Coxeter Quaternions and Reflections AMM 1946
Coxeter Quaternions and Reflections AMM 1946
Coxeter Quaternions and Reflections AMM 1946
Author(s): H. S. M. Coxeter
Source: The American Mathematical Monthly, Vol. 53, No. 3 (Mar., 1946), pp. 136-146
Published by: Mathematical Association of America
Stable URL: http://www.jstor.org/stable/2304897
Accessed: 21/09/2010 01:01
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at
http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless
you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you
may use content in the JSTOR archive only for your personal, non-commercial use.
Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at
http://www.jstor.org/action/showPublisher?publisherCode=maa.
Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.
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
1. Introduction.It is just a hundred years since Cayley began to use quaternions for the discussion of rotations. He was followed by Boole, Donkin,
Clifford,Buchheim,Klein, Hurwitz, Hathaway, Stringham,and Study. Apparently none of these men thoughtof consideringfirstthe simpler operation of
and deducing a rotation as the product of two reflections.This proreflection
cedure will be described in ??3 and 5, and its consequences developed in the
later sections.
Every quaterniona =ao+a1i+aZj+a3k determinesa point P. = (ao, a,, a2, a3)
in euclidean 4-space, and a hyperplaneaoxo+a1xl+a2x2+a3x3 =0. The reflection
in that hyperplaneis found to be the transformationx---axa/Na. This leads
easily to the classical expression
x - axb
(Na = Nb = 1)
i2=
ijk
k2=
=-1,
which implyjk = =i-kj, ki=j =-ik, ij=k =-Ji. Thus quaternions forman
associative but non-commutativealgebra.
It is often convenient to split a quaternion into its "scalar" and "vector"
parts:t
a = Sa + Va,
Sa
ao,
Va
ao
ali
a2j
a3k
ad
a 2+
a2+
a2+
a3
136
137
Va= 2(a-a).
In termsof a and d, we have Sa=2(a+d),
=f, etc.,we easily verifythat
and i=-k
Since = -1-I
ab=
whence Nab =aab = baab = b(Na)b = NaNb. If Na = 1, we call a a unit quaternion. To every non-vanishingquaterniona therecorrespondsa unit quaternion*
Ua
a/N.
X3:+X3
X2
xy=-(xIyI+X2y2+x3y3)+
Y2
Y3
Y3
XI j+
Yi
Xl
yl
X212
Y2
we see that -Sxy and Vxy are the ordinary 'scalar product" and 'vector
product" of the two vectorsPoPe,and PoP11.If x and y are pure unit quaternions,
then Px and P, lie on the unit sphere around the origin Po, and we have
Z PxPoPv=O, where
cos
=+ yx).
Sxy=
-(xy
Thus the condition forP. and P, to lie in perpendiculardirectionsfromPo is
xy + yx = 0.
LEMMA 2.1.
ay =yy.
p(Sa)
By addition,
ap
- p=
0.
0.
138
AND REFLECTIONS
QTJATERNIONS
[March,
)b = a(b -)b,
Nb) = (ab
Na)b.
so we can
Proof.Since Npt = (Np) t, it will be sufficientto prove this fora unit quaternion, a. Since (Sa)2 - (Va)2 = Na =1, such a quaternion may be expressed as
a
cos a + p sin a,
wherep
pty
Since yxy= yy = -yxy, yxy is pure for any position of P,. Thus the linear
transformation
x-->yxy(where Ny = 1) representsa collineationwhich leaves invariant every point P. in the plane throughPo perpendicularto PoPV,i.e., the
plane
ylXl +
Y2X2 +
y3X3 =
0.
y.
* Two such quaternions satisfythe same "rank equation" x2- 2mx+n = 0, where m = Sa = Sb
1946]
139
in the plane of
But the only collineationhaving these propertiesis the reflection
invariant points. Hence
THEOREM
formation
x -yxy
-*
zyxyz.
- Syz.
-=ay-yd
= O,
(Na=
1)
\/
(i + j)(1-k)
2
We sum up in
THEOREM
(Pl, P2, p3)
cos 21 +
is anotherrotation,
Since the product of two rotations,x-*axd and x--*bxb,
viz., x->baxba, we can immediatelydeduce
THEOREM 3.3. All therotationsaboutlines through
theoriginin ordinaryspace
to thegroup of all unit quaternions.
forma group,homomorphic
140
[March,
x - ax4/Na, or x -+ axa-1.
(A homogeneous quaternion is the class of all scalar multiples of an ordinary
quaternion.) To make this an isomorphismrather than an anti-isomorphism,
we must agree to multiplygroup elements fromrightto left.
