Applied Mathematics and Computation 115 (2000) 7788

www.elsevier.com/locate/amc

Sedenions: algebra and analysis

K. Imaeda a, M. Imaeda

b,*

507-150, Yokoikami, Okayamashi, 701-1145, Japan

IFUP, 24201, Rue De Cezanne, Laguna Niguel, CA 92677, USA

Abstract
A 16-dimensional CayleyDickson algebra is presented. Its unique algebraic properties, its zero-divisors, and the solutions to a general linear equation are found. A
theory of function is developed in terms of the regularity (monogenicity) conditions and
some such functions are constructed. 2000 Elsevier Science Inc. All rights reserved.
Keywords: CayleyDickson algebra; Sedenions; Monogenicity; Quaternions Octonions

1. Introduction
What we mean here by a sedenion algebra is a non-commutative, non-associative, non-alternative, but power-associative 16-dimensional Cayley
Dickson algebra with a quadratic norm and whose elements are constructed
from real numbers, R, by iterations of CayleyDickson process [1,2], where 1
is chosen as the eld parameter at each step of the construction. In such process
2N -dimensional algebras of complex numbers, C (N 1), quaternions, Q
(N 2), octonions, O (N 3), sedenions, S (N 4), and other power-associative hypercomplex numbers are successively obtained.
The fact that sedenions do not necessarily constitute a composition nor a
division algebra, the reason for which can be traced to the non-associativity of
O, and possessing properties that are not obvious within our current physical
theories, may have been the motivation for introducing concepts and modications to achieve or eliminate some specic metric and algebraic properties.
Some parts of our work paralells that of other authors [35]; but unlike the

Corresponding author. E-mail: mari_imaeda@bigfoot.com

0096-3003/00/$ - see front matter 2000 Elsevier Science Inc. All rights reserved.
PII: S 0 0 9 6 - 3 0 0 3 ( 9 9 ) 0 0 1 4 0 - X

78

K. Imaeda, M. Imaeda / Appl. Math. Comput. 115 (2000) 7788

sedenions discussed in [3,4] our sedenions are non-alternative, non-modular

and pocess real norm. In [5] alternativity is restored by introducing ternary
products. Should Hurwitz theorem be one possible indication of non-conformity of sedenions to current mathematical methods in physics? It is clear from
Muses [6] and G
ursey [7], and other authors that some new algebraic and
analytical structures, such that might be oered by a 16-dimensional algebra,
may hold a key to better formulation of physical theories. Furthermore, its
relation to the Lie group G2 remains compelling under any circumstances. We
had found that the zero-divisors of sedenions are conned to some hypersurfaces and that the linear equations are solvable under certain conditions. Our
rst task is to list some basic algebraic properties of sedenions in Section 2.
Find the zero-divisors and the conditions for their existence, in Section 3. The
non-trivial task of solving a linear equation is treated in Section 4. We also
develop a theory of function of a sedenion variable based on Fueter's Moreratype regularity (monogenicity) conditions and construct intrinsic, polynomial,
and exponential functions, in Section 5, as the special cases of more general 2N dimensional theory studied earlier [8,9]. A brief account of octonions that we
refer to in the main text is found in Appendix A.

2. Basic algebra
In this paper we denote sedenions by S, T, Z, in bold capital letters, and
octonions by capital letters, A, B, X, Y, etc... We dene a sedenion,
S : A; B 2 S;

A; B 2 O;

as an ordered pair of two real octonions, A and B. The conjugation of S is then

 the conjugate of A, and of B by
dened in terms of A,
S : A; B:
 S, and that the automorphism is imAs with octonions this implies that S
proper. The product of two sedenions, S : A; B and T : C; D, under
CayleyDickson process, is
 BC DA:
ST A; BC; D : AC qDB;

We choose the eld parameter, q: 1.

