Gauss-Jordan Reduction: A Brief History

Author(s): Steven C. Althoen and Renate McLaughlin

Source: The American Mathematical Monthly, Vol. 94, No. 2 (Feb., 1987), pp. 130-142
Published by: Mathematical Association of America
Stable URL: http://www.jstor.org/stable/2322413 .
Accessed: 23/09/2011 09:41

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.

Mathematical Association of America is collaborating with JSTOR to digitize, preserve and extend access to
The American Mathematical Monthly.

http://www.jstor.org
130 STEVEN C. ALTHOEN AND RENATE MCLAUGHLIN [February

A BriefHistory
Reduction:
Gauss-Jordan

CollegeandmyPh.D.in
STEVEN C. ALTHOEN: I receivedmyB.A. at Kenyon -

ofNewYorkunderthedirection
1973at TheCityUniversity ofEldonDyer.
StephenB. Maurerof Swarthmore
I wouldliketo thankProfessor College
forpreventing
mefromcrediting thewrongJordan in mytextsby relaying
My primary
Tucker'sobservation. researchinterest of
is theclassification
real divisionalgebras.
finite-dimensional

RENATE MCLAUGHLIN: I did my undergraduate workin Germanyand

myPh.D.in complexanalysis
received (1968)at theUniversity
ofMichigan
underthedirectionof PeterL. Duren.My primary research is still
interest
complexanalysis.Butwithtwochildren school,I havealso
in elementary
inthemathematics
becomeinterested doesnotgoon)
thatgoeson(orbetter:
schoolclassrooms.
in elementary d 3 _

A. W. Tucker[28] has pointedout thatit was thegeodesistWilhelmJordan

(1842-1899)and not themathematician CamilleJordan(1838-1922)who intro-
ducedtheGauss-Jordan methodof solvingsystems of linearequations.Thereis a
naturaltendency to the
attribute method to CamilleJordan,whois justlycredited
withanotherlinearalgebratopic,theJordan normalform.Recallthatin Gaussian
Elimination,rowoperations are usedto changethecoefficientmatrixto an upper
matrix.
triangular The is
solution thenfound by backsubstitution,
startingfromthe
lastequationin thereducedsystem. In Gauss-JordanReduction,rowoperations are
usedto diagonalizethecoefficient matrix,and theanswersarereaddirectly.
Thisarticleis aboutWilhelm Jordan ofhismethod.
and theintroduction
1. Howdo weknowitwasWilhelm
Jordan?Thereis littledoubtas totheidentity
toin"Gauss-Jordan
oftheJordanreferred reduction."It seemsthatthenameswere
analysts.For example,Householder
attachedto themethodby numerical [13,p.
141]states,
The Gauss-Jordan method, so-called,seemsto havebeendescribedfirstby
Clasen[7].Sinceitcanbe regardedas a modification
ofGaussianelimination,
thenameof Gaussis properly applied,butthatofJordanseemsto be due to
an error,sincethe methodwas describedonlyin the thirdeditionof his
HandbuchderVermessungskunde prepared afterhisdeath.
is clear,butso is thefactthatthethirdeditionofJordan's
Thisreference bookwas
preparedbyJordanhimself,wellbeforehisdeath.Theforeword to thethirdedition
1987] GAUSS-JORDAN REDUCTION: A BRIEF HISTORY 131

is dated May 1888 and signedby Jordan.Furthermore, thebook reviewby C.

Muller[23]statesthatJordan himself
preparedthefourth edition, whichappeared
in 1895.Incidentally,
thefrontispieceofthatparticular
editionis a picture
ofGauss,
sincehe was famousforhis geodesyas wellas his mathematics and physics.It
shouldalso be notedthatsinceClasen'sarticle[7] appearedin thesameyearas
Jordan'sthirdedition, it seemsthattheirdiscoveries
wereindependent.
Johnson [15,p. 66] states:

So faras I know,thismethod
ofviewingthereducednormalequationsdidnot
in anytreatise
appearexplicitly uponLeastSquarespriorto thethirdedition
of W. Jordan'sVermessungskunde,....

