iley, Chich-
44135-7. ester: xviii + 395 pp., ISBN O-471-98632-llhbk, 0-471-98633-
)Upbk.
D. J. Higham & N. J. Higham (2000). Merr. As Guide. SIAM, Philadel-
phia: xxii +283 pp., ISBN 0-89871-469-9/hbk. I. A. Mal'tsev (2001). Linear Algebra (rn Russian).lzdatel'stvo Insti-
tuta Matematiki Im. S. L. SobolevaSO RAN, Novosibirsk: 315 PP.o
R. R. Hill & B. Kolman (2001). Modern Matrix Algebra.. Prentice
ISBN 5-86 134-092-7/Pbk.
Hall, Upper SaddleRiver, NJ: xv + 51 1 pp., ISBN O-L3-948852-9.
Mathematical Sciences ResearchInstinrb, Berkeley (1999). Random
V. A. Il'in & Eh. G. Poznyak (2001). Linear Algebra (rn Russian).
Matrix Models and their Applications. Lectures from the Work-
5th edition. Kurs Vysshej Matematiki i Matematicheskoj Fiziki 4.
shop held in Berkeley, CA, February 22-26,1999. Selections from
Fiziko-Matematichesk ay aLiteratura, Moscow : 3 17 pp.
MSRI's Video Archive 5. Mathematical SciencesResearchInsti-
K. J[nich (2000). Lineare Algebra (in German). 8th edition. Springer-
tute, Berkeley, CA: CD-ROM (Windows, Macintosh & UNDQ.
Lehrbuch. Springer, Berlin, xii + 277 pp., ISBN 3-540-66888-
A. Meister (1999). Numerik linearer Gleichungssysteme: Eine
8/pbk.
Einfiihrung in modeme Verfahren (in German). Friedr. Vieweg &
M. C. Jain (2001). Vector Spacesand Matrices in Physfcs. Chapman &
Sohn, Braunschweig:x + 222 PP.,ISBN 3-528-03135-2.
HaIVCRC, Boca Raton, FL: c. 160 pp., ISBN 0-8493-0978-6.
yu. A. Melnikov (1999). InfluenceFunctions and Matrices. Mechanical
L. W. Johnson, R. D. Riess & & J. T. Arnold (2W2). Introduction
Engineering LL9. Dekker, New York: x + 469 PP., ISBN 0-8247-
to Linear Algebra. 5th edition. Addison-Wesley, Boston: 600 pp.,
t94l-7.
ISBN 0-201-65859-3.
fMAGE 27: October 2OO1 page 15
K. Murota (2000). Matrices and Matroids for SystemsAnalysis. Algo- S. R. Searle&L. S. Willett (2001). Matrix Algebrafor Applied Eco-
rithms and Combinatorics 20. Springer, Berlin: xii + 483 PP.,ISBN nomics, Wiley Series in Probability and Statistics. Wiley, New
3-540-66024-0thbk. York: xx + 402 pp., ISBN 0-471-32207-5.
W. K. Nicholson (200l). Elementary Linear Algebra. McGraw-Hill Ry- T. Shifrin & M. Adams (2001). Linear Algebra. W. H. Freeman, New
erson, Toronto: xx + 586 pp., ISBN 0-07-089229-6. York, c. 464pp., ISBN O-7167-4337-X.
V. Olshevsky, ed. (2001). Structured Matrices in Mathematics, Com- A. Shuchat& F. Shultz (2000). The Joy of Mathematica: Instant Math-
puter science, and Engineering, I: Proceedings of the AMS-IMS- ematicafor Calculus, Dffirential Equations, and Linear Algebra.
SIAM Joint Surnmer Research Conference on Structured Matrices 2ndedition. Harcourt/Academic Press:San Diego: xxiii + 576 pp.,
in Operator Theot!, Numerical Analysis, Control, Signal and Im- ISBN 0-12-640730-4.
age Processing held at the University of Colorado, Boulden CO, L. E. Spence,A. J. Insel & S. H. Friedberg (2000). Elementary Linear
June 27-JuIy I, 1999. Contemporary Mathematics 280. Ameri- Algebra: A Matrix Approach. Prentice Hall, Upper Saddle River,
can Mathematical Society, Providence, R[: xiv + 327 pp., ISBN NJ: xiv + 477 pp., ISBN O-13-716722-9lhbk.
0-8218-192r-6. G. W. Stewart (2001). Matrix Algorithms, Vol. II: Eigensystems. SIAM,
O. T. O'Meara (2000). Introduction to Quadratic Forms.Reprint of the Philadelphia:xix + 469 pp., ISBN 0-8987I-503-2.
1973 edition. Classics in Mathematics. Springer, Berlin: xiv + 344 J. Stoer & R. Bulirsch (2000). Numerische Mathematik 2, Eine
pp., ISBN 3-540-66564-1/pbk. Einfiihrung: unter Beriicksichtigung von Vorlesungen von F. L.
V. Y. Pan (2001). Structured Matrices and Polynomials: Unified Super- Bauer (in German). Springer, Berlin: x + 376 pp., ISBN 3-540-
fast Algoritltm.s. Birkhduser Boston & Springer-Verlag, New York: 67644-9.
xxvi + 278 pp., ISBN 0-8176-4240-4. H. Stoppel & B. Griese (2001). tlbungsbuch zur linearen Algebra:
B. Pareigis (2000). Lineare Algebra fiir Informatiker I: Grundlngen, Aufgaben und lisungen (rn German). 3rd revised edition. Vieweg
diskrete Mathematik; II: Lineare Algebra (in German). Springer- Studium: Grundkurs Mathematik. Vieweg, Wiesbaden, ix + 286
Lehrbuch. Springer, Berlin: viii + 274 pp., ISBN 3-540-67533- pp., ISBN 3-528-27288-0/pbk.
7lpbk. F. Szabo (2000). Linear Algebra: An Introduction using Mathematica.
M. Paz & W. Leigh (2001). Integrated Matrix Analysis of Structures: Academic Press, San Diego: ISBN 0-12-680135-5/hbk.lStudent
Theory and Computation. Kluwer, Boston: xii, 320 pp., ISBN 0- SolutionsManual, Academic Press,SanDiego: ix + 267 pp., ISBN
7923-7308-r. 0-12-680137-rlpbk.l
R. C. Penney (1999?). Linear Algebra: Ideas and Applications. Wiley, F. Szabo (2001). Linear Algebra: An Introduction using Maple. Har-
Toronto (for the University of Ottawa): c. 400 pp., ISBN 047l- court/AcademicPress,San Diego: ISBN 0-12-680140-1.
