Functions of multivectors in 3D Euclidean geometric algebra via

spectral decomposition (for physicists and engineers)

Miroslav Josipović

[email protected]
December, 2015

Geometric algebra is a powerful mathematical tool for description of physical

phenomena. The article [3] gives a thorough analysis of functions of multivectors in Cl3
relaying on involutions, especially Clifford conjugation and complex structure of algebra.
Here is discussed another elegant way to do that, relaying on complex structure and
idempotents of algebra. Implementation of Cl3 using ordinary complex algebra is briefly
discussed.

Keywords: function of multivector, idempotent, nilpotent, spectral decomposition, unipodal

numbers, geometric algebra, Clifford conjugation, multivector amplitude, bilinear
transformations

1. Numbers

Geometric algebra is a promising platform for mathematical analysis of physical

phenomena. The simplicity and naturalness of the initial assumptions and the possibility of
formulation of (all?) main physical theories with the same mathematical language imposes the
need for a serious study of this beautiful mathematical structure. Many authors have made
significant contributions and there is some surprising conclusions. Important one is certainly
the possibility of natural defining Minkowski metrics within Euclidean 3D space without
introduction of negative signature, that is, without defining time as the fourth dimension ([1,
6]).

In Euclidean 3D space we define orthogonal unit vectors e1 , e2 , e3 with the property

ei2  1 , ei e j  e j ei  0 ,

so one could recognize the rule for multiplication of Pauli matrices. Non-commutative
product of two vectors is ab  a  b  a  b , sum of symmetric (inner product) and anti-
symmetric part (wedge product). Each element of the algebra (Cl3) can be expressed as linear
combination of elements of 23 – dimensional basis (Clifford basis)
1, e1 , e2 , e3 , e1e2 , e3e1, e2e3 , e1e2e3 ,
where we have a scalar, three vectors, three bivectors and pseudoscalar. According to the
number of unit vectors in the product we are talking about odd or even elements. If we define

1
j  e1e2e3 it is easy to show that pseudoscalar j has two interesting properties in Cl3: 1)
j 2  1 , 2) jX  Xj , for any element X of algebra, and behaves like an ordinary imaginary
unit, which enables as to study a rich complex structure of Cl3. This property we have for n =
3, 7, … [3]. Bivectors can be expressed as product of unit pseudoscalar and vectors, jv .
We define a general element of the algebra (multivector)
M  t  x  jn  jb  z  F , z  t  jb, F  x  jn

where z is a complex scalar and an element of center of algebra, while F , by analogy, is a

complex vector. Complex conjugation ( j  ) is z   z †  t  jb, F   F †  x  jn , where
dagger means reversion, to be defined later in the text. The complex structure allows a
different ways of expressing multivectors, one is
M  t  x  jn  jb  t  jn  j  b  jx  ,

where multivector of the form a  vjv̂ belongs to the even part of the algebra and can be
associated with rotations, spinors or quaternions. Also we could treat multivector as ([12])
3
M   0   i ei , k  and implement it relying on ordinary complex numbers.
i 1

Main involutions in Clifford algebra are:

1) grade involution: M̂  t  x  jn  jb
2) reverse (adjoint): M †  t  x  jn  jb  z   F 
3) Clifford conjugation: M  t  x  jn  jb  z  F  z  F ,
where an asterisk stands for a complex conjugate. Grade involution is the transformation
xˆ   x (space inversion), while reverse in Cl3 is like a complex conjugation,
x  x, j   j .
† †
Clifford conjugation is combination of two involutions
M  Mˆ † , x   x, j  j . Bivectors given as a wedge product could be expressed as
x  y  jx  y , where x  y is a cross product. An application of involutions is easy now.

Defining paravector p  t  x we have pp  t  x   t  x  t  x   t 2  x 2 and we

2

have a usual metric of special relativity.

From M  Mˆ  M  t  jn , the even part of the algebra (spinors).
 