It should be pointedout thatalthoughthe firsteditioncontainsno hintof

Gauss-Jordan reduction,thegermof the idea is alreadypresentin the second
edition(1877)in theexampleon pages34 and 35.
Kunz [18, p. 221] has a sectionwhosetitle"Gauss-Jordan Method"has the
footnote:W. Jordan, "HandbuchderVermessungskunde." BothBartlett[5] and
Leland [19] mention Jordan'sHandbuch der Vermessungskundein theirbibliogra-
phies.In Gewirtz,Sitomer, and Tucker[10,p. 246]we find:

WilhelmJordan(1819-1904
[sic])devisedthepivotreduction known
algorithm,
as Gauss-Jordan forgeodetic
elimination, reasons.

Writersof current
linearalgebratextsgenerallyapplythename"Gauss-Jordan
reduction"withoutreferenceto its origins.However,it was throughWilhelm
Jordan'sHandbuchder Vermessungskunde thatGauss-Jordanreduction
was intro-
ducedto theworld.TodayClasen,whomwe discussbelow,is largelyforgotten.
2. Who was W. Jordan?We knowaboutWilhelmJordanthrough severalob-
ituaries[26],[16],[24](see also [11]and [25]).He was bornMarch1, 1842,in the
townof Ellwangen in southern Germany. Following highschool,he attendedwhat
todaywouldbe calledan engineering collegein Stuttgart.
He workedfortwoyears
as an engineering assistanton preliminary workforrailroadconstruction and as a
" trigonometer" on measuring elevations.He spenttwofurther yearsas an "Assis-
tant"in geodesyat thecollegein Stuttgart, andin 1868,whenhe wasonly26 years
old,he becamefullprofessor ofgeodesyat thetechnical collegein Karlsruhe.
Jordanwas actively involved in surveyingseveralareasofGermany. In 1873he
became editorof the Zeitschrift (Journalfor Surveying)and
fur Vermessungswesen
in thiscapacity
he remained untilhisdeath.Jordan
was a prolific His major
writer.
work startedin 1873 as Taschenbuchder PraktischenGeometrie(Pocket Book of
Practical Geometry).In latereditionsthisbecame the Handbuchder Vermessungs-
kunde(HandbookofGeodesy). By thetimeof Jordan's
death,fiveeditionsof this
book had appeared,and it had beentranslated
intoFrench,Italian,and Russian.
132 STEVEN C. ALTHOEN AND RENATE MCLAUGHLIN [February

Jordan'sabilityto present abstract

ideasin livelywayswas credited
withthewide
distribution
of thisbook.
In 1882JordanleftKarlsruhe andwentto theTechnicalUniversity in Hanover.
He continued to be activein surveyingandin publishing
hisworks.Apparently he
was a first-rate
teacher whohad a particulartalentforbringing
outtheconnections
betweentheoryand thereal-world problems confronting
him.His fieldtripswith
students werepraisedas examples ofhowone oughtto teach.
Althoughapparently physicallyfit,Jordanhad suffered foryearsfromheart
disease and otherproblems.In 1899,at the age of 57, he died in a stateof
depression.
3. WhatwastheProblem?Gaussinvented theMethodofLeastSquarestofinda
best linearfunctionto approximate observeddata. The methodwas naturally
attractive Hereis a quickdescription
to geodesists. of themathematicsinvolved,
alongwithan interesting theorem, whichwas firstexplicitlystatedbyGauss.
Supposewe makem observations, eachofwhichdependson n inputs.Following
Jordan,we call the observedvalues -l1,..., -in, but to undertake a modem
it willbe usefulto labeltheinputvectors
analysis,
(ai1,..., ain), fori = 1,..., m.
Our m observationsli are a function of theinputvectors. It is desirableto havea
good linearapproximation forthat function.Let
y = L (yl , *- - YJ = YlXl + * *YnXn
denotea generallinearfunction withcoefficients xl,..., xn. Let videnotetheerror
thatarisesfromthelinearity assumption:
Vi= L(ail,..., ain)-li
= ai1x1+ ** - +ainxn + 4i, fori = 1, ... , m. (1)
We selectxl,..., xn to minimize
thesumofthesquarederrors
E = v2 + *-+VM2.
Thus,the Methodof Least Squaresreducesto a simpleproblemof multivariate
calculusin whichwe solvethesystem
of normal
equations:
dE/lxj = 0, forj = ,...,n.
Let A = (aij), v = col(vl,..., vm), 1 = col(ll,..., in), and x = col(xl,..., xn),
wherecol meanscolumn vector. System 1) can thenbe written as
v = Ax + 1.
Directmultiplication
yields:
E-= va2+ _ +v2