In constructing the sedenion basis, el ; l 0; . . . ; 15, from the octonions
basis, il ; l 0; . . . ; 7, we choose the generator, e8 (0;1), and dene
el : il

l 0; 1; . . . ; 7;

e8l : il e8

i0 e0 1:

K. Imaeda, M. Imaeda / Appl. Math. Comput. 115 (2000) 7788

79

The multiplication of the two basal elements, ej and ek (the Greek (Latin) indices take some or all values of 0; 1; . . . ; 15 1; . . . ; 15), satises
ej ek djk ejkm em ;

where the value of the structure constant, ejkm , which is totally antisymmetric in
its indices, is given by the usual permutation rule for a 3-indexed antisymmetric
tensor with values 0 or 1, and determined by the appropriate 35 sedenion
cycles j; k; m. In terms of the basis a real sedenion, S, may be written componentwise as
S : s0

15
X
sj e j ;

sl 2 R;

j1

P15
where s : j1 sj ej is termed as the ``pure sedenion''. The structure constant,
ejkm , is associated with the non-commutativity of sedenions by
X
sj tk ejkm em ;
3
S; T : ST TS 2
jkm

where the summation is over all possible permutations of sedenion cycles j; k; m. The non-associativity of three sedenions, S, T, and U, is measured
by the non-zero (not necessarily zero) associator, (S, T, U) : (S T) U S(T
U).
Although the multiplication of basal elements is strictly alternative (true for
the basis of any 2N -dimensional CayleyDickson algebra with eld parameter
1) it is not necessarily true for arbitrary sedenions. For two sedenions, S and
T, the associator
 S; T
S; S; T A; B; D; A; B; C S;

is not necessarily zero. Furthermore, unlike octonions,

ej ek em ej em ek

m 6 k

does not always vanish. This translates to the fact that the second structure
constants, which deals with triple products of sedenions, are non-antisymmetric in its indices. The norm of a sedenion S, dened as
 SS
 nA nB
nS : SS

15
X
l0

s2l 2 R;

is a non-degenerate quadratic form associated with a symmetric bilinear form

1
S
T nS T nS nT:
2
Sedenions do not form a composition algebra. Indeed, n(ST) n(S) n(T) does
not vanish in general. Furthermore, sedenions do not constitute a division

80

K. Imaeda, M. Imaeda / Appl. Math. Comput. 115 (2000) 7788

algebra. For a sedenion S A; B, which satises certain norm and ``vectorclosure'' conditions, there exist a sedenion, Z, which is both left and right zerodivisor of S,
ZS 0 SZ:

3. Zero-divisors
One cannot solve a general linear equation of sedenion in a usual manner
due to the non-alternativity of the algebra. However, we can write a sedenion
as an ordered pair of octonions. We rst deal with the case of nding the zerodivisors of sedenions. For the right zero-divisor, Z, of S,
SZ 0:

Representing S and Z by A; B and X ; Y , respectively, where A; B; X ; Y are

non-zero octonions, for the non-trivial case, we have
AX Y B 0;

BX YA 0;

from which we obtain

A

1
B; X ; Y ;
nZ

1
Y ; A; B;
X
nS

1
A; X ; Y ;
nZ

1
Y
X ; A; B:
nS

We can now prove the following.

Proposition (Zero-divisor conditions). Eq. (6) implies the norm condition:
(I) nA nB, and nX nY .
Eq. (7) implies the following \vector-closure" conditions:
(II) A; B; X ; Y are pure octonions; thus the left and the right zero-divisors are
same for a given sedenion, S.
(III) Any product of A, B, X, Y, such as AB, AX, AY, are pure sedenions,
(IV) The triple products arising from A, B, X, Y, such as X(AB), Y(AB),
A(XY), and B(XY), are pure octonions.
(V) Non-vanishing of a determinant, as discussed below.
By suitable substitution and by using octonion properties (see Appendix A)
we obtain the following coupled equations from (7):
xX A; B X ; A; B 0;

xY A; B Y ; A; B 0;

K. Imaeda, M. Imaeda / Appl. Math. Comput. 115 (2000) 7788

81

where
2

x :

nS
2 R:
nA; B

One can show that x 1, which implies that Eq. (8) for X reduces to
XAB X BA 0:

Componentwise this equation is written as

7
X

k0j xj 0;

10

j1
7
X

!
x j a j br

j1

7
X

!
x k b k ar 0

8r 1; . . . ; 7;

k1

P
where k0j : km ak bm ejkm :
The vector-closure conditions for zero-divisors imply
x0 0;
AX 0

7
X

aj xj 0;

j1

X AB0

7
X

BX 0

7
X

bj xj 0;

11

j1

k0j xj 0;

j1

and thus Eq. (9) is automatically satised.

