An Introduction to

Lanczos Potential Theory

Dominic Reiss
Department of Applied Mathematics and Theoretical Physics
University of Cambridge
This essay is submitted as a part of Part III of the Mathematics Tripos
Master of Advanced Study
Fitzwilliam College May 2, 2014
Overview
In this essay, we present a brief review of the development of the theory of the Lanczos
potential, a three index tensor which arises as a Lagrange multiplier when considering
action principles for four dimensional Riemannian and pseudo-Riemannian geometries.
The essay is composed of ve chapters, the structure of which we will now give
an overview. The rst chapter will give a brief introduction to various geometric
objects which will be important in our later discussions. To supplement this any text
on dierential geometry or general relativity, such as Nakahara [10] or Hawking and
Ellis [4], should suce.
We then proceed, in the second chapter, to investigate the motivations and his-
torical development of the Lanczos tensor, following the work of Cornelius Lanczos
in geometric action principles from the 1930s up to his landmark paper in 1962. The
source material for this chapter can be found in papers by Lanczos [59] as well as
sections 4.1-3 of ODonnell [11].
In chapter 3, the algebra and analysis of two component spinors is introduced. Most
importantly, the embedding of the tensor algebra within the spinor algebra, and the
correspondence between a number of tensor and spinor operations is discussed. This
chapter follows material found in sections 2.3-5 and 3.3-4 of Penrose and Rindler [13],
as well as sections 2.2-8 and 3.1-2 of ODonnell [11].
Applying the two-component spinor formalism, we obtain the spinor equivalent of
the Lanczos tensor in the fourth chapter. We continue to nd the spinor equivalent of
the Weyl-Lanczos equations, nding that they simplify signicantly. Here, we follow
sections 4.3-4 of ODonnell [11].
In the nal chapter, we investigate the potential physical signicance of the Lanczos
tensor. Considering the Jordan form of general relativity brings forth many similari-
ties between general relativity and electromagnetism, and suggests that the Lanczos
tensor may play a similar role in gravity as the electromagnetic vector potential does
in electrodynamics. We follow an investigation by Zund [18] which considers transfor-
iv
mations akin to the U(1) gauge transformations of electrodynamics. We also comment
on work by Roberts [15] which suggests that an eect similar to the Aharonov-Bohm
eect of electrodynamics may come into play due to the Lanczos potential.
Contents
Overview iii
Contents v
1 Introduction 1
1.1 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Covariant derivative . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Riemann curvature tensor . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.4 Weyl tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2 Historical formulation of the Lanczos tensor 7
2.1 Motivations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2 Early work in action principles of geometric quantities . . . . . . . . . 8
2.3 Anti-self-dual and self-dual splitting of the Riemann tensor . . . . . . . 12
2.4 Lanczos canonical Lagrangian . . . . . . . . . . . . . . . . . . . . . . . 13
2.5 Lanczos gauge and normalisation conditions . . . . . . . . . . . . . . . 16
2.6 The fundamental nature of H
abc
. . . . . . . . . . . . . . . . . . . . . . 17
2.7 The Weyl-Lanczos equations . . . . . . . . . . . . . . . . . . . . . . . . 19
3 Spinor algebra and analysis 21
3.1 Spinor algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.2 Spinor analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.3 Spin coecients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4 The Lanczos spinor 39
4.1 Lanczos spinor and decomposition . . . . . . . . . . . . . . . . . . . . . 39
4.2 Spinor form of Weyl-Lanczos equations . . . . . . . . . . . . . . . . . . 43
vi Contents
5 Interpretation of the Lanczos tensor 45
5.1 Jordan form of general relativity . . . . . . . . . . . . . . . . . . . . . . 45
5.2 Gauge tensor candidates . . . . . . . . . . . . . . . . . . . . . . . . . . 47
5.3 Aharonov-Bohm-esque eects . . . . . . . . . . . . . . . . . . . . . . . 48
Conclusion 49
References 51
Chapter 1
Introduction
Before examining the history and theory of the Lanczos potential, we discuss a number
of preliminary topics. In this chapter, we dene the fundamental geometric quantities
which are manifest in physical theories, namely the Riemann curvature tensor and
related quantities. In addition, a number of conventions to be used throughout this
paper are dened and justied.
1.1 Preliminaries
1.1.1 Units
We will assume natural gravitational units unless otherwise stated. That is
G = c = 1. (1.1)
1.1.2 Conventions
Dierentiation shorthand
Ordinary partial dierentiation of a tensor with respect to a basis vector will be
represented by an ordinary comma (e.g. R
abcd,e
is shorthand for

x
e
R
abcd
). Similarly,
covariant dierentiation with respect to a basis vector will be represented by a semi-
colon (e.g R
abcd;e
is shorthand for
e
R
abcd
).
2 Introduction
(Anti-)Symmetrisation shorthand
Symmetrisation over a set of indices will be represented by ordinary round brackets
over the set of indices to be symmetrised (e.g. T
(ab)
=
1
2
(T
ab
+T
ba
)). Similarly, anti-
symmetrisation will be represented by square brackets over the set of indices to be
anti-symmetrised (e.g. T
[ab]
=
1
2
(T
ab
T
ba
)). If a set of indices within the (anti-)
symmetrisation is to be omitted from the operation, it shall be separated by vertical
bars (e.g. T
(a|b|c)
=
1
2
(T
abc
+T
cba
)).
Duals of R
abcd
The Riemann tensor, due to its anti-symmetries (1.10) and (1.11), can be dualised
over the rst two, the last two, or both pairs of indices. In the literature during the
early and middle of the 20th century, on which the majority of this review is based, a
single asterisk, *R
abcd
, was often used to indicate the dual over both pairs of indices.
We shall employ a dierent convention in which the dual over the rst pair of indices
shall be represented by *R
abcd
, over the second pair as R*
abcd
, and over both pairs as
*R*
abcd
.
1.1.3 Abstract index notation
This essay will be employing the abstract index notation described in section 2.1 of
ODonnell [11] for describing tensor and spinor equations in a basis invariant formal-
ism, while still being able to use the conveniences of Einstein notation. For a more
rigorous description of abstract index notation, chapter 2 of Penrose and Rindler [13]
gives a complete and formal formulation.
Briey, unbolded Roman indices (a, b, c, d, . . . ) shall represent basis independent
indices, while bold Roman indices (a, b, c, d, . . . ) shall represent basis dependent in-
dices. If an equation contains only unbolded indices, then the equation is valid in
all bases. On the other hand if an expression contains uncontracted bold indices, it
is only valid in a specic basis (often the case if we move to local inertial frame or
employ some symmetry of a system).
In general we shall adopt the notation that lower-case letters represent tenso-
rial indices, while unprimed upper-case letters represent spinorial indices and primed
upper-case letters represent conjugate spinorial indices.
1.2 Covariant derivative 3
1.2 Covariant derivative
The gradient of a scalar function on a manifold Mis given a unique invariant meaning
through the dierential structure of M(see [13], eq. (4.1.32)). However, the denition
of a gradient of any arbitrary valence tensor
1
requires an additional structure on M,
known as a connection which denes the notion of the covariant derivative.
A covariant derivative operator
a
is dened as a linear map from the (p; q) tensors
to the (p; q + 1) tensors which, given an arbitrary valence (p; q) tensor T
b
1
b
2
...b
p
a
1
a
2
...a
q
and (r; s) tensor U
b
1
b
2
...b
r
a
1
a
2
...a
s
, satises the following properties:
1. Leibniz Rule:

a
_
T
b
1
b
2
...b
p
a
1
a
2
...a
q
U
b
1
b
2
...b
r
a
1
a
2
...a
s
_
= T
b
1
b
2
...b
p
a
1
a
2
...a
q

a
U
b
1
b
2
...b
r
a
1
a
2
...a
s
+U
b
1
b
2
...b
r
a
1
a
2
...a
s

a
T
b
1
b
2
...b
p
a
1
a
2
...a
q
. (1.2)
2. Commutation with index substitution.
3. Commutation with contraction:

a
T
b
1
b
2
...c...b
p
a
1
a
2
...c...a
q
= g
d
c

a
T
b
1
b
2
...c...b
p
a
1
a
2
...d...a
q
=
a
g
d
c
T
b
1
b
2
...c...b
p
a
1
a
2
...d...a
q
. (1.3)
4. The metric tensor g
ab
is covariantly constant:

a
g
bc
= 0 and
a
g
bc
= 0. (1.4)
5. The torsion of the connection vanishes
2
:
(
a

a
) = 0, (1.5)
for any scalar function .
1
Besides, of course, a valence (0; 0) tensor, i.e. a scalar.
2
This is not a strict axiom of a general covariant derivative in Riemannian geometry, however
we will henceforth make the assumption that the connection is torsion-free, and is therefore the
Levi-Civita connection.
4 Introduction
6. The covariant derivative acts as an ordinary partial coordinate derivative on
scalars:

a
=
a
, (1.6)
for any scalar function .
1.3 Riemann curvature tensor
The concept of curvature arises naturally if we question whether (1.5) holds for a
general tensor rather than simply scalar functions. We shall see that in the case where
the covariant derivative does not commute with itself, then the manifold has non-zero
curvature. The curvature is measured by a quantity known as the Riemann curvature
tensor R
a
bcd
, which we shall dene below.
1.3.1 Denition
We dene the Riemann curvature tensor to be the mapping of an arbitrary co-vector
eld A
a
under the commutator of a covariant derivative, as
R
a
bcd
A
a
= [
c
,
d
] A
b
= A
b;cd
A
b;dc
. (1.7)
In terms of the Christoel symbols, one can write the Riemann curvature tensor ex-
plicitly as
R
a
bcd
=
c

a
db

a
cb
+
a
ce

e
db

a
de

e
cb
. (1.8)
Lowering the rst index with the metric tensor yields the Riemann curvature tensor
with all covariant indices:
R
abcd
= g
ae
R
e
bcd
. (1.9)
1.3.2 Symmetries
From the denition of the Riemann tensor R
abcd
in terms of the Christoel sym-
bols (1.8) combined with the lowering of the rst index (1.9), a number of symmetries
are evident:
1. Anti-symmetry in a and b:
R
abcd
= R
bacd
. (1.10)
1.3 Riemann curvature tensor 5
2. Anti-symmetry in c and d:
R
abcd
= R
abdc
. (1.11)
3. Symmetry in the exchange of the index pair ab with the index pair cd:
R
abcd
= R
cdab
. (1.12)
4. First Bianchi identity (often called cyclic identity):
R
abcd
+R
acdb
+R
adbc
= 0, (1.13)
or equivalently
R
a[bcd]
= 0. (1.14)
5. Second Bianchi identity (often simply referred to as the Bianchi identity):
R
abcd;e
+R
abde;c
+R
abec;d
= 0, (1.15)
or equivalently
R
ab[cd;e]
= 0. (1.16)
1.3.3 Ricci tensor and scalar
We dene the Ricci tensor R
ab
, often known as the contracted curvature tensor,
through a contraction of the Riemann tensor as
R
ab
= R
c
acb
. (1.17)
As a consequence of the rst Bianchi identity (1.14), the Ricci tensor is symmetric:
R
ab
= R
ba
. (1.18)
It is interesting to note that due to the symmetries of the Riemann curvature ten-
sor, the Ricci tensor (up to a sign) is the only possible non-trivial tensor which one
can achieve through a single contraction of any two indices of the Riemann tensor.
6 Introduction
Performing a second contraction yields a quantity
R = R
a
a
(1.19)
which is known as the Ricci scalar and is the trace of the Ricci tensor. The Einstein
tensor G
ab
of general relativity can be constructed through a trace reversal of the Ricci
tensor as
G
ab
= R
ab