+2ailailx2 + 21,aila aila

+2 aillixl +
1987] GAUSS-JORDAN REDUCTION: A BRIEF HISTORY 133

wherewe havelistedtheterms
involving runfromi = 1
xl and all thesummations
to i = m. Then

dE/dxl = 2Eailaixl + 2Eailai2x2 + * +2>ailaia,xn + 2Failli = 0,

or
Eailailxl + Iailai2x2 + * + IailainXn + Eailli = 0.
Gauss denotes Eaijaik by [ajak] and Yaijli by [ajl]. Thus, he would writethis
equationas
[alal]xl + *- +[alan]Xn + [all] = 0.
The remainingn - 1 equationsareobtainedsimilarly.
It is now easy to verifythatthe entiresystemof normalequationscan be
in matrix
represented formas
AtAx = -Atl
(whereAtdenotesthetranspose
ofA) withsolution
x = -(At4) -lAtl
Note thatAtAis a symmetric
matrix.
Hereis a modernstatement thatoccursin Article13 of
and proofofa theorem
Gauss's Disquisitiode ElementisellipticisPalladis [8].
is used to diagonalizethesymmetric
reduction
GAUSS'STHEOREM.If Gauss-Jordan

[
matrix
AtA -Al

--(-Atl)t itl

thenthenumberin thelowerrightcorneris thesumof thesquarederrorsincurred by

usingthecoefficients
ofthesolutionvectorx as coefficients
forthelinearapproximation
thelowerrightcornerwill
and thegivendata points.Thatis, aftertriangularization,
contain
E 2 + * +V2 Vivfv

Proof.Viewthesystem
AtA Atl]
(_Atl)t 0
as thelinearprogramming
problem:
Maximizew = (Atl) tx (2)
subjectto
AtAx = -Atl.
Thenthenumber in thelowerright
aftertriangularization ofthe
willbe thenegative
valueoftheobjective
function(Atl)txat thesolutionto thesystem. So ifwe begin
134 STEVEN C. ALTHOEN AND RENATE MCLAUGHLIN [February

with iNiin thatcorner,we will obtain

il + (Atl)tx,
wherex = -(AtA)`Atl.
On the otherhand,
vtv= (Ax + 1)t(Ax + 1)
= (xtAt + lt)(Ax + 1)

= xtAtAx+ xtAtl+ ltAx+ lt

= xtAtA[-(AtA4) -lAtl] + xtAtl+ ltAx+ itl

= -xtAtl+ xtAtl+ ltAx+ lt

= ti + (Atl)tx.

Thus, if we triangularize
thesymmetric
system

L AtA -Atl1
L(-Atlit iti J'

we obtain not only the solutionto the normalequations,but also the value of the
sum of the squared errorsfortheparticularobservations.
4. What was Gauss's Methodand Notation?In 1810, in his Disquisitiode
ElementisellipticisPalladis, Gauss [8], [9, p. 123] sets out to determinedetails
("elliptical elements")about the orbitof Pallas, the second-largestasteroidof the
solar system.He obtainsa systemof linearequationsin six unknowns,wherenot all
equations can be satisfiedsimultaneously. Hence he needs to determinevalues for
the unknownsthatwill minimizethe totalsquared error.Instead of merelysolving
theproblemat hand,Gauss digressesand introducesa methodfordealingwithsuch
systemsof linear equations in general.This is where his characteristic notation
appears forthe firsttime.
To save space, we use the symbolsof Section 3 above (Gauss lets p, q, r,...
denote the variables and writes 2 = w2 + w'2 + w",2+ for the sum of the
squared errors,etc.). Gauss's startingpointis thesystem
{ allxl + ***+ alnXn + 11 Vl X

