Twistor Algebra

R. Penrose

Citation: J. Math. Phys. 8, 345 (1967); doi: 10.1063/1.1705200

View online: http://dx.doi.org/10.1063/1.1705200
View Table of Contents: http://jmp.aip.org/resource/1/JMAPAQ/v8/i2
Published by the American Institute of Physics.

Additional information on J. Math. Phys.

Journal Homepage: http://jmp.aip.org/
Journal Information: http://jmp.aip.org/about/about_the_journal
Top downloads: http://jmp.aip.org/features/most_downloaded
Information for Authors: http://jmp.aip.org/authors

Downloaded 13 Nov 2012 to 139.184.30.132. Redistribution subject to AIP license or copyright; see http://jmp.aip.org/about/rights_and_permissions
JOURNAL OF MATHEMATICAL PHYSICS VOLUME 8. NUMBER 2 FEBRUARY 1967

Twistor Algebra
R. PENROSE
Department of Mathematics, Birkbeck College, London, England
(Received 16 May 1966)

A new type of algebra for Minkowski space-time is described, in terms of which it is possible
to express any conformally covariant or Poincare covariant operation. The elements of the algebra
(twistors) are combined according to tensor-type rules, but they differ from tensors or spinors in that
they describe locational properties in addition to directional ones. The representation of a null line
by a pair of two-component spinors, one of which defines the direction of the line and the other, its
moment about the origin, gives the simplest type of twistor, with four complex components. The
rules for generating other types of twistor are then determined by the geometry. One-index twistors
define a four-dimensional, four-valued ("spinor") rep);esentation of the (restricted) conformal group.
For the Poincare group a skew-symmetric metric twistor is introduced. Twistor space defines a complex
projective three-space C, which gives an alternative picture equivalent to the Minkowski space-time
M (which must be completed by a null cone at infinity). Points in C represent null lines or "complexi-
fied" null lines in M; lines in C represent real or complex points in M (so M, when complexified, is the
Klein representation of C. Conformal transformations of M, including space and time reversals (and
complex conjugation) are discussed in detail in twistor terms. A theorem of Kerr is described which
shows that the complex analytic surfaces in C define the shear-free null congruences in the real space M.
Twistors are used to derive new theorems about the real geometry of M. The general twistor descrip-
tion of physical fields is left to a later paper.

I. INTRODUCTION of (1.1) involves integrals). The invariance of all

expressions under the translation
I N the study of Lorentz covariant local field theo-
ries, field quantities (with spin) are normally de-
scribed by vectors, tensors, or, more generally, by
(1.2)

spinors, all finite-dimensional representations of the is generally ensured by the fact that the (Minkow-
Lorentz group' being expressible in terms of spinors skian) coordinates Xi must, themselves, never enter
(vectors and tensors being regarded as effectively into the field equations explicitly, only the operators
special cases). A local (restricted) Lorentz trans- a/ ax' being permitted to occur.
formation then takes the form I t is curious that the invariance under two types of
transformation (1.1) and (1.2), each of which simply
t-(-I) v'
J' , refers to symmetries of the Minkowski space-time,
should find mathematical expression in so different
(1.1) a way. This is, moreover, not just a property of a
(t;) being a complex unimodular (2 X 2) matrix, difference in group structure between the translation
with inverse (t<-l);). Except where general rela- and rotation elements of the Poincare group, but
tivity is involved, normally it is further required rather a consequence of what is meant by a local
that physical quantities be suitably covariant under field theory. In fact, mathematical formalisms do
the full Poincare group.' However, this covariance exist2 in which all the Poincare transformations are
is expressed (locally) in a totally different way from represented according to an algebraic" tensor type"
the local Lorentz covariance. Whereas the depend- law similar to (1.1). It is the purpose of the present
ence on space-time direction is expressed algebrai- paper to exhibit and discuss in some detail one such
cally [ef., (1.1)], the dependence on position is de- formalism, the basic elements of which are re-
scribed by differential equations (so that the analog
2 The use of homogeneous coordinates in space-time would
afford a simple example of such a formalism (but apparently
I Throughout this paper, "Lorentz group" always refers to one of limited physical interest). More significant, of course
to the six-parameter homogeneou8 group and "Poincare group" is the representation of physical quantities in terms of Hilbert
to the ten-parameterinhomogeneou8 group. "Conformal group" space, which has the advantage that the (infinite demensional)
refers to the fifteen-parameter group of transformations which analogs of (tM A ) are unitary [cf., particularly E. P. Wigner,
preserve the local conformal structure of Minkowski space- Ann. Math. 40, 149 (1939) V. Bargmann and E. P. Wigner,
time. If in any particular context it is important to exclude the Proc. Nat!. Acad. SCl. U. S. 34, 211 (1948)]. However, it will
space reversing and time reversing transformations, then this be essential here to preserve the finite dimensionality of the
is made explicit, for example, by the use of the term "re- operations at thi8 stage, so that geometrical questions can be
stricted". kept in the forefront.
345