Combiningx->axd with the inversionx- -x, we obtain the transformation
x --axCa
(Na=1)
141
1946]
PZPy= (yo -
XO)
cos 0 = XOyo
+ xlyl + x2y2+ x3y3= Sx9 = 2(xY + yx).
Thus the condition for P. and Pv to lie in perpendiculardirectionsfromPo is
xy + yx = 0.
If Ny = 1, this condition may be expressedas
x = - yxy.
Thus the linear transformationx-* - yxy (where Ny =1) representsa collineation which leaves invariant every point P. in the hyperplane throughPo perpendicular to PoPS, i.e., the hyperplane
Y2X2 +
ylXl +
YOXO +
Y3X3 =
0.
THEOREM
yyy =-y.
Y-
Hence
bythe
in thehyperplane
5.1. The reflection
>y,x, =0 is represented
transformation
(Ny= 1).
x->-yxy
and x-?-zxz
(with
zyxyz.
E Y'x=
throughangle p,where cos
THEOREM
angleq5(abouta plane) is
through
5.2. Thegeneralrotation
x
->
axb,
j4.
a= zy,
b = yz.
142
[March,
-+ a2xI
xa2.
a2a.
* FelixKlein,Vorlesungen
uibernicht-Euklidische
Geometrie,
Berlin,1928,p. 240.
t WilliamThrelfalland HerbertSeifert,TopologischeUntersuchung
der DiskontinuitItsRaumes,Mathematische
bereicheendlicher
Bewegungsgruppen
des dreidimensionalen
spharischen
Annalen,vol. 104,1931,p. 10.
19461
143
(Na=Nb=1).
Of course,axb is the same as (- a)x(- b). With this exception,each displacement has a unique expressionx->axb. For, the equation axb =a'xb' would imply
a'Ilax -xb'b-l foreveryx, whence a'-'a =b'b-1 = ? 1. In otherwords,every displacement (with a fixedorigin) is the product of a left translationand a right
translationin just two ways. Thus the direct product of the groups of all left
translationsand of all righttranslationsis homomorphicto the groupof all fourdimensionaldisplacementspreservingthe origin(with two elementsof the direct
product foreach displacement), and this in turn is homomorphicto the direct
square of the group of all three-dimensionaldisplacementspreservingthe origin
Xwithtwo displacementsforeach element of the direct square).
8. The general orthogonaltransformationin four dimensions. The general
"opposite" or "sense-reversing"transformationleaving the origin invariant is
the product of an odd number of reflections.Hence, in four dimensions,it is
eithera single reflectionor a product of three. But in the latter case the three
reflectinghyperplanesintersectin a line of invariant points, and every hyperplane perpendicular to this line is invariant; so this scarcely differsfrom a
ro'tatory-reflection
in ordinaryspace. As such, it has an axis and a special reflectingplane. Its product with the special reflectionx-x- - is a displacement
itself must be x- >-axb, or, after changing
x->axb; so the rotatory-reflection
the sign of a,
x -> axb.
Since a a+-b b =a+b, the line of invariant points is
- (a - b), the axis is POPa_b.
We sum up our conclusion in
THEOREM
POPa+b.
Since a a-b b
8.1. Everyorthogonaltransformation
in four dimensionsis either
x -axb
or x-> axb.
144
(March,
where p and q are pure unit quaternions. This is the half-turnabout the plane
containingall points P. forwhich x =pxq, or
px + xq = 0.*
Since p - q and 1+pq are particular solutions of this equation for x, we may
describethe plane as PoPp_sP1+p.A rotationthrought7rabout the same plane is,
of course, x-*ptxqt. Replacing q by -q (= =q1), we deduce that x?ptxq-t
is a rotation through thrabout the completely orthogonal plane PoPp+Pl-p,.
(The fact that these two planes are completelyorthogonalis most easily verified
by observing that the product of the half-turnsx-*pxq and x- pxq-1 is the
inversionx->p2x= -x.)
about the respectiveplanes
Thus the productof rotationsthroughtr and uwx
PoPprqPi?pqis x+pt+uxqt-u. Setting tr-=a+13, ur==a-f, and observing that
cos a + P sina
=p2a/-
(see the proofof Lemma 2.3), we deduce that the product of rotations through
angles a?/3 about planes POPP,P1?ip is
x -*
In other words,
through
x->axbis thedoublerotation
9.1.1 Thegeneraldisplacement
THEOREM
aboutplanesPoPpi,Pi?pq, where
anglesa
cos a = Sa,
cos P = Sb,
p = UVa,
q = UVb.
