arXiv:physics/0308027v3 [physics.

class-ph] 10 Sep 2003

What is spin?
Andr Gsponer
Independent Scientific Research Institute
Box 30, CH-1211 Geneva-12, Switzerland
e-mail: isri@vtx.ch
ISRI-03-10.13 February 2, 2008
Abstract
This is a late answer to question #79 by R.I. Khrapko, Does plane wave
not carry a spin?, Am. J. Phys. 69, 405 (2001), and a complement (on
gauge invariance, massive spin 1 and 12 , and massless spin 2 fields) to the
paper by H.C. Ohanian, What is spin?, Am. J. Phys. 54, 500505 (1985).
In particular, it is confirmed that spin is a classical quantity which can be
calculated for any field using its definition, namely that it is just the non-local
part of the conserved angular momentum. This leads to explicit expressions
which are bilinear in the fields and which agree with their standard quantum
counterparts.

The problem of defining and calculating the intrinsic spin carried by plane
waves is a recurring question (see, e.g., [1, 2]) even though it has been definitely
clarified sixty years ago by J. Humblet who recognized that such a definition is
only possible for a wave-packet, that is a superposition of plane waves confined
to a bounded four-dimensional region of space-time [3, p.596]. This concept is in
agreement with standard text books which make clear that plane waves of infinite
extent are just mathematical idealizations and that physical waves correspond to
fields produced a finite time in the past (and so localized to a finite region of
space) [4, p.333], [5, p.99]. More generally, this concept agrees with Julian
Schwingers fundamental idea of source theory, namely that a physical description refers only to the finite space-time region which is under the experimenters
control [6, p.78]. Finally, in practical calculations, this concept leads to a simple
recipe, applicable in many cases, which consists of approximating a wave-packet
by a plane wave provided it is assumed that all contributions coming from infinitely
remote space-time points (i.e., from so-called surface terms) are zero.
It is therefore not surprising that the paper of H.C. Ohanian [1], as well as
both answers to R.I. Khrapkos question [2], i.e., [7, 8], are essentially based on
this concept. These papers focus on the spin 1 electromagnetic field: What about
spin 21 electrons, massive spin 1 particles, and possibly higher spin fields such
1

as spin 2 gravitation? Is there a simple, classical, i.e., non-quantical approach

showing that these fields have indeed spin 12 , 1, and 2 respectively?
In the case of massless spin 1 particles, i.e., the electromagnetic field, Ohanian
~ of a wave-packet can be decomposed into
showed that the angular momentum M
~ (~r) = L(~
~ r) + S,
~ where by use of Gausss theorem one of the parts
two part, M
~
is found to be non-local, i.e., independent of the spacetime point ~r. This part, S,
therefore corresponds to an intrinsic angular moment carried by the wave-packet
~ and the vector potential A,
~ the
as a whole. Written in terms of the electric field E
~
dS
~ A,
~ while the energy
volume density of this spin angular momentum is dV = E
dU
~ E
~ +B
~ B)
~ where B
~ is the magnetic field. The normalized
= 12 (E
density is dV
spin content of a wave-packet is therefore
~
~ A
~
dS
E
= 1
~ E
~ +B
~ B)
~
dU
(E
2

(1)

which is essentially Humblets result [3], confirmed by Jackson [4, p.333], Rohrlich
[5, p.101], Ohanian [1], and many others.
~
The occurrence of the non-observable and gauge-dependent vector potential A
in a measurable quantity deserves an explanation: The crucial step in the derivation
~ B)
~ of the
~ as B
~ =
~ A
~ in the definition dM~ = ~r (E
of (1) is to express B
dV
total angular momentum density. This enables the use of the Gauss theorem to
~ A
~ where, therefore,
discard a surface term and to isolate the non-local term E
~
~ 1 B,
~
A can be considered as an abbreviation for the operational expression ()
~ A
~ in a gauge transformation
[9]. As for the non-invariance of the spin density E
~ A
~ + ,
~ the answer is that by using Maxwells equations and discarding
A
surface terms the additional contribution to the total spin content of a wave-packet
RRR 3
RRR 3 A~
~
~ or
~ . This means that
can be written either
d xE
d x t
the definition (1) is consistent with Maxwells equations and the notion of a wave~ = 0. This
~ = A~ , that is, only if the scalar potential is such that
packet if E
t
condition, which is equivalent to the statement = (t), defines a gauge which by
Maxwells equations for a source-free electromagnetic field yields the constraint
~ A
~ = 0. Therefore, while the separation of the total angular-momentum density

into spin and orbital parts is not gauge invariant, the definition of spin for a wave~ = 0 must be used. This makes this definition
packet implies that the gauge
unique because for a source-free electromagnetic field the gauge function is
~ 2 = 0, whose only solution which is regular
then restricted by the requirement
~ = 0, and (1) is then gauge invariant.
everywhere is a constant, so that
In the case of a massive spin 1 field one has to use Procas equations instead
of Maxwells. It turns out that the calculations made in the case of Maxwells
2

~ and B,
~ as well as
field can be repeated with the difference that the vector fields E
~ and , are now complex. The normalized spin
the vector and scalar potentials A
content of a Proca wave-packet is then
1 ~
~ + E
~ A)
~
~
(E A
dS
2
= 1
~ E~ + B
~ B~ ) + 1 m2 ( + A
~ A~ )
dU
(E
2
2

(2)

which reduces to (1) for a real field of mass m = 0.