amlxl + *+amnXn + Im Vm=

He merelystatesthat"it is easy to see" [facilequidemperspicitur]

thatin orderfor
the total squared errorE = vl2+ + vmto be a minimum,the followingcondi-
tions mustbe satisfied:

allvl + a21v2 + ** +amlvm = 0,

alnvl + a2nv2 + +amnVmV

1987] GAUSS-JORDANREDUCTION: A BRIEF HISTORY 135

Gauss does notuse thewords"methodof leastsquares"in thesectionwherehe

considersthissystem butearlierin thepaper[8, p. 16] he indicates
of equations,
practica]thatmakethe
thathe has longused "certaintricks"[quaedamartificia
applicationof themethodof leastsquaresmoreconvenient.At anyrate,thelast
systemis (exceptfora commonfactorof 2) identicalto the systemof normal
equations dE/lxl = 0,..., dE/dxn= 0. Now Gauss introduceshis characteristic
notation
[all]= allll + a2112 + +amllm'

[aiak =-alialk + a2ja2k + - - +amiamk,

fromthe
x1,..., xn need to be determined
and so forth.Hence the unknowns
equations
{
[alal]xl + ?[alan]xn + [all] = 0,
(3)
[anal]xl + *+ ?[anan]Xn + [anl] 0-
Gauss nowusesa procedure (butnotmatrix notation) thatis essentially
equivalent
to whattodayis knownas Gaussianelimination. He expresses in terms
E explicitly
ofx1,..., x and showsthatE - R2/[alal] is independent ofx1.HereR1 denotes
theleft-handsideofthefirstrowofthesystem above.Nexthe eliminates x2 from
E(1) - E - R /[alal], and so forth.In thisway he obtainsa representation

E (Al(xl ... - Xn))2 (A2(X2, ... , Xn))2 (An(Xn))

F -- + *v+ +?A
al a2 an
HereA represents
numbers.
wherea1,...., an arepositive theminimumvalueofE,
and theunknowns aredetermined fromtheequations
bybacksubstitution
Al(XI ... , Xn) = ,
A2(x2, ... , Xn) = 0

An(Xn)
= 0.
in Section3 above.
Thisis whatwe calledGauss'sTheorem
5. WhatwasJordan's MethodandNotation?On page 83 of thethirdeditionof
his Handbuchder Vermessungskunde [17],Jordanpresentsa numericalexample,
ingeodesy,
derivedfroma leastsquaresapplication themethodthathas
toillustrate
cometo be knownas Gauss-Jordan reduction.The particular he considers
system
wouldnowbe written as
{
17.50x- 6.50y- 6.50z = 2.14,
-6.50x + 17.50y - 6.50z = 13.96,
- 6.50x - 6.50y + 20.50z = - 5.40,
-2.14x - 13.96y + 5.40z = w - 100.34,
of
wherethe w in thelast lineis fromequation(2). However,sinceall systems
136 STEVEN C. ALTHOEN AND RENATE MCLAUGHLIN [February

normalequations are symmetric,

Jordanadopts an abbreviatedrepresentation:
17.50x - 6.50y - 6.50z - 2.14 = 0
+ 17.50y- 6.50z - 13.96 = 0
+ 20.50z + 5.40 = 0
+ 100.34,
where the number PI just "floats" at the lower right.The modernmethod for
solvingsystem(4) uses row operationsas follows.

17.50 - 6.50 - 6.50 2.14

- 6.50 17.50 - 6.50 13.96
- 6.50 -6.50 20.50 - 5.40
- 2.14 -13.96 5.40 -100.34
17.50 - 6.50 - 6.50 2.14
R2 + (6.5/17.5)R1 J0 15.09 -8.91 14.75
R3 + (6.5/17.5)R1 0 -8.91 18.09 -4.61
R4 + (2.14/17.5)R1 L0 -14.75 4.61 -100.081
R1 + (6.5/15.09)R2 17.50 0 -10.34 8.491
0 15.09 - 8.91 14.75
R3 + (8.91/15.09)R2 0 0 12.83 4.10
R4 + (14.75/15.09)R2 [0 0 -4.10 -85.66]
R1 + (10.34/12.83)R3 17.50 0 0 11.791
R2 + (8.91/12.83)R3 0 15.09 0 17.60
0 0 12.83 4.10
R4 + (4.10/12.83) R 3 0 0 0 -84.35]

x = 11.79/17.50 = 0.67
y = 17.60/15.09 = 1.17
z = 4.10/12.83 = 0.32
Total squarederror= 84.35