Downloaded 13 Nov 2012 to 139.184.30.132. Redistribution subject to AIP license or copyright; see http://jmp.aip.org/about/rights_and_permissions
346 R. PENROSE

ferred to here as twistors. It will turn out that the tance in the asymptotic analysis of such fields,7 and it
twistor algebra will have the same type of univer- would appear that a "manifestly conformally in-
sality, in relation to the conformal group, l that the variant" formalism should be valuable to this type
well-known and highly effective two-component of analysis. Furthermore, although the presence of
spinor algebra of van der Waerden3 has, in relation mass breaks conformal invariance, the conformal
to the Lorentz group. Twistors are, in fact, the group appears to have relevance as an approximate
"spinors" which are relevant to the six-dimensional symmetry in very high energy physics. 8 When the
space whose (pseudo-) rotation group is isomorphic energies of the particles are high enough, their
with the conformal group of ordinary Minkowski rest masses can be neglected.
space-time.' The simplest (non-scalar) twistors con- However, the applicability of the twistor formal-
stitute a four-dimensional, four-valued representation ism is not restricted to situations in which the con-
of the restricted conformal groupO (eight-dimensional formal group is relevant. Since the Poincare group
if reflections are included). The general twistor is is a subgroup of the conformal group, it follows that
then a many-index quantity constructible from the Poincare covariant operations can be expressed in
above basic twistors by means of the usual" tensor twistor terms. It merely becomes necessary to in-
type" rules. Any finite-dimensional representation troduce a fixed (skew-symmetric) "metric twistor"
of the conformal group is thus expressible as a direct r P which singles out a particular Minkowski struc-
sum of twistor representations.' ture (consistent with the given conformal structure).
The emphasis here will be on the geometrical By including operations involving r P with the basic
aspects of twistors. It will, in fact, be possible to twistor operations, a formalism invariant under the
give a fairly complete geometrical picture of twistors Poincare group is obtained, in terms of which, any
and of their basic operations. Ordinary space-time Poincare covariant operation is, in principle, ex-
concepts can then be translated into twistor terms. pressible. Furthermore, by making slightly different
However, the geometrical expressions of the most alternative choices for r P, the corresponding formal-
immediate twistor concepts have a somewhat non- isms for a class of space-times including the de
local character. Thus, the primary geometrical object Sitter and Einstein cosmologies are obtained.
will not be a point in Minkowski space-time, but Perhaps the most obvious drawback of the twistor
rather a null straight line or, more generally, a formalism, however, from the point of view of a
twisting congruence of null lines. Points do, in fact, possible fundamental applicability in physics, is that
emerge, but only at a secondary stage. (It also turns it is so tied to the idea of a conformally flat back-
out that a natural description of physical fields in ground space-time, that it is difficult to conceive
twistor terms is given by quantities having a non- of how to incorporate the formalism (as it stands)
local space-time interpretation.) However, any vec- completely into general relativity. It is not impos-
tor, tensor, or spinor operation can be translated sible that some modification of twistor algebra might
into twistor terms, if desired, and vice versa. be applicable to general curved background space-
All the basic operations of the formalism are con- times. Indeed, some of the basic ideas of the formal-
formally invariant. This is of relevance in the study ism, namely those concerning the null line congru-
of zero rest-mass fields, since, for each spin, the ences (cf., the results of Robinson and Kerr referred
free-field equations are all effectively conformally to in Sec. VIII) have, as their origins, some re-
invariant. 6 • 7 This latter fact has particular impor- searches into general relativity. However, any such
modification would have to be of a different order
lB. L. van der Waerden, Nachr. Ges. Wiss. Gottingen 100, from the comparatively straightforward adaption
1 (1929).
'F. Klein, Gesammelte Mathern. Abhandlungen (J. SEringer, of spinor algebra into general relativity. 9 The point
Berlin, 1921); cf. also H. Weyl, The Classical GrOUP8 (Prince- of view one must apparently take, is that whereas
ton University Press, Princeton, New Jersey, 1939); R. Brauer
and H. Weyl, Am. J. Math. 57, 425 (1935) . the vector-tensor-spinor algebra refers to the im-
• Quantlties which are essentially twistors have been des-
cribed by W. A. Hepner, Nuovo Cimento 26, 351 (1962). See
also Y. Murai, Nucl. Phys. 6, 489 (1958); Progr. Theoret. 8 H. A. Kastrupp, Phys. Letters (Amsterdam) 3,78 (1962);
Phys. (Kyoto)~.. 147 (1953); 11,441 (1954). Similar quantities Phys. Rev. 142, 1060 (1966). A recent additional suggestion is
have also recently gained prOlninence in, for example, the work that the mass splittings of strong interaction physics may be
of A. Salam, R. Delbourgo, and J. Strathdee, Proc. Roy. Soc. derivable from conformal group symmetry: D. Bohm, M.
(London) A284, 147 (1965). Flato, D. Sternheimer, and J. P. Vigier, Nuovo Cimento 38
8 E. Cunningham, Proc. London Math. Soc. 8, 77 (1910); 1941 (1965). For a discussion of the relevance of the conformal
H. Bateman, ibid. 8, 223 (1910); P. A. M. Dirac, Ann. Math. group in physics, see T. Fulton, F. Rohrlich, and L. Witten,
37,429 (1936); J. A. McLennan, Jr., Nuovo Cimento 10, 1360 Rev. Mod. Phys. 34, 442 (1962).
(1956); H. A. Buchdahl J"bid. 11,496 (1959). 9 L. Infeld and B. L. van der Waerden, S. B. Preuss. Akad.
7 R. Penrose, Proc. Roy. Soc. (London) A284, 159 (1965). 9, 380 (1933).