1
2
Rg
ab
. (1.20)
1.4 Weyl tensor
The Weyl tensor C
abcd
can be thought of as the traceless part of the Riemann curvature
tensor. In general relativity, it is the only part of the Riemann tensor which is left
locally undetermined by the Einstein eld equations.
1.4.1 Denition
We dene the Weyl tensor C
abcd
through a decomposition of the Riemann tensor into
its various traces and the Weyl tensor, known as the Ricci decomposition, as
R
abcd
= C
abcd

1
2
(R
ac
g
bd
+R
bd
g
ac
R
ad
g
bc
R
bc
g
ad
)
+
1
6
R(g
ac
g
bd
g
ad
g
bc
) (1.21)
(see [11], eq. 4.29). The denition is such that tracing over any pair of indices of C
abcd
vanishes.
1.4.2 Symmetries
The Weyl tensor shares symmetries (1)(4) of the Riemann tensor in section 1.3.2 due
to its denition (1.21). In addition, we may express its traceless property concisely as
C
a
bac
= 0. (1.22)
Chapter 2
Historical formulation of the Lanczos tensor
In this chapter we will present the historical development of the Lanczos potential.
In 1962 Lanczos released a paper entitled The Splitting of the Riemann Tensor [9] in
which the Lanczos tensor was derived though an action principle. This groundbreak-
ing paper was motivated heavily by the work of Einstein in his famous 1916 paper
connecting Riemannian geometry and physics [1], and in his less famous 1928 paper
on how the classical theories of electromagnetism and general relativity might be uni-
ed under action principles of purely geometric quantities [3]. Lanczos preceded his
1962 paper by his exploration of action principles in geometric quantities in the early
1930s [58].
2.1 Motivations
2.1.1 Einsteins theory of general relativity
In 1916 Einstein released his review on the theory of general relativity [1], asserting
that space-time may not be simply at Minkowski space, but rather a curved mani-
fold for which the contracted Riemann tensor is sourced by the stress-energy tensor
describing the matter present in space-time. This was not the rst theory of physics
to introduce geometry as the source of gravitation; Nordstrms 1912 and 1913 the-
ories were predecessors of general relativity but were experimentally veried to be
incorrect. In his 1962 paper [9], Lanczos observes that since Einsteins theory was so
revolutionary and supported the fundamental signicance of the contracted curva-
ture tensor for the description of the geometrical properties of the physical universe,
it eclipsed the more general aspects of Riemannian geometry for physicists. Lanczos
even mentions that when searching for a unied geometric theory of electromagnetism
8 Historical formulation of the Lanczos tensor
and gravitation, Einstein was compelled to drop the classical Riemannian geometry
in favour of other mathematical tools due to the inability of the Ricci tensor to contain
any information which could be correlated to electric quantities.
2.1.2 Unied theories of classical gravity and classical elec-
tromagnetism
After the onset of Einsteins theory of general relativity, geometry became somewhat
of a fad in developing new theories of physics. Most notable of this trend are theories
such as Einsteins theory of distant parallelism, Veblens, Homans and Paulis
theory of projective reality, as well as the famous ve-dimensional Kaluza-Klein
theory. These theories all had in common the desire to unify the classical eld theories
of electromagnetism and gravity, and express all of known physics in terms of space-
time structure.
2.2 Early work in action principles of geometric
quantities
Lanczos was highly motivated to nd a theory which explained both electromagnetic
and gravitational phenomena through purely geometric means, essentially encoding
the information about either the eld strength tensor or the vector potential into the
curvature coecients, alongside general relativity. Lanczos released several interesting
articles on Riemannian geometry, particularly concerning the determination of four-
dimensional Riemannian geometries through extremal action principles, for several
decades before writing his 1962 paper highlighting the Lanczos tensor as a fundamental
geometric quantity.
In 1938 Lanczos investigated the potential types of scalar invariants which might
appear in a geometric Lagrangian in A Remarkable Property of the Riemann-Christoel
Tensor in Four Dimensions [6]. The paper considers the properties of a scale inde-
pendent
1
action principle in the components of the Riemann tensor. Lanczos assumes
1
Weyl [17] developed the notion of a scale independent geometry which requires a "gauge-
symmetry" of metric scaling invariance. That is the transformation g

ab
= g
ab
where is an
arbitrary function of the coordinates, leaving physical quantities on the manifold invariant. This
invariance is motivated as it has the eect of changing the scale of calibration of the innitesimal
yard-stick. That is, we are only changing the units by which we are measuring other invariants of
the theory.
2.2 Early work in action principles of geometric quantities 9
that the fundamental independent quantities of the theory were the components of the
metric tensor and the curvature tensor. This scale-invariance insists that the scalars
from which the Lagrangian is built are quadratic functions of R
abcd
. Lanczos con-
tinues by analysing all possible quadratic scalars and nds that any scale-invariant
Lagrangian may be constructed from the scalars
I
1
= R
ab
R
ab
, (2.1)
I
2
= R
2
(2.2)
(see [6], eq. 6.5) with all other quadratic scalar terms either being reducible to some
linear combination of (2.1) and (2.2), contributing only as boundary terms, or van-
ishing identically. This work conrms the importance of the Ricci tensor in physical
theories of geometry asserted by Einstein, assuming that all physical theories can be
derived from some extremal action principle.
The generalised Lagrangian L determining a geometry is thus given by
L = R
ab
R
ab
+cR
2
(2.3)
(see [7], eq. 2.1), where c is an undetermined quantity following the investigation
in [6]. Lanczos continued this investigation in 1942 [7] by considering a tensor
S
ab
def
= R
ab

1
4
Rg
ab
(2.4)
(see [7], eq. 2.2) for which is undetermined. The generalised Lagrangian (2.3) can
now be rewritten as
L = S
ab
S
ab
(2.5)
(see [7], eq. 2.3) under the condition:
(2 ) + 4c = 0 (2.6)
(see [7], eq. 2.4). For any c
1
4
, real solutions for can be found. Since the Lagrangian
only contains a term quadratic in S
ab
, essentially treating it as a free-eld, it is clear
that
S
ab
= 0 (2.7)
is a potential solution. Lanczos argues that this is the most stable solution as it
10 Historical formulation of the Lanczos tensor
minimizes the action integral to zero. Comparing the solution (2.7) with (2.4) we nd
R
ab

1
4
Rg
ab
= 0 (2.8)
(see [7], eq. 2.5) and, after contracting through with g
ab
:
(1 ) R = 0 (2.9)
(see [7], eq. 2.6). Here, R = 0 or = 1. The former case, together with with (2.8),
gives the trivial eld equations R
ab
= 0. More interesting is the latter case, = 1,
which when combined with (2.6) gives c =
1
4
and simplies (2.4) to
S
ab
= R
ab

1
4
Rg
ab
(2.10)
(see [7], eq. 2.12). This is precisely the quantity suggested by Einstein in 1919 [2]
to replace the stress-energy tensor in his model of the electron given below in (2.22).
The 1942 article [7] continues by suggesting the stable solution (2.7) was a metrical
plateau over which a high-frequency superposition of carrier elds could be added as
a perturbation. Such a construction would result in a microscopic scale average curva-
ture radius, potentially allowing for the explanation of fundamental particles through
curvature. The basic eld equations of this theory were fourth order in g
ab
, which
troubled Lanczos as the equations of nature seem to oer themselves as dierential
equations of rst and second order.
In 1949 Lanczos revisited the problem in Lagrangian Multipliers and Riemannian
Spaces [8] and considered the metric components g
ab
and the curvature components
R
abcd
, together with Lagrangian multipliers to impose the Bianchi identity, as indepen-
dent metrical quantities in order to reduce the order of the fundamental eld equations
presented in [7] to second order. The Bianchi identity is more naturally expressed in
terms of the dual of the Riemann tensor. We dene the dual of the Riemann tensor
as
*R*
abcd
= R
ijkl

ijab

klcd
(2.11)
(see [8], eq. II.1) where
abcd
is the totally anti-symmetric Levi-Civita tensor. The
Bianchi identity is then expressed as
*R*
a
bcd;a
= 0 (2.12)
2.2 Early work in action principles of geometric quantities 11
(see [8], eq. II.2). Lanczos argues that instead of the scalar given in equation (2.1),
one may dene a dierent scalar as
I

1
= *R*
abcd
*R*
abcd
(2.13)
(see [8], eq. III.1) through a linear combination of (2.1) and (2.2) without losing
information. He then states that for the case of innitesimal elds the contribution
of the scalar (2.2) is an innitesimal of the second order and thus negligible for our
present purposes. This leaves us with an action integral of the following form
1
2
_
(*R*
abcd
)
2
d (2.14)
(see [8], eq. III.3). The Bianchi identity (2.12) is incorporated into the action (2.14)
through a Lagrangian multiplier H
a
bc
, which is anti-symmetric in b and c due to
the properties of the Bianchi identity, but with the symmetry of the rst index left
unspecied [12]. This constraint is added to the action integral as