Turning to the case of a spin 21 field one meets the difficulty that the Dirac field
is not a classical field in the sense that it is usually not expressed (as the Maxwell
and Proca fields) in terms of scalars and three-dimensional vectors. This is the
reason why in the paper of Ohanian there is no explicit calculation showing that the
spin of Diracs field for an electron is indeed 21 . Fortunately, there is a formulation
of Diracs theory (available since 1929!) which enables such a calculations to be
made in a straight forward manner, and which allows a direct comparison with the
Maxwell and Proca fields. In this formulation [10] the four complex components
of the Dirac field correspond to a scalar s and a three-dimensional vector ~v , which
are combined into a complex quaternion D = s + ~v obeying the Dirac-Lanczos
equation
D = mD i~
(3)
where is the 4-dimensional gradient operator and ~ a unit vector. This equation,
which is fully equivalent to Diracs equation, has many remarkable properties
which stem from the fact that it is written in the smallest possible algebra in which
the Dirac electron theory can be expressed [11, 12]. In the present application,
following the procedure which lead to equations (1) and (2), and without using
any quantum mechanical operator or rule such as projecting onto a subspace or
calculating a Hilbert space scalar-product, one directly finds that the normalized
spin content of an electron wave-packet is given by
~
dS
1
=
dU
2

V[Di~ D ]

1
S[( it
D)i~ D Di~ ( it
D )]
2

(4)

which has an overall factor 12 explicitly showing that the electron field has spin 21 . In
(3, 4) the operator ( ) means changing the sign of the vector part, i.e., s + ~v = s~v ,
while S[ ] and V[ ] mean taking the scalar, respectively vector, part of the quaternion
expression within the square brackets.
We see that there is a great similarity between expressions (1, 2) and (4). The
denominator is always the energy density which is obtained from the scalar part
of the Poynting 4-vector, while the numerator is the vector part of the spin angular

momentum pseudo 4-vector, i.e., Di~ D in Lanczoss formulation of Diracs

3

theory [13]. Moreover, if 21 i~ is interpreted as the spin operator, expression (4)

is exactly the same as the one that would be written in quantum theory for the ratio
between the spin and energy densities.
In principle, the method used here can be generalized to any spin. This
requires to know the explicit form of the corresponding energy-momentum tensor
from which the analogue of the Poynting vector can be derived and used to
calculate the non-local part of the angular momentum. Unfortunately, there is
no unique definition for this tensor, and there is much debate even for the spin 2
case which corresponds to gravitational radiation. Nevertheless, using Einsteins
energy-momentum pseudotensor, standard text books show how to derive the
average rates of loss of angular momentum and energy of a point mass emitting
gravitational waves in terms of its reduced mass quadrupole moments Qjk , e.g.,
[14, p.357] and [15, p.994]. This leads to the expressions
...
... ...
2
dSj
kn Qnl > , dU = 1 m2 < Qjk Qjk >
= m2 jkl < Q
(5)
dt
5
dt
5
which after division by one another show that gravitational radiation has spin 2, as
is confirmed by comparing with the corresponding expressions for the electromagnetic angular momentum and energy loss rates of a point charge of 4-velocity Z
and electric charge e, see, e.g., [14, p.175, p.206]
dSj
2
= e2 jkl < Z k Zl >
dt
3

dU
2
= e2 < Zj Zj > .
dt
3

(6)

In conclusion, the spin content of any field can be defined and calculated
without any reference to quantum theory. In particular, if a polarized plane wave is
used to calculate expressions (1, 2) or (4) one obtains a result that is 1 or 12 times a
normalized unit of angular momentum. If the corresponding wave is attributed to a
single quantum such as a photon, a weak interaction boson, or an electron, this unit
can be taken as the measured value of ~. However, in order to consistently deal
with fields containing a single or a small number of quanta, the classical theory
is not sufficient: It must be supplemented by a quantum interpretation in which
the fields themselves become dynamical variables [4, p.751]. Finally, it is clear
that spin has nothing to do with a vortex or a whirl which would be carried by a
wave or a wave-packet: It is simply the non-local part of the angular momentum
that derives from the dynamics implied by the wave-equations defining the field.