Jordan'spresentationuses thesame arithmetic in a different

arrangement. In the
firstplace, he takes advantage of the symmetry that remains as the reduction
size typeso thatnumbersto be added can be
proceeds. Secondly,he uses different
placed convenientlyone above the otherwithoutconfusion.Finally,to achievethis
convenience,the entriesabove the diagonal are placed to the rightof the matrix.
Here thenis Jordan'slayoutof his method.We have insertedlettersnextto several
of the numbersso thattheycan be identifiedin the notes that follow.Of course
theseletterswerenot includedby Jordan.
1987] GAUSS-JORDAN REDUCTION: A BRIEF HISTORY 137

a b c 1
+ 17.50 - 6.50 -6.50 -2.14
+ 17.50 -6.50 - 13.96
-2.41a -2.41a - 0.79a
+ 20.50 + 5.40
- 2.41a - 0.79a
+ 100.34
-0.26b

+ 15.09c - 8.91c - 14.75c - 6.50d

+ 18.09e + 4.61e -+.50f
- 5.26g - 8.71g -3.84h
+ 100.08i -2.14j
- 14.42k -6.35h

+ 12.831 -4.101 -10.34m -8.91n

+ 85.66o -8.49m - 14.75p
- 1.31q - 3.30r - 2.85s

+ 84.35t - 11.79u - 17.60v -4.10w

= [vv] -17.50 -15.09 - 12.83 neg.denominator
+0.67 + 1.17 +0.32
=x =y =z

6.5 2.14 4.10

(a) l R (i) R4+ -7R1 (q) 12.83R3
17.5 17.5
2.14 10.34
(b) 17R5 of
(i) (1,4)-entry (r) ,R
17.5 12.83
original
system
6.5 14.75 8.91
(c) R2+ 175R -- R2
(k) 15.09 (s)()12.833R
8.91 4.10
of
(d) (1, 2)-entry (I) R3? ~R (t) R4 +~~
15.09 12.83
originalsystem
6.5 6.5 10.34
(e) R3+ 17R (m) RI ? (u) R + R
15.09 12.83
8.91
of
(f) (1, 3)-entry after
(n) (2,3)-entry (v) R2 + R
originalsystem reduction
first
8.91 14.75
(g) ~ 2 (o) R4+ R (w) (3,4)-entryafter
15.09 15.09 secondreduction
6.5
(h) 1. 2 after
(p) (2,4)-entry
15.09 firstreduction
138 STEVEN C. ALTHOEN AND RENATE MCLAUGHLIN [February

To describeproblems usesandextendsthenotation
Jordan
in general, ofGauss.
He attempts, in thefashionofthetimes,
togiverecursive
formulae withoutthehelp
ofan indexor subscripts.Thus,theinputdataarelabeleda, b, c, etc.and 1.In [17,
p. 77] he writesthegeneralnormalequationsas:

[aa]x + [ab]y + [ac]z + [all = 0,

[bb]y + [bc]z + [bl] = 0,
[CC]z + [cl] - 0,
[ii].