We proceed to solve for X in a usual manner: when aj and bj j 1; . . . ; 7
are given xj are determined only by the three independent equations (11). By
re-indexing the equations (for notational purposes only) so that x1 ; x2 ; x3 are the
three independent unknown, and xk 4 6 k 6 7 are the free parameters, such
that the determinant,


 k01 k02 k03 


D :  a1 a2 a3 
 b1 b2 b3 
does not vanish, we nd
 P
 f :p k0k xk

 P
1 
x1 D  f :p ak xk
 P
 f :p bk xk

k02
a2
b2


k03 

a3 :

b3 

82

K. Imaeda, M. Imaeda / Appl. Math. Comput. 115 (2000) 7788

Similarly for x2 and x3 . The corresponding solution for Y can now be obtained
using the appropriate equation or simply from
Y

X A; B
:
nS

We may dene a left or a right inverse of S as

S1 :


S
Z;
nS

however, this does not automatically imply that S1 1 S.

4. Linear equations
We now solve for Z, in terms of S and T, in a linear equation,
SZ A; BX ; Y T;

A 6 0 6 B:

 : F ; G, we
Using the identity (4) and dening octonions, F and G, by ST
obtain
nSX A; B; Y F ;

12

nSY A; B; X G:
By substitution we have for X,
13

xX A; B A; B; X 2K;
where
K :

nSF A; B; G A; B
;
2nA; B

x :

nS
:
nA; B

Writing this equation componentwise we have the following simultaneous

equations:
7
X
k0j xj k0 ;
x
j1
14
7
X
X
kjk xj 1 x
ejkm k0m xj kk ;
xk0k x0
j1

jm

where we had used the fact that for each xed j and k,
X
X
ap bq ejkpq
k0m ejkm kjk ; kjk : aj ; bk :
pq

K. Imaeda, M. Imaeda / Appl. Math. Comput. 115 (2000) 7788

83

These equations may be written in a matrix form as

alm xm kl ;

15

where alm is a skew-symmetric matrix with

all 0;

a0j xk0j aj0 ;

which satises
X
ajk a0k 0;

ajk kjk 1 xejkm k0m ;

j 1; . . . ; 7:

Both xm and kl are column matrices. Through a tedious but straightforward

calculation we obtain the determinant
!4
7
X
4
2
k0j :
16
jalm j x4 1 x
j1

From Eq. (16) we have the following cases:

Case 1: x 1.
1. K 0: then X 0 ) Z 0.
2. K 0: the solution for xl l 0; . . . ; 7 is uniquely determined by A; B; F
and G.
Case 2: x 1.
1. K 0: the case is that of the zero-divisor discussed in Section 3.
2. K 0: from Eq. (14) we have
k0k x0 bk v ak w kk ;
where
7
X
aj x j ;
v :
j1

w :

7
X
j1

17
bj x j ;

7
X

k0j xj k0 :

j1

The equations are now solved in a conventional manner as follows. Note that
the seven equations of (17) are imposed on x0 , v, and w, and that the rank of D
and M, where
0
1
0
1
k01 a1 b1 k1
k01 a1 b1
M : @ k
02 a
2 b
2 k
2 A;
L : @ k
02 a
2 b
2 A;
k07 a7 b7
k07 a7 b7 k7
should be equal to or be less than three.
(i) The rank of L is three: by re-indexing the ak and bk (strictly for notational
purposes only), so that

 k01

D  k02
 k03

a1
a2
a3


b1 
b2  6 0;
b3 

84

K. Imaeda, M. Imaeda / Appl. Math. Comput. 115 (2000) 7788

we nd
x0

D1
;
D

D2
;
D

D3
;
D

where Dj j 1; 2; 3 is the determinant in which the three entries of the jth

column in D are replaced by k1 , k2 and k3; respectively. Now to solve for xk we
have the following system of equations:
7
X

k0k xk k0 ;

k1
7
X

D2
;
D

aj x j

j1

7
X

bj x j

j1

D3
:
D

The solution is given by,

xj

Dj
D

8j 1; 2; 3;

where


 k01

D :  a1
 b1

k02
a2
b2


k03 
a3 ;
b3 

and Dj is the determinant D in which the three entrices in the jth column are
replaced by
7
7
7
X
D2 X
D3 X

k0j xj ;
aj x j ;
bj x j ;
k0
D
D
j1
j1
j1
where xj j 4; . . . ; 7 denote the free parameters.
(ii) The rank of L is two: we consider B aA; a 2 R, and conclude from
Eq. (12) that
X