64341,6. F. Szidarovszky & S. Moln6r (2001). Introduction to Matrix Theory:
G. L. Peterson & J. S. Sochacki (2002). Linear Algebra and Dffiren- With Applications to Business and Economics. World Scientific,
tial Equations. Addison-Wesley, Boston: x + 463 pp., ISBN 0-201- River Edge, NJ: 420 pp., ISBN 981-02-4504-1/hbk,981-02-4513-
66212-4. O/pbk.
A. Popescu (1999). Linear Algebras and Applications: Fundamentals L. N. Trefethen(2000). SpectralMethods in MarLAB. SIAM, Philadel-
(in Romanian). Editura Universitifii Bucuregti, Bucharest: 384pp., phia: xviii + 165 pp., ISBN 0-89871465-6.
rsBN 973-575-340-5. D. A. Turkington (2001). Matrix Calculus & Zero-One Matrices:
A. Quarteroni (20@). Mdthodes Numdriques pour le Calcul Scien- Statistical and Econometric Applications. Cambridge University
tifique: Programmes en MntLAB (in French). Springer, Paris: xv Press,264 pp., ISBN A-52L-80788-3.
+ 444 pp., ISBN 2-287-59701-8. F. Uhlig (2002). Transform Linear Algebra. Prentice Hall, Upper Sad-
H. Radjavi & P. Rosenthal (2000). Simultaneous Triangularization. dle River, NJ: 528 pp., ISBN 0-13-041535-9/hbk.
Universitext. Springer, New York, xii + 318 pp., ISBN 0-387- Universidade de Coimbra, Departamento de Matemdaca (1999). Ma-
98466-6/pbk. trices and Group Representations: Papersfrom the Workshop held
A. R. Rao & P. Bhimasankaram (2000). Linear Algebra.2nd edition. in honour of Graciano N. de Oliveira on the Occasion of his 60th
Texts and Readings in Mathematics 19. Hindustan Book Agency, birthday at the University of Coimbra, Coimbra, May 6-8, 1998.
New Delhi: xiii + 4I4 pp., ISBN 81-85931-26-7. Textos de Matemdtica S6rie B 19. Coimbra: vi + 128 pp., ISBN
Research and Education Association: M. Fogiel, chief editor (1999). 972-8564-19-8.
The Linear Algebra Problem Solver: A Complete Solution Guide R. Vein & P.Dale (1999). Determinants andtheir Applications in Math-
to Any Textbook The Association, Piscataway,NJ: xii + 1011 pp., emntical Physics. Applied Mathematical Sciences I34. Springer,
ISBN 0-8789I -5 I 8-alpbk. New York: xiv + 376 pp., ISBN 0-387-98558-1.
V. N. Sachkov & V. E. Tarakanov (2000). Combinatorics of Nonneg- G. Weinreich(1998). GeometricalVectors.ChicagoLectures in Physics
ative Matrices (in Russian). Progress in Theoretical and Applied Series.Univ. of Chicago Press:x + 115 pp., ISBN 0-226-89048-1,
Discrete Mathematics 2. Nauchnoe lzdatel'stvo TVP, Moscow: viii 0-226-89047-3.
+ 448 pp., ISBN 5-85484-011-1. Y. X. Yang (2001). Theory and Applications of Higher-Dimensional
L. A. Sadun (2001). Applied Linear Algebra: The Decoupling Princi- Hadamard Matrices. Combinatorics and Computer Science 1. Sci-
ple. Prentice Hall, Upper Saddle River, NJ: xvii + 349 pp., ISBN ence Press,Beijing & Kluwer, Dordrecht: 332 pp., ISBN 0-7923-
0-13-085645-2. 7061-9.
J. T. Schwartz (200t). Introduction to Matrices and Vectors.Reprint
edition. Dover, New York: 176 pp., ISBN O-486-42000-0/pbk.
Have YouSeenthe Intest
slHoF,PeK$W
Templates for the Solution Matrix Algorithms
of Algebraic Eigenvalue Problems Yolume I I: Eigensystems
A Practical Guide G. V1y'.
Stewart
Thaoiun Bai,JamesDemmel, This book is the secondvolume in
Jack DongaFf?,Axel Ruhe, a projectedfive-volume surveyof
and Henk van der Vorst
numericallinear algebraand matrix
Sofnvare, Environments, andTools I I
algorithms.This volume treatsthe
Large-scaleproblems of engineering numericalsolution of denseand large-scaleeigenvalue
and scientific computing often problemswith an emphasison algorithms and the
require solutionsof eigenvalueand theoreticalbackgroundrequiredto understandthem.
relatedproblems.This book givesa Stressingdepth over breadth,ProfessorStewart treatsthe
unified overview of theory, algorithms,and practical derivation and implementationof the more important
softwarefor eigenvalueproblems.It organizesthis large algorithmsin detail. The notesand referencessections
body of material to make it accessiblefor the first time to containpointers to other methodsalong with historical
the many nonexpertuserswho needto choosethe best comments.
state-of-the-artalgorithms and softwarefor their The book is divided into two Parts:dense
problems.Using an informal decisiontree,just enough eigenproblemsand large eigenproblems.The first part
theory is introducedto identify the relevantmathematical gives a futl treatmentof the widely used QR algorithm,
structurethat determinesthe best algorithm for each which is then applied to the solution of generalized
problem. eigenproblemsand the computationof the singular value
The algorithms and softwareat the "leaves" of the decomposition.The secondpart treatsKrylov sequence
decisiontree rangefrom the classicalQR algorithm, methodssuch as the Lanczos.andArnoldi algorithms and
which is most suitablefor small densematrices,to presentsa new treatmentof the Jacobi-Davidson method.
iterativealgorithmsfor very largegeneralizedeigenvalue The volumes in this surveyare not intended to be
problems.Algorithms ile presentedin a unified style as encyclopedic.By treatingcarefullyselectedtopics in
templates,with different levels of detail suitablefor depth,each volume gives the readerthe theoretical and
readersrangingfrom beginningstudentsto experts.The practicalbackgroundto readthe researchliterature and
authors' comprehensivetreatrnentincludes a treasureof implementor modify new algorithms.The algorithms
further bibliographic information. treatedare illustrated by pseudocodethat has been tested
Royaltiesfrom the sale of this book are contributedto in MnTLAB implementations.
the SIAM studenttravel fund. 2001 . xix + 469 Pages' Softcover
2000 . xxx + 410 pages' Softcover ISBN0-8987I -503-2 ' List Price$50.00
I S B N0 - 8 9 8 7l - 4 7 l - 0 ' L i s t P r i c e$ 6 2 . 0 0 S f A MM e m b e rP r i c e$ 3 5 . 0 0' O r d e r C o d e O T 7 7
S I A MM e m b e rP r i c e$ 4 3 . 4 0' O r d e r C o d e S E I I
To ORDER
. Useyour credit card (AMEX, MasteCard,
Order online (nowwith shoppingcart!): www.siam.org/catalog
andVISA): Call SIAM Customer Service at worldwide
215-382-9800 or toll freeat 800-447-SIAM in USA and
Canada:fax:215-386-7999;e-mail:service@siam.org. Sendcheck or money order to: SIAM, Dept. BKIL0I'
PA 19104-2688
3600UniversityCity ScienceCenter,Philadelphia, . If yourelec$onic retailerclaimstheyareoutof
stock,orderdirectfrom SIAM!