Acknowledgments
The author thanks Prof. J.D. Jackson, Prof. F. Rohrlich, Dr. J.-P. Hurni, and
Dr. A.B. van Oosten for valuable comments and suggestions.
4

References
[1] H.C. Ohanian, What is spin?, Am. J. Phys. 54 (1986) 500505. See also:
W. Gough, The angular momentum of radiation, Eur. J. Phys. 7 (1986)
8187.
[2] R.I. Khrapko, Does plane wave not carry a spin?, Am. J. Phys. 69 (2001)
405.
[3] J. Humblet, Sur le moment dimpulsion dune onde lectromagntique,
Physica 10 (1943) 585603.
[4] J.D. Jackson, Classical Electrodynamics (Wiley, 1975) 848 pp.
[5] F. Rohrlich, Classical Charged Particles (Addison-Wesley, 1965) 305 pp.
[6] J. Schwinger, Particles, Sources, and Fields, Volume I (Addison-Wesley
Pub. Co., 1969) 425 pp.
[7] L. Allen and M.J. Padgett, Response to question #79, Am. J. Phys. 70 (2002)
567568.
[8] V.B. Yurchenko, Answer to question #79, Am. J. Phys. 70 (2002) 568569.
[9] I am indebted to Professor Rohrlich for an illuminating correspondence on
this point. See equation (4-182) in his book [5, p.100].
[10] C. Lanczos, Die tensoranalytischen Beziehungen der Diracschen
Gleichung, Z. f. Phys. 57 (1929) 447473, 474483, 484493.
Reprinted and translated in W.R. Davis et al., eds., Cornelius Lanczos Collected Published Papers With Commentaries (North Carolina
State University, Raleigh, 1998) Vol. III pages 2-1133 to 2-1225.
http://www.physics.ncsu.edu/lanczos .
[11] In Diracs formulation the 4-complex-component electron field is taken as a
4 1 column vector , and the linear operators are 4 4 complex matrices.
In Lanczoss formulation the same 4-complex-component field is a biquaternion D B
= C3,0 . The linear operators are
= C1,2
= M2 (C)
then linear biquaternions functions of biquaternions, which are isomorphic
to the algebra of 4 4 complex matrices M4 (C)
= C4,1 .
= M2 (B)
In both formulations the operator space has 4 4 2 = 32 dimensions
over the reals. The difference is that in the Dirac formulation the field is an
abstract 4-component column vector, while in the Lanczos formulation the
5

field is directly related to the algebraic structure of spacetime because any

biquaternion D = s + ~v is the direct sum of a scalar s and a 3-component
vector ~v .
Lanczoss formulation is therefore more suitable than Diracs for studying
and demonstrating the classical aspects of the electron field, and for
making comparisons with the Maxwell and Proca fields which are usually
expressed in terms of scalars and vectors.
Finally, in terms of Clifford algebras, the Dirac field is a degenerate 8real-component element of the 32-dimensional Clifford algebra C4,1 (i.e.,
an element of an ideal of that algebra) while the Lanczos field D is any 8real-component element of the 8-dimensional Clifford algebra C1,2
= B,
which is therefore the smallest algebra in which Diracs electron theory can
be fully expressed.
For more details, see: A. Gsponer and J.-P. Hurni, Comment on formulating
and generalizing Diracs, Procas, and Maxwells equations with biquaternions or Clifford numbers, Found. Phys. Lett. 14 (2001) 7785. Available
at http://www.arXiv.org/abs/math-ph/0201049 .
For an introduction and overview of the modern use of quaternions
in classical and quantum physics, see: A. Gsponer and J.-P. Hurni,
The physical heritage of Sir W.R. Hamilton, presented at the conference commemorating the sesquicentennial of the invention of quaternions, Trinity College, Dublin (1719 August 1993) 35 pp. Available at
http://www.arXiv.org/abs/math-ph/0201058 .
[12] A. Gsponer, On the equivalence of the Maxwell and Dirac
equations, Int. J. Theor. Phys. 41 (2002) 689694. Available at
http://www.arXiv.org/abs/math-ph/0201053 .
[13] A. Gsponer and J.-P. Hurni, Lanczos-Einstein-Petiau: From Diracs equation to non-linear wave mechanics, in W.R. Davis et al., eds., Cornelius Lanczos Collected Published Papers With Commentaries (North Carolina State University, Raleigh, 1998) Vol. III pages 2-1248 to 2-1277.
http://www.physics.ncsu.edu/lanczos .
[14] L.D. Landau and E.M. Lifshitz, The Classical Theory of Fields (Pergamon,
1975) 402 pp.
[15] C.W. Misner, K.S. Thorn, and J.A. Wheeler, Gravitation (Freeman and Co.,
San Francisco, 1973) 1279 pp.