Generalized Madelung Transformations For Quantum Wave Equations I: Generalized Spherical Coordinates For Field Spaces
Generalized Madelung Transformations For Quantum Wave Equations I: Generalized Spherical Coordinates For Field Spaces
Generalized Madelung Transformations For Quantum Wave Equations I: Generalized Spherical Coordinates For Field Spaces
D. H. Delphenich
The Madelung transformation of the space in which a quantum wave function takes its
values is generalized from complex numbers to include field spaces that contain orbits of
groups that are diffeomorphic to spheres. The general form for the resulting real wave
equations then involves structure constants for the matrix algebra that is associated with
the group action. The particular cases of the algebras of complex numbers, quaternions,
and complex quaternions, which pertain to the Klein-Gordon equation, the relativistic
Pauli equation, and the bi-Dirac equation, resp., are then discussed.
1 Introduction………………………………………………………………….. 2
2 Examples of associative algebras…………………………………………… 4
2.1 The complex numbers 4
2.2 The quaternions 6
2.3 The complexified quaternions 9
3 Wave equations………………………………………………………………. 16
3.1 Forms for wave equations 16
3.2 Generalized spherical coordinates on field spaces 18
3.3 Wave equations in generalized spherical coordinates 19
4 Physical examples……………………………………………………………. 22
4.1 Klein-Gordon equation 22
4.2 Relativistic Pauli equation 25
4.3 Bi-Dirac equation 27
4.4 Dirac equation 30
5 Discussion……………………………………………………………………. 31
Appendix - Associative algebras……………………………………………….. 32
A.1 Basic concepts 32
A.2 Polarization 33
A.3 Structure constants 34
A.4 Polynomials 35
A.5 Complexification 37
A.6 Representations 40
References…………………………………………………………………… 41
Generalized spherical coordinates for field spaces 2
1 Introduction
The two fundamental problems at the interface between physics and mathematics are the
problem of the mathematical modeling of physical phenomena and the converse problem
of the interpretation of mathematical results in terms of physical phenomena. Although
the former problem is the one that defines the mathematical methods of physics most
essentially, nevertheless, in the last century the latter problem has also been increasingly
dominant, especially in the realm of quantum phenomena, which, by their nature, lie
beyond the scope of human intuition or direct observation.
The problem of interpretation is complicated by the fact that the same mathematical
models generally describe many analogous, but nonetheless distinct, physical
phenomena. For instance, the same mathematical methods, namely, the calculus of
exterior differential forms, can be applied with only revisions of the notations to the
description of Maxwell’s theory of electromagnetism, relativistic hydrodynamics, and the
description of weak-field gravity, when one includes the “gravito-magnetic” contribution
to the gravitational field. Hence, unless one knows in advance the physical context of a
given mathematical model, choosing a proper physical interpretation for a mathematical
model is a much deeper problem than it sounds like it should be on the surface of things.
If one examines the early history of quantum physics, one sees that this was precisely
the nature of the greatest problem that early quantum physicists were confronting. In
effect, the mathematical models were preceding the physical interpretations, so there was
considerable debate about what the proper interpretations should be.
The dominant interpretation of wave mechanics that emerged was the statistical
interpretation that was developed by the Copenhagen school of Bohr, Born, Heisenberg,
and others. Basically, this stemmed from regarding the modulus squared ρ = ψ *ψ of the
quantum wave function ψ as a probability density function for the position of a point
particle. However, not everyone agreed that the statistical interpretation represented a
definitive theoretical statement. Notably, Einstein, who was himself distinguished in
statistical physics, felt that the nature of statistical methodology was too inherently
empirical to represent a fundamental statement of natural law
Another interpretation that came out in roughly the same year as the statistical
interpretation was the hydrodynamical interpretation of Madelung [1], who suggested
that instead of interpreting ρ as a probability density function for a point particle, one
should regard it as proportional to the mass density function of an extended matter
distribution. The Schrödinger equation in the complex wave function ψ could then be
converted into a pair of real equations that took the form of the conservation of mass for
ρ and the conservation of energy for the extended distribution. In the latter equation, the
distinction between classical and quantum continuum mechanics took the form of a
“quantum potential” function that was proportional to ∆ ρ / ρ , by way of Planck’s
constant. Hence, the problem of interpretation was reduced to the problem of correctly
accounting for the appearance of the quantum potential.
The general nature of Madelung transform is that one attempts to convert a wave
equation in a complex-valued wave function into a pair of real equations by means of
expressing the values of the wave function in polar coordinates in the usual method of
setting ψ = Reiθ. Although the extension of the Madelung transformation from the
Schrödinger equation to the Klein-Gordon equation was carried out by Takabayasi [2],
Generalized spherical coordinates for field spaces 3
with analogous results, nonetheless, when later researchers attempted to give a “neo-
classical” form to the Pauli equation [3] and the Dirac equation [4], the path that they
chose was the related, but not identical, method of bilinear covariants, which is the
generally accepted method that is taught in modern wave mechanics. (For excellent
surveys on the causal, or “hydrodynamical,” interpretation of wave mechanics, see Vigier
[5], de Broglie [6], and Halbwachs [7].)
It is in the context of the Klein-Gordon equation that yet another interpretation for ρ
emerged, namely the Pauli-Weisskopf interpretation. Since the original reasons for
rejecting the Klein-Gordon equation as the physically correct relativistic form of the
Schrödinger equation included the fact that the conserved current that one derives from
its Lagrangian is not positive definite, and thus cannot be consistent with the nature of
probability currents, it is interesting that some time later Pauli and Weisskopf [8] decided
that the Klein-Gordon equation was a reasonable candidate for a relativistic wave
equation if one chose to interpret ρ as an electric charge density.
Now, an interesting aspect of the wave functions for the Pauli equation, which is the
Schrödinger equation when the particle has spin, and the Dirac equation is that they take
their values in vector spaces, namely, C2 and C4, which take the form of the product
manifolds R+ × S3 and C* × S C3 when one ignores the origin. (Here, our notation is that
R+denotes the positive real numbers, S3 denotes the (real) 3-sphere, C*denotes the non-
zero complex numbers, and S C3 refers to the complex 3-sphere.) If one considers that C*
takes the form of R+ × S1 then one sees that all of the physically interesting wave
functions – at least at the elementary level – seem to take their values in field spaces that
take the form of a radial coordinate and a sphere.
Furthermore, the spheres in question are also the underlying manifolds of Lie groups
in each case. In fact, if one regards the motion of the wave as involving not only a group
of motions that acts on the spacetime manifold, but another – possibly the same – group
that acts on the space of field values then it is natural to transfer the values of the wave
functions to the group of motions that acts on field space.
The purpose of the present article is then to generalize the Madelung transformation
by generalizing the spherical coordinates introduced on the field space. The resulting
system of real equations is then derived in general form and then specialized to each of
the cases of interest to wave mechanics, namely, C, C2 and C4.