10th ruAS
Conference
Departmentof Mathematics
Auburn University
Auburn, AL 36849-5310,
IJSA
June 10 - 13, 2002
Call for papers: Abstracts for contributedand i,nuitedpa- Conference Site: Auburn University Hotel and Dixon
pers on any aspect of Linear Algebra and Matrix Theory are due ConferenceCenter on the campus of Auburn University.
March 15, 2OO2
electronicallv at ilas20O2@auburn.edu . Travel Information: Auburn lies about 100 miles south-
west of the Atlanta Airport along I-85. Auburn is in the
Program: The task of this conferenceis to bring together the- Central Time Zone, while Atlanta is on Eastern time. There
oreticians and practitioners of Linear Algebra and Matrix The- is frequent limousine service between Auburn and the Atlanta
ory to describe and exchange ideas about current research and airport. However, departures before noon from Atlanta require
o develop future challengesin the field. a previous night stop-over at some Atlanta Airport hotel since
nvited Speakers: T. Ando, Japan; H. Bart, Netherlands; the earliest (7 am) limousine arrives at 9:45 am in Atlanta due
. Benzi, Emory; R. Brualdi, Wisconsin; D. Calvetti, Case to the time zone change.
estern; G. Chartier, France; M. Choi, Canada; M. Chu, NC Partial Fundirg Auburn University,
tate; G. Dahl, Norway; P. v. d. Driessche,Canada; S. Fallat,
Oak Ridge Associated Universities,
anada; N{. Goldberg, Israel; J. Grof3,Germany; M. Kilmer,
National Security Agency,
ts; L. Lerer, Israel; M. Neubauer, CalState; O. Nevan-
others pending.
nna, Finland; M. Overton, Courant Institute; L. Qui, China;
With the cooperation of the SIAM Special Interest Group on
. Rao, Penn State; F. Roush, Alabama State; H. Schneider,
Linear Algebra.
isconsin; A. Sierpinsh, Canada; J. d. Silva, Portugal; W.
, San Jose State; \,I. Tsatsomeros, Washington State; E. Support for Graduate Students etc.: Supportfor
yrtshnikov, Russia; H. v. d. Vorst, Netherlands graduate students and participants in need has been requested.
It is available upon request to the organizers.
nvited Mini-Symposia: Six invited Mini-Symposia:
atrir Ertensions and Interpolation Problems: L. Rodman: Excursion: We are planning an excursion for the late
dm@math.wm.edu and H. Woerdeman: hugo@math.wm. afternoon of Tuesday, June 11, 2002. We will visit Tuskegee
University and the Carver Museum there. After dinner we
l,trir"r'in Mar Algebras:S. Gaubert: Stephane.Gaubert@in- travel on to the Alabama ShakespeareFestial in Montgomery
ia.fr and G. J. Olsder: G.J.Olsder@its.tudelft.nl for a choice of two plays.
onlinear Matrir Equati,ons:A. Ran: ran@cs.vu.nl Details for Registration etc.: The I$TEX tem-
atrices in Control Problenls: A. C. Antoulas: aca@rice.edu pt't. f ilable at our website
omplerity 'in l{umerical L'inear Algebra: V. Olshevsky: www.auburn.edu/i1as2002. Further details, such as regis-
olshevsky@cs.gsu.edu tration forms and hotel and travel information will be made
'inear Algebra Educat'ion: M. Trigueros: trigue@itam.mx and
available early in 2002 at the website and over ILASNET.
Weller: wellerk@bethel-in.edu
Contact: For the conference: ilas2002@auburn.edu
nyone wanting to contribute to a mini-symposium is invited to
or Frank Uhlig, conferencechair: uhligfd@auburn.edu
ntact the mini-synlposium organi zers directll..
IMAGE27: October2001
andTweW Haifa
TheNinth ILASConference
fMAGE 27: October 2OO1
in LinearAlgebraandMatrixTheory
andWorkshops
Conferences
The Ninth ILAS Conference and the
Twelfth Haifa Matrix Theory Conference
The purpose of the workshop was to stimulate researchmd, group of scientists and engineers with diverse interests in com-
in an informal setting, to foster the interaction of researchersin binatorics and matrix theory.
the interface between matrix theory and statistics. A total of 54 The conference chairs are: Richard A. Brualdi and Suk-
participantsfrom 14 countriesjoined the workshop. Geun Hwang. The international advisory committee comprises
The Workshop wits openedby Marcel P. R. Van den Broecke, Arnold Krluter, Bryan Shader, and Jia-Yu Shao. The local co-
Director of the Permanent Office of the International Statistical ordinator at Postechis Hyun Kwang Kim: hkkim@postech.ac.kr
Institute. This was followed by sessions of invited and con- Invited plenary speakersare Steve Kirkland: 'An Approach
tributed papers. Participants can submit their papers for pub- to Algebraic Connectivity Via Nonnegative Matrices" and Jian
lication, after a refereeing process,in a Special Issue on Linear Shen: "Some problems on digraphs". other invited speak-
Algebra and Statistics of Linear Algebra and its Applications ers include: Kazuyoshi Okubo, Jason Zrongshan Li, Miroslav
(LAA). The keynote speakerswere E. Kontoghiorghes and T. 'W. Fiedler, Bolian Liu, Qiao Li, Willem Haemers, Ian Wanless,
Anderson. There were nine invited speakers:J. K. Baksalary,H. Richard Brualdi, Arnold Kriiuter, Bryan Shader,Charles John-
Bozdogffi, N. R. chaganty, s. Ghosh, J. c. Gower, J. GroB, T. son, Jennifer Seberry,ild Jia-yu Shao.
Mueller, A. C. Vandal, and F. Atang. Another 28 contributed Participants presenting papers are invited to submit them
paperswere presentedin 9 sessions. for possible publication in a special issue of the international
The lunch buffets at the CBS provided a good atmosphere journal , Linear Algebra and lts Applications (LAA), to one of
to stimulate contacts and exchange ideas. The social progftrm the special editors: Suk-Geun Hwang, Arnold Krluter, Bryan
included a welcome reception on August 2, ZOOLat the How- Shader, and Jia-yu Shao. Papers will be refereed in accordance
ijck mansion in Voorburg, where Christi aan Huygens (1629- with the normal high standards of LAA. April 30, 2002 is the
1695)had his residencefor 16 years. The workshop dinner was deadline for papersto be submitted to one of the special editors.
held on August 3 in the Kurhaus Hotel in Scheveningen. An An excursion to Kyongju will take place on Wednesday af-
excellent Indonesian rice table was served. Ttle exceptionally ternoon. Kyungju was the capital of the Shilla Kingdom (57
good weather made it possible to enjoy great beach views on BC - 935 AD) and is Korea's tourist highlight. The cost for the
the Kurhaus terrace. On August 3, the gounnet dinner was en- excursion including transportation, enffance fees, and dinner is
joyed by thirteen participants at the one Michelin star Savelberg us$50.