F
;
nS

G
:
nS

Thus, Z is uniquely determined by S as

Z


ST
:
nS

5. Theory of functions
As found by many authors both past and present the usual Cauchy
Riemann type dierentiability conditions, when extended to functions of a
hypercomplex variable, are too restrictive to yield interesting, if not useful,

K. Imaeda, M. Imaeda / Appl. Math. Comput. 115 (2000) 7788

85

functions. Our path is to follow one of the most successful approach in constructing functions, that of regular functions; a theory originally developed for
quaternions by Fueter [10], which was then rened, generalized, and extended
to the cases of a variable in other CayleyDickson algebras [1116] as well as in
Cliord algebras, and which are successfully used in physics by authors too
many to list them here. We dene the regularity (monogenicity) condition
for sedenions in terms of the variation of an integral of a function in a 16dimensional
Euclidean space. For a function, F(Z), of a sedenion variable,
P15
Z: l0 zl el ; zl 2 R, consider an integral over a closed hypersurface S15 ,
Z
15
dZ F Z:
J S :
S 15

The surface element dZ can be written in terms of the dierentials, dzl , or of

the directional cosines, cl , and the volume element, dr, of S15 as
dZ

15
X

ozl el

l0

15
X
l0

cl drel :

Using Stoke's theorem we can write J[S15 ] as a volume integral over V16 enclosed in S15 as
Z
DF ZdV ;
J S 15
V 16

where
D :

15
X

ol e l ;

l0

dV :

15
Y

dzl :

l0

to S15
in
Now consider an innitesimal variation of the surface S15
A
B
15
the neighbourhood of a point Z on SA . The volume enclosed by the two
surfaces is
Z
dV 16 2 R:
dV :
SA SB

Dene
dJ S
J SA15 J SB15
: lim
lim
dV !0
dV !0
dS
dV

R
dV

DF dV
dV

DFZ:

Then the function F(Z) is left regular at Z0 on S15 if J[S] is stationary under the
innitesimal variation of the surface at Z0 ,



dJ S 


DFZ
0:



dS ZZ0
ZZ0

86

K. Imaeda, M. Imaeda / Appl. Math. Comput. 115 (2000) 7788

The denition is independent of the choice of S15 for DF(Z) 0. The right D
regularity as well as D-regularity
conditions can be dened in a similar manner.
We now outline the construction of some regular functions of a sedenion
variable.
(I) Intrinsic functions: Starting with an analytic function,
gz ux0 ; x ivx0 ; x;
of a complex variable, z x0 ix, and replacing i
P15
j1 xj ej
;
i)
x

p
1 by a pure sedenion

where

v
u 15
uX
x2j ;
x : t
j1

the function assumes the form

P15
j1 vx0 ; xxj ej
;
GX ux0 ; x
x
where X is a sedenion variable. Then the function
FX : 7 GX;


 : DD

15
X
l0

o2l ;

is left-regular.
(II) Polynomial functions: Consider a function
"
#n
15
X
XT TXn

x0 ej xj tj ;
Fn X; T
2
j1
where
X :

15
X

xl el ;

l0

T :

15
X

tj ej 2 S:

j1

The power associativity ensures without ambiguity that Fn (X, T) can be written
as a power series in tk ,
X
n15
n!Pn1 n2 . . . n15 x0 ; x1 ; . . . ; x15 t1n1 t2n2 . . . t15
;
Fn X; T
P15

nr

nj , and Pn1 n2 ...n15 is a polynomial function of a sedenion variable

1 X
x0 ek1 xk1 . . . x0 ekn xkn :
Pn1 n2 ...n15 x0 ; x1 ; . . . ; x15
n! k

where n :

j1

K. Imaeda, M. Imaeda / Appl. Math. Comput. 115 (2000) 7788

87

The summation is over all possible permutations of the series kr

k1 ; . . . ; kn ; fk1 ; . . . ; kn g f1; 2; . . . ; 15g, and the number of times each of
1; 2; . . . ; 15 appears in the series is n1 ; n2 ; . . . ; n15 , respectively. It can be easily
shown that Fn (X, T) and, henceforth, the polynomials Pn1 n2 ...n15 , are both side
D-regular.
(III) Exponential
functions: By again considering the hypercomplex variable
P15
FX : j1 x0 ej xj , we may dene exponential function of S as
n

ExpiFX; T :

i n
F X; T:
n!