Since the resulting system of real differential equations involves the products of the
matrices that generate the field space motions, not the Lie brackets, it is necessary to
summarize some elementary notions from the theory of associative algebras, which is
presented in an Appendix. The specific results for the algebras in question, namely, C,
the quaternions H, and the complex quaternions HC, are derived in section 2. In section
3, various forms for wave equations are presented, the basic idea of generalized spherical
coordinates on the field spaces is defined, and the general form is obtained for the system
of real equations that is associated with essentially the Klein-Gordon equation when one
Generalized spherical coordinates for field spaces 4
allows the wave function to take its values in a Lie group of the type in question. In
section 4, this general form is specialized to each of the algebras under discussion.
This article stops short, however, of proposing a new interpretation for wave
mechanics as a result of the analysis. This is because the analysis of the wave equations
is incomplete, at this point, in advance of any discussion of the role of spin in the waves
being described, along with the relationship between the present method and the method
of bilinear covariants. These topics will then be addressed in subsequent efforts, since
the presentation of the generalization of the Madelung transformation is already going to
occupy a considerable volume of space.
We first consider how the abstract concepts that are discussed in the Appendix apply to in
the specific cases of interest to quantum wave equations.
1 0 0 −1
I = , J = . (2.1)
0 1 1 0
a −b
aI + bJ = . (2.2)
b a
Generalized spherical coordinates for field spaces 5
1 0 0 1
ηab = , αab = , (2.5)
0 −1 1 0
which are seen to be the matrix of the two-dimensional Minkowski scalar product in an
orthonormal frame and the matrix of the permutation of the x and y axis, or, what
amounts to the same thing, the real and the imaginary numbers.
Hence, we can summarize the multiplication table for the algebra of C in the form:
0
aab = 2bab0 = ηab, a1ab = 2bab
1
= αab, c
cab = 0. (2.7)
The exponential map for the algebra C is the usual one that is given in complex
analysis. Because the algebra is commutative, one can say that exp(a + b) = exp(a)exp(b)
for any a, b ∈ C, and, in particular:
The image of C under exp is then C*, which denotes the non-zero complex numbers.
Hence, all numbers exp(a + ib) will have multiplicative inverses and, indeed, define the
Abelian Lie group of non-zero complex numbers under multiplication. When one
chooses an appropriate interval on the imaginary line, the map exp is also invertible, and
Generalized spherical coordinates for field spaces 6
the previous expression for exp(a + ib) represents the polar coordinate form for any non-
zero complex number z, namely:
z = Reiθ. (2.9)
One can then say that the multiplicative group C* = exp(C) is isomorphic to the direct
product group R+ × U(1), where R+ represents the multiplicative group of positive real
numbers.
If one regards z = a + ib as a complex number, and chooses {1, i} as a basis for C as a
real two-dimensional algebra then the expansion of exp z into a linear combination of
basis elements is given simply by de Moivre’s theorem:
We now extend the vector space C to two complex dimensions, which then represents
four real dimensions. However, although the real vector space R4 can be regarded as the
complexification of the real vector space C = R2, nonetheless, when one examines the
structure constants of the algebra H as an algebra over R4, one will see that it is not, in
fact, the complexification of the algebra C. In fact, we shall see that the subalgebra of
M(2; C) that we shall be ultimately concerned with is essentially real, not complex.
We denote the canonical basis elements of R4 by 1, i, j, k, respectively, and give them
the multiplication table:
We then recognize that we are dealing with the algebra of the quaternions (see, e.g.,
Porteus [9] or Artin [10]). It is associative, but from the last three sets of relations, one
can see that it is not commutative.
Furthermore, since the squares of i and j are proportional to 1, while ij gives the basis
element k, it is clear that the set {i, j} represents a minimal set of generators for the
algebra. One can represent the basis set in the form {−ii, i, j, ij}.
Generalized spherical coordinates for field spaces 7
We see that the subalgebra spanned by {1, i} is isomorphic to the algebra C, but
although we have extended the vector space from two real dimensions to four,
nevertheless, we cannot regard H as the complexification of the algebra C, nor any other
algebra over C2. The key to understanding this is to recall (cf., Appendix) that the
product map for a complex algebra must be C-bilinear.
Hence, suppose we represent the vector space of H as C ⊕ Cj. Let u = a + bj and v =
c + dj, with a, b, c, d ∈ C. If the algebra product were C-bilinear then one would have:
However, as we saw from the multiplication rules for the basis elements, the algebra is
not commutative, in general.
The representation that we shall use in M(2; C) is the real subalgebra that is spanned
by the matrices {ea , a = 0, …, 3} with:
1 0 0 1
e0 = I, e1 = J, e2 = i , e3 = i . (2.13)
0 −1 1 0
The matrices ei, i = 1, 2, 3 are proportional to the Pauli matrices σi by way of the
factor i, since we are using skew-Hermitian matrices, instead of Hermitian ones. This is
because it is really the skew-Hermitian matrices that define the infinitesimal generators of
one-parameter subgroups of unitary transformations, not the Hermitian ones. Of course,
given a Hermitian matrix a one can associate it with a skew-Hermitian one in the form of
ia, so the distinction is mostly one of convenience. This is essentially a higher-
dimensional analogue of regarding an element of U(1) as eθ with θ imaginary or as eiθ
with θ real. We choose the present convention since we shall be using both Hermitian
and skew-Hermitian matrices in the next section.
We have also made a cyclic permutation of the basis elements 1, 123 → 231, in order
to make the basis for H be an extension of the basis for C. That is, in order relate our e
matrices to the Pauli σ matrices, one must substitute:
1
Sakurai [11] refers to this particular permutation of the Pauli matrices as the “non-conformist”
representation. However, the author of the present work is choosing it for the sake of mathematical
convenience, not adolescent iconoclasm.
Generalized spherical coordinates for field spaces 8
e0 ea = eae0 = ea , (all a)
ei ej = εijk ek (i, j = 1, 2, 3).
which makes the representation faithful, as one observes. For this algebra, it is sufficient
to define the matrices that represent i and j in order to completely define the
representation.
One can then see that the matrix that corresponds to an arbitrary element (x0, …, x3)
in R4 for this choice of basis will take the form:
x 0 + ix 2 − x1 + ix3 z1 −z 2
xaea = 1 ≡ . (2.15)
x + ix x 0 − ix 2 z 2
3
z1
The latter expression shows that although one can define an invertible map that takes
(u, v) ∈ C2 to a matrix in M(2; C) that is in the image of H, nevertheless, the map is not
complex linear. In particular, under complexification the quaternion u + vj corresponds
to the 2×2 complex matrix:
u −v x 0 + ix 2 − x1 − ix 3
uI + vJ = = 1 , (2.16)
− x − ix x 0 + ix 2
3
v u
if we also set u = x0 + ix2 and v = x1 + ix3. Hence, this representation of the vector space
H = C2 in M(2; C) differs fundamentally from the one described in (2.15).