1
2
_
H
a
bc;d
*R*
abcd
(2.15)
(see [8], eq. III.2), after performing an integration by parts and discarding boundary
terms. After performing the variation to extremize the action, Lanczos found the
following expression for the dual of the Riemann tensor in terms of the Lagrangian
multiplier:
*R*
a
bcd
= H
a
bc;d
H
a
dbc;
+H
a
b d;c
H
a
c d;b
(2.16)
(see [12], eq. 2; [8], eq. III.5). This is immediately analogous to the denition of
the Riemann tensor given in terms of the Christoel symbols, however H
a
bc
and
a
bc
have signicantly dierent properties as noted in [12]. Firstly,
a
bc
transforms as a
connection while H
a
bc
transforms as a tensor. Secondly, H
a
bc
is anti-symmetric in b
and c while
a
bc
, being dened as the torsion-free Levi-Civita connection, is symmetric
in b and c.
Lanczos nally derives eld equations for H
a
bc
by substituting the form found for
the dual Riemann tensor in terms of H
a
bc
(2.16) into the Bianchi identity (2.12),
nding
H
a
bc
H
d a
bc;d
+H
a d
b ;dc
H
a d
c ;db
= 0 (2.17)
12 Historical formulation of the Lanczos tensor
(see [8], eq. IV.1) where is the Laplacian operator
2
. Hence, using the action inte-
gral (2.14) and the Bianchi identities, a system of 24 second order partial dierential
equations for the determination of the 24 independent components of H
a
bc
has been
found.
Although the investigations of his 1949 (and earlier 1932) paper attempting to
unify electricity and magnetism through geometrical quantities in space-time were not
fruitful as a valid theory, the insights Lanczos made into dening geometries via action
principles was an important precursor to the realization of the Lanczos tensor as a
fundamental quantity in four-dimensional Riemannian geometry.
The tensor H
abc
was revisited by Lanczos in his 1962 paper, albeit under a slightly
dierent, yet consistent, denition. It is the same quantity which is now commonly
referred to as the Lanczos tensor. The historical development of this tensors properties
will be the subject of the remainder of the chapter.
2.3 Anti-self-dual and self-dual splitting of the Rie-
mann tensor
We dene the anti-self-dual and self-dual parts of the Riemann tensor respectively as
A
abcd
def
= R
abcd
*R*
abcd
(2.18)
S
abcd
def
= R
abcd
+ *R*
abcd
(2.19)
so that
R
abcd
=
1
2
(A
abcd
+S
abcd
) (2.20)
(see [11], eq. 4.1-3). Clearly, taking the double dual of (2.18) reverses the sign and
similarly leaves (2.19) invariant. This splitting was noted by Rainich in a short let-
ter [14], in which the self-dual tensor (with 11 independent components) was attributed
to gravitational phenomena while the anti-self-dual part (with 9 independent compo-
nents) was attributed to electromagnetic phenomena.
The anti-self-dual tensor (2.18) was shown by Einstein to be reducible to the Ricci
2
Rather than , as Lanczos was working in imaginary time.
2.4 Lanczos canonical Lagrangian 13
tensor through the following decomposition
A
abcd
=
_
R
ac

1
4
Rg
ac
_
g
bd
+
_
R
bd

1
4
Rg
bd
_
g
ac

_
R
ad

1
4
Rg
ad
_
g
bc

_
R
bc

1
4
Rg
bc
_
g
ad
(2.21)
(see [11], eq. 4.4), and is easily veried by moving to normal coordinates (where g

reduces to

), however a fully covariant derivation of (2.21) can be found in section

1 of [9]. Einstein used this identity to derive the following eld equations in [2] in an
attempt to model a stable electron:
R
ab

1
4
Rg
ab
= KT
ab
(2.22)
(see [11], eq. 4.5) where T
ab
is the energy-momentum tensor of Maxwell theory. Only
9 of the 10 components of the energy-momentum tensor are determined by (2.22).
This ambiguity in the eld equations caused Einstein to abandon this analysis of the
anti-self-dual component of the Riemann tensor.
Although Einstein was not fruitful in his development of the eld equations (2.22),
his analysis of the anti-self-dual tensor helped elucidate the nature of four-dimensional
geometry. The structure of the self-dual tensor (2.19), however, was left unanalysed.
Is there a generating function of lower order for the self-dual tensor, similar to
the Ricci tensor in the case of the anti-self-dual tensor? Lanczos [9] makes the bold
statement that without this analysis we cannot claim to have fully understood the
structure of four-dimensional Riemannian geometry.
2.4 Lanczos canonical Lagrangian
Lanczos, taking the investigation of the anti-self-dual curvature tensor as motivation,
revisited action principles in Riemannian geometry by considering, in analogy to (2.14)
and (2.15), the Lagrangian
L

= L(*R*
abcd
, g
ab
) + H
abc
*R*
abcd
;d
+P
ab
c
_

c
ab

_
c
ab
__
+
ab
_
R
ab
+
c
bc,a

c
ab,c
+
c
ad

d
bc

c
ab

d
dc
_
(2.23)
14 Historical formulation of the Lanczos tensor
(see [11], eq. 4.9), where again H
abc
is a Lagrangian multiplier ensuring the Bianchi
identity (2.12) is enforced, P
ab
c
is a Lagrangian multiplier enforcing the equality be-
tween the torsion-free ane connection and the Christoel symbols

c
ab

_
c
ab
_
= 0 (2.24)
(see [11], eq. 4.7), and nally
ab
is the Lagrangian multiplier enforcing the denition
of the Ricci tensor
3
in terms of the Christoel symbols
R
ab
+
c
bc,a

c
ab,c
+
c
ad

d
bc

c
ab

d
dc
= 0 (2.25)
(see [11], eq. 4.8). The Lagrangian (2.23) has canonical variables g
ab
,
c
ab
, and *R*
abcd
,
with conjugate variables
ab
, P
ab
c
, and H
abc
respectively. The Lagrangian multipliers
P
ab
c
and
ab
are symmetric in a and b, however H
abc
is anti-symmetric in a and b
4
.
Lanczos proceeds with his investigation of the Lagrangian multipliers by preparing
to perform the variation with respect to the canonical variables. First he notes that
H
abc
, the conjugate of the dual Riemann tensor (which has 20 independent compo-
nents), with its anti-symmetry in the rst two indices has 24 independent components.
He therefore adds the additional condition, without loss of generality,
*H
ad
a
=
1
2
H
abc

abcd
= 0 (2.26)
(see [9], eq. 2.9), in order to restrict the number of components to 20, which is
equivalent to imposing the cyclic identity:
H
abc
+H
bca
+H
cab
= 0 (2.27)
(see [9], eq. 2.10). Also in preparation of the variation, he splits
ab
into its trace and
its trace-free parts by dening a scalar q as
q
def
=
1
4

ab
g
ab
(2.28)
3
In [6], Lanczos shows that the full Riemann curvature tensor does not enter into the action prin-
ciple, so here only the denition of the contracted curvature tensor must be enforced as a constraint.
4
This denition of the multiplier H
abc
diers from that in Lanczos 1949 paper, however they are
equivalent, and we shall henceforth use the multiplier with these symmetry properties.
2.4 Lanczos canonical Lagrangian 15
(see [9], eq. 2.12) and a tensor Q
ab
as
Q
ab
def
=
ab
qg
ab
(2.29)
so that

ab
= Q
ab
+qg
ab
(2.30)
(see [9], eq. 2.11). Finally, we rewrite the second term of (2.23) using integration by
parts and neglecting the boundary term (as its variation vanishes by construction)
H
abc
*R*
abcd
;d
= H
abc;d
*R*
abcd
(2.31)
(see [11], eq. 4.17). We can now rewrite the canonical Lagrangian as
L

= L(*R*
abcd
, g
ab
) H
abc;d
*R*
abcd
+P
ab
c
_

c
ab

_
c
ab
__
+
_
Q
ab
+qg
ab
_ _
R
ab
+
c
bc,a

c
ab,c
+
c
ad

d
bc

c
ab

d
dc
_
. (2.32)
The variation with respect to
c
ab
can be performed to obtain a relation for P
ab
c
without specifying L(*R*
abcd
, g
ab
) as it does not depend on
c
ab
explicitly. Lanczos
does this (see [9], eq. 2.14) and nds that P
ab
c
can be written in terms of other
variables and thus is not a fundamental quantity.
Next, Lanczos performed the variation with respect to g
ab
to obtain the following
relation
5
L
g
ab
+
1
2
Lg
ab
= Q
ij
*R*
iajb
+Q
ai
*R*
b
i
+Q
bi
*R*
a
i

1
2
RQ
ab
q*R*
ab

1
2
Rqg
ab

1
2
_
P
aib
;i
+P
bia
;i
P
abi
;i
_
(2.33)
(see [9], eq. 2.15).
Finally when performing the variation with respect to *R*
abcd
we notice that the
terms of the Lagrangian with Christoel symbols drop out, and the variation *R*
abcd
becomes
L
*R*
abcd
H
abc;d
+Q
ac
g
bd
qg
ac
g
bd
(2.34)
5
Here we have made use of that fact that *R*
c
a bc
= *R*
ab
=
_
R
ab

1
2
Rg
ab
_
and *R*
a
a
=
*R* = R.
16 Historical formulation of the Lanczos tensor
(see [9], eq. 2.16). Due to symmetries of *R*
abcd
, we must impose a symmetrisation
(represented by square brackets)
6
before allowing the variation to vanish. The resulting
equations from the variation are thus given by
_
L
*R*
abcd
_
=
_
H
abc;d
(Q
ac
qg
ac
) g
bd
_
(2.35)
(see [9], eq. 2.19).
It is important to note that the preceding analysis is completely general and de-
termines the form of the Lagrangian multipliers for a geometry corresponding to the
form of L, and not only for innitesimal geometries as previously considered in the
1942 and 1949 papers [7, 8].
2.5 Lanczos gauge and normalisation conditions
Lanczos noticed that (2.35) possessed a particular gauge invariance through the fol-
lowing transformations
H
abc
H
abc

b
g
ac
+
a
g
bc
(2.36)
Q
ab
Q
ab
+
a;b
+
b;a

1
2

c
;c
g
ab
(2.37)
q q
1
2

a
;a
(2.38)
(see [9], eq. 2.20). Since this invariance leaves the content of (2.35) physically equiv-
alent, we have the freedom to choose a gauge to further restrict (or normalise) the
form of H
abc
through choosing the form of
a
. If we choose

b
=
1
3
H
abc
g
ac
(2.39)
(see [9], eq. 2.21), then the resulting transformation leads to the following identity,
now known as the trace-free condition,
H
b
a b
= 0 (2.40)
6
Lanczos denes the bracketed version of a four-index tensor A
abcd
as:
[A
abcd
] =
1
2
(A
abcd
+A
cdab
+A
badc
+A
dcba
A
bacd
A
cdba
A
abdc
A
dcab
) . Note that this spe-
cial symmetrisation leaves the Riemann tensor and the double dual Riemann tensor invariant (up to
a constant of proportionality).
2.6 The fundamental nature of H
abc
17
(see [9], eq. 2.23) which reduces the number of independent components of H
abc
from
20 to 16.
We can further restrict the form of H
abc
by introducing the divergence-free gauge
condition
H
c
ab ;c
= 0 (2.41)
(see [9], eq. 3.15). These six additional gauge conditions do not respect the general
form of the variation equations (2.33) and (2.35), and reduce the number of indepen-
dent components from 16 to 10. However, for certain explicit forms of L (2.35) is
overdetermined, and we can use (2.41) if we are only concerned with solving (2.35)
and disregard (2.33).
2.6 The fundamental nature of H
abc
In his exploration of the self-dual tensor, Lanczos revisits the scalar invariant K =
R
abcd
*R*
abcd
, which he had previously shown in [6] to not generate eld equations in
action principles for four-dimensional geometries as its variation vanished identically.
However, Lanczos makes the realization that it is exactly for this reason we have
here a variational property which characterizes all Riemannian geometries of four
dimensions.
We can thus make the choice L =
1
8
R
abcd
*R*
abcd
without placing constraints on
the resultant geometry. We nd
L
*R*
abcd
=
1
4
R
abcd
(2.42)
(see [9], eq. 3.2-3). Inserting this into (2.35), and using symmetries of the Riemann
tensor, we obtain
R
abcd
=
_
H
abc;d
(Q
ac
qg
ac
) g
bd
_
(2.43)
(see [9], eq. 3.4). Multiplying (2.43) through by g
ad
and performing the contraction
gives the Ricci-Lanczos equations:
R
bc
= H
a
bc;a
H
a
ba;c
+H
a
c a;b
H
a
c b;a
2Q
bc
+ 6qg
bc
(2.44)
(see [11], eq. 4.26), and after contracting through again by g
bc
we nd
R = 4H
ab
b;a
+ 24q (2.45)
18 Historical formulation of the Lanczos tensor
(see [11], eq. 4.27). Rearranging (2.44) and using (2.45) we arrive at
Q
bc
qg
bc
= H
i
(bc);i
+H
i
(c|i| ;b)