Downloaded 13 Nov 2012 to 139.184.30.132. Redistribution subject to AIP license or copyright; see http://jmp.aip.org/about/rights_and_permissions
 TWISTOR ALGEBRA 347

mediate neighborhood of a "point" in space-time, ings in terms of space-time geometry. For whether
the twistor algebra refers to a more extended" non- or not twistors have a significant role to play in
local" region. It seems to be mathematically more future physical theory, the results so far obtained
difficult to piece together such regions into a curved suggest strongly that the formalism ought at least
space-time, than to build a space-time out of points. to be thoroughly explored from both geometric and
Finally, one of the most important initial motiva- analytic points of view.
tions for considering a formalism of this kind should
be briefly mentioned, although implications of this II. TWISTORS AND NULL LINES
idea are not discussed in this paper. It concerns As a starting point for the geometrical description
possible applications to quantum field theory. One of a twistor, consider a null straight line L in
of the most significant quantum mechanical opera- Minkowski space-time M. Choose a set of Minkow-
tions is the splitting of field amplitudes into their ski coordinates12 (Xi) for M with origin O. Let Z'
positive and negative frequency parts. Although this be the position vector of some point P on L and let
is generally expressed in terms of Fourier transforms, n' be a future-pointing tangent vector to L (see
there is the alternative description in terms of Fig. 1). If we wish to assign a set of coordinates
singularity-free analytic extensions into the upper to the line L, we may do this (Plucker-Grassmann
or lower complex half-planes. 1o In the present for- coordinates) by selecting ni and the moment
malism, the idea is to use this second approach
and to combine it with global properties of com- (2.1)
plexified Mlnkowski space-time. Similarly to the
way that the real axis divides the complex plane of the vector n' (acting at P) about O. Then, the
ratios of the ten quantitites (n" mif) will uniquely
into two disconnected halves, the space of "null"
twistors (describing real Minkowskian null lines) define L. (Note that this is independent of the choice
of P on L.) This is, however, a highly redundant
divides twistor space into two disconnected halves,
representation. In addition to the requirement here
namely the "right-handed" and the "left-handed"
twistors (which may be thought of as describing that n' be null
"complexified" null lines). n'n, = 0, (2.2)
Free fields of zero rest mass and arbitrary spin
can be described particularly conveniently in twistor there are the consistency relations for (2.1)
terms. Moreover, it turns out that they can be (2.3)
generated in a remarkably simple way as contour
integrals of arbitrary analytic functions defined in [and also EiikZmifmkZ = 0, which is implied by (2.3)].
twistor space. (A null field is generated when the Equation (2.3) in fact represents just three inde-
contour surrounds a simple pole of the function. This pendent conditions which, together with (2.2), re-
leads to a certain generalization of a theorem of duce the set of nine ratios in (n" mil) to just five
Robinson. 11) If the relevant singularities of this independent real numbers. This is consistent with
function are associated with one half, rather than
the other, of the twistor space, this ensures that the
field is of (say) positive frequency. Other properties
of zero rest-mass fields also find a natural expression FIG. 1. The reyresentation of a
in twistor terms, notably the interrelation with nul line.
certain types of potential fields 7 and the structure of
their total energy-momentum-angular-momentum
(which give ten of the fifteen components of a
trace-free" Hermitian" twistor En). The discussion 12 World vectors and tensors are labeled by lower case Latin.
indices running over 0,1,2,3, the Minkowski metric being
of these matters is left to a later paper. given by (gil) = diag(1,-1,-1,-1). Capital Latin index
Care has been taken here to illustrate that twistor letters, primed or unprimed, denote spinor indices and run
over 0,1 or 0',1'. Greek indices are twistor indices and run
algebra is not merely an abstract formalism, but over 0,1,2,3. The summation convention applies to each of
that the algebraic operations have well-defined mean- these four types of index separately. Thus, in particular, no
summation takes place between primed and unprimed spinor
10 See any standard work on quantum field theory. The es- indices even when the same letter is used, e.g., J and J' are
sential matters referred to here can be found in R. F. Streater regarded as distinct letters in xJJ'. This allows us to write the
and A. S. Wightman, peT, Spin and StatistiC8, and All That tensor-spinor correspondence in a definite way by simply using
(W. A. Benjamin, Inc., New York, 1964). the two capital versions of a tensor index as its spinor trans-
11 I. Robinson, J. Math. Phys. 2, 290 (1961). lation: xiHxJJ', etc. (Rindler's convention).