†
From M  M †  M  t  x , paravector; reverse is an anti-automorphism MM †  MM † ,
so MM † (square of multivector magnitude, [2]) is a paravector.
From M  M  M  t  jb  z , a complex scalar. Clifford conjugation is anti-automorphism,
MM  MM , so MM (square of multivector amplitude, [2]) is a complex scalar and there is
no other “amplitude” with such a property ([1]).
We define a multivector amplitude M (hereinafter MA)

MM  M  t 2  x 2  n2  b2  2 j  tb  x  n  , M 
2
(3)
which we could express as

2
 MM  M   z  F  z  F   z 2  F 2 , F 2  x 2  n2  2 jx  n  .

For F  0, or F 2  N 2  0 fallows M  z ( N 2  x 2  n 2 , x  n  0 is a nilpotent in the

algebra). For F 2  c  ( x  n  0 , whirl [1]) here it is used a designation F c  . From

F2  we have F1  F / F 2 , F12  1, and Fˆ  F / F 2  F / F   jF1 , Fˆ 2  1

(complex unit vector). With f  F1 we also define u  1  f  / 2, u 2  u , u u  0 ,
idempotents of the algebra ([1, 9]). Every idempotent in Cl3 can be expressed as
u  p  N  , N 2  0 , where p are simple idempotents, like p  1  e1  / 2. For example,
u  1  e1  e2  je3  / 2, N  e2  je3 (figure below).

Implementation

3
From M   0   i ei   0  A,  k  , it is
i 1
easy to implement the algebra on computer using
ordinary complex numbers only. In [15] are
defined products:
3
A B  i i (generalized inner product),
i 1

A  B  det  ei ,  i , i  , (generalized outer

product), and

AB  A B  A  B (generalized geometric
product).

Now we have
 A  A B  B    A B   B A   A B  AB .
We can find  i for multivector
M  t  x  jn  jb using linear independency.

Spectral decomposition

Starting from multivector

F(1)2  1

"whirl" M  z  F  z  F 2 f  x  yf , F 2  0
we see the form of unipodal-like numbers. Defining
M   x  y and recalling a relation (easy to proof)
fu  u follows
Mu   x  yf  u   x  y  u  M  u ,

3
so we have a projection. Part of complete spectral basis u  is very useful because the
binomial expansion of a multivector is very simple

M 2   M u  M u   M 2u  M 2u  M n  M nu  M nu , n  ,

2

where n < 0 is also possible for M  0 ( M 1  M /  MM  ).

Defining conjugation  a  bf   a  bf (obviously the Clifford conjugation) we have



 a  bf   a  bf   a 2  b2 , where

a 2  b2  MM   MM is a multivector amplitude. In
spectral basis using u  u we have MM    M  u  M  u  M  u  M  u   M  M  .

Starting from M  z  F we have

x y 
M  z  F 2 f  x  yf     f     cosh   f sinh   ,   x 2  y 2 ,
  
obtaining the polar form of a multivector. If the multivector amplitude is zero we have light-
like multivector and there is no a polar form. Now defining tanh    (“velocity”) we have

M    cosh   f sinh     1   f  ,  1  1   2

and in the spectral basis

M   1   f   k u  k u  k   1      K 1 ,

where K  1    / 1    is a generalized Bondi factor (   log K ). Now we have

K u 1   K11u  K1u  K11u   K1 K2 u  K11 K2 1u  Ku  K 1u  K  K1 K 2 ,
or
 1 2 1  1 f 1  2 f    1 2 1  1  2  f  12  
 1 2 1  12  1  1  2  / 1  12  f  
   1 2 1  12  ,   1  2  / 1  12  ,

and we have “velocity addition rule”. So, every multivector could be mathematically treated
like an ordinary boost in special relativity. For   1 we have a “boost”
 1   f   Ku  K 1u as transformation that preserves multivector amplitude and should
be possible considered as the extension of the Lorentz group ([1, 2]). As a simple example of
a unit complex vector we already mentioned f  e1  e2  je3  e1  e2  e1e2 , completely in
Cl2, suggesting that one could analyze problem in basis 1, F(1)  or related spectral basis in
e1e2 plane and rotate all elements to obtain relations for an arbitrary orientation of a plane,
using powerful apparatus of geometric algebra for rotations.
Mapping basis (1, f ) to (e f , fe f ) we obtain a new orthogonal basis and a new
components of multivector
a  bf  a e f  b fe f  (a   b f )e f ,
a   ae  f , b  be  f , a  bf  a   b f .