From the multiplication table, we see that:
a0cb =1/2 b0cb = δcb , aij0 =1/2 bij0 = −δij , aijk =1/2 cijk =εijk . (2.19)
The latter notation is appropriate since exp(φ0e0) will be a positive real number (times
the identity matrix) and exp(φi ei) will be an element of SU(2). That is, the group that we
are dealing with is R+ × SU(2).
Since the squares of the matrices ei are proportional to I = e0 any polynomial in them
− in particular, exp(φi ei) − will have to be expanded into four components relative to the
full set of basis elements of H, not merely three components relative to the set of three
that are defined by ei . In other words, the vector space that is spanned by the set {ei, i =
1, 2, 3} does not define a subalgebra of H when one restricts the product. Hence, in
general, one will have:
For instance, if one is considering a rotation of angle θ about only one of the axes –
say, i = m – then one reverts to the previous result for the algebra C, namely:
where the asterisk refers to the Lorentz adjoint of a, which we define by:
Generalized spherical coordinates for field spaces 10
a* = ηaTη, (2.24)
in which η = diag[+1, −1, −1, −1] and the “T” signifies the matrix transpose. Hence, one
can also represent the defining relation as:
ηa + aTη = 0, (2.25)
ATηA = η, (2.26)
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0
1 0 0 −1 0
J1 = , J2 = , J3 = , (2.27a)
0 0 0 −1 0 0 0 0 0 1 0 0
0 0 1 0 0 −1 0 0 0 0 0 0
0 1 0 0 0 0 1 0 0 0 0 1
1 0 0
0 0 0 0
0 0 0 0 0
K1 = , K2 = , K3 = . (2.27b)
0 0 0 0 1 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 1 0 0 0
[Ji, Jj] = 2εijk Jk , [Ji, Kj] = 2εijk Kk , [Ki, Kj] = − 2εijk Jk . (2.28)
The first set defines the Lie algebra of so(3; R). The second set says that the adjoint
action of so(3; R) on the vector space of infinitesimal boosts takes boosts to other boosts.
The last set is particularly interesting from the standpoint of Thomas precession since it
leads to a purely relativistic rotation of a spin vector when it is not in a state of rectilinear
motion.
Now, observe what happens when one allows the Ji basis vectors to be multiplied by
complex scalars in such a way that:
Ki = iJi . (2.29)
Generalized spherical coordinates for field spaces 11
These three vectors then span a three-dimensional complex vector space over the
same set of vectors as the six-dimensional real one that so(3, 1) is defined over. In fact,
one can associate each 4×4 real matrix in so(3, 1) that takes the form ωiJi + βiKi with a
complex 3×3 matrix of the form ( ωi + iβi)Ji , with:
0 0 0 0 0 1 0 −1 0
J1 = 0 0 −1 , J2 = 0 0 0 , J3 = 1 0 0 , (2.30)
0 1 0 −1 0 0 0 0 0
this time. These are clearly the infinitesimal generators of three-dimensional Euclidian
rotations about the x, y, and z axes, respectively, although now we are also allowing them
to be multiplied by complex scalars.
Furthermore, if one makes these replacements in the commutation rules, one sees that
the second and third sets take the form:
Now, the Lie bracket that is associated with M(3; C) is complex bilinear, so these
commutation relations are consistent with the assumption that we are dealing with the
complex Lie algebra spanned by the three Ji with the commutation relations of so(3).
Hence, the assignment of ( ωi + iβi)Ji with ωiJi + βiKi says that the complexification of the
Lie algebra so(3; R) is (real) isomorphic to the Lie algebra so(3, 1).
It is interesting to note that since the vector cross product on R3 gives it the structure
of the Lie algebra so(3; R), all that one has to do to make the vector cross product useful
in relativistic mechanics is to define it over C3, instead of R3, and in the same manner.
Hence, one does not necessarily have to abandon the cross product when going from non-
relativistic to relativistic physics, merely complexify it.
Now, let us look at the Lie algebra sl(2; C), which is also isomorphic to so(3, 1). It
consists of 2×2 complex matrices with trace zero, so it is a three-complex-dimensional
complex vector subspace of M(2; C), as well as a Lie subalgebra under the associated Lie
bracket. One can then regard it as a six-dimensional real vector subspace. Indeed, we
have already defined a three-real-dimensional subspace of sl(2; C) in the form of su(2),
which is also isomorphic to the Lie algebra so(3). Hence, all that one must to do to
complete su(2) to sl(2; C) is complexify it.
As we pointed out above, when one multiplies a skew-Hermitian matrix by i one
obtains a Hermitian one. Now, one can polarize any a ∈ sl(2; C) with respect to the
Hermitian conjugate operator:
Generalized spherical coordinates for field spaces 12
a = 1, …, n} amounts to a collective rotation of each pair {ea, ea } in the plane that they
span:
ea → cos θ ea – sin θ ea , ea → sin θ ea + cosθ ea . (2.33)
Thus, the eiθ part of λ represents a duality rotation for the expression of C4 in terms of R4
⊕ iR4.
Notice that the projection C* → R*, Reiθ ֏ R defines a homomorphism of the
multiplicative groups, so it is the U(1) part of C* that is basically lost under the
projection.
Analogous to the way that we complexified H itself, we can also complexify the
representation of H in M(2; C) by means of the basis matrices {ea , a = 0, 1, 2, 3}. Since
these matrices span a four-dimensional real vector subspace, they will span a four-
complex-dimensional subspace under complexification. However, M(2; C) is four-
complex-dimensional to begin with, so we are faithfully representing HC by way of the
entire algebra of M(2; C).
The missing four-real-dimensional subspace in M(2; C) that is added during
complexification is spanned by the ea when they are given imaginary coefficients, or
rather, by the iea when they are given real coefficients. Hence, ie0 represents the
imaginary axis in C, while the iei , i = 1, 2, 3 are Hermitian matrices. Here, we see why it
is necessary to be consistent in our use of anti-Hermitian matrices to represent
infinitesimal rotations, instead of Hermitian ones, since we are also using the Hermitian
matrices to represent other things.
As a complex algebra over C4, the multiplication table of HC, and therefore the set of
structure constants, is same as for H, namely, (2.18), but its multiplication table as a real
algebra over R8 must take into account the fact that i2 = − 1, since the real vector a and
the imaginary vector ia are linearly independent over the real scalars, but not over the
complex scalars. Hence, we extend our real basis {ea , a = 0, …, 3} for H to a real basis
{ea , ea ≡ iea , a = 0, …, 3} for HC and include the extra entries in the multiplication table:
c
ea eb = ea eb = i ea eb = aab ec , ea eb = − ea eb = − aab
c
ec . (2.34)
c
Therefore, we must extend the structure constants aab for H accordingly:
Generalized spherical coordinates for field spaces 14
c
aab c
= aab = − aab
c c
= aab , (2.35)
0 − I
J = . (2.37)
I 0
ea 0
Ea =
ea
. (2.39)
0
0 −ea
Ea = JEa =
0
. (2.40)
ea
One sees that the matrices Ei , i = 1, 2, 3 have essentially the same form as the Dirac αi
matrices:
0 σi
αi = i = µ = 1, 2, 3.