This function is both side D-regular.

(IV) Cauchy-type kernel function: For a function
H X :

X
;
nX

of a variable X, in a 2N ( m)-dimensional CayleyDickson algebra, one can

readily show that,
n HX
where

Cn HX
;
nXn

Cn 1 2n n!

n 2 Z ;

n
Y
m 2k;
k1


 : DD

15
X
o2
;
ox2l
l0

and that  HX is a both side regular function of an m-dimensional Cayley

Dickson algebra if n m=2 1. Thus, one can construct a Cauchy-type kernel
function of a sedenion variable which is both side regular by choosing n 7.
However, a general integral formula for sedenions based on this Cauchy-type
kernel function cannot be constructed due to the non-antisymmetric nature of
the second structure constant.
Acknowledgements
The author K. Imaeda would like to thank Mr. Susumu Ohta for his
stimulating talks on the relations of hypercomplex numbers to the theories of
superstrings and supergravity.
Appendix A. Octonions
We now list octonion properties and relations that are relevant to our main
work. Let A, B, C 2 O. Componentwise the octonion A, the conjugate, the
norm, and the inverse may be written, respectively, as

88

K. Imaeda, M. Imaeda / Appl. Math. Comput. 115 (2000) 7788

A : a0

7
X

aj ij ;

A : a0

15
X

j1


nA : AA AA

aj i j ;

j1
7
X
l0

a2l ;

A:1

: A=nA;

where the multiplication of basal elements, ij j 1; . . . 7, is governed by the

same relations as that of the sedenion basis, ej j 1; . . . ; 7, dened in Eq. (2).
The non-zero (not necessarily zero) associator, which is a pure octonion, is
dened by
X
aj bk cm ejkmn in ;
A; B; C : ABC ABC 2
jkmn

where the second structure constant, ejkmn , is antisymmetric in its indices, and is
determined from the seven distinct associator cycles, j; k; m; n by the usual 4indexed tensor rule: ejkmn (+1, 1, or 0), depending on the (even, odd, or
otherwise) permutation of the octonion associator cycle.
The following identities hold for octonion associators:
 B; C B; A; C;



A; B; C A;
A; B; A; B; C A; BA; B; C A; B; CA; B;
A; B : AB BA;
A; AB; C A; B; CA:

A:2

References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]

R.D. Schafer, An Introduction to Nonassociative Algebras, Academic Press, New York, 1966.
L.E. Dickson, J. de Math. Pures et Appliq. 2 (1923) 281.
K. Carmody, Appl. Math. Comput. 84 (1997) 27.
K. Carmody, Appl. Math. Comput. 28 (1988) 47.
L. Sorgsepp, J. L~
ohmus, J. Hadronic 4 (1981) 327.
C. Muses, Appl. Math. Comput. 60 (1994) 25.
F. G
ursey, in: Y. Ne'eman (Ed.), To Fulll a Vision, Addison-Wesley, Reading, MA, 1981.
M. Imaeda, in: J.S.R. Chisholm, A.K. Common (Eds.), Cliord Algebras and their
Applications in Mathematical Physics, Reidel, Dordrecht, 1986.
K. Imaeda, M. Imaeda, Bull. Okayama Univ. Sci. 20A (1985) 133.
R. Fueter, Comm. Math. Helvetici. 7 (1935) 307.
K. Imaeda, H. Tachibana, M. Imaeda, S. Ohta, Nuovo Cim. 100B (1987) 53.
A. Sudbery, Math. Proc. Camb. Phil. Soc. 85 (1979) 199.
K. Imaeda, Nuovo Cim. 32 (1976) 138.
K. Imaeda, Memoirs of the Faculty of Liberal Arts and Education 8 (1957) 131.
K. Imaeda, Memoirs of the Faculty of Liberal Arts and Education 2 (1951) 111.
P. Dentoni, M. Sce, Rend. Sem. Mat. Univ. Padova 50 (1973) 251.