4
 2. Functions of multivectors

Using series expansion it is straight forward to find a closed formulae for (analytic at
least) functions. If F  0 we have f  M   f  z  and it is easy to find closed form using the
theory of functions on the complex field. Otherwise, from the series expansion
f  n  0  xn
f  x   f  0  
n n!
using M  M u  M u we have
n n
 
n
 

f  M   f  M   u  f  M   u 

and, again, it is “easy” to find a closed form because of M   . For M  F  F 2 f we

have M    F 2  f  M   f    
F 2 u  f  F 2 u . If a function is even we have

f F   f  F2  u  
 u   f F2  and similarly for the odd functions

f F   f  F2  u  u   f 
 F2  f . For M  z  F , F 2  N 2  0 there is no a

spectral decomposition ( f is not defined), but we have M n   z  N   z n  nz n1 N , giving

n

f  z  N   f  z   f   z  N . We also have the special cases

f  u   f  1 u ,
f  f   f  u  u   f 1 u  f  1 u ,

 
f Fˆ  f   ju  ju   f   j  u  f  j  u .

 
Obviously, for an odd function we have f  f   f 1 f , f Fˆ   f  j f and for the even

 
functions f  f   f 1 , f Fˆ  f  j  f .

For the inverse function we have

f 1  y   x  f  x   y  f  x   y  x  f 1  y  .

For a light-like multivectors ( MM  0 ) we have

M  z  F f =z+z F f , z 2 - F 2  0   z-z F  z+z F  ,

and two possibilities:

1) z  zF  M   2 zF , M   0  f  M   f  2 zF  u
2) z   zF  M   0, M   2 zF  f  M   f  2 zF  u

Once a spectral decomposition of a function is analyzed there remains just to use the
well known properties of functions of a complex variables.

5
Examples

Here we (again) define M  z  F  z  F 2 f  z  zF f and

M   z  zF , M   z  zF .

For the inverse of M we have ( M  0 )

1 M  u  M  u M u  M  u u u
M 1          ,
M  u  M  u  M  u  M  u  M  u  M  u  MM M M

as expected. Now it is obvious that

1 u u
M n    .
 M  u  M  u  M  M 
n n n

We can find square root using

M  S  S u  S u  M  M  u  M  u   S  u   S  u  S   M  .
2 2

So, generally we have M 1/ n  S  S   M  

1/ n
, n . As a simple example (an
interested reader could compare with [3]) ei    j  ei  / 2 j .

The exponential function is easy one, eM  eM u  eM u and now we have
exponentials of complex numbers (just use ordinary i  1 and replace i  j at the end).
Logarithm is the inverse function to exponential, so we have
log M  X  e X  M  M  u  M  u  exp  X   u  exp  X   u  X   log M  .
In [3] is derived formula log M  log M   Fˆ ,   arctan  F / z  , but those are equivalent:

zF  F 2   j F , Fˆ   jf 
log  M    log  M   log  M    log  M  
log  M   u  log  M   u   f 
2 2
log M  jFˆ log  
1  j  F / z  / 1  j  F / z   log M  Fˆ arctan  F / z   log M   Fˆ .

Now we can find, for a  , M a  X  log X  a log M  X  e a log M ,

but the same appears to be correct for a  z  F and one can find, for example,
M v  X  log X  v log M  X  ev log M ,

although here some caution is needed because of possibility log X   log M  v 

X  e log M v . Also, relation M  elog M is generally not valid and needs some care due to the
multivalued nature of the logarithm operation. Nevertheless, expressions like
j e1  exp  e1 log j   exp  j e1 / 2   je1 , or  e1 e2  3  j are quite possible in Cl3. Simple
e

examples ( u  1  e1  / 2 ):