0
, (2.41)
σ i
I 0
β = , (2.42)
0 − I
namely, the β matrix does not belong to the algebra that is spanned by the eight basis
elements Ea and Ea , which must all take the form:
φ a ea −φ a ea
φ Ea + φ Ea = a
a a
, φa, φ a ∈ R. (2.43)
φ ea φ a ea
There is clearly no such linear combination that will result in the matrix β. Hence, the
Clifford algebra generated by the set {β, αi , i = 1, 2, 3} is quite distinct from the algebra
HC, even though the anti-commutators of the basis elements of HC seem to have much in
common with those of the aforementioned Clifford algebra.
We can, however, say that:
αi = i β Ei , i = 1, 2, 3 (2.44)
0 σi
γ0 = β, γi = = βαi ,
0
(2.45)
−σ i
2
Since various conventions exist for the gamma matrices, which generally depend upon the choice of sign
convention for the Minkowski scalar product, we are presenting the Bjorken and Drell form [12], which is
consistent with our choice of convention on g that it be “timelike positive.” The form presented in Sakurai
[11], which differs by a factor of i for the spatial indices, is based on representing the Minkowski scalar
product with the “imaginary time” convention that makes it Euclidian.
Generalized spherical coordinates for field spaces 16
so we can relate the spatial matrices to our corresponding basis elements by way of:
γi = i Ei , (2.46)
3 Wave equations
The physics of waves is more general in scope than any particular wave equation,
whether linear or nonlinear, and strictly begins in the definition of constitutive laws for
the wave medium, along with field equations for the wave field. However, this generally
leads to dispersion laws for the medium, which eventually come down to a Lorentzian
structure g (or possibly more than one) on the tangent or cotangent bundle of the
configuration manifold M; for simplicity, we assume that M is an open submanifold of
Rn, which amounts to a choice of coordinate system on M.
Due to the nature of the dispersion laws for most wave media, most wave equations
will generally involve the resulting d’Alembertian operator:
∂ ∂
□≡ g µν (3.1)
∂x µ ∂xν
in some way.
In the quasi-linear case, the wave equation for ψ will take the form:
In the linear case, these equations often take the form of the eigenvalue equation:
□ ψ = λψ, (3.3)
in which the wave function ψ: M → V takes its values in a vector space V, which might
be real or complex, and λ ∈ C is a scalar constant.
Generalized spherical coordinates for field spaces 17
One can put the general second order quasilinear equation into first-order form by
essentially the “1-jet prolongation” of ψ, namely, the addition of its first partial
derivatives as supplementary variables:
If the components gµν are constant on M then one can put this system of equations
into “conservation law” form:
In fact, if the components of g are not constant in this coordinate system then it is
more correct to make the left-hand side of the middle equation in (3.5) take the form:
1
kµ; µ = ∇µ ( | g | k µ ) , (3.6)
|g|
The notation used here is: d refers to the exterior derivative operator, which acts on
exterior differential forms, δ to the divergence operator, which acts on multivector fields,
j1ψ to the 1-jet prolongation of ψ, which locally takes the form (xµ, ψ(x), kµ(x)), k denotes
the vector field that is metric-dual to k, and ig: T*(M) → T(M) to the isomorphism of
cotangent spaces and tangent spaces that is define by g, and is locally referred to as
“raising an index.”
Furthermore, if one substitutes the locally correct replacement of the exactness of p
that is expressed by the first equation of (3.7) with the condition that it be closed, then the
equations take the form:
As has been discussed elsewhere by the author [13], system of equations of this form
basically take the generalized form of an integrability condition on a field – ψ, in this
case, – a dual integrability condition on the dual of the field − namely, k − and a
constitutive law that relates the original field to its dual. Hence, one can think of the field
k as the kinematical variable and the field k as the dynamical variable in the dynamics of
the wave. Although the constitutive law is purely geometrical in this case, nevertheless,
these equations can be generalized to define the fundamental field equations for wave
motion at a “pre-metric” level, in such a way that the form (3.8) follows from the
dispersion law for the wave medium. Indeed, the notion of integrability seems to play an
Generalized spherical coordinates for field spaces 18
even more fundamental role in physics than perhaps the least action principle itself. (For
more discussion of this see [13] and references cited therein.)
The wave equations that we are concerned with for the remainder of this study all take
the form of the Klein-Gordon equation, with the difference between them being in the
vector space V in which the wave function takes its values, which we then call the field
space of the wave functions.
Interestingly, the vector spaces that seem to be most interesting in the eyes of
elementary wave mechanics are V = C, C2, C4, although C3 certainly plays a role in
quantum chromodynamics, since it supports the fundamental representation of SU(3).
Furthermore, the groups of interest to physics that act on these spaces are U(1), SU(2),
and SU(2; C), which correspond to phase rotations, Euclidian rotations or isospin
rotations, and Lorentz transformations, respectively.
In fact, since all of the aforementioned groups can be embedded as manifolds in the
corresponding spaces, one can say that V – {0} in each case can be represented as the
manifolds R+ × U(1), R+ × SU(2), and C* × SL(2; C), which underlie the Lie groups C*,
the Weyl group of SO(3), and GL(2; C), resp. Since these spaces are also diffeomorphic
to R+ × S1, R+ × S3, and C* × S C3 , respectively, we can say that we introducing
generalized spherical coordinates into V – {0}, in effect. Actually, for the sake of
differential equations, we will generally be working with the corresponding coordinates
on their Lie algebras R ⊕ u(1), R ⊕ su(2), and C* ⊕ sl(2; C). Moreover, we will find
that we are really more concerned with the structure constants of the associative algebras
over these vector spaces than the Lie algebras themselves.
In the case of V = C, the spherical coordinates in question on C* = R2 – {0} are (R,
eiθ). The coordinates on the Lie algebra R ⊕ u(1) are then the usual polar coordinates (r,
θ) for R2 – {0}, as long as one sets R = er.
In the case of V = C2, since SU(2) is diffeomorphic to the real three-sphere S3, we are
essentially extending polar coordinates from R2 – {0} to R4 – {0}, by using (R, exp(φ1e1),
exp(φ2e2), exp(φ3e3)) as our spherical coordinates. The corresponding coordinates on R
⊕ su(2) are (r, φ1, φ2, φ3). Since the φ i generate the Euler angles as coordinates on
SU(2), they are to be distinguished from the usual right ascension and codeclination
angles that one uses for the spherical coordinates of R3 – {0}.
Generalized spherical coordinates for field spaces 19
We now come to the main subject of the present article, which is what happens to the
form of the wave equation, in any of the above forms, when one gives the vector space V
of values taken by the wave functions the generalized spherical coordinates that we just
introduced. We assume the V is a defined over the field K of scalars, which we assume
to be either R or C, for our purposes. Note that the process of introducing generalized
spherical coordinates into V – {0} is generally distinct from the process of giving the
manifold M itself a system of spherical coordinates, unless, for instance, the vector space
V is represented by the tangent spaces to M. Hence, we are more concerned with
symmetries of the field space than with symmetries of the spacetime manifold itself.