Downloaded 13 Nov 2012 to 139.184.30.132. Redistribution subject to AIP license or copyright; see http://jmp.aip.org/about/rights_and_permissions
 348 R. PENROSE

the fact that the null lies in M form an 00 5 sys- Equation (2.6) implies a geometrical realization of
tem (for, choosing P as the intersection of L with a spinor {3J, up to a phase multiplier, namely as a
a fixed spacelike hyperplane, we have 00 3 choices future-pointing null vector. Moreover, {3J can be
for P and 00 2 choices for the null direction at P). completely realized geometrically14 up to sign, in
So far, the fact that L is null has not played an terms of the bivector
essential part. But null directions are very con- 7I"ik ~ {3J{3K EJ'K' + EJKPJ·pK'. (2.7)
veniently represented in terms of two-component
spinors. It will emerge that, by using spinors, we This is real (and null), and it determines a half-
can greatly simplify the description of the null plane element (the "flag plane") tangent to the light
line L; the set of four conditions (2.2), (2.3) being cone along the vector OI i • However, the sign of {3J
effectively replaced by a single condition. Since the cannot be realized locally15 geometrically, since a
following discussion depends essentially on the use continuous rotation through 271" changes {3J into
of spinors, a very brief review of the ideas required -{3J. If we are interested in (3J only up to propor-
will now be given. 13 The translation from world tionality, then the complete geometrical realization
tensors to spinors is achieved using a quantity12 is simply as a null direction.
u;J' [a Hermitian (2 X 2) matrix for each ij and Let us now represent, in spinor terms, the quanti-
its inverse u~J' subject to ufJ' U~K' EJKEJ'K' = g;k, ties n\ mii, which define the null line L. We have
ufJ'u~J' = 5;. The EJK, EJ'K' are skew-symmetric
(2.8)
Levi-Civita symbols and are used for raising and
lowering spinor indices: ~AEAB = ~B' EAB~B = e, and, from (2.1),
Le., ~o = h, ~1 = -~o, and similarly for primed mik ~ lJJ''AK'XK' _ 'AJ'XJ'lKK'
indices. Any tensor (e.g., Xiii) has a spinor transla-
tion, which is written using the same base symbol, = ieJKJL(J''X K') - iJL(J'AK)EJ'K', (2.9)
but with each tensor index replaced by the cor-
where
responding pair of spinor indices, e.g.,
(2.10)
(Round brackets denote symmetrization. The factor
Under complex conjugation, the roles of primed and
- i is for later convenience.) Thus 'AA and JLA'
unprimed indices are interchanged, so that real-
together determine li and mik. We may think of
ity of tensors is expressed as Hermiticity of spinors.
'A A as defining the direction of Land JLA' as effec-
Since M is flat, we can here choose u;J' constant
tively giving us the moment of 'A A (" acting" at P)
and equal to T' times the unit matrix and Pauli
about O. It is clear from (2.10) that, if 'A A is multi-
matrices. Then, say,
plied by any complex factor, then L is unchanged
00' OI
o I 2 3 X if JLA' is multiplied by the same factor. Furthermore,
(x ,x ,x ,x) ~ X ']
[ 10' II' (2.10) implies that JLA' is independent of the choice
x
of P on L (Le., if lAA' ---7 lAA' + a'AA'X A', then JLA' is
X