Restaurantin Voorburg. About 16 participants joined the Delft The Web site of the conference contains up-to-date infor-
excursion on Saturday, August 4. Without any doubt, the visit mation including elecffonic forms for submission of individual
to the working windmill Molen de Roos and the explanations presentations.Pleasevisit
by the miller were the highlights of this excursion. Pleasevist
http://matrix .skku .ac .krAg Iee/postech/postech.htm
http://matrix.fsw.leidenuniv.nl
for severalphotographs taken during
this Workshop.
Kelley, Gerard Meurant, Dianne O'Le?Tg Chris Paige, Youcef Confirmed Particiapants. T. Ando, Natdlia Bebiano, Chan-
Saad, Gene Wachspress, Henk van der Vorsto Charbel Farhat, dler Davis, CharlesDolberry, Charlie Johnson,Tom Morley, Hi-
Michael Griebel, Van Henson, Wei-Pai Thng, Ray Ttrminnfo, roshi Nakazato, Peter Nylen, Yiu-Tung Poon, Jo6o da Provi-
Gabriel Wittum, FritzL. Bauer,Urs Hochstrasser,John Todd. dencia, Leiba Rodman,Ilya Spitkovsky, Bit-Shun Thm, Michael
For more information pleasevisit the Web site Tsatsomeros,Frank Uhlig, and Wen Yan.
http:/ lwww.cg50.ethz .c h There will be no registration fees for this Workshop. Lim-
ited financial support to participants based on need may be re-
quested. To confirm your participation and reserveyour accom-
WesternCanadaLinearAlgebraMeeting
modation : April l, 2002. To submit the title and abstract of
your talk : April 15, 2002.
Regitr?rSaskatchewan:10-1 1 May 2OO2
The Western Canada Linear Algebra Meeting (\M-CLAM) pro-
vides an opportunity for mathematicians in western Canada
ComputationalLinearAlgebrawith Applications
working in linear algebra and related fields to meet, present
accounts of their recent research, and to have informal discus-
Milovy, CzechRepublic: 4-10 August2OO2
sions. While the meeting has a regional base, it also attracts
people from outside the geographical area. Participation is open Following the successfulmeeting on Iterative Methods and Par-
to anyone who is interested in attending or speaking at the meet- allel Computing in 1997 we will organize the second Milovy
ing. Previous W-CLAMs were held in Regina (1993), Leth- meeting (Milovy 20A2) from August 4th to 10th, 2002. The
bridge (1995), Kananaskis (1996), Victoria (1998), and Win- meeting will be held at the Devet Skal Hotel (literally "Nine
nipeg (2000). Rocks") in Milovy, a small village located in the Bohemian-
'02 will be held at the University of Regina Moravian Highlands about 100 km (c. 62.5 miles) from Prague.
The W-CLAM
in Regina on May 10-1 1,2002. fiie participation fee is C$20, The meeting will concentrate on computational methods,
to be collected at the meeting. This fee will be waived for partic- namely on methods of numerical linear algebra, their (parallel)
ipating students and postdoctoral fellows. We anticipate being implementations and applications. Based on the contributions,
able to offer support of up to C$300 per person for a limited we plan to structure the progam into subthemes such as ma-
number of participating studentsand postdoctoral fellows to be trix theory and its applications, sparse direct solvers, iterative
distributed on a first-coille, first-servedbasis. methods for linear systems,preconditioning techniques,large-
The deadline for the submission of titles and absffacts is scale eigen-problems,inverse problems and regularization. Spe-
April l,2OO2. We expect to have online registration availableby cial attention will be paid to application areas not restricted to
January 2002; the department's website http://www.math.uregina.ca PDEs, control, optimi zationor image processing.The scientific
will then have a link to the W-CLAM Web site. For further in- progrutmis planned for five full days (Monday through Friday)'
formation contact either SteveKirkland: kirkland@m ath.uregina.ca with a free Wednesdayafternoon reserved for an excursion.
or ShaunFallafi sfallat@math.uregina.ca The Program Committeeand the Local Organization' Mario
Arioli, Michele Benzi, Iain Duff, Michael Eiermalul' Roland
Freund, Anne Greenbaum, Martin Gutknecht, Ivo Marek,
The SixthWorkshoPon Volker Mehrmann, Gerard Meurant, Jim Nugy, Henek Sffakos,
NumericalRangesand NumericalRadii Daniel Szyld, Henk van der Vorst, Olof Widlund, Mirek Tuma
and Miro Rozloznik will be in charge of the local organuzation.
Auburn University,Alabama:7-8 f une 2OO2 Registration and Submissionof Absffacts: The registration
form can be found on the conference Web site
The Sixth Workshop on "Numerical Ranges and Numerical -m
http://www.cs.cas.czf i lovy/main97.htm
Radii" (6-WONRA) will be held at Auburn University, A1-
abama, from June 7 to June 8, 2002, in conjunction with the The meeting facility holds up to 150 people. Therefore, early
ILAS conference on "Challenges in Matrix Theory" (10-13 regisffation is recommended.The deadline for submitting an ab-
June 2002, seepage l7). stract is January 15,2002. All participants will be notified about
Purpose. The purpose of the workshop is to stimulate re- the acceptanceof their contributions by February 15,2002. Con-
searchand foster interaction of researchersinterested in the sub- ference proceedingswill be published as a special issue of the
ject. The informal workshop atmospherewill guaranteethe ex- research journ al BIT. All submitted papers will be subject to
change of ideas from different research areas md, hopefully' the standard refereeing procedures. The submission deadline will
participants will leave informed of the latest developmentsand be November 30,2002. The conference fees include registra-
newest ideas. For some background about the subject and pre- tion fee, full board in Milov], bus transportation from and to
vious meetingspleasevisit the WONRA Web site: Prague and excursions on Wednesday afternoon. The early reg-
istration fee (paid before April 30, 2OO2)is US$450 (double
http://www.math.wm.edu/ ckl i/won ra.htm I room), US$530 (singleroom).
fMAGE 27: October2001 page 23
Before and after the conferencethe participants will have an will provide a forum through which statisticians working in the
opportunity to visit the historical parts of Prague. We havemade field of linear algebra and matrix theory may be better informed
several group-reservations in a variety of hotels, with relevant of the latest developments and newest techniques and may ex-
information available on the Web site. All participants will be change ideas with researchersfrom a wide variety of countries.