This process begins with assuming that the motion of the wave ψ is due to the action
of a Lie group G on V, which we shall assume to be a linear action; i.e., we have a
representation of G as a subgroup of GL(V). This has the advantage of allowing us to
denote the action of a group element A ∈ G on a vector v ∈ V by means of the
multiplication of matrices for both when one chooses a basis {ei, i = 0, …, n) for V, which
is not the same thing as choosing a coordinate system for M. The vector Av will then
have the components Aij v j relative to this basis.
Really, it is not the group G that plays the key role here, but the group of all smooth
functions A: M → G, which then acts on the smooth functions ψ: M → V pointwise; that
is:
(Aψ)(x) = A(x)ψ(x). (3.9)
a subalgebra of the Lie algebra gl(n; K). However, as we shall see, it is not only the Lie
algebra g that is generated by these basis elements that we need to deal with, but the
associative algebra A, as well, which will also be a sub-algebra of M(n; K), which is
defined over the same vector space of matrices as gl(n; K).
The starting point for our conversion of the wave equations is given by a
generalization of the Heisenberg representation for wave mechanics, namely, we let by
ψ0 be a constant field on M and define the more general field ψ as:
ψ = exp(φaea)ψ0 . (3.10)
□ ψ = □ (exp(φaea))ψ0 . (3.11)
It is in the left-hand expression that we find the justification for the mathematical
abstractions and generalizations that are discussed in the Appendix. One sees that is it
the product ea eb that enters into the expression, not the Lie bracket. Hence, if one knows
c
the structure constants aab of the associative algebra A that is generated by the ea then one
can re-express the term in brackets as:
□ ψ = { [ □ φc + gµν c
aab φa,µ φb,ν ] ec }ψ, (3.14)
Generalized spherical coordinates for field spaces 21
so the eigenvalue equation then gives rise to the following set of n quasi-linear second-
order partial differential equations for the φa:
Note that the nonlinear contribution to what started out as a linear equation is solely due
to the nature of the algebra A.
We can further convert this system of n second-order equations into a system of 2n
first-order equations in the usual way:
or, to put this into the quasi-linear first order form of (3.4):
By further defining the 1-forms ka = kµa dxµ and assuming that the gµν are constant for
the chosen coordinate system, we can put this into coordinate-independent form (but not
independent of the set of generators on A) as in (3.7):
One can also generalize the first set of equations to put this into the form (3.8):
One must notice that these equations do not refer to the φa explicitly anywhere, and can
thus be regarded as equations in the 1-forms ka.
Since we shall be exclusively concerned with algebras that decompose into the direct
sum of a 1-dimensional center and the Lie algebra of a Lie group, we shall then re-
express (3.19) in terms of k0 and ki, explicitly:
In the next section, we shall examine the form that these equations take for the
various algebras of interest to quantum wave mechanics.
Generalized spherical coordinates for field spaces 22
4 Physical examples
In order to apply the method of generalized spherical coordinates on the field space to the
specific cases of V = C, C2, C4 that are of interest to us, one need only go back to section
2 to obtain the appropriate structure constants for the algebras A = C, H, and HC,
respectively.
When one chooses A = C, the resulting equation is the usual Klein-Gordon equation:
□ψ = − k02ψ , (4.1)
which basically represents the eigenvalue equation for the d’Alembertian operator, under
the assumption that the one is looking for negative eigenvalues3. In wave mechanics,
where ψ is a complex-valued C2 function on spacetime that represents the wavefunction
of a moving pointlike particle, one assumes that k0 = m0 c / ℏ is the Compton wave number
that is associated with the rest mass m0 of the particle.
The extension of the Madelung transformation that was carried by Takabayasi [2]
began with expressing ψ in polar form:
ψ = R eiS / ℏ , (4.2)
in which R and S are then real-valued C2 functions on spacetime, and then obtaining real
differential equations by annulling the real and imaginary parts of the equation that
followed from (4.1) by the substitution (4.2).
We shall apply the slightly generalized results that we obtained above by the Ansatz:
ψ = exp(ln ρ + iθ )ψ 0 . (4.3)
ψ = exp(ln ρ I + θ J )ψ 0 . (4.4)
3
The negative sign appears by starting with the d’Alembertian operator and replacing the operators ∂/∂xµ
with their Fourier transforms ikµ – i.e., their symbols. This associates the eigenvalue equation for the
d’Alembertian with a dispersion law for the frequency-wave number 1-form k = kµdxµ, namely k2 = k0 .
2
Generalized spherical coordinates for field spaces 23
0
We recall that the non-zero structure constants of the algebra C are a00 = − a110 = a101 =
1
a10 = 1. Substitution in the general equations (3.20a, b) gives the pair of equations:
d ρ dρ
k0 = d (ln ρ ) = = , k1 = dθ ≡ k (4.6)
ρ 2ρ
and take into account that if f is a 0-form and and α is a 1-form then one has:
□ ρ
dk = 0, δJ = 0, J ≡ iρgk, k2 = k02 + . (4.8)
ρ
k = dθ = dS / ℏ = (1/ ℏ) p , (4.9)
in which p represents the energy-momentum density 1-form of the motion then (4.8)
become:
δJ = 0, J ≡ iρgp = ρp,
p2 = m02 + ℏ 2
□ ρ c2 .(4.10)
dp = 0,
ρ
The usual argument for regarding the second equation as the law of conservation of
mass is based on the assumption that p/m0 can be given the interpretation of the
covelocity 1-form u of the motion. However, this argument is conceptually flawed by the
fact that m0 is the total rest mass of the distribution, and a more natural relation between p
and u is defined by means of the rest mass density ρ0 ≡ m0ρ, namely:
Generalized spherical coordinates for field spaces 24
p = ρ0u. (4.11)
However, the overall effect on equations (4.10) is moot if one sets p = m0u, since that
makes J = ρ0u, which then represents the energy-momentum vector field.
This puts the middle equation into the form:
δ(ρ0u) = 0, (4.12)
which then says either that the motion is dynamically incompressible4 or, equivalently,
that rest mass is conserved along the flow of u.
With the assignment (4.11), the first equation in (4.11) does not say that the flow is
kinematically irrotational, which amounts to the vanishing of du, but that its dynamic
vorticity dp vanishes. Its kinematic vorticity du obeys:
That is, the flow is kinematically irrotational if and only if the gradient dρ0 of the rest
mass density is collinear with the covelocity.
As for the last equation in (4.10), it usually gets the interpretation of the Hamilton-
Jacobi equation that is associated with the Hamiltonian:
□ ρ
H = T +U ℏ = p 2 c 2 − ℏ 2 c 2 , (4.14)
ρ
□ ρ
U ℏ = − ℏ 2c2 (4.15)
ρ
p2 = ρ 02 c 2 , (4.16)
in which we are dealing with the rest mass density ρ0, instead.
4
The terminology of “kinematic” and “dynamic” quantities used here is that of Carter [14]. Kinematic
quantities pertain to the covelocity 1-form u, while dynamic quantities pertain to the energy-momentum 1-
form p.