1
3
H
ij
j;i
g
bc

1
2
R
bc
+
1
12
Rg
bc
. (2.46)
We can substitute (2.46) into (2.43) to nd the following expression for the Riemann
tensor, also known as the Riemann-Lanczos equations:
R
abcd
= H
abc;d
+H
cda;b
+H
bad;c
+H
dcb;a
+
_
H
i
(ac);i
+H
i
(a|i| ;c)
_
g
bd
+
_
H
i
(bd);i
+H
i
(b|i| ;d)
_
g
ac

_
H
i
(ad);i
+H
i
(a|i| ;d)
_
g
bc

_
H
i
(bc);i
+H
i
(b|i| ;c)
_
g
ad

2
3
H
ij
j;i
(g
ac
g
bd
g
ad
g
bc
)

1
2
(R
ac
g
bd
+R
bd
g
ac
R
ad
g
bc
R
bc
g
ad
)
+
1
6
R(g
ac
g
bd
g
ad
g
bc
) (2.47)
(see [11], eq. 4.28). Employing the trace-free and divergence-free gauge conditions (2.40)
and (2.41) simplies these equations to the following form:
R
abcd
= H
abc;d
+H
cda;b
+H
bad;c
+H
dcb;a
+H
i
ac;i
g
bd
+H
i
bd;i
g
ac
H
i
ad;i
g
bc
H
i
bc;i
g
ad

1
2
(R
ac
g
bd
+R
bd
g
ac
R
ad
g
bc
R
bc
g
ad
)
+
1
6
R(g
ac
g
bd
g
ad
g
bc
) (2.48)
(see [11], eq. 4.35). Lanczos revisited the splitting of the Riemann tensor (2.20),
balancing the splitting by dening
U
abcd
def
= A
abcd
+
1
6
R(g
ac
g
bd
g
ad
g
bc
) (2.49)
V
abcd
def
= S
abcd

1
6
R(g
ac
g
bd
g
ad
g
bc
) (2.50)
so that R
abcd
= U
abcd
+ V
abcd
, with each U
abcd
and V
abcd
having 10 independent com-
2.7 The Weyl-Lanczos equations 19
ponents. A similar expression to (2.21) can be found for U
abcd
by simply plugging the
denition (2.49) into (2.21) to obtain
U
abcd
=
_
R
ac

1
6
Rg
ac
_
g
bd
+
_
R
bd

1
6
Rg
bd
_
g
ac

_
R
ad

1
6
Rg
ad
_
g
bc

_
R
bc

1
6
Rg
bc
_
g
ad
(2.51)
(see [9], eq. 3.12). Multiplying (2.47) by a factor of two and using the fact that
2R
abcd
= U
abcd
+V
abcd
and subtracting (2.51), we can nd an expression for V
abcd
:
V
abcd
= 2(H
abc;d
+H
cda;b
+H
bad;c
+H
dcb;a
+
_
H
i
(ac);i
+H
i
(a|i| ;c)
_
g
bd
+
_
H
i
(bd);i
+H
i
(b|i| ;d)
_
g
ac

_
H
i
(ad);i
+H
i
(a|i| ;d)
_
g
bc

_
H
i
(bc);i
+H
i
(b|i| ;c)
_
g
ad

2
3
H
ij
j;i
(g
ac
g
bd
g
ad
g
bc
)). (2.52)
Although H
abc
had appeared before in Lanczos 1949 investigation of a theory of
innitesimal elds, he shows here that it holds fundamental signicance to all Rie-
mannian geometries of four dimensions. Lanczos found that H
abc
is a generating
function of V
abcd
. Lanczos made the realization that the tensor V
abcd
contains ex-
actly those components of the full Riemann tensor which are not reducible to the
contracted tensor R
ab
. This highlights the fundamental nature of H
abc
as it con-
tains the exact information necessary to complete the description of the curvature
components of four-dimensional geometries provided g
ab
and R
ab
.
2.7 The Weyl-Lanczos equations
Although Lanczos realized the fundamental signicance of (2.52), it wasnt until
Takeno investigated further in 1964 [16] that it was realized that V
abcd
was in fact
proportional to the Weyl tensor, C
abcd
. We use the Ricci decomposition (1.21) of the
20 Historical formulation of the Lanczos tensor
Riemann tensor along with (2.47) to obtain an expression for the Weyl tensor,
C
abcd
= H
abc;d
+H
cda;b
+H
bad;c
+H
dcb;a
+
_
H
i
(ac);i
+H
i
(a|i| ;c)
_
g
bd
+
_
H
i
(bd);i
+H
i
(b|i| ;d)
_
g
ac

_
H
i
(ad);i
+H
i
(a|i| ;d)
_
g
bc

_
H
i
(bc);i
+H
i
(b|i| ;c)
_
g
ad

2
3
H
ij
j;i
(g
ac
g
bd
g
ad
g
bc
) (2.53)
(see [11], eq. 4.30), known as the Weyl-Lanczos equations. From these it is clear
that V
abcd
= 2C
abcd
. If we again employ the trace-free and divergence-free gauge
conditions (2.40) and (2.41), the Weyl-Lanczos equations simplify to
C
abcd
= H
abc;d
+H
cda;b
+H
bad;c
+H
dcb;a
+H
i
ac;i
g
bd
+H
i
bd;i
g
ac
H
i
ad;i
g
bc
H
i
bc;i
g
ad
(2.54)
(see [11], eq. 4.36).
Chapter 3
Spinor algebra and analysis
In the context of theoretical physics, spin-vectors are often rst dened as a represen-
tation of an orthogonal group (such as the rotations or the Lorentz group) such that
they extend the properties of vectors in a natural way. This is traditionally motivated
by the fact that the properties arising from this representation naturally occur in the
description of physical phenomena, such as Dirac fermions. In this chapter a very
brief introduction to the algebra and analysis of two-component spinors is given. For
a more rigorous and complete description of the two-spinor formalism, one should look
at Penrose and Rindler [13] and ODonnell [11].
3.1 Spinor algebra
3.1.1 Spin-vectors and spin-space
A real valence (r; s) tensor A
a
1
...a
r
b
1
...b
s
at a point p in a manifold is often thought of as
a multi-linear map from r vectors in T
p
(Minkowski space for a Lorentzian manifold)
and s co-vectors in T

p
to R. We can construct algebras analogous to the tensor algebra
by changing the vector space to which the vectors belong, rather than simply choosing
it to be T
p
.
We dene spin space S as a two-dimensional vector space over C which possesses
a bi-linear, skew-symmetric and non-degenerate inner-product (see [11], section 1.5).
Elements of S are known as spin-vectors. If we consider spin-vectors , , , S
and C, then the properties of the inner product [ , ] directly imply the following
22 Spinor algebra and analysis
relations:
[, ] = [, ] , (3.1)
[, ] = [, ] , (3.2)
[ +, +] = [, ] + [, ] + [, ] + [, ] . (3.3)
Consider two spin-vectors , S which are linearly related by complex factor ,
that is = . It is clear from (3.1) and (3.2) that
[, ] = 0. (3.4)
Let (o, ) be a spin-basis for S where o and are arbitrary spin-vectors, under the
condition
[o, ] = 1 (3.5)
(see [11], eq. 1.42). We can expand any spin-vector on this basis
=
0
o +
1
(3.6)
where
0
= [, ] and
1
= [, o] are the components of in this spin basis. In terms
of components, the inner product takes the form
[, ] =
0

0
(3.7)
(see [11], eq. 1.46).
3.1.2 Spin transformations
We consider now general linear transformations of the components of a spin-vector
=
_
_

1
_
_
:

0
= a
0
+b
1

1
= c
0
+d
1
(3.8)
(see [11], eq. 1.29), where a, b, c, d C and ad bc = 1 (so that the inner product is
preserved). These transformations are known as the spin transformations and can be
3.1 Spinor algebra 23
written in matrix form as
_
_

1
_
_
=
_
_
a b
c d
_
_
_
_

1
_
_
. (3.9)
These matrices form a faithful representation of the group SL(2, C), and the spin-
vectors can be thought of as a representation space of this transformation group.
3.1.3 General valence spinors
We now introduce the dual spin-space S

. For every spin-vector S we can con-

struct the map [, ] S

, which we call spin-co-vectors. Similarly to how tensors of

valence (p; q) are constructed as maps from p co-vectors and q vectors to the scalars,
we can construct valence (p; 0; q; 0) spinors as bilinear maps from p spin-co-vectors
and q spin-vectors to C.
We now introduce the operation of complex conjugation of a spinor index. We
represent the complex conjugate of a spin-vector S by S. Similarly, we can
conjugate spin-co-vectors [, ] S

simply by dualising the corresponding element in

S, that is [, ] = [, ].
With complex conjugation dened, we can dene the most general valence (p, q; r, s)
spinor as a bilinear map from p spin-co-vectors, q conjugate spin-co-vectors, r spin-
vectors, and s conjugate spin-vectors to the complex numbers. We introduce abstract
index notation for spinors analogously as we did for tensors, however rather than
lower-case letters, upper-case letters represent spinor indices. If an index is primed, it
represents a conjugate index. Hence a (p, q; r, s) spinor, M, can be represented by the
notation M
A
1
...A
p
B
1

...B
q

C
1
...C
r
D
1

...D
s
. In the case of two-spinors, all indices run over
the set {0, 1}.
3.1.4 Levi-Civita () spinors
The properties of the inner product (3.1), (3.2), and (3.3) ensure the existence of a
spinor
AB
, called the Levi-Civita spinor, such that for any two spin-vectors , S
we have
[, ] =
AB

B
(3.10)
[, ] =
AB

A
(3.11)
(see [11], eq. 2.13-4), from which it is obvious that
AB
=
BA
.
24 Spinor algebra and analysis
The Levi-Civita spinor plays an analogous role to the metric tensor g
ab
. We can
use
AB
to lower indices as is apparent from (3.10) or (3.11) so that
AB