responsiblefor making their own hotel reservations. We would This Workshop is the eleventh in a series. The previous ten
appreciate it if you send us information about your reservation Workshops were held as follows: (1) Tampere,Finland: August
(for the purpose of transport to and from Milovy). Those partic- 1990, (2) Auckland, New Zealand: December 1992, (3) Tiutu,
ipants who wish to stay in a student dormitory should indicate Estonia: May 1994, (a) Montr6al (Qu6bec),Canada: July 1995,
this interest on the registation form. conference organuzers. (5) Shrewsbury,England: July 1996, (6) Istanbul, Turkey, Au-
gust 1997, (7) Fort Lauderdale, Florida, USA, December 1998:
Fifth China Conference on in Celebration of T. W. Anderson's 80th Birthday, (8) Thmpere,
Matrix Theory and lts Applications Finland, August 1999, (9) Hyderabad, India, December 2000:
in Celebrationof C. R. Rao's 80th Birthday, (10) Voorburg, The
Shanghai,China: 14-18 August2OO2 Netherlands, August 2OOl.
ManyWith Solutions
IMAGEProblemCorner:Old Problems,
We presentsolutionsto IMAGEproblems 26-2 through 26-5 publishedin IMAGE26 (April 2001),pp. 24-25. We are still hoping to receive
solutionsto Ftoblems lg-3b, 23-l and to Problem 26-1, which are repeatedbelow. In addition,we introducesix new problems(page32 below)
and invite readersto submit solutionsas well a$ new problemsfor publication in IMAGE.Pleasesubmit all material (a) in macro-ft'eeI4$
by e-mail,preferablyembeddedas text, to werner@united.econ.uni-bonn.de andO) trro papercopies(nicely printedplease)by classicalp-mail
to Hans JoachirnWerner,IMAGEEditor-in-Cbief, Inetitute for Econometricsand OperatioasResearch,EconometricsUnit Universiryof Bonn'
(in
AdenaueralleeU.42,D-S3ll3 Bonn,Germany.Pleasemakesurethat your nameaswell asyoure-mail andclassicalp-mailaddresses full) are
includedin both (a) and(b)!
(a) Characterize square matrices A and B satisfying AB = pA * qB, where p nd q are given scalars'
(b) More generally, chancterizelinear operators A and B acting on a vector space ,t satisfying ABr € span(Ar, Br) for every
r€N.
Spaces
Prcblem 25-1: Degenerate Complex Quadratic Forms on Real Vector
drury@math'mcgill'ca
Proposedby S. W. Dnuny, McGilt University, MontrCal, Qudbec,Canad.a:
andimaginarypartsof C respectively (i'e''
I-et C be acomplex symmetric n x n matrix. SupposethatA and B arethereal
- dl = 0 for all real t. Prove or find a counterexampleto the following statement'
ajk ftcix, bil,-= sc7*). Supposethat det(B
= n andq > r andarealinvertiblen x n matrixP' such
There necessarilyexistnonnegativeintegersP,s mdr suchthatp * q*r
I
that
P,CP -
[: Y,
= 0 and r = 0 ateallowed' Forinstance'p: n-ro
wherex is apxpcomplexmatrixandY is aqxrcomplex matrix' ThecasesP
real vector in the null space of C'
Q= !, r = 0 is the case where there is a non-zero
Ptoposer lws not yet found a solution!
The Editors have not yet receiveda solution to thisproblem-indeed arcn the
IMAGE27: October2001 page 25
By adding an extra first row and column to the matrix of hoblem 25-1 (see IMAGE26 (April 2001), p. 2), we obtain the 8 x 8 matrix
(seealso Solution 25-1.3 on pp. 34 above). Denote by 47, its principal submatrix of order /c, and observethat A2, Aa and As dre
skew symmetric and (apart from a scalar multiple) orthogonal for any real o; with oo = 0. Can this construction be continued? Note
that only the sign pattern of A16 is to be properly chosen, for the distributionof the o; is easy to follow: if the sign-free matrix ,47,is
denotedby Mr, then
/ m* 'n/*\
LI2p= | |
\lr* Mx/
Sofution 26-2.1 by F6tima Su.ve LeIrE, Departamznto de Materruitica, (Jniversidade de Coimbra, Coimbra, Portugal:
fleite@mat.uc.pt
The answer is rol Indeed, that construction cannot be continued, due to the following result in algebraic topology, which has been
an open problem for many years and was completely solved in the early 1960s; seeAdams (1962):
Refercnces
[1] J. Adams(1962).Vectorfieldson spheres. AzzalsMathematics:SecondSerics,75,6O3-632.
[2] I. Eganova & M. Shirokov (1984). Orttromatricesandoctonions.Reportson MathcmaticalPhysics,2O,l-ll.
[3] F. Silval,eite (1993).The geometryof hypercomplex matrices.Linear and MultilinearAlgebra,34,123-132.
page 26 IMAGE27: October2001
Solution 26-2.2 by Yongge TttrN, Queen's University, Kingston, Ontario, Canada: ytian@mast.queensu.ca
The matrix 47, proposed in the problem is closely linked to real representations of complex numbers, quatenrions and octonions. For
Az, it cn be regarded as a real matrix representation of the complex number e, = ao a o 1f through the mapping
a o* a 1 i e C + - Az: (1)
dz: a: 6z@)
[::
I ,, oo
d+:o,= 0o* aj*a2j *oak+Q+@) = | (2)
ae onorr.
la2
I
This mapping satisfies L ou -a2 ;:tl
(a)a-beQ+@)=Q+(b) .
+ b) = d+(o)+ d+Q).
(b) da,fu
(c) dnfub): d+@)Qn(b), ,in()o)= \d+(o).
(O d+(a)= 6T@).
(e)4+@-L) = 4i' @),fi a I a.
(f; det ldn(o)]= lola.
Moreoverthe diagonalmatrixala is universallysimilarA dn(o), seeTian (1999).