Generalized spherical coordinates for field spaces 25
□ ρ
ρ 02 = m02 + ℏ 2 . (4.17)
ρ
When the rest frequency k0 is no longer a constant, but a function on spacetime, one
sees that the Klein-Gordon equation itself takes on a subtly more complicated form:
□ψ = − k02 ( x µ )ψ (4.18)
Hence, the most natural wave equation for extended matter distributions is more general
than the usual eigenvalue equation for the d’Alembertian, since the eigenvalues are
functions, rather than constants. The appearance of constants would have to follow from
the process of reducing from an extended body to a pointlike body; e.g., from mass
density to total mass.
One can also take the position that the dispersion law for k is what is most
fundamental and that a more general form that includes the higher moments of ρ involves
the generalization of k0 to some function k0(k). However, we shall defer a detailed
analysis of this possibility for a subsequent study, since it also implies that the Klein-
Gordon equation might require a different modification than the one suggested in (4.18).
Actually, what one refers to as the “Pauli equation” is the non-relativistic wave equation
that one obtains from the Schrödinger equation by substituting wave functions that take
their values in C2, which carries a representation of SU(2), as a first step towards
including spin in the wave functions, such as one would desire for electrons, for instance.
However, since the relativistic form of the Schrödinger equation, namely, the Klein-
Gordon equation, is mathematically more concise to deal with, although it is physically
more debatable in its interpretation, we shall proceed from the Klein-Gordon equation
(4.1) with the wave function ψ taking its values in the algebra A = H.
Hence, we make the Ansatz:
Thus we use φ0 = ln ρ , as before, along with the three φi as the components of the wave
function ψ for this choice of A.
The real form that the Pauli equation takes as a result of this substitution is given by
(3.20a,b). If we recall that the structure constants of H are given by (2.18) a0γ β =1/2 b0γβ =
δγβ, aij0 =1/2 bij0 = −δij , aijk =1/2 cijk =εijk , in which we now index the basis elements for H by
α, β, γ = 0, …, 3 to distinguish them from the coordinates of M, then the (3.20a,b) give
Generalized spherical coordinates for field spaces 26
the following set of four real second order quasi-linear partial differential equations for
the kα = dφα:
In the middle equation of (4.20b), we have taken into account that the symmetry of g
clashes with the anti-symmetry of ε to annul the corresponding term of the form εijk g(kj,
kk).
If we make the same substitution k0 = d(ln ρ ) as we did for the case of the algebra
C then these equations take the form:
□ ρ
dki = 0, δJi = 0 . Ji = iρgki δij g(ki, kj) = k02 + , (4.21)
ρ
Now, we see that we have three currents Ji to deal with, not just the one, and rather than a
separate dispersion law for each ki, we have a single equation for all three.
However, one must realize that collectively the three 1-forms ki define the
components of a 1-form K on M with values in R3, relative to its canonical basis δi,
namely:
K = ki ⊗ δi . (4.22)
Thus, one can find another orthonormal basis ei(x) for R3 that generally depends upon
which point x ∈ M one considers and differs from δi by a rotation Rɶ ij ( x) , so:
ej(x) = Rɶ ij ( x) δi , (4.23)
K = k ⊗ e1 = Ri1k i ⊗ e1 . (4.24)
in which we have defined the angular velocity of the frame ei relative to δi by:
ω ij = dRki Rɶ kj , (4.26)
Generalized spherical coordinates for field spaces 27
we see, taking into account that δJi vanishes for each i, that:
What we are calling the bi-Dirac equation here is simply the Klein-Gordon equation
when the wave function ψ takes its values in C4. The Dirac equation, properly speaking,
is obtained when one essentially takes the “square root” of the operators that act on ψ in
order to reduce the order of the differential equations from two to one:
∇ψ ≡ γ µ ∂ µψ = ik0ψ . (4.31)
∇ψ ∗ = − ik0ψ∗ . (4.32)
Generalized spherical coordinates for field spaces 28
dk 0 = 0, δ k 0 = − 2ηαβ g (k α , k β ) , k 0 = ig k 0 , (4.33c)
dk i = 0, δ k i = −2 g (k 0 , k i ) − 2 g (k 0 , k i ) , k i = ig k i . (4.33d)
Note that one could also have obtained these from (4.20a, b) by substituting kα + ik α for
kα and separating the real and imaginary equations. That is, although we are representing
C* − {0} by complex spherical coordinates, nevertheless, the algebra HC, which
exponentiates onto it, is still represented by Cartesian coordinates.
By the substitution of k0 = d (ln ρ ) , we can then put (4.33a-d) into the form:
Generalized spherical coordinates for field spaces 29
□ ρ
δij [g(k i, k j) − g (k i , k j ) = k02 − (k 0 ) 2 + , (4.34a)
ρ
2
dk i = 0, δJi = g (J 0 , J i ) , Ji = iρpki, (4.34b)
ρ
2
dk 0 = 0, δ J 0 = − δ ij g (J i , J j ) , J 0 = iρ g k 0 , (4.34c)
ρ
2
dk i = 0, δ Ji = − g (J 0 , J i ) , J i = iρ g k i . (4.34d)
ρ
We can do the same thing with ki and k i that we did with ki for the relativistic Pauli
equation, namely, define an adapted frame, except that now we must be careful to
recognize that generally we must define two independent orthonormal 3-frames ei = Ri j δ j
and fi = Ri j δ j in R3 that are adapted to the vectors k = kiδi and k = k i δ i , resp. Note that a
possible source of confusion is that although these two real orthonormal 3-frames − or
rather, ei and ifi − belong to separate subspaces of the complex field space, we are
nonetheless defining both of them in the same real vector space. Hence, in particular, the
scalar products δ(ei, fj) do not all have to vanish, when we let δ = δij θi ⊗ θj (θi(ej) = δij)
represent the Euclidian scalar product on R3. In fact, since both ei and fi are both real
orthonormal, one has:
which shows that the set {ei, ifi} of six real 3-vectors does not define a complex
orthonormal frame.
We then define the following 1-forms and vector fields on M:
We can then put our basic system of real equations into the form:
□ ρ
k2 − k 2 = k02 − (k 0 ) 2 + , (4.38a)
ρ
2
dk = 0, δJ = δ (e1 , f1 ) g (J 0 , J ) , J = iρpk, (4.38b)
ρ
Generalized spherical coordinates for field spaces 30
2
dk 0 = 0, δ J 0 = − δ (e1 , f1 ) g (J , J ) , J 0 = iρ g k 0 , (4.38c)
ρ
2
dk = 0, δ J = − δ (e1 , f1 ) g (J 0 , J ) , J = iρ g k . (4.38d)
ρ
One must avoid the temptation to confuse the scalar product g, which is defined on
the tangent spaces to M and the scalar product δ, which is defined on the field space.
However, since a 3-frame in a tangent space TxM can be redefined to be the image of a 3-
frame in R3 under a linear injection of R3 in TxM, it is possible to relate the two scalar
products, at least locally.