is givenby thetableau[17,p. 82]

reduction
Then,Gauss-Jordan

[aa]3 [ab]o [ac]o [ad]O [al]O

[bb] [bc] [bd] [bl]
[cc] [cd] [cl]
[ddl [dl]
[Ull
[bb.1]3[bc.1]1[bd.l] [bl.1]1 [ab]o
[cc.1] [cd.1] [cl.1] [ac]o
[dd.1] [dl.1] [ad]o
[11.1] [al]O
[cc.2]3[cd.2]2 [cl.212 (ac.1) [bc.1]1
[dd.2] [dl.2] (ad.1) [bd.1]1
[11.2] (al.1) [bl.1],
[dd.3]3[dl.3]3 (ad.2) (bd.2) [cd.2]2
[11.3] (al.2) (bl.2) [cl.2]2
[11.4] (al.3) (bl.3) (cl.3) [dl.313 Numer.
[aa]3 [bb.1]3 [cc.2]3 [dd.3]3 Denom.
- x -y -z -t Quotient

by formulaethat yield the usual

The entriesin this tableauare determined
Gauss-Jordanreduction.Forexample,

[bb.1] = [bb] - [a] [ab],

(ac.1) = [ac] - [bc1 ab].

[bb.1]
1987] GAUSS-JORDAN REDUCTION: A BRIEF HISTORY 139

6. WhataboutClasen?In thequotation fromHouseholder givenabove,we find

theclaimthattheso-calledGauss-Jordanmethodseemstohavebeenfirst described
byClasenin [7].In [22],ThomasMuiralsowrites
aboutthissamearticlebyClasen:
In the solutionof a set of linearequationswitharithmetical coefficients
considerablelatitudeis availablein the choice of the seriesof derived
equationswhichis toendup withthevalueoftheunknowns. Thechoicemade
by the writer... is made withreal skilland deservesattention.

Clasen's name appears as "Clasen (abbe B.-I.), cure-doyend'Echternach

(Grand-Duche de Luxembourg)" in themembership listsofTheSocieteScientifique
de Bruxelles from1887-1901[2].His nameappearson thelistofdeceasedmembers
in 1902[3] and it wouldseemfromthehistorical volume[4,p. 62] thathe diedthat
year.He publishedonlyone articlein theAnnalesde la SocieteScientifique de
Bruxelles[4, pp. 121 and 234].In thissectionwe takea closerlookat thatarticle.
First,it is worthnotingthatClasengivesno references, and thenameof Gauss
(or Jordan)is nevermentioned. The onlyothermathematician in thearticleis Paul
Mansion,whoapparently Clasenthankshimin a brieffootnote.
servedas referee.
In today'snotation, Clasen'smethodappliedto thesystem

allxl + a12x2 + * +alnxn b

a21x1 + a22X2 + *- = b2
+a2nXn

anlxl + an2X2 + * +annxn=bn

becomes

all a .2 al,, b, m 0 af() ... b(l)

a21 a22 *-- a21, b2 0 m a111 ... b
a31 a32 a33 .
3

anl
. . a ,,, b,l ..
an anl, ant2 an3 .

m 0 a(f) a (1) * . b(l) R 0 0 a (2) ... b(2)

0 m a(1) a () ... b(l) 0 R 0 a(24) ... b(2)
0 0 R a 11) b) | 0 0 R a34) b'
l e
aR7l at72 aRZ 3 as74 *-- b, a,7lR3 R72 34 - , etc.
a,, til a a2n2 a, 0 ... b,, a,, a,, a n~4I . .
a 0~~~~~~~~, b
n4 til n2
140 STEVEN C. ALTHOEN AND RENATE MCLAUGHLIN [February

Here,
all a12
m = alla22 - a2la12 = a2l a22

and
all a12 a13
R = a2l a22 a23
a31 a32 a33

The arrowsindicatetheobviousrowoperations.
Clasen points out repeatedlythat his method does not involve division-a
definitesign of precalculatordays. He also insertsextraequationsas checks.
Clasen himselfdoes not use matrixnotation,ratherhe writesthe successively
occuringequations in a sequence that seems ratherlaborious.For example,if he
were to solve a 3 x 3 system,he would proceedas follows.