Although the matrix ,4s given in the problem is skew-symmetricand orthogonal(apartfrom a scalarmultiple), it cannotbe
of an elementin the well-knowneight-dimensionalClifford or Cayley-Dicksonalgebra.
regardedas a matrix representation
An alternativeexpressionof As is
d a ( a )=
Ar -01 -Qz -43
a.+ Ag AO
which,is skew-symmetric and orthogonal (apart from a scalar multiple). This matrix is a matrix representation of the octonion
e = a0.l- afit+...+ azet (i.e.,anelementintheeight-dimcnsionalCayley-Dicksonalgebra). Themultiplicationrules forthe
fMAGE27: October2001 page 27
e1 €Z eS e4 e5 eO eZ
Q1 -1 es -e2 e5 -e 4 -ez e6
e2 -es -1 €1 e6 e7-e4-e5
n{nr-
eq -es -e6 -eZ -1 et €2 eg
and
ab- ( o' + atte)(b'+ b"e) - (o'b'- U,a") + (b"o' + a"b,)e,
where b/, 6// denote the conjugates of the octonions bt and b" . Through the matrix representation of the octonion presented above,
one can easily get a 16 x 16 real matrix representationfrc(o) for any sedenion a = a' I a" e which is skew-symmetricand 6"@')
is its principal submatrix. Various interesting properties for /16(o) can be derived through the sedenion algebra. It is well known,
however, that the sedenion algebra is not a division algebra, i.e., there exist two nonzero sedenions o and 6 such that ab = Q. In that
case,it is easy to prove that the correspondingreal matrix representations/16(o) and {16(6) are singular. Thus the matrix fru(")
and /16(6) cannot be orthogonal. This fact tells us that one cannot construct a skew-symmetricand orthogonal matix A2x (k >_4)
through matrix representations of Clifford and Cayley-Dickson algebras.
Our final conclusion is that there does not exist 42* (k > 4) which is skew-symmetric and orthogonal (apart from a scalar
multiple). If so, then one can constuct a 2k-dimensional algebra ..4 from the matrix ,42r such that A is a division algebra. But
it is well knowu that there do not exist 2*-dimensional division algebras over the reals when k ) 4; see, e.g., Schafer (1966) and
Zhevlakovetal. (1982). tr
References
[] R. D. Schafer(1966).An Introductionto Non-Associative Algebras.AcademicPress,NewYork.
[2] Y. Tian (1999). Universalfactorizationequalitiesfor quatemionmatricesand their applications.MathematicalJounnl of Ol<ayanaUniversity,
42,45-62.
[3] Y. Tian (2000). Matrix representationsof octonionsandtheir applications.Advancesin Applied Clifford Algebras, 10,6l-90.
l4l K. A. Ztevlakov,A. M. Slin'ko, I. P.Shestakov& A. I. Shirshov(1982). Ringsthat areNearlyAssociative.Translated from theRussianby
Harry F. Smith. Pure andApplied Mathematics104.AcademicPress,New York.
page28 IMAGE27: October2001
and
range(.\11,- A') o range(\zln - A') + 1011
thenthe setequality
{ [ ( r r / " - A ) ( ^ 2 r , - A ) ] - ] = -A-2l- A. l { ( 1 , r , - A ) - - ( \ z r ^ - A ) - }
(or inner)inverseof a matrix.
holds,where(.)- denotesa generalized
and
- A') fl range(\zI^-
range(.\11, n') n...t'l range(\xI,- A\ + l0I,
thenproveor disprovethat
Solution 26-3.1 by Hans Joachim WenNBn: University of Bonn, Bonn, Germany.' werner@united.econ.uni-bonn.de
We prove the followingmore general result
THsonnMI.IztAbearealorcomplexnxnmntrix,andlet\t(i-1,2,'..,k)bescalarssuchthat\*\iwhenarcrifj.
For A7, ;= ),,i1- A, i = I,2,' ' ' , k, we thenhave
p*-r(tr)
r ^ - r \ " / ' ,=i lr-',ll:l
- )r)'
?.-.lli*o}i
It is obviousthat the degreeof this polynomialcannotexceedk - l. For all i = 1,2,. . .,k, clearlypr-r()t) = 1. This in turn
imptiesthatp;-1(.)= 1. Sothedegreeof thepolynomialp;.-1()) is0andwedohavep*-1(A) = lforeverysquarcmatrixA.
=
SincethematricesAy (i I,2,. . ., k) commutewith eachother,we now obtain
lk \ /h
= o^,)or-'(o)
(I'^')(Ea"JE'^'J=E;'^'(9",J
,
=fIA^,
\ /ft \ ft ., / \ ro \ k
LI
thus completing the proof of inclusion ) in (3). To prove the converse inclusion, observe first that according to Theorem 2.4.I n
Rao & Mitra (1971) we now know that a general solution to the matrix equation
lk
'(g o^,)(.I'^',)
(,Io^')
\ /k \ lk \
IMAGE27: October2001 page 29
is
G- (r *^r,)+
\,( r - CC- ) + ( r - C- C) w, (4)
w h e r e C , = f l f = r A 1 , , w h e rAe1 Q = L , 2 , . . . , / c ) a n d C - r e p a r t i c u l a r g - i n v e r s e s o
( if=A1l , 2 , . . . , k) and C, respectively,
andwhereV andW are arbitrarymatrices.Evidently,wheneveri + j, N(A^,) I R(Axi) and,A/(,41,)n Ar(A1r) - {0}; here
7?(.)and,A/(.)denotethe range(columnspace)andthenull space,respectively, of (.). Therefore,
^{ = Ar(Ax,)
and r (,trr,)- O.^r(Ai,), (s)
EA^,) O i-r
with O indicating a direct sum. With this in mind, it is clear ttrat (4) can be rewritten as
for suitable choices of % and Wt Q = I,2,. . ., /c). Since the expressionsin the bracketsare, again according to Theorem 2.4.1 in
Rao&Mitra(1971),g-inversesofthecorrespondingmatricesAl,(i=L,2,...,k),theproofofourtheoremiscomplete. tr
Although our Theorem 1 contains, of course, as a special case also an answer to part (a) of Problem 26-3, we offer below an
alternativeproofforthatparticularcase. Tothisend, let,\1 and)2betwoscalarswith,\r f )2. Moreover,letAbeagivensquare
matrix.Put11,:=)r1 -AandA^2'.=),zI-A.ThenAl, - A ^ 2 = ( , \ 1- A 2 ) / , a n d i t i s c l e a r t h a t - 4 1 , a n d - A l , i s a p a i r o f
parallel summable matrices; for a detailed discussion of parallel summable matrices we refer the interested reader to Rao & Mitra
(1971, p. 188 tr). We note that the parallel sum of ,41, and -As", which is defined as
The squarecomplex matrix A is said to be EP wheneverthe column space(range)R(A) = R(A-). Let ,4 and B be n x n complex
matrices and let r{+ denote the Moore-Penrose inverse of A.
Solution 26-4.1 by Jerzy K. BnxsaLlrr^, Zielona Gdra University,Zielona G6ra, Poland: jbak@lord.wsp.zgora.pl
and Oskar Maria Blrsntl.nx, Adam Mickiewicz University,Pogtafi, Poland: baxx@main.amu.edu.pl
Since an orthogonal projector (a Hermitian idempotent matrix) is uniquely determined by a subspace onto which it projects, and
MM+ and M+M are the orthogonal projectors o\toR(M)
sincrc, andB(M*), respectively,it is clear that M € Cn,n is an EP
matrixifandonlyif MM+ = M+M.