6
 1. e1e1  X  e1 log e1  log X ,
e1  u  u  log e1  u log1  u log  1  j u 
e1 log e1   j u  X  exp   j u   exp   j  u  u
(solution e1 is not valid because of e1 log e1   j u   log e1 ).
2. e1e2  X  log X  e2 log e1  j e2u ,
but e2 u e2 u  e2 e2 u u  0 , so X  exp  j e2 u   1  j e2 u ,
or log X   log e1  e2  j u e2  X  1  j e2 u ,
and finally X  1  j e2 u and X n  1  jn e2 u , n 
(solution 1 is not valid because of e2 log e1  0 , it is a nilpotent).
Trigonometric and hyperbolic trigonometric functions are straightforward and ctg one
could obtain as inverse of tan. For example

sin F  sin F  u  sin   F  u  sin  F  u  u   F sin  F  ,

2

2

2
  1
2

cos F  cos  F  u  cos   F  u  cos  F   u  u   cos  F  .

2

2

2
 
2

For F  N there is no spectral decomposition, but using exp(zN) = 1+ zN we have

sin N  N , cos N  1, sinh N  N , cosh N  1 . There is no N-1, so ctgN is not defined.
Series in powers of argument are crucial for all analytic functions (in geometric
algebra too) and we can use presented spectral decomposition to obtain components of such
functions in spectral basis.

Conclusion

Geometric algebra of Euclidean 3D space (Cl3) is really rich in structure and gives the
possibility to analyze functions defined on multivectors, extending thus theory of functions of
real and complex variables, providing intuitive geometrical interpretation also. From simple
fact that for a complex vector ( F 2  0 ) we can write F / F 2  f , f 2  1, F 2  follows
nice possibility to explore idempotent structure u  1  f  /2 and spectral decomposition of
multivectors. Using the orthogonality of the spectral basis vectors (part of it, idempotents) u 
it is shown that all multivectors ([1]) can be treated as the unipodal numbers (i.e.
hypercomplex numbers over a complex field). A definition of functions is then quite simple
and natural and strongly counts on the theory of functions of complex variable. Complex
numbers and vectors (bivectors, trivectors) are thus united in the promising system.

7
Appendix

A1. Bilinear transformations

Regarding that bilinear transformations of multivectors do not change some property

of multivectors one could ask yourself: what property? In [2] was shown that the multivector
amplitude, defined using the Clifford conjugation, is the unique involution that is
commutative ( MM  MM ) and belongs to the center of the algebra. From text we see that
multivector amplitude is defined as M  M   using just natural conjugation
a  bf  a  bf , where f is our “hypercomplex unit”. But this is just the Clifford
conjugation and we see now new meaning for it: it is just a “hypercomplex conjugation”. This
is a strong argument to regard a restricted Lorentz transformations to be a subgroup of bilinear
transformations that preserve multivector amplitude. It is verified on paravectors giving the
known special relativity, but now we should extend it on a whole multivector including in this
way all multivector symmetries of Euclidean 3D space. For multivector X (transformation)
now we have X  X   1 and

X  e M  M  log X  log X   Fˆ  log1   Fˆ   Fˆ  F ,

giving thus a general bilinear (twelve parameters) transformation M   e p  jq Me r  js .

We started from a new definition of a vector multiplication (Clifford or geometric

product) and we see that 3D Euclidean space has a rich complex and hyper-complex structure.
The Clifford conjugation is the natural choice to define a multivector amplitude. A bilinear
transformations that preserve multivector amplitude form a group containing the ordinary
Lorentz transformations (restricted), but it is really natural to assume that the physical reality
should be richer. Some consequences of that assumption are discussed in [1] and [2].

A2. Hyperbolic inner and outer products

Given two multivectors M 1  z1  z1F f and M 2  z2  z2 F f we define a square of

multivector distance (conjugate products, [13]) as
M1 M 2   z1  z1F f  z2  z2 F f   z1 z2  z1F z2 F   z1 z2 F  z2 z1F  f  hi  ho f ,

where hi and ho stands for hyperbolic inner and hyperbolic outer products. If M1  M 2  M
we have M  M  z 2  zF2   zzF  zzF  f  z 2  zF2 , just a square of multivector amplitude.
This suggests that hi and ho have to do something about being “parallel” or “orthogonal”,
besides being “near” and “close”. For complex and hypercomplex plane (with real
coordinates) meaning is obvious (fig A2).