X1: allx + a12y + a13z + b, = 0

Y1: a2Ix + a22y + a23z + b2 = 0
Z1: a3Ix + a32y + a33z + b3 = 0

X1: allx + a12y +aI3z +b= 0

aj1Y1: ajja2Ix + ajja22Y +alla23z +a1lb2 = 0
-a2IXl: -a2lalIx - a2laI2y -a2lal3z -a2-lb = 0
Y2: my +(alla23 - a2lal3)z +allb2 - a2-b, = 0

mXl:malx +mal2y +ma13z +mb1 = 0

-a12Y2: -al2my - al2(alla23 - a2lal3)z -al2(a,,b, - a2-b,) = 0
malx +[mal3 - al2alla23 + a12 a2lal3]z +mb, - al2(allb2 - a2lb1) = 0
mallx +al,la3z +a , b' = 0
mx +at3z +bI = O

mZ,: ma3lx +ma32y + ma33z + nb3 = 0

- a32Y2: -aa32my - a32a'3z = 0
-a32b
- a3l: -a31mx -a3all3z - a3bl = 0
Rz +bI=0

RY2: Rmy +Ra3z +Rb =0

-at3Z3: -a'3Rz -a'3b = 0
Y3: Rmy +bI = ?

RX2: Rmx +Ra3z Rbt = 0

-at3Z3: -a'3Rz - a'3b = 0
X3:Rmx +bI'=O
1987] GAUSS-JORDAN REDUCTION: A BRIEF HISTORY 141

At the end of the article,Clasen emphasizes thathis methodrequiresfewer

calculationsthana methodusingdeterminants, and he showshowhismethodcan
be usedto computedeterminants.
The same volumeof theAnnalesde la SocieteScientifique de Bruxellesthat
containsClasen'sarticlealsohasa review byP. Mansion[20].Mansion(1844-1919)
was a well-known Belgianmathematician and,from1865,professor at theUniver-
sityof Ghent[1], [27]. Mansioncalls Clasen'smethod" the methodof equal
coefficients,"and he givesa fairlydetaileddescription of Clasen'swork.Mansion
also givesno referenceto Gaussor anyoneelse.
Later,in 1930,R. Mehmke[21]describes the"traditionaland stillcustomary"
methodof solvingsystems of linearequations-itis clearlyGaussianelimination,
althoughthe name Gauss is nevermentioned. Mehmkedoes observethatback
substitutioncan be troublesome, especiallyin largesystems.He statesthatClasen's
methodis superior to Gaussianelimination, eventhoughtheworldhas completely
ignoredtheformer.
The purposeofMehmke's articleis to remindthereadersofClasen'swork.But
Mehmkeintroduces a moreformalized notation (e.g.arraysofnumbers whereeach
variablehas a column)thatdoes not appearin Clasen'swork.To aid withthe
computations, Mehmkeevenusespaperstrips!Finally,Mehmke"improves" Cla-
sen'smethodby adaptingit to accelerated whereseveralvariablesare
elimination,
eliminated at eachstep.Mehmkementions neither GaussnorJordan.

7. Whataboutothermethods ofsolving linearsystems? In 1947,E. Bodewig[6,

p. 931]writesconcerning largesystems oflinear equationsthat noneoftheavairable
will
methods remove the inherent and
difficulties thatcalculating machinesneedto
be used. He continues, "The newly famous electroniccalculatingmachines solve
suchsystems automatically, so thatthemethoduseddoesnotmatter. However, the
priceof suchmachines is so highthatonlyone specimen of eachof thetwotypes
exists(or willeverexist).Everyone elsein theworldmustbe satisfied withtheusual
calculatingmachines forwhich themethod does matter."
The lack of adequatecalculators made everylittleshortcut important.Even
thoughthesolution of linear systems was understood in principle, processwas
the
unpleasant. Differentmethods fordealingwith thisunpleasantness werepopularat
varioustimes.At leasttwosurvey articles([6] and [14]) listby name manylong
forgotten methodsofsolving linearsystems.