The first part of our solution corrects the assertion in part (a) of the problem. The correction consists in deleting the implication
thatif AB = BA and A+ B - BA+ , then A is an EP matrix. TWo simple counterexamples are obtained by taking B as the zero or
identity matrix, in which casesthe two equalities above are fulfilled by ny A € Cn,n.
L E M M A .I f M e C n , n i s a n E P m a t r i 4 t h e n a r r y N e C n , n s u c h t h a t M N = N M satisfiesM+N - NM+.
whichconcludestheproof. tr
The secondpart of our solution extendsthe assertionin part (b) of the problem.
Pnoor. It is clear that M € Cn,, is an EP matrix if andonly if M+ hasthe sameproperty.In view of this observation,the chain of
implications(i) + (ii) + (iii) + (iv) + (i) follows immediatelyfrom kmma by substitutingM =,8 andtrf = Ato showthat (i)
+(ii), M: A andl{ = B* toshowthat(ii)+ (iii),M - B+ andl[ = A+ toshowthat(iii)+(iv),and,finally,bysubstituting
M = A+ and N = B to showthat(iv) + (i). D
1. The polynomials
- AAI, (At A). - At A, AAt A - A, AI Ant - At} (7)
{{aat).
which when set to zero define the pseudo-inverse of A.
2. T\epolynomial
- ae). (8)
{ea
3. The fact that.,l is EP gives
- A ' , G G - L- r , c - l c - r } .
{o,
4. OnewaytoshowthatAtandB commute*ao*r" Q ndweadjoinQ -
anadditionalvariable AIB+ BAr,sincewe
achieveour goal ifQ is 0.
5. We adjoin the adjoints of the polynomials above. (NCAlgebra has a convenient command that accomplishes this.)
The second part of the input is an order on monomials which typically is induced by an order on the variables in the problem. In our
case the order is degree lexicographic order induced by the following order on variables:
(e)
Recall that in a Griibner basis algorithm variables which are high in the order tend to be eliminated before variables which are low
in the order.
We use the command NCProcess, with options set so that it makes and sorts an NC Grtibner basis by running Mora's algorithm
for four iterations and removes some redundant polynomials. This makes a long list of polynomial equations, one of which is - 0.
Q
This shows that part (a) is true. The run took 49 minutes on a 1 Gigallertz Pentium III machine with one gigabyte of RAM.
(b) Suppose A and B cornmute. Again we use the Grijbner basis package applied to the starting relations:
1. The hypothesis that both A and B are EP is equivalent to the existence of G and f1 satisfying
- A * , G G - L- r , G - l G - r , B H - B * ,H g - t - r,r-1r - t}.
{o"
2. Applyingpart (a) to thepairs A and,Bwhich commute, B and,4 which commute, A and itself, -U -U itself allows us to
adjoin the relations "
{re " , -BIA.BAt - A t-B , A A t -AtA,BBt - a-t n -) .) '
3. The polynomial
rl
_ BAj. (10)
IAB
4. We adjoin the defining polynomials for the pseudo-inverse of A given in equation (7) and the corresponding polynomials for
B.
5. Once again, our objective is achieved if the additional variable Q satis$ing the equation
Q-finI+BtAt
is zero.
6. We adjoin the adjoints of the above equations.
We choose a degree lexicographic order with Q x the top. (We found, when Q is placed at the top of the order, the result to be
insensitive to the rest of the order in 4 experiments.) We ran NCProcess with options set to 2 iterations and arnong the output was
Q = 0' It follows that -4t and Bt commute, thus establishingthe "if' part of (b). Theseruns took less than a minute.
The converse is obtained similarly. In the input polynomials we merely replace
- rr}
{^"
page 32 fMAGE27: October2001
by
- sret)
{etar
andredefineQ byadjoiningthepolynomialQ - AB + BA insteadof Q - At At + gt At;we dothis sinceourgoalis nowto show
that A and B commute. Again we choose a degree lexicographic order with @ at the top and find after 3 iterations and about 15
minutes thatQ - 0. AgainwetriedsomevariationsontheordersandineachcasefoundthatQ - g.
This solves problem Problem26-4by fairly automatic use of NCAlgebra's noncommutative Griibner basis package and suggests
the followingproblem to us:
Now that some noncommutative computer algebra is available, which of the purely algebraic problems (with a modest
number of variables and equations) that have been posed over the past years in IMAGEsuccumb in an automatic fashion
to computer algebra?
Since amatix M isFP iff M+ is EP, and since (M+)' = M,the followingtheorem is a direct consequenceof Theorem 1.
THEoREM 2. Izt A and B be a pair of EP rnatrices and let AB be defined. Thefollowing conditions arc then equivalent:
We conclude with mentioning that if B = 0, then for each matrix A of the same size as B, clearly AB = 8,4 = 0 and
A+ B = BA+ - 0, even when ,4 is not EP. This shows that claim (a) of Problem 264is incorrect.
Show that for any Hermitian nonnegative definite matrices ,4 and B of the same size and for any real number.\ satisfying 0 < .\ < 1,
(b) hA + (1 - A)Bl+ <L ^A+ + (1-,\)B+ + [1.4+ (1- l)B]+ {m lxe+ (1- ))Bl+m\- lx,++(1- A)Bl+,
where P standsfor the orthogonal projector onto /t(A) nR(B), ?(.) denotesthe range (column space) of (.), (.)+ indicates the
Moore-Penroseinverse of (.), M '.= I - P, and (1- standsfor the Liiwner ordering.
fMAGE 27: October 2OO1 page33
A s-
(: andBs= (1-
,:,) ,:r)
Since01y A,itisclemthat0 (r A+ and7P-(A+) =R(A), thusimplyineR(P)gR(A+). Moreover, P(A+)+P - pAp 1r
PAP, which completesthe setof conditions(11) concerningAg. Henceit follows that 0 (; ,4g and,by analogousarguments,
0 {'1, Ba. Consequently,
+(1 -))r+ P \
o ( r , \ A * + ( 1- ^ ) B + - ( ^ ^ .
\ P P l ^ A +( 1- , r ) B,l ) ,
and therefore, on account of the third condition in (12),
p { p [ l A + ( 1- l ) B ] p ] + p < y ^ A + + ( 1 - l ) B + (13)
TuBonerra. Let A and B be Hennitiannonnegative definite matrices of the sameordet let ), be a real number satistying 0 < ,\ < 1,
andletC = \A+ (1 - \)8. Morcovenlet P denotetheorthogonalprcjectorontoR(A)nR(B) and.letM = I - P.Then
Pnoor. The latter inequality in (15) is just part (a) of the problem, which has been proved above, while the former one follows when
we apply (11) and (12) with (14) to the matrix
where the matrices on the right-hand side are modified according to (14). Let Co = C1/2. Then R(Cs) : R(C), and since
the column spaces of all matrices appearing in (16), as well as the column spaces of their conjugate transposes, are contained in
R(C) = R(C*), it follows that (16) is equivalentto
4P : c+ M) - r(P)+ r(MC+M).