A
is the dual
of
B
and thus

B
=
AB

A
(3.12)
(see [11], eq. 2.16). We can thus write the inner product [, ] in index form as
A

A
,
or as
0

0
+
1

1
in an explicit spin basis. The Levi-Civita spinor with indices upstairs
is similarly dened and raises indices, for example

A
=
AB

B
(3.13)
(see [11], eq. 2.20). If we insert (3.12) into (3.13) we nd

A
=
CB

AB

C
=
A
C

C
(3.14)
where
A
C
is the usual Kronecker delta, and so we arrive at the following properties

AB

AC
=
B
C
=
B
C
, (3.15)

CB

BA
=
C
A
=
C
A
(3.16)
(see [11], eq. 2.21-2). We can thus replace the Kroneker delta
B
A
=
B
A
with the
Levi-Civita spinor
B
A
, if we respect the anti-symmetry properties

B
A
=
B
A
, (3.17)

AB
=
BA
(3.18)
(see [11], eq. 2.23-4).
3.1.5 Spinor dyad bases
We introduced the idea of a spin basis above in (3.5), however the relationship between
the basis spin-vectors can now be written equivalently using (3.10) as
o
A

A
= 1,
A
o
A
= 1,
o
A
o
A
= 0,
A

A
= 0, (3.19)
3.1 Spinor algebra 25
(see [13], eq. 2.5.39-41). We say that the two basis spinors, o
A
,
A
form a dyad in spin-
space in analogy to how the four basis vectors t
a
, x
a
, y
a
, z
a
form a tetrad in Minkowski
space. We can write the dyad o
A
,
A
collectively as
A
A
, where

A
0
= o
A
,
A
1
=
A
(3.20)
(see [13], eq. 2.5.44). The dual basis,
A
A
, must satisfy the condition

A
A

B
A
=
B
A
=
_
_
1 0
0 1
_
_
(3.21)
(see [13], eq. 2.5.47). The components of a spinor with respect to a dyad can be found
using
A
A
, for example the components of
AB
are

AB
=
AB

A
A

B
B
=
_
_
0 o
A

A
o
A

A
0
_
_
(3.22)
(see [13], eq. 2.5.45).
3.1.6 Spinor representation of tensors
The algebra of tensors is embedded in the spinor algebra we have introduced. This
is directly related to the fact that there exists a homomorphism between the linear
transformation group of spin-vectors, SL(2, C), and that of vectors, L(4), although
because this is a homomorphism (and not an isomorphism) there is not always a
tensorial equivalent of a spinorial object (although the spinor equivalent of a tensor
can always be found). If we represent tensors in terms of their equivalent spinors, we
shall nd that many complicated tensorial expressions are simplied by the additional
structure present in the spinor algebra.
We introduce the Infeld-van der Waerden symbols,
a
AB
, as the objects which
connect tensorial indices to their spinor counterparts. Here the index a is a tensorial
index which runs over the tetrad in the tangent space, and A, B

are spinor indices

which run over the spinor dyad. Leaving the spinor indices free, each tensorial com-
ponent is dened as a 2 2 Hermitian matrix which in a normalised spin-basis (as
dened in (3.5)) are the identity matrix and the three Pauli matrices (up to a factor
26 Spinor algebra and analysis
1

2
):

0
AB
=
1

2
_
_
1 0
0 1
_
_
,
1
AB
=
1

2
_
_
0 1
1 0
_
_
,

2
AB
=
1

2
_
_
0 i
i 0
_
_
,
3
AB
=
1

2
_
_
1 0
0 1
_
_
(3.23)
(see [11], eq. 2.69). The Infeld-van der Waerden symbols are an explicit description of
the group homomorphism described above. From their form it is clear that a vector
can be represented by a spin-vector and a conjugate spin-vector. Similarly, any tensor
index is matched to an unprimed and primed spinor index.
We now give an example of how the relationship between tensors and spinors is
realized explicitly. If T
b
a
is a valence (1; 1) tensor, then its spinor-equivalent is
T
b
a
= T
BB

AA
= T
b
a

a
AA

BB

b
(3.24)
where we have used the Infeld-van der Waerden symbols with spinorial indices raised
using the Levi-Civita spinor, and tensorial indices lowered using the metric tensor.
Through explicit calculation, one can verify the following equations:

AA

a

b
AA
=
b
a
, (3.25)

a
AA

BB

a
=
B
A

B

A
(3.26)
(see [11], 2.74-5). The rst property (3.25) is just the statement that the Pauli matrices
(and identity) form an orthonormal basis for the faithful representation of SL(2, C).
The second property (3.26) is that the four world-vectors obtained by keeping A and
A

xed and allowing the tensorial index to vary, form an orthonormal basis for the
faithful representation of L(4).
We dene the metric tensor in terms of the Infeld-van der Waerden symbols as
g
ab
=
AB

B

AA

a

BB

b
(3.27)
(see [11], 2.73), from which the explicit relationship between the Levi-Civita spinor
and the metric tensor can be found by applying (3.25) and (3.26) to (3.27), obtaining
g
ab

a
AA

b
BB
=
AB

B
(3.28)
3.1 Spinor algebra 27
(see [11], 2.76).
From this point forward omission of explicit Infeld-van der Waerden symbols is
standard and it is understood when tensorial indices are replaced by spinor indices (or
vice-versa) that Infeld-van der Waerden symbols have been invoked.
3.1.7 Symmetry operations
One of the chief reasons which cause spinor equivalents of tensorial expressions to
be more elegant or natural algebraically is due to the symmetry properties of two-
dimensional spin-space. Due to these symmetries, any spinor which is anti-symmetric
in three or more primed or unprimed indices vanishes identically. That is, for any
spinor of the form A
AIJK
or B
AI

K
, where the index A represents an arbitrary set
of spinorial indices, we have
A
A[IJK]
= 0, B
A[I

]
= 0 (3.29)
(see [13], eq. 3.3.24). This property is due to the fact that spin-space is two-
dimensional; hence in any set of three anti-symmetrised indices at least two of the
indices must refer to the same dyad component. Let us consider the particular case
of (3.29) for the Levi-Civita spinor

A[B

CD]
= 0 (3.30)
(see [11], eq. 2.54), which we can rewrite as

AB

CD
+
AC

DB
+
AD

BC
= 0 (3.31)
(see [11], eq. 2.55). After raising the indices C and D and rearranging we have

C
A

D
B

C
B

D
A
= 2
C
[A

D
B]
=
AB

CD
(3.32)
(see [11], eq. 2.56).
Another convenient property of spinors is that they can, in a sense, be reduced to
symmetric spinors. Contracting (3.32) through with an arbitrary spinor
ACD
, gives
2
A[AB]
=
C
AC

AB
(3.33)
28 Spinor algebra and analysis
(see [13], eq. 2.5.24). We can then write the spinor
AAB
as

AAB
=
A(AB)
+
A[AB]
=
A(AB)
+
1
2

C
AC

AB
(3.34)
(see [11], eq. 2.59-60).
We dene the equivalence relation between two spinors if their dierence is an
outer-product of Levi-Civita spinors and symmetric spinors of lower valence than the
originals. We rst show that

AA...Z

A(A...Z)
(3.35)
holds for each
AA...Z
. We express the symmetrisation of
A(A...Z)
as

A(A...Z)
=
1
n
_

AA(BC...Z)
+
AB(AC...Z)
+ +
AZ(AB...Y )
_
(3.36)
(see [13], eq. 3.3.50). If we consider the dierence between the rst term and any other
term in the bracketed expression on the right-hand-side of (3.36), and using (3.34),
we nd

AC(AB...Z)
=
AA(BC...Z)
+
AC

X
A (BX...Z)
(3.37)
(see [13], eq. 3.3.51). If we substitute this form of
AC(AB...Z)
into (3.36) and repeat
for all other terms on the right-hand-side, we establish the equivalence

A(AB...Z)

AA(B...Z)
(3.38)
(see [13], eq. 3.3.52). We can now absorb the index A on the right-hand-side of (3.38)
into the index set A and repeat the argument to obtain

A(AB...Z)

AA(B...Z)

AAB(C...Z)

AABC...Z
(3.39)
establishing the desired result. This argument is equally valid for sets of primed
indices, and if we have an arbitrary spinor we have the general result

...F

PQR...Z

(A

...F

)(PQR...Z)
. (3.40)
Furthermore, although the algorithm above was done on a spinor with all indices
downstairs, the argument is general as we may simply lower all the indices of a mixed
3.1 Spinor algebra 29
valence spinor and perform the algorithm.
The decomposition above is actually the decomposition of general spinors into the
direct sum of irreducible representations of SL(2, C), which are in fact the spaces
of symmetric spinors. Another way of expressing this irreducibility of completely
symmetric spinors is by realizing that if we impose any additional (anti-)symmetry,
we either nd the symmetry is redundant or we destroy the spinor.
We give an example of the decomposition to help elucidate the idea of the reduction
to symmetric spinors. Consider the spinor
ABA

B
. We can decompose it as follows

ABA

B
=
(AB)(A

1
2

AB

C
C(A

1
2

B

C

(AB) C

+
1
4

AB

B

C C

C C
(3.41)
(see [13], eq. 3.3.56).
3.1.8 Tensor equivalents of spinor operations
It is clear from the way we have constructed the connection between tensors and spinors
that every algebraic tensor operation has a spinor analogue. However if we have a
spinor associated with some tensor, the richer spinor algebra allows certain tensorial
operations to be represented elegantly. The spinorial operations of exchanging two
unprimed or primed indices do not have a simple tensorial counterpart at rst glance,
and we nd that such operations arise much more naturally when working in the spinor
framework.
Trace reversal over symmetric indices
Consider an arbitrary symmetric tensor of valence (0; 2), T
ab
= T
ba
. We can convert
this to the spinor formalism as
T
AA

BB
= T
BB

AA
(3.42)
30 Spinor algebra and analysis
(see [13], eq. 3.4.2). Adding and subtracting T
ABBA
from the right-hand-side and
rearranging
1
gives
T
ABA

B
=
1
2
(T
ABA

B
+T
ABB

A
) +
1
2
(T
BAB

A
T
ABB

A
) (3.43)
(see [13], eq. 3.4.3). The rst term on the right-hand-side is symmetric in A and B
and in A

and B

through (3.42), and can be written as T

(AB)(A

)
. The second term
on the right-hand-side is anti-symmetric in A and B and in A

and B

by (3.42), and
can be written as T
[AB][A

]
. Using (3.33) twice on each set of antisymmetric indices
in the second term, we get
2
T
ab
= T
ABA