unchanged since AAAA = 0).

(2.4)
Note a particular choice of P which is of interest,
namely the intersection of L with the null cone of
The spinor translation of a complex null vector O. (This intersection exists uniquely provided L
c/ (i.e.,(/01; = 0) has the form does not lie in any null hyperplane through 0.)
Then li or _Zi has the form (2.6) and it follows from
(2.5)
(2.10) that
If Oil is real and future pointing, then we can take
'",J' to be the complex conjugate of {3J, i.e., (2.11)
14 E. T. Whittaker, Proc. Roy. Soc. (London) A158, 38
(2.6) (1937); W. T. Payne, Am. J. Phys. 20, 253 (1952); R. Penrose,
"Null Hypersurface Initial Data", in P. G. Bergmann's
13 For more complete accounts see Refs. 3 and 9 and, for Aeronautical Research Lab. Tech. Documentary Rept. 63-56
example, W. L. Bade and H. Jehle, Rev. Mod. Phys. 25, 714 (Office of Aerospace Research, U. S. Air Force, 1963).
(1953); E, M. Corson, An Introduction to Tensors, Spinors 16 To give a rigorous definition of a spinor which takes into
and Relativistic Wave Equations (Blackie & Son Ltd., London, account its sign, it is usual to appeal to the theory of fibre
1953); F. A. E. Pirani, in Brandeis Summer Institute in bundles. This is not essential, however, and an elementary
Theoretical Physics, 1964, Lectures on General Relativity (nonlocal) geometrical description will be given in an appendix
(Prentice-Hall Inc., Englewood Cliffs, New Jersey, 1965), to a forthcoming book by R. Penrose and W. Rindler on the
Vol. 1. applications of spinors in relativity.

Downloaded 13 Nov 2012 to 139.184.30.132. Redistribution subject to AIP license or copyright; see http://jmp.aip.org/about/rights_and_permissions
 TWISTOR ALGEBRA 349

Thus, the null direction defined by }.IA,[cf., (2.6)] is [~]) and if Re (),.AilA) < 0, a left-handed twistor.
that of the null line through 0 which meets L (Fig. The reason for this terminology will become evident
1). The exceptional case ),.EilB = °
corresponds to
L lying in a null hyperplane through O. [This follows
later. When),.A = 0, we shall still refer to L" as a
null twistor even though no finite null line L is
from (2.10), since ni would be necessarily orthogonal represented. The role of such twistors will emerge
to any choice of lj.] In this case, AA and il A are pro- shortly.
portional, so the null direction of }.lA' is that of L.
More exceptionally still, L might pass through O. m. INCIDENCE OF NULL LINES IN
TWISTOR TERMS
This is the case }.lA' = 0. On the other hand, for
the present, we do not allow),.A =
will be reconsidered shortly).
°
(although this The algebraic rules for the manipulation of twis-
tors will have, as their basis, the idea of incidence
The null line L can now be assigned, for coordi- between null lines. Let us consider, in addition to
nates, the three complex ratios of the four complex L, a second null line X, described by the null twistor
quantities X":
(3.1)
(2.12) where i/ and 1/A' are the analogs of ),.A, }.lA' above.
That is, ~A defines the direction of X and we have
which we write
(3.2)
(2.13)
where x IS the position vector of a point on X
These three complex ratios are equivalent to six real
(see Fig. 2). Suppose now that X and L do intersect.
parameters, so we must expect to find one real
Then we may choose li = Xi for the coordinate
relation connecting ),. A and }.IA" This is obtained
vector of this intersection point, whence by (2.10)
from (2.10), since the reality of the vector li implies
and (3.2),
lAA' is Hermitian, whence
(3.3)
(2.14)
Let us define the complex conjugate of a twistor
Provided ),.A ¢ 0, condition (2.14) is also sufficient
L" to be L" (valence [m, where
to ensure the existence of a null line L associated
with ),.A and }.IA" For, if ),.AilA is pure imaginary and (3.4)
nonvanishing, then (2.11) gives us a point P, through
which we choose L with direction given by ),. A. On when L" is given by (2.13) (irrespective of the con-
the other hand, if ),.AilA = 0, we can easily solve dition (2.14), or whether or not ),.A vanishes]. In
component form l7 :
(2.10) to obtain P, with OP spacelike (lH' of the
form )"Aii A , + VAX A ,). Lo = L2,
A quantity with components (U) given as in
(2.13) will be called a null twistor (of valence if [m (3.5)
(2.14) holds. 16 We shall also be interested in such Then, (3.3) tells us that a necessary condition for
quantities when (2.14) does not hold (and also when
),. A is allowed to vanish). We may regard such a