Moreover,
- r ( M C + ) - r ( M P c )= r ( P c - P P c ) -
4MC+M) 4Pc - P).
SincePcP = P, it follows from the Theoremin $42of the Halmos(1958)that Pg - P is a projector,andtherefore
Pnoor: SinceR(P) g R(V), (a) follows immediatelyfrom (2.6)in Werner& Yapar(1996),andso it is evidentthat 2(PlI/) is
not empty.From Theorem2.5 in Werner& Yapar(1996)we evenknowthat P(PV+ P)+ PV+ is inP(PlV); this provesthefirst
partof (e). Theorem2.7(i)& Theorem2.7(ni)inWerner &Yapar (1996)giveus(b) and(c). EquationQV(I - Q)* = 0 from
s V = Q V Q . - V Q . a n d ( 1- Q l V = ( / - Q ) V ( I - Q ) * = V g - Q ) . . S i n c e i n v i e w of f = Q + 1 t - O ) w e
( c ) i m p l i eQ
trivially haveV = Q\' + (I - Q)V , theproof of (d) is now complete.Combiningthe first part of (b) with the first part of (e) gives
- P. By checkingthe definingequationsof the Moore-Penrose
QV = P(PV+ p1+p sincePV+V inverseof PV+ P it is seen
thatP(PV+P)*p - (PV+P)+, andsowealsohavethesecondpart(e). SinceM - I - P,R(M) = "4/(P).lVenotethat
VM(MVM)+Misaprojectoronto?-(VM) =VN(P) alongasubspacethatincludes N(M) =R(P). firerefore,inviewof
(a),(b)&(c), (1 -Q)V = VM(MVM)*UV isinvariantforanychoiceof Q en1flV). Thisis(f).From(d),(e)&(f)we
obtain(g). Since?(V+ ) = R(V),7i-e) gR(V+), andso,in viewof (g)and(V+;+ = lz, it is clearthat(h) alsoholdstrue. tr
The proof of this theorem is left as an exercise to the interested reader. In this context, we mention that Theorem 10.1.8 in Rao &
Mitra (1971) deals with a less restrictive situation; we also note that A(A + B)- B is often called the parallel sum of A and B.
We are now in a position to prove the claimed inequalities. First, let ) be such that 0 ( ^ < 1. We recall that for each matrix C
and each scalarc I 0, and we have (C*)- = C and (cC)+ : c+C+ with cr = c-|. Hence, in view of Theorem 2(d),
l p ( ^ A + ( 1- . \ ) B ) p l + : ^ +A + [ l + A + + ( t - ) ) + 3 + 1 +( 1- r ) + B +
: A * [ ( r - . \ ) . 4 ++ 1 g + ] r r r + .
Forconvenience, W = A++^(B+ - A+) - S+ +(1-l)(A+ - B+). Thematrices
wenowputW ,= (1-,\)A+ +^B+. Clearly,
A+ andB+ areHermitiannonnegativedefinitewith?-(A+) =R(A) andR(B+) :R(B). Therefore,since0
< ^ < l,according
WW+ (B+ - A+ ) - B+ - A+
definrtewithB(W) = R(4 +n(B). Consequently,
to Theorem2(a),W is Hermitiannonnegative
and (A+ - 9+ 1lil+ W - A+ - B+. with thisin mind, we get
A+W+B+ = [ . r a ++ ( 1 - r ) , 4 ++ ] B + - ) B + ] 1 4 t + l S B++( 1 - ^ ) B + + ( 1 - . \ ) , 4 +- ( 1 - ] ) , 4 + ]
-
= lw + ^(A+ B+)l W* lW + (1 - l)(B+ - ,4+)l
: w + \ ( A + - B + ) + ( 1 - l ) ( B + - A + ) + l ( 1 - l ) ( A + - B + ) w + @ +- A + )
= A + + ( 1- r ) ( B + - A + ) - l ( 1 - r ) ( A +- B + \ , y 1 r + ( A - +B + )
= - -
.\A++ (1 ))B+ l(1 - ))(,4+- n+)reLr+@+ -.8+)
andthus,sincetr4/*is Hermitiannonnegative
definite,also
. \ A ++ ( 1 - . f 1 4 + - n + y 1 7 + 3=+ , l ( 1 - f ) ( A + - B + Y y v + ( A-+B + ) > r -0 .
Thiscompletestheproofofinequality(a)incaseof0 < ^ < 1. For) = 0,inequality(a)reducesto(PBP)+ St B+,whichisthe
result of Theorem 1ft). For A = 1, inequality (a) becomes (PAP)+ (l ,4*, which again holds accordingto Theorem 1(h).
So it remains to show that inequality (b) also holds true. To this end, let 0 < A < 1 be arbitrary but fixed. For convenience, put
H := ),A +(1-))8. T h e n , a c c o r d i n g t o T h e o r e m l ( h ) , f I += ( P H P ) + + H + M ( M H + M ) + M H * . Frominequality(a),we
alreadyknowthat (PH P)+ (1 )A+ + (1 - ^)B+. By combiningthesetwoobservationsweobtaintheclaimed inequality (b).
Refercnces
lll C. R. Rao& S. K. Mitra (1971).GeneralizedlwerseofMatricesand ltsApplications.Wiley, New York.
[2] H. J. Werner& C. Yapar(1996). On inequalityconstrainedgeneralizedleastsquanesselectionsin the generalpossiblysingular Gauss-Markov
modet a projectortheoreticalapproach.Linear Algebraand its Applications,2371238,359-393.
page 36 IMAGE27: October2001
IMAGEProblemCorner:New Problems
Probfem 27-22SpecificGeneralizedInverses
Proposedby JiirgenGnoB & Giitz TRENKLER,UniversitiitDortmwrd,Dortmrntd'Germany:
k.uni-dortm
trenkler@statisti
k.uni-dortmund.de,
gross@statisti und.de
t)
L-(l:) and (r-(t
\0 r/
* r (r-9#6o
Ao A-L A)-',
where o stands for the Hadarnard product, ( stands for the Iiiwner ordering, and c and d are the largest and smallest eigenvalues
of A, respectively. This inequaliry is a count€rpart to an inequality established by G. P. H. Styan (1973): Hadamard products and
multivariate statistical analysis, Linear Algebra and lts Applications, 6, 217 -24O.
How are the complex orthogonal projectors P aad Q (i.e., Hermitian idempotent matrices with complex entries) related if they
satisfy the following inequality in the Ltiwner ordering
,/re + q+ <LP+ Q,
where (P + q+ denotesthe uniquenonnegativedefinite squareroot of P * Q ?