One sees that the quantity δ(e1 , f1) plays a key role in equations (4.38a, b, c, d) since
its vanishing would imply the conservation of all three of the currents J 0 , J, and J .
Although it is most commonplace for physics to represent geometrical objects, such
as ei and fi, in the tangent spaces to the spacetime manifold, we must nonetheless point
out that the dispersion law (4.38a) seems to be more naturally associated with the scalar
product (F ^ *F)(V) = E2 – B2 that one defines on 2-forms by means of the Hodge
operator * than with the scalar product on Minkowski space itself. This is simply due to
the fact that the Cartan-Killing scalar product on so(3; C) is isometric to the complex
Euclidian scalar product on R3, and when one gives the vector space Λ2(R4) a complex
structure by way of a * isomorphism that obeys *2 = −I the vector space of such 2-forms
becomes C-linear isomorphic to the vector space underlying so(3; C). Hence, we see that
it is probably physically incorrect to represent the field space in terms of the tangent
spaces to M directly, since the field space is more directly related to the Lie algebra
so(3,1) than it is to Minkowski space.
At first glance, it would seem that the Dirac equation (4.31) would admit a natural
transformation under the imposition of spherical coordinates on C4, since the Dirac γ
matrices belong to the matrix algebra M(4; C), along with the Eµ and Eµ matrices that we
have been using to define the spherical coordinates.
However, on closer inspection, one notices that we are dealing with two different
actions of the matrices in M(4; C) on C4 that refer to two different representations of the
points of C4. In the case of the Dirac equation, one is essentially regarding the points of
C4 in Cartesian coordinates, whereas in the case of generalized spherical coordinates,
those points are associated with matrices in M(4; C). Although the action of matrices in
M(4; C) is defined on both column vectors in C4 and the other matrices of M(4; C),
Generalized spherical coordinates for field spaces 31
nevertheless, the effect is not the same. Hence, it would incorrect to represent the values
of the wave function ψ in terms of matrices and then multiply them by the Dirac matrices
as if they were column vectors.
The conventional way of representing the Dirac equation by an equivalent set of real
equations (see, e.g., Takabayasi [4]) is to use the method of bilinear covariants.
However, a thorough description of this method and the search for any possible
relationship with the present method of generalized spherical coordinates is of sufficient
scope as to warrant a separate treatment from the present one. Hence, we shall defer that
analysis to a subsequent article.
5 Discussion
In this article, we have shown that the most common wave functions of elementary
quantum mechanics, both non-relativistic and relativistic, take their values in complex
vector spaces − namely, C, C2, and C4 − that all share the common property that the
complement of the origin can be represented as the product manifold of a line and a
sphere. In the case of C and C2, the line is R+ and the spheres are S1 and S3, respectively;
in the case of C4, the line is C* and the sphere is S C3 .
In all cases, the spheres are the underlying manifolds of Lie groups, namely, U(1),
SU(2), and SL(2; C). Hence, one can regard the punctured complex vector spaces as
diffeomorphic to the manifold of the groups R+ × U(1), R+ × SU(2), and C* × SL(2; C) ≅
GL(2; C), resp. Their Lie algebras are then associated with the algebras of C, H, and HC,
resp., by way of the commutator bracket of the algebra product. Interestingly, other
research into the nature of the Madelung-transformed wave equations of wave mechanics
has indicated that they are closely related to the geometry of Weyl-Cartan spaces, which
deals with connections on the bundle of Weyl frames over a Riemannian or pseudo-
Riemannian manifold. (See, for instance, Delphenich [15], Carroll [16], or Shojai and
Shojai [17].)
If we let the wave function in the generalized Klein-Gordon wave function take its
values in any of these aforementioned spaces then we find that the resulting system of
real differential equations involves the structure constants of the associative algebra that
gives the Lie algebra in question by way of the commutator bracket. We proceeded to
derive the specific forms for these systems of equations when one substitutes the specific
structure constants.
In the cases of C and H, the resulting system of equations can be put into the form of
an integrability condition on a 1-form k, a constitutive law that associates it with a vector
field J, a conservation law for J, and a dispersion law for k, at least when one passes to a
co-spinning frame in the field space. In the case of HC, one must also introduce 1-forms
k 0 and k , along with associated currents J 0 and J , as a result of the complexification, and
Generalized spherical coordinates for field spaces 32
the conservation laws for the three currents involve possibly non-vanishing cosines of the
angles between the three currents. Furthermore, the dispersion law also seems to be more
relevant to the scalar product on so(3; C) than the one on Minkowski space.
To summarize, there are four fundamental issues that we have deferred to later
efforts, due to the length of the present paper:
i) The role of spin in the aforementioned systems of equations and the equation that it
adds to the system.
ii) The relationship of the equations that we obtained in the cases of C2 and C4 to
those obtained from the Pauli and Dirac equations by means of the method of bilinear
covariants.
iii) The issue of whether the most physically meaningful extension of the theory from
point particles to extended ones is to replace the constant eigenvalue k0 in the Klein-
Gordon equation with functions of spacetime position or to replace the k0 in the
dispersion relation k2 = k02 with a function of k.
iv) The ultimate discussion of the nature of the problem of associating the general
system of equations with various physical interpretations.
As far as the last issue is concerned, we regard the essential problem as that of
interpreting the nature of the density function that one obtains from the wave function by
way of ψ†ψ. The leading possibilities then seem to be a probability density function for a
point particle, a particle number density function for an extended particle, and an electric
charge density function for an extended particle. These three possibilities then lead to the
statistical, hydrodynamic, and electromagnetic interpretations for wave mechanics.
What seems to be emerging from the generality presented above is simply the idea
that since the same mathematical formalism may be applied to otherwise unrelated
physical phenomena, the problem of starting with a mathematical formalism and looking
for its point of application in physics is fundamentally ambivalent. One is essentially
trying to find an “optimal” section of a projection, namely, all of the physical phenomena
that can be described by the same mathematical formalism.
It is the author’s hope that perhaps by generalizing the nature of converting the Klein-
Gordon equation into a set of real equations one can gain a clearer idea of what exactly
would constitute such an optimal section when one starts with the original problem of
wave mechanics, namely, the structure of the electron/positron at the elementary level.
It might prove convenient for some readers to introduce a few elementary notions from
the general theory of algebras that we shall apply in the discussion above.
A.2 Polarization
In general, one can polarize the product of any two elements a and b of an algebra
over V into a commutative part and an anti-commutative part:
in which the expression {a, b} is called the anti-commutator of a and b and the
expression [a, b] is called the commutator of a and b, or also the Lie bracket of a and b.
If a and b commute then [a, b] vanishes, whereas, if they anti-commute then {a, b}
vanishes.
The anti-commutator bracket defines a commutative algebra over V that we shall call
the symmetric algebra associated with the associative algebra over V that we started with.
If that associative algebra also has a unity element then if one includes the factor of ½ in
the anti-commutator bracket the resulting algebra is commutative and has a unity
element.