REFERENCES
1. L'AcademieRoyaledes Sciences, et des Beaux-arts
des Lettres de Belgique, National
Biographie
Bruxelles,
Bruylant, 30 (1959)540-542.
11 (1887)16,35; 25 (1901)19.
de Bruxelles,
2. Annalesde la SocieteScientifique
26 (1902)46.
de Bruxelles,
3. Annalesde la SocieteScientifique
142 STEVEN C. ALTHOEN AND RENATE MCLAUGHLIN [February

4. Annales de la Societe Scientifiquede Bruxelles,Table Analytiquedes vingt-cinqpremiersvolumes

1875-1901, procedeede l'Histoiredocumentaire de la SocieteScientifiqueet de la Liste generalede
ses membres,pp. 62, 121, and 234.
5. D. Bartlett,General Principlesof theMethodof Least Squares,Boston,1915.
6. E. Bodewig,Berichtiuberdie verschiedenen Methodenzur Losung eines SystemslinearerGleichun-
gen mit reellen Koeffizienten, Koninkl. Akad. Wetensch.,50 (1947-48) 930-941, 1104-1116,
1285-1295, and 51 (1948) 53-64, 211-219.
7. B.-I. Clasen, Sur une nouvellemethodede resolutiondes equationslineaireset surl'applicationde
cette methode au calcul des determinants, Ann. Soc. Sci. Bruxelles,12 (1888) 251-281; also
Mathesis(Suppl. 2) 9 (1888) 1-31.
8. C. F. Gauss, Werke,SechsterBand, Disquisitiode ElementisellipticisPalladis ex Oppositionibus
Annorum1803, 1804, 1805, 1807, 1808, 1809, KoniglicheGesellschaftder Wissenschaften, Gottin-
gen (1874) 3-24.
9. C. F. Gauss, Abhandlungenzur Methodeder KleinstenQuadrate,Berlin,1887.
10. A. Gewirtz,H. Sitomer,and A. W. Tucker,Constructive Linear Algebra,Prentice-Hall, Englewood
Cliffs,1974.
11. Harbert,Ruhestaitte ffirWilhelmJordan,Zeitschrift ffirVermessungswesen, 85 (1960) 138.
12. Helmert,Wilhelm Jordan,ZeitschriftffirVermessungswesen, 28 (1899) 321-328. (The official
announcementof Jordan'sdeath(April 17, 1899) is on page 265 of thisvolume.)
13. A. S. Householder,The Theoryof Matricesin NumericalAnalysis,Blaisdell,New York, 1953.
14. H. Jensen,An Attemptat a SystematicClassificationof the Methods forthe Solutionof Normal
Equations, GeodaetiskInstitut,Meddelelse,No. 18 (1944).
15. W. W. Johnson,On Gauss's Methodof Substitution, Annalsof Mathematics,7 (1893) 65-69.
16. H. Jordan,WilhelmJordan,Lebensbildeines deutschenGeodaten,Zeitschrift ffirVermessungswe-
sen, 51 (1922) 577-585.
17. W. Jordan,Handbuch der Vermessungskunde, ErsterBand, Dritte verbesserteAuflage,Metzler,
Stuttgart, 1888.
18. K. Kunz, NumericalAnalysis,McGraw-Hill,New York, 1957.
19. 0. M. Leland, PracticalLeast Squares,McGraw-Hill,New York, 1921.
20. M. P. Mansion, Sur le memoirede M. l'abbe B.-I. Clasen,Ann. Soc. Sci. Bruxelles,12 (1888) 50-59.
21. R. Mehmke,Uber die zweckmiissigste Art,lineareGleichungendurchEliminationaufzulosen,Z.
Angew.Math. Mech., 10 (1930) 508-514.
22. T. Muir,The Theoryof Determinants in theHistoricalOrderof Development,Macmillan,London,
1906-1923; Dover reprint, New York, 1960.
23. C. Muller,Bucherschau,Zeitschrift ffirVermessungswesen, 51 (1922) 143-144.
24. A. NUsse,WilhelmJordan.Eine Erinnerung, AllgemeineVermessungsnachrichten, 54 (1942) 66-70.
25. J. C. Poggendorff, Biographisch-LiterarischesHandworterbuchder exaktenNaturwissenschaften,
Akademie-Verlag, Berlin7a (1969) 305.
26. C. Runge,WilhelmJordan,Zeitschrift furArchitektur und Ingenieurwesen 45, (1899) 265-66.
27. E. de Seyn,Dictionnairebiographiquedes Sciences,des Lettreset des Artsen Belgique,volume2,
L'Avenir,Bruxelles,1939,p. 713.
28. A. W. Tucker,Privatecommunication.