The transformation
whosecoordinates(xO, X1l X2, x3) are homogeneous.
x -*
- yxy
is the reflectionin the polar plane of Pv, and this is the same as the inversion
in P, itself. We now say that the group of all displacements is precisely the
direct product of the groups of left and rightdisplacements; accordingly,it is
isomorphicto the direct square of the group of displacements preservingthe
origin. Instead of Theorem 9.1, we say that the general displacement x->axb
is the product of rotations through angles a? + about the respective lines
PpT-P,?,,, where
(1)
cosa = Sa,
cos, = Sb,
p = UVa,
q = UVb.
* IrvingStringham,
Transofplanesin parabolicspace of fourdimensions,
On thegeometry
actionsof theAmericanMathematicalSociety,vol. 2, 1901,p. 194.
de 1'espace,
et les divisionsr6guliRres
orthogonales
t tdouard Goursat,Sur les substitutions
de l'gcole NormaleSup6rieure(3), vol. 6, 1889,p. 36.
AnnalesScientifiques
19461
QUATERNIONS
AND REFLECTIONS
145
denotedby {p, q},* or equally well by { -p, -q}. Thus any two pure unit
a line {p, q}. The absolute polar line is { -p, q} or
determine
quaternions
{P, -q}.
(2)
Thus they are the two axes of this double rotation. Any point on eitheraxis
perpendiculars,are
(3)t
Jo ?
cosa
where
:|
-
Sp'p
--
Spp',
cosf8 =-Sqq';
(4)
(5)
+
? UVpp',UVqq'1.
is
It followsfrom(3) thatthe conditionfor {p, q} and {p', q'} to intersect
Spp' = Sqq',
and then the angle between them, being half the angle of the rotation (2), is
the conditionfor {p, q } and {p', q' } to be perpenarc cos (C+Spp'). Similarly,
(6)
Spp'+ Sqq'= 0.
(7)
SpP' = Sqq' = 0.
(8)
{UVpp', UVqq'},
where p' may range over all unit pure quaternions (except +p). To verify
* This notationwas usedby A. S. Hathaway,Quaternion
oftheAmerican
space,Transactions
of a
MathematicalSociety,vol. 3, 1902,p. 53. It is closelyassociatedwiththe representation
line in ellipticspace by an orderedpair of pointson a sphere;see Eduard Study,Beitriigezur
vol. 29, 1907,pp. 121-124.
AmericanJournalof Mathematics,
Geometrie,
nichteuklidische
t Here we are using (1) witha= -p'p, b= -qq'. Plus signswouldhave giventhe suppleto use the signthat makes
mentarydistances,whichare equallyvalid; but it seemspreferable
a and 63smallwhenp' and g' are nearlyequal to p and q.
146
this we merely have to observe that the line (8), which intersects {p, q } at
right angles, also intersects {p, q'} at right angles. The distance between
these left parallel lines, measured along any of the common perpendiculars,is
2arc cos (-Sqq').
Similarly,rightparallel lines {p, q } and {p', q } are distant 2 arc cos (- Spp')
along any of an infinityof common perpendiculars(8), only now it is q' that
can vary. Thus the common perpendicular lines of right parallels are left
parallel, and vice versa.
By the remarkat the beginningof ?9, the condition for the line {p, q } to
contain the point P. is
(9)
px+xq=O.
and {- xqx-1,q}.
namely{-x-1px}
Thus the set of all left (or right) parallels to a given line is an elliptic congruence: thereis just one memberof the set throughevery point of space.
If {p, q } contains P,,, its polar line lies in the polar plane of P. Replacing
q by - q, we deduce that the condition for the line {p, q} to lie in the plane
= Ois
EY,x1V
py= yq.
(10)*
Instead of insistingthat the p and q of the symbol {p, q } shall be unit pure
quaternions,we could just as well allow them to be any two pure quaternions
of equal norm.Then {p, q } is the same line as {Xp, Xq} forany non-zeroscalarX;
i.e., the two pure quaternions are "homogeneouscoordinates" for the line. Instead of (3) we must now write
cos a =-Spp'/v'}Npp,
cos 3
-Sqq'/VNN;
but instead of (4) we findthat the commonperpendicularsto {p, q } and {p', q'}
are simply { ?Vpp', Vqq'}, or
{Vpp', Vqq'} and { Vp'p, Vqq'}.
Formulas (5), (6), (7), (9), (10) remain valid, but (8) takes the simpler form
{ Vpp', Vqq' } .
To expressthe line PaPb in the form {p, q } we have to findpure quaternions
p and q satisfying
pa + aq = pb + bq = 0.
We may take p=ab -b4 and q=aZb-Da, or, halving these, p=Vab and q=VcVb.
Thus theline PaPb is {Vab, Vab }.
Similarly, the line of intersection of planes ,x = 0 and Eb,x, = 0 is
{Vab, Vba}.
* Hathaway,QuaternionSpace, p. 52.