The case of a Clifford algebra over an orthogonal space (V, g) is defined by imposing
the following condition on the anti-commutator of elements in V:
by starting with sub-algebras of the associative algebra with unity M(n, K) and giving
them the Lie bracket.
Note that one can always associate a Lie algebra with any Clifford algebra by way of
the commutator [a, b], but the restriction (A.3) that was placed on the product ab does not
actually imply that this Lie algebra will be the orthogonal Lie algebra of (V, g); i.e., the
one associated with infinitesimal generators of one-parameter families of orthogonal
transformations.
When V is n-dimensional and given a basis {ei, i = 1, …, n}, from the bilinearity of
the product for any algebra A over V, it is sufficient to know all of the expressions eiej in
order to determine the products of more general elements. Since these elements will all
be in V they can be expressed in terms of the chosen basis in the form:
and the scalar constants aijk are called the structure constants of the algebra.
By polarizing eiej , we get:
eiej = 1
2 {ei, ej} + 12 [ei, ej] ≡ 12 bijk e k + 12 cijk e k = 12 (bijk + cijk )ek , (A.5)
Generalized spherical coordinates for field spaces 35
in which the structure constants bijk and cijk belong to the symmetric algebra and the Lie
algebra associated with the given algebra, respectively. This gives us the relation:
Given such a basis for an algebra A, one can then express the components of the
product ab of two elements a, b ∈ A in terms of the effect of the structure constants on
their components with respect to the basis:
since the square of any element clearly represents a symmetric product and one sums
over the indices of a twice.
A.4 Polynomials
Associative algebras are especially important, since they allow one to make sense of
powers of algebra elements. For instance:
makes sense only when one has the associativity of the multiplication.
One sees that by iterating (A.8) one can express any power of a in terms of a
homogeneous polynomial in the components ai and the symmetric structure constants bijk .
For instance:
N
P(a) = ∑ α n a n , α ∈ K, a ∈ V, (A.11)
n= 0
which will make rigorous mathematical sense for an associative algebra. (Here, if we are
also assuming the existence of a unity element I then we define a0 = I.) Polynomials in
more than one independent variable are likewise defined, although one must respect the
order of multiplication in the individual terms.
Generalized spherical coordinates for field spaces 36
n
P(a) = ∑ P ( a) e ,
i =1
i
i (A.12)
∞
1 n
exp(a) = ∑ a . (A.13)
n= 0 n !
When the series converges, exp(a) is another element of A; when this series does not
converge, it is to be interpreted as a formal power series.
In either event, if A has a finite basis then one can express A = exp(a), when it
converges, as a finite linear combination:
n
A = ∑ Ai ei . (A.14)
i =1
Each component Ai will be defined by a power series in a variable that takes its values
in K. Furthermore, the nature of the power series will involve not only a, but also the
structure constants of A, which relate to the way that successive powers of a expand into
components themselves.
In the example of M(n; K), we have an associative algebra whose unity element is
given by the identity matrix I. When the underlying vector space is given the product
2
topology of K n , all expressions of the form exp(a) will converge. Furthermore, since
one has that:
any matrix in the image of the map exp will be invertible. In particular, exp(0) = I and
exp(−a) = [exp(a)]−1. Hence, if A is a sub-algebra of M(n; K) then the image A of A
under exp will be a subgroup of GL(n; K). However, the map exp: A → A is not
Generalized spherical coordinates for field spaces 37
A.5 Complexification
jv. One can easily see that if the real dimension of V is n then the real dimension of VC is
2n.
In order to make VC into a complex vector space one must define the multiplication of
complex scalars time the elements of VC. This obtained by the use of the isomorphism
that is defined by j:
J: V × V → V × V, (v, w) ֏ (− w, v).
Addition of vectors in V × V is defined by the usual Cartesian product rule. One can
define scalar multiplication by complex scalars on vectors v ∈ V × V by way of:
0 − I
J = , (A.18)
I 0
aI −bI
aI + bJ =
aI
. (A.19)
bI
This allows us to extend the action of the group Z2 = {I, J} to an action of U(1) ≅
SO(2) on CV that one refers to as the action of duality rotations, by analogy with the
corresponding operations on 2-forms on a four-dimensional vector space when that
vector space of 2-forms is given a complex structure (see Delphenich [19]).
A duality rotation R(θ) through an angle θ is then represented by the rule:
cos θ I − sin θ I
R(θ) = cos θ I + sin θ J =
cos θ I
. (A.21)
sin θ I
This allows one to always define a smooth (indeed, analytic) circular loop that connects
any a to its dual image ja.
Another canonical linear isomorphism that is always associated with any vector space
of the form V×V, and which plays a recurring role in quantum wave equations, is the
transposition isomorphism T: V×V →V×V, which takes (v, w) to (w, v), and therefore has
the block matrix form:
0 I
T = . (A.22)
I 0
ei e j = ei e j = jei ej , ei e j = − ei ej . (A.25)
Generalized spherical coordinates for field spaces 40
A.6 Representations
for all a, b ∈ A. Hence, every element of A will correspond to an m×m matrix with
entries in K and the multiplication of elements in A will correspond to the multiplication
of matrices in M(m, K).
The representation D is faithful if it is a linear injection. In such a case, the algebra
over the image of A under D will be isomorphic to the algebra A.
If {ei, i = 1, …, k} is a set of generators for A (k ≤ n) then it is sufficient to know the
matrices D(ei) in order to determine the images of all products ab under the
representation D.
If a = aiei and b = bjej and the structure constants for A for this basis are aijk then one
has:
D(ab) = aijk aibj D(ek). (A.28)
If {ea, a = 1, …, m2} is a basis for M(m, K) then one can express the representation D
in terms of its matrix Dai with respect to the choice of bases on A and M(m, K). By
definition:
ea = Dai ei . (A.29)
5
Here, we are using the “kernel-index” notation for the structure constants, as preferred by Schouten and
his followers. That is, the kernel in this case is the letter a, whereas the overbar on the indices denotes
indices that refer to basis elements that have overbars.
Generalized spherical coordinates for field spaces 41
Generally, in physics applications when the basis elements are matrices themselves one
usually adds extra indices to Dai for the rows and columns.
The particular representations that we shall be concerned with are the defining
representation, which only makes sense when A is a subalgebra of M(n; K) to begin with,
and the left and right multiplication representations. That is, if A is an algebra over a
vector space V then each a ∈ A defines a left multiplication operator La: A → A, x ֏ ax
and a right multiplication operator Ra: A → A, x ֏ xa. These, in turn, define maps L: A
→ End(A), a ֏ La, and R: A → End(A), a ֏ Ra, which are easily seen to define
representations of A in the algebra End(A) of all linear maps of A to itself. Choosing a
basis for A then defines an isomorphism of End(A) with M(n; K).
It is easy to see that if A is a real algebra that is represented in a real vector space V
by way of D: A → M(V) then there is a canonical extension of D to a representation of D:
AC → M(VC), D(a + jb) ֏ D(a) + jD(b).
References