B
= S
ABA

B
+
AB

B
(3.44)
(see [13], eq. 3.4.4), where S
ABA

def
= T
(AB)(A

)
and
def
=
1
4
T
CC

CC
=
1
4
T
c
c
. We can
rewrite (3.44) in tensor form as
T
ab
= S
ab
+g
ab
(3.45)
(see [13], eq. 3.4.7), which is the canonical decomposition of a tensor into its trace-free
part S
ab
and its trace .
We perform the operation of the trace reversal to T
ab
, eectively leaving the trace-
free part invariant and negating the trace, by

T
ab
def
= T
ab

1
2
T
c
c
g
ab
(3.46)
(see [13], eq. 3.4.10). In spinor form we have

T
ab
=

T
AA

BB
= S
ABA

B

AB

B
(3.47)
(see [13], eq. 3.4.12) which upon comparison with (3.44), and recalling the anti-
symmetry of the Levi-Civita spinor gives the simple relation

T
AA

BB
= T
BAA

B
= T
ABB

A
(3.48)
1
Note that primed and unprimed indices may pass through each other with no eect. Only the
exchange of (un)primed indices with other (un)primed indices has a non-trivial eect.
2
Note that this decomposition is equivalent to the one presented in (3.41) with additional sym-
metry.
3.1 Spinor algebra 31
(see [13], eq. 3.4.13). We thus see that trace reversal applied to a pair of symmetric
tensorial indices a and b is realized in the spinor formalism simply by interchanging
the spinor indices A and B or equivalently interchanging A

and B

.
Dualisation over anti-symmetric indices
Consider now an arbitrary anti-symmetric tensor of valence (0; 2), F
ab
= F
ba
, often
called a bivector. We can convert this to the spinor formalism as
F
AA

BB
= F
BB

AA
(3.49)
(see [13], eq. 3.4.15). Adding and subtracting F
ABB

A
from the right-hand-side and
rearranging gives
F
ABA

B
=
1
2
(F
ABA

B
F
ABB

A
) +
1
2
(F
ABB

A
F
BAB

A
) (3.50)
(see [13], eq. 3.4.16). The rst term on the right-hand-side is anti-symmetric in A

and
B

and so we can apply (3.33) to rewrite it as

1
2
F
C

ABC

A

B
. Similarly, the second
term on the right-hand-side is anti-symmetric in A and B and through applying (3.33)
we can rewrite it as
1
2
F
C
C A

B

AB
. Rewriting (3.50) in terms of these decompositions,
we have
F
ABA

B
=
1
2
F
C

ABC

A

B
+
1
2
F
C
C A

B

AB
(3.51)
(see [13], eq. 3.4.17).
The Hodge dual of a bivector F
ab
is given by
*F
ab
def
=
1
2

cd
ab
F
cd
(3.52)
(see [13], eq. 3.4.21), where
cd
ab
is the alternating tensor which has spinor form

cd
ab
= i
C
A

D
B

D

A

C

B
i
D
A

C
B

C

A

D

B
(3.53)
(see [13], eq. 3.3.44). Applying (3.52) and (3.53) to (3.51) gives
*F
ab
= *F
ABA

B
= i
1
2
F
C

ABC

A

B
+i
1
2
F
C
C A

B

AB
(3.54)
32 Spinor algebra and analysis
(see [13], eq. 3.4.22), from which it is clear upon comparison with (3.51) that
*F
ABA

B
= iF
ABB

A
= iF
BAA

B
(3.55)
(see [13], eq. 3.4.23). Hence we see that dualisation of a pair of anti-symmetric
tensorial indices a and b is realized in the spinor formalism simply by interchanging
the spinor indices A and B and multiplication by i or equivalently interchanging A

and B

and multiplication by i.
Procedure over a general pair of indices
Now suppose we have an arbitrary world-tensor of valence (0; 2), G
ab
,with no symmetry
in a and b. We can decompose G
ab
into its symmetric and anti-symmetric parts:
G
AA

BB
= G
(ab)
+G
[ab]
. (3.56)
Using (3.48) and (3.55) on the symmetric part and antisymmetric part of G
ab
respec-
tively, we arrive at
G
BA

AB
=

G
(ab)
+i*G
[ab]
, (3.57)
G
AB

BA
=

G
(ab)
i*G
[ab]
(3.58)
(see [13], eq. 3.4.51-2). If we write these expressions out explicitly in tensorial form,
we have
G
BA

AB
=
1
2
_
G
ab
+G
ba
G
c
c
g
ab
+i
abcd
G
cd
_
, (3.59)
G
AB

BA
=
1
2
_
G
ab
+G
ba
G
c
c
g
ab
i
abcd
G
cd
_
(3.60)
(see [13], eq. 3.4.53-4). The tensorial complexity which arises from the simplest
spinor operation, namely the exchange of two indices, is remarkable. Moreover, the
operations of trace reversal and dualisation appear with high frequency in physics,
often allowing the spinor formalism to be a more natural setting to derive and express
physical expressions.
3.2 Spinor analysis 33
3.2 Spinor analysis
So far we have introduced the algebra of spinors at a point on a manifold. If we want
to express geometric theories, such as general relativity, in a spinor formalism we must
dene some sort of covariant derivative on the manifold by which spinors at dierent
points can be related.
3.2.1 Spinor covariant derivative
We dene the spinor covariant derivative
a
=
AA
, in analogy to the tensor covariant
derivative in section 1.2, by a linear map from the (p, q; r, s) spinors to the (p, q; r +
1, s + 1) spinors which satises the following properties, given an arbitrary valence
(p, q; r, s) spinor T
C
1
...C
p
D
1

...D
q

A
1
...A
r
B
1

...B
s
:
1. Leibniz rule as in (1.2).
2. Commutation with index substitution.
3. Commutation with contraction as in (1.3), however noting that in the spinor
case the contraction may be over a single pair of spinor indices and is performed
with the Levi-Civita spinor
D
C
or
D

C
(rather than the metric tensor).
4. Commutation with complex conjugation:

a
T
C
1
...C
p
D
1

...D
q

A
1
...A
r
B
1

...B
s

=
a
T
C
1
...C
p
D
1

...D
q

A
1
...A
r
B
1

...B
s
(3.61)
(see [11], eq. 3.7).
5. The Levi-Civita spinor (rather than the metric tensor in (1.4)) is covariantly
constant:

BC
= 0 and
a

BC
= 0 (3.62)
(see [11], eq. 3.9).
6. The torsion of the connection vanishes as in (1.5).
7. The spinor covariant derivative acts as an ordinary partial coordinate derivative
on scalars as in (1.6).
34 Spinor algebra and analysis
3.2.2 The curvature spinors
Consider the spinor equivalent of the Riemann tensor:
R
AA

BB

CC

DD
= R
abcd
(3.63)
(see [11], eq. 3.15). Recalling the anti-symmetry of the Riemann tensor in a and b
in (1.10) and using the splitting in (3.50), we have
R
abcd
=
1
2
(R
ABA

CDC

D
R
ABB

CDC

D
)
+
1
2
(R
ABB

CDC

D
R
BAB

CDC

D
)
= R
AB[A

]CDC

D
+R
[AB]B

CDC

D
(3.64)
(see [11], eq. 3.16). Using the anti-symmetry in the second set of indices c and d
in (1.11) and performing the same procedure to each term in (3.64) leads to
R
abcd
= R
AB[A

]CD[C

]
+R
AB[A

][CD]C

+R
[AB]B

CD[C

]
+R
[AB]B

[CD]C

D
(3.65)
(see [11], eq. 3.19). We can then apply (3.33) twice to each term in (3.65) to get
R
abcd
= X
ABCD

B

C

D
+
ABC

D

A

B

CD
+
A

CD

AB

D
+X
A

D

AB

CD
(3.66)
(see [11], eq. 3.20), where X
ABCD
and
ABC

D
are dened as
X
ABCD
def
=
1
4
R
E

ABE

CDF
, (3.67)

ABC

def
=
1
4
R
E

F
ABE

F C

D
(3.68)
(see [11], eq. 3.21-2). X
ABCD
and
ABC

D
are referred to as curvature spinors and
contain the complete information of the original Riemann curvature tensor.
3.2 Spinor analysis 35
3.2.3 The Weyl spinor
Let us now investigate the properties of X
ABCD
further. Under simultaneous exchange
of A and B as well as A

and B
3
, we have anti-symmetry in the Riemann tensor as
in (1.10); contracting over one pair of these indices forces symmetry in the other. Like-
wise, exchange symmetry in the index sets AA

BB

and CC

DD

translates through
to the curvature spinors directly.
Hence, X
ABCD
has the symmetries:
X
ABCD
= X
(AB)(CD)
, (3.69)
X
ABCD
= X
CDAB
(3.70)
(see [11], eq. 3.23, 3.25). Investigating the manifestation of the cyclic symmetry (1.14)
in the curvature spinor X
ABCD
yields the following property
4
:
X
B
ABC
= 3
AC
(3.71)
(see [11], eq. 3.36), where
def
=
1
6
X
AB
AB
is proportional to the trace of X
ABCD
. Hence,
after contracting over the second and fourth indices, we are left with a completely anti-
symmetric spinor
5
.
In an attempt to decompose the anti-symmetric parts of X
ABCD
, we expand using
the following identity
6
:
X
ABCD
= X
(ABCD)
+X
[ABCD]
+
11
12
X
ABCD

1
12
(X
ACDB
+X
ADBC
+X
BCAD
+X
BADC
+X
BDCA
+X
CDAB
+X
CABD
+X
DACB
+X
DCBA
+X
DBAC
) (3.72)
(see [11], eq. 3.54). Clearly, X
[ABCD]
= 0 by (3.69), as anti-symmetrising a pair of
3
This also holds when exchanging both C and D as well as C

and D

.
4
To derive this one should express the cyclic identity in terms of the right dual of the Riemann
tensor as R*
cb
ab
= 0 and nd the corresponding spinor equation in terms of the curvature spinors.
One nds that
ABC

D
drops out of the expression due to its symmetries (see [11], sec. 3.2.1).
5
Note that this is not violating the rule of decomposition of anti-symmetric spinors into symmetric
spinors as here we have the direct product of the completely symmetric zero valence spinor with

AC
.
6
One can verify this identity by simply expanding the terms X
(ABCD)
and X
[ABCD]
.
36 Spinor algebra and analysis
symmetric indices annihilates the object. Similarly, (3.69) gives X
(ABCD)
= X
A(BCD)
,
as symmetrising an index set with a member of a symmetric index set is equivalent
to simply symmetrising over the union of both index sets. With this, (3.72) can be
simplied to
X
ABCD
= X
A(BCD)
+
1
3
(X
ABCD
X
ACBD
) +
1
3
(X
ABCD
X
ADCB
)
= X
A(BCD)
+
1
3
X
A[BC]D
+
1
3
X
A[B|C|D]
(3.73)
(see [11], eq. 3.56), which is now in a form where we can apply (3.33) to arrive at
X
ABCD
= X
A(BCD)
+
1
3

BC
X
E
AE D
+
1
3

BD
X
E
AEC
(3.74)
(see [11], eq. 3.57). Applying (3.71) to (3.74), we can complete the decomposition as
X
ABCD
=
ABCD
+ (
BC