8
 With ho = 0 multivectors are said to be
“h-parallel”, while for hi = 0 they are
“h-orthogonal”. For hi = ho = 0
multivector distance is null and we said
it to be “h-light-like”, where h- stands
for hyperbolic.
In “boost” formalism we have
hi  12 1 2 1  12  ,
ho  12 1 2 2  1  / 1  
1 2 .

So, “h-parallel” multivectors have

h-orthogonal h-parallel equal “velocities” and hi  1 2 , while
for “h-orthogonal” multivectors
fig A2: Yellow line represents h-numbers parallel “velocities” are reciprocal and ho
or orthogonal to given h-number (red dot). becomes infinite (orthogonal
multivectors belong to different hyper-
quadrants delimited by light-like hyper-
planes).
Lema: Let M1 M 2  0 for M 1  0 and M 2  0 .Then M 1 M 2  0 , and vice versa.

M1 M 2   M1 u  M1 u  M 2 u  M 2 u   M1 M 2 u  M1 M 2 u ,

M1 M 2   M1 u  M1 u  M 2 u  M 2 u   M1 M 2 u  M1 M 2 u ,
and we have M 1 M 2  0 or M 1 M 2  0 , which means M1  0 and M 2  0 or
M1  0 and M 2  0 , but either case gives M 1 M 2  0 . Converse proof is similar.

A3. Polynomials

Suppose we have simple equation M 2  1  0 , then objects that squares to -1 are

solutions. Using spectral decomposition we could explore it further, so,
M 2  1   M  u  M  u   u  u   M 2  1 u   M 2  1 u  0 
2

M 2  1  0, M 2  1  0 , so we have two equations over complex numbers. Obvious solutions

are 1, j, jei , Fˆ , but it is possible to investigate further. There is an infinite number of
solutions, obviously, due to algebraically expanded paradigm of number.
Another simple equation is M 2  M 2u  M 2u  0 . One obvious solution is M  N
(nilpotent) which we cannot obtain using spectral decomposition (because there is no one for
a multivector N ), so, some caution is necessary.
In physics we are using a lot of special polynomials and their roots and we probably
should reconsider those in geometric algebra.

9
References

[1] Josipović, Some remarks on Cl3 and Lorentz transformations, vixra: 1507.0045v1
[2] Chappell et al, Exploring the origin of Minkowski spacetime, arXiv: 1501.04857v3
[3] Chappell, Iqbal, Gunn, Abbott, Functions of multivector variables, arXiv:1409.6252v1
[4] Chappell, Iqbal, Iannella, Abbott, Revisiting special relativity: A natural algebraic alternative to Minkowski
spacetime, PLoS ONE 7(12)
[5] Dorst, Fontijne, Mann, Geometric Algebra for Computer Science (Revised Edition), Morgan Kaufmann
Publishers, 2007
[6] Hestenes, New Foundations for Classical Mechanics, Kluwer Academic Publishers, 1999
[7] Hestenes, Sobczyk, Clifford Algebra to Geometric Calculus: A Unified Language for Mathematics and
Physics, Kluwer Academic Publishers, 1993
[8] Hitzer, Helmstetter, Ablamowicz, Square Roots of -1 in Real Clifford Algebras, arXiv:1204.4576v2
[9] Mornev, Idempotents and nilpotents of Clifford algebra (russian), Гиперкомплексные числавгеометрии
и физике, 2(12), том 6, 2009
[10] Sobczyk, New Foundations in Mathematics: The Geometric Concept of Number, Birkhäuser, 2013
[11] Sobczyk, Special relativity in complex vector algebra, arXiv:0710.0084v1
[12] Sobczyk, Geometric matrix algebra, Linear Algebra and its Applications 429 (2008) 1163–1173
[13] Sobczyk, The Hyperbolic Number Plane, http://www.garretstar.com
[14] Sobczyk, The Missing Spectral Basis in Algebra and Number Theory, JSTOR
[15] Sobczyk, A Complex Gibbs-Heaviside Vector Algebra for Space-time, Acta Physica Polonica, B12, 1981

10