AD
+
BD

AC
) (3.75)
(see [11], eq. 3.58), where
ABCD
is dened as

ABCD
def
= X
A(BCD)
(3.76)
and is often called the Weyl spinor or gravitational spinor.
Rewriting (3.66) using the Weyl spinor to yields
R
abcd
=
ABCD

B

C

D
+
A

D

AB

CD
+
ABC

D

A

B

CD
+
A

CD

AB

+ 2(
AC

BD

C

B

D

AD

BC

D

B

C
) . (3.77)
This decomposition is the spinor equivalent of the Ricci decomposition (1.21), and
through further analysis it is possible to show the equivalence of the Weyl tensor and
the Weyl spinor as
C
abcd
=
ABCD

B

C

D
+
A

D

AB

CD
(3.78)
(see [11], eq. 3.61).
3.3 Spin coecients 37
3.3 Spin coecients
The spin-coecient formalism, also known as the Newman-Penrose formalism, is a
notation equivalent to the spinor formalism often used numerical relativity. Although
somewhat beyond the scope of this essay, we shall provide a brief summary of the
formalism here.
A set of four null vectors is chosen: two real null vectors and two complex-
conjugates. The idea is that the information in tensors in the theory are projected
onto this null tetrad. In essence, the spin-coecient formalism gives the relationship
between:
1. Twelve complex spin-coecients which store information about the directional
covariant derivatives along the tetrad vectors.
2. Five complex functions which store information about the Weyl tensor in the
tetrad basis.
3. Four real functions and three complex functions (and their conjugates) which
store information about the Ricci tensor in the tetrad basis.
Although seemingly complicated and arduous, working with a specic null tetrad
to exploit the symmetries of the space-time can cause several of the spin-coecients
to be trivial. Working in the spin-coecient formalism is often a preferred method in
numerical relativity. The reader is suggestion to consult section 3.5-6 of [11] for more
information on the spin-coecient formalism.
Chapter 4
The Lanczos spinor
We saw in the Lanczosian development of H
abc
in chapter 2 that the Lanczos tensor is
a fundamental quantity in four-dimensional Riemannian geometry. Shortly after this
realization was made, it was shown that in fact the Lanczos tensor was a potential
for the Weyl tensor through the Weyl-Lanczos equations (2.54). However, these ex-
pressions are quite complicated and it can be hard to work with them analytically.
Furthermore, suppose we wish to solve these equations numerically for the quantity
H
abc
; the Weyl-Lanczos equations are 16 non-linear equations in 16 unknowns in a
convoluted form, and are thus computationally expensive.
In this chapter we make use of the properties of the 2-spinor formalism introduced
in chapter 3 to express the theory of the Lanczos potential in spinor form, closely
following the procedure in ODonnell [11].
4.1 Lanczos spinor and decomposition
We convert the Lanczos tensor H
abc
to spinor form
H
AA

BB

CC
= H
abc
(4.1)
in the usual way, omitting the explicit invocation of the Infeld-van der Waerden sym-
bols. The Lanczos tensor has an anti-symmetry in the a and b index, which is expressed
in the spinor formalism as
H
AA

BB

CC
= H
BB

AA

CC
(4.2)
40 The Lanczos spinor
(see [11], eq. 4.43). We split (4.2) using (3.50) as
H
abc
= H
ABA

CC

=
1
2
(H
ABA

CC
H
ABB

CC
)
+
1
2
(H
ABB

CC
H
BAB

CC
) (4.3)
(see [11], eq. 4.44). The rst term on the right-hand-side is anti-symmetric in A

and
B

, which when combined with (4.2) indicates symmetry in A and B. The second
term on the right-hand-side is anti-symmetric in A and B, which similarly indicates
symmetry in A

and B

. We can thus write H

abc
as
H
abc
= H
ABA

CC
= H
(AB)[A

]CC
+H
[AB](B

)CC
(4.4)
(see [11], eq. 4.45). Recalling the identity (3.33) for a pair of anti-symmetric spinor
indices, we can rewrite each term in (4.4) as the direct product of a Levi-Civita spinor
and a spinor of lower valence, and hence (4.4) can be rewritten as
H
abc
= H
ABA

CC
=
1
2

B
H
D

(AB)D

CC
+
1
2

AB
H
D
D (B

)CC
(4.5)
(see [11], eq. 4.47).
We now dene the spinors H
ABCC
and
A

CC
for convenience as
H
ABCC
= H
(AB)CC

def
=
1
2
H
D

(AB)D

CC
, (4.6)

CC
=
(A

)CC

def
=
1
2
H
D
D (B

)CC
(4.7)
(see [11], eq.4.48), allowing us to write (4.5) as
H
abc
= H
ABA

CC
=
A

B
H
ABCC
+
AB

CC
(4.8)
(see [11], eq. 4.49). Taking the conjugate of (4.8) and noting the fact that the
conjugate of a real tensor is itself gives
H
abc
=
AB
H
A

CC
+
A

B

ABCC
(4.9)
(see [11], eq. 4.50), which when equating with (4.8) gives the consistent relations
4.1 Lanczos spinor and decomposition 41

CC
= H
A

CC
and
ABCC
= H
ABCC
. Thus, we may write (4.8) in terms of
H
ABCC
and its conjugate H
A

CC
alone:
H
abc
= H
ABA

CC
=
A

B
H
ABCC
+
AB
H
A

CC
(4.10)
(see [11], eq. 4.51). The complete information contained in the Lanczos tensor is also
contained in the quantity H
ABCC
, which we henceforth will refer to as the Lanczos
spinor.
So far, we have only imposed the anti-symmetry of the Lanczos tensor in its rst
two indices, resulting in the following symmetry of the Lanczos spinor:
H
ABCC
= H
(AB)CC
. (4.11)
Without imposing additional symmetry, the Lanczos spinor as it stands has 12 inde-
pendent complex components, containing the same information as the 24 real inde-
pendent components of the Lanczos tensor before imposing (2.26), (2.40), and (2.41).
We proceed to nd the form of these additional symmetries for the Lanczos spinor.
First we investigate the trace-free gauge condition (2.40),
H
b
a b
= H
abc
g
bc
= H
AA

BB

CC

BC

= 0, (4.12)
which after substituting the form of H
AA

BB

CC
in (4.10) gives
H
b
a b
=
_

B
H
ABCC
+
AB
H
A

CC

BC

= H
ABCA

BC
H
A

= 0, (4.13)
or, equivalently,
H
ABCA

BC
= H
A

(4.14)
(see [11], eq. 4.53).
Now we impose (2.26), which is equivalent to the cyclic property (2.27), in spinor
form as
*H
abc
g
bc
=
_
i
A

B
H
ABCC
+i
AB
H
A

CC

BC

= H
ABCA

BC
+H
A

= 0, (4.15)
42 The Lanczos spinor
or, equivalently,
H
ABCA

BC
= H
A

(4.16)
(see [11], eq. 4.54) by recalling (3.55). We see that the right-hand-side of (4.14)
and (4.16) are equal, and thus equating the left-hand-sides implies
H
D
AD A
= 0 (4.17)
(see [11], eq. 4.55). Once again calling upon the identity (3.33), although this time in
reverse, we rewrite (4.17) as
1
2

BC
H
D
AD A
= H
A[BC]A
= 0. (4.18)
Since, when anti-symmetrised on indices B and C, H
ABCA
vanishes identically, it
must instead be symmetric in B and C, that is
H
ABCC
= H
A(BC)C
(4.19)
(see [11], eq. 4.56), which when combined with the original symmetry of the Lanczos
spinor (4.11) can be written as
H
ABCC
= H
(ABC)C
(4.20)
(see [11], eq. 4.57). At this point, there are eight remaining independent complex
components of the Lanczos spinor which match the 16 real components of the Lanczos
tensor after imposing the cyclic and trace-free conditions. We do not yet impose
the divergence-free condition, but note that it has the following form in the spinor
formalism
1
:
H
c
ab ;c
=
CC

_
H
CC

AB

A

B
+H
C

C
A

B

AB
_
=
A

B

CC
H
CC

AB
+
AB

CC
H
C

C
A

= 0 (4.21)
(see [11], eq. 4.59). Contracting through with
A

and recalling the symmetry of the

1
The second expression follows since we have assumed the Levi-Civita spinor is covariantly con-
stant in (3.62).
4.2 Spinor form of Weyl-Lanczos equations 43
Lanczos spinor and its conjugate in the rst two indices in (4.6) gives

CC
H
CC

AB
= 0 (4.22)
(see [11], eq. 4.60). Contracting (4.21) through with
AB
, we nd a similar expression
for the conjugate Lanczos spinor:

CC
H
C

C
A

B
= 0. (4.23)
4.2 Spinor form of Weyl-Lanczos equations
Assuming the divergence-free gauge condition, however not invoking it explicitly and
simply holding (4.22) and (4.23) as auxiliary conditions, we investigate the form of
the Weyl-Lanczos equations (2.54) in the spinor formalism, recalling the decomposi-
tion (3.78) of the Weyl tensor into the Weyl spinor and its conjugate, as
C
abcd
=
ABCD

B

C

D
+
A

D

AB

CD
=
DD

_
H
ABCC

A

B
+H
A

AB
_

CC

_
H
ABDD

A

B
+H
A

AB
_
+
BB

_
H
CDAA

C

D
+H
C

CD
_

AA

_
H
CDBB

C

D
+H
C

CD
_

EE

_
H
E
A CC

E

A
+H
E

E
A
_

BD

EE

_
H
E
B DD

E

B
+H
E

E
B
_

AC

+
EE

_
H
E
A DD

E

A
+H
E

E
A
_

BC

+
EE

_
H
E
B CC

E

B
+H
E

E
B
_

AD

D
. (4.24)
Contracting through with
A

, recalling that the Weyl spinor and its conju-

gate inherit symmetry in their rst and last index pairs from the curvature spinor
X
ABCD
and that the Levi-Civita symbol is covariantly constant, leads to the following
simplication:
2
ABCD
=
E

D
H
ABCE
+
E

C
H
ABDE

+
E

B
H
CDAE
+
E

A
H
CDBE
(4.25)
44 The Lanczos spinor
(see [11], eq. 4.58). These are the Weyl-Lanczos equations expressed in the spinor
formalism, together with the divergence-free gauge condition (4.22).
We can express the divergence-free condition in a way which can be directly incor-
porated into (4.25). We apply (3.33) in reverse to (4.22) to write the gauge condition
as
1
2

DE

CC

H
ABCC
=
C

[E
H
|AB|D]C
= 0, (4.26)
or equivalently

E
H
ABDC
=
C

D
H
ABEC
(4.27)
(see [11], eq. 4.61). Finally, we can use two applications of (4.27) with dierent index
permutations to simplify (4.25) to its nal form, with both gauge conditions invoked:

ABCD
= 2
E

D
H
ABCE
(4.28)
(see [11], eq. 4.62).
Although elegant, the nal spinor form of the Weyl-Lanczos equations (4.28) are
still not easy to work with. A common approach to solving for the Lanczos spinor in
a given space-time is by applying the spin-coecient formalism, briey described in
section 3.3, to nd the Lanczos coecients. For more information, one should consult
section 4.4-12 in [11].
Chapter 5
Interpretation of the Lanczos tensor
We have shown the fundamental signicance of the Lanczos tensor in describing four-
dimensional Riemannian geometries, however its physical interpretation is often over-
looked. In a 2010 review of Lanczos potential theory [12], ODonnell and Pye state
that the interpretation of the Lanczos potential remains largely uninvestigated and
is also arguably the most important factor of the theory that requires delineation.
In this chapter, we rst explore Zunds 1975 investigation [18] of the ramications
of the Weyl-Lanczos equations in the Jordan form of general relativity. We then
briey discuss the ideas behind Roberts 1995 investigation of the possibility of an
eect analogous to the Aharonov-Bohm eect arising in the quantum realm of gravity.
5.1 Jordan form of general relativity
The eld equations of general relativity are most commonly expressed as a local de-
termination of the Ricci tensor R
ab
in terms of the stress-energy tensor T
ab
present
in the space-time. In particular we have a proportionality between the trace-reversed
Ricci tensor and the stress-energy tensor, known as the Einstein eld equations:
R
ab

1
2
g
ab
R = 8T
ab
. (5.1)
These equations are a set of 10 coupled non-linear partial dierential equations in the
metric and its rst and second derivatives
1
.
As can be seen from the Ricci decomposition of the Riemann tensor (1.21), the
1
Note, however, that the contracted Bianchi identity and the fact that the metric is covariantly
constant implies that the Einstein equations are divergenceless, and thus are equivalent to 6 inde-
pendent dierential equations.
46 Interpretation of the Lanczos tensor
Ricci tensor and Weyl tensor are algebraically independent parts of the curvature, and
thus the Weyl tensor is not determined locally by the Einstein equations. However,
applying the second Bianchi identity (1.16) to the Ricci decomposition (1.21) leads to
a relation between the rst derivatives of the Weyl tensor and the rst derivatives of
the Ricci tensor. Hence, the global form of the Weyl tensor is determined indirectly
by the matter distribution.
A once contracted form of the second Bianchi identity can be written using the
Weyl tensor as
C
abcd
;d
= J
abc
, (5.2)
where we dene
J
abc
def
= R
c[a;b]
+
1
6
g
c[b
R
;a]
(5.3)
(see [4], eq. 4.28-9). Since the Ricci tensor is completely determined locally by the
stress-energy tensor, it is possible to rewrite (5.3) fully using T
ab
and its trace T.
Therefore, we can think of (5.2) as the eld equations of general relativity encoded as
rst-order dierential equations for the components of the Weyl tensor, with source
J
abc
. This form of general relativity bears striking resemblance to the rst Maxwell
equations in terms of the eld tensor:
F
ab
;b
= J
a
. (5.4)
The second Maxwell equation in terms of the dual eld tensor,
*F
ab
;b
= 0, (5.5)
has an analogous expression corresponding to the left dual
2
of the Weyl tensor, by
rewriting the traceless property (1.22) as
*C
ab
cd;a
= 0, (5.6)
(see [18], eq. 14).
The eld tensor F
ab
is generated dierentially in terms of the vector potential A
a
as
F
ab
=
[a
A
b]
. (5.7)
In a similar way, the Weyl-Lanczos equations (2.53) show that the Weyl tensor is
2
Or, equivalently due to the symmetries of the Weyl tensor, the right dual.
5.2 Gauge tensor candidates 47
generated dierentially in terms of the Lanczos tensor H
abc
.
5.2 Gauge tensor candidates
In a 1975 article entitled The Theory of the Lanczos Spinor [18], Zund observes the
above correspondences between general relativity and electromagnetism and asks the
obvious question of whether there also exists a gauge group under which the equations
of general relativity, as they are posed in (5.2), remain invariant; similar to how
Maxwells equations are invariant under the U(1) transformation:
A
a
A

a
= A
a
+
a
(5.8)
which leaves the eld tensor invariant. More precisely, Zund asks whether there exists
a tensor L
abc
, such that the transformation
H
abc
H

abc
= H
abc
+L
abc
(5.9)
(see [18], eq. 18) leave the Weyl tensor invariant:
C

abcd
= C
abcd
. (5.10)
In his investigation Zund sought forms of L
abc
that leave the Weyl tensor invariant.
One particular form he considered, as the closest analogy to electromagnetism, was
L
abc
=
c

ab
(5.11)
(see [18], eq. 19).
ab
must be a bivector as the Lanczos tensor is anti-symmetric in
its rst two indices. Furthermore, the cyclic property and trace-free condition become
*
ab;c
= 0 and
ab
;a
= 0, (5.12)
which are equivalent to the source-free Maxwell equations, implying that
ab
is a
singular bivector. Unfortunately, the divergence-free condition causes diculties, and
thus we make the further assumption that
ab
is recurrent, that is

ab;c
=
c

ab
, (5.13)
48 Interpretation of the Lanczos tensor
(see [18], eq. 20). With this assumption, the conditions (5.12) require the co-vector
eld
c
to be a principal null direction of
ab
. The divergence free equation then
becomes,
a
;a
= 0. Finally Zund found that C

abcd
= C
abcd
if

c
[b

a];c
= 0, (5.14)
which is satised if
a
is a parallel null vector eld. Hence, if
a
is both a parallel null
vector eld and principally null with respect to
ab
, which itself is a singular recurrent
bivector with recurrence vector
a
, then (5.11) is a possible form of the gauge eld.
There are certainly other forms of L
abc
which leave the Weyl tensor invariant.
Zund concluded, however, that only a deeper physical study of the gravitational eld
will indicate what kind of gauge tensor is appropriate. It will certainly be fruitful
to consider the algebraic structure of the full group of gauge transformations, and
consider possibly restricting it to a subgroup which can be physically motivated by
general relativity. Furthermore, considering the Weyl-Lanczos equations in spinor
form may lead to simplications in this analysis.
5.3 Aharonov-Bohm-esque eects
In his paper, entitled The physical interpretation of the Lanczos tensor [15], Roberts
gives a potential eect of the presence of the Lanczos potential, analogous to the
Aharonov-Bohm eect of electrodynamics, occurring in the quantum realm of gravity.
The Aharonov-Bohm eect arises from the fact that Maxwells equations are cast as
a gauge theory, and thus we can replace partial derivatives with a gauge covariant
derivative with respect to the gauge group of electrodynamics. Replacing the partial
derivatives by gauge covariant derivatives in the Schroedinger equation gives rise to
several factors of the vector potential, which aects charged particles in the system
directly. Consider, for example, a space-time with vanishing eld tensor, but non-
trivial vector potential. Classically such a scenario would not have any electromagnetic
eect on charged particles in the space-time. However, in the quantum realm this is
not the case and the vector potential is shown to be, in a sense, physical.
Roberts constructed many potential gauge covariant derivatives for the Jordan
form of general relativity in terms of the Lanczos potential. In the end, the results of
the study were inconclusive as to whether the Lanczos tensor may produce an eect
similar to the Aharonov-Bohm eect.
Conclusion
Although Lanczos failed to realize his intentions of producing a unied geometric the-
ory of electromagnetism and gravity, his work in geometric action principles helped elu-
cidate a fundamental quantity in four-dimensional Riemannian and pseudo-Riemannian
geometry. The Weyl-Lanczos equations highlight that we should not consider the Weyl
tensor as being fundamental, but rather the Lanczos tensor should be considered a
fundamental constituent of four-dimensional Riemannian geometry. Moving to the
spinor formalism, we see that the complex relationship of the Weyl-Lanczos equations
simplies incredibly in terms of the Weyl spinor and the Lanczos spinor, highlighting
the power of simple operations in the spinor formalism in replicating complex tensorial
operations.
It seems that the Lanczos tensor is currently largely unnoticed, possibly in part
due to our knowledge of classical gravity being regarded as somewhat complete while
other elds often recognized as more fundamental or applicable take the stage. How-
ever, there is still signicant work to be done regarding the Lanczos tensor, such as
further investigating its role as a potential in the Jordan form of general relativity, or
generalizing solutions for the Lanczos tensor to arbitrary space-times.
References
[1] Einstein, A. (1916). Die Grundlage der allgemeinen Relativitatstheorie. Annalen
der Physik, 49(7):769822.
[2] Einstein, A. (1919). Spielen Gravitationsfelder im Aufbau der materiellen Ele-
mentarteilchen eine wesentliche Rolle? Preussische Akademie der Wissenshchaften,
(1):349356.
[3] Einstein, A. (1928). Neue Mglichkeit fr Eine Einheitliche Feldtheorie von Gravi-
tation und Elektrizitt. Preussische Akademie der Wissenshchaften, 18(1):224227.
[4] Hawking, S. and Ellis, G. (1973). The Large Scale Structure of Space-Time. Cam-
bridge University Press.
[5] Lanczos, C. (1932). Electricity as a natural property of Riemannian geometry.
Physical Review, 39(4):716736.
[6] Lanczos, C. (1938). A remarkable property of the Riemann-Christoel tensor in
four dimensions. Annals of Mathematics, 39(4):842850.
[7] Lanczos, C. (1942). Matter waves and electricity. Physical Review, 61(11):713720.
[8] Lanczos, C. (1949). Lagrangian multiplier and Riemannian spaces. Reviews of
Modern Phyiscs, 21(3):497502.
[9] Lanczos, C. (1962). The splitting of the Riemann tensor. Reivews of Modern
Physics, 31(3):379389.
[10] Nakahara, M. (2003). Geometry, Topology, and Physics. IOP Publishing.
[11] ODonnell, P. (2003). Introduction to 2-Spinors in General Relativity. World
Scientic.
[12] ODonnell, P. and Pye, H. (2010). A brief historical review of the important
developments in Lanczos potential theory. Electronic Journal of Theoretical Physics,
7(24):327350.
[13] Penrose, R. and Rindler, W. (1984). Spinors and Space-time; Volume 1: Two-
spinor calculus and relativistic elds. Cambridge University Press.
[14] Rainich, G. (1925). Electricity in curved space-time. Nature, 115(2892):498.
52 References
[15] Roberts, M. (1995). The physical interpretation of the Lanczos tensor. Il Nuovo
Cimento B, 110(10):11651176.
[16] Takeno, H. (1964). On the spintensor of Lanczos. Tensor, 104:103119.
[17] Weyl, H. (1919). Eine neue Erweiterung der Relativittstheorie. Annalen der
Physik, 364(10):101133.
[18] Zund, J. (1975). The theory of the Lanczos spinor. Annali di Matematica Pura
ed Applicata, 15(1):239268.