quantity as describing, in some sense, a" complexi-

fied" null line [when (2.14) fails], since if we allow
l' to be a complex vector, then (2.14) would, in FIG. 2. Incidence of
general, be violated. [This is not a straightforward two null lines.
complexification of the null lines in M in the ordinary
sense, however, since the real dimensionality of the
system of lines is only increased from five to six
by dropping (2.14); but cf., Sec. VL] If Re (),.ApA) >
0, we refer to L" as a right-handed twistor (valence 17 The operation of twistor (or spinor) complex conjugation
is denoted by a bar extending only over the base symbol
16 Any temptation to identify the twistor (2.13) with a Dirac involved and not over the indices. If the bar extends also over
spinor should be rejected here, since their transformation the indices, this denotes simply the complex conjugate of the
properties are quite different [cf., for example, (7.13), (7.17)]. complex number that the symbol represents.

Downloaded 13 Nov 2012 to 139.184.30.132. Redistribution subject to AIP license or copyright; see http://jmp.aip.org/about/rights_and_permissions
350 R. PENROSE

the null lines L and X to meet, in terms of their

corresponding twistors, is
ing of this condition X" La = °
when X and L are
real parallel null lines. Perhaps the simplest way to
(3.6) see the result is to use a limiting argument. Keep
La fixed but vary xa, so that xa La = 0 throughout
since the motion. Suppose L and X are initially not paral-
X"La = ~AjZA + l1A.X A'. (3.7) lel, but become parallel in the limit. The intersection
point recedes to infinity along L. The possible posi-
Note, further, that the condition (2.14) for L~ to
represent a real null line is simply tions of X, for each finite position of the intersection
point, are the generators of the null cone of this
(3.8) point. As the point recedes to infinity, the null cone
becomes, in the limit, the null hyperplane through
We can think of (3.8) as a special case of (3.6) since
L intersects itself! We also have, here,
X. Thus, when L and X are parallel, the condition
x"L" = 0 becomes the condition that L and X lie
Xax" = 0. (3.9) in the same null hyperplane.
Condition (3.6) is, conversely, also sufficient for In fact, it is convenient to regard all the null
the null lines L and X to intersect, where we assume, lines of a null hyperplane as intersecting in a single
for the moment, that XA and ~A are not proportional, point at infinity. Thus, we adjoin to our space M a
so that L and X are not parallel. For, we then have set of 3 such points at infinity, one for each of
(Xl

~AXA ~ 0 and we can construct the (complex) vector the null hyperplanes in M. The geometrical role
played bythetwistorsL" withX A = 0 (but/-l A ' ~ 0)
Pi ~ PJJ' = (i/~AXA)(XJ7JJ' - ~JJl.J')' (3.10) now emerges. For, if XA = 0, the condition Xa L" =
We observe that o becomes ~.APA = 0 [cf., (3.7)]. That is to say,
the direction of X coincides with the null direction
represented by /-lA', so that for fixed La the correspond-
Thus, when Pi is real, we can satisfy (2.10) and (3.2) ing lines X are all parallel. Each null hyperplane of
by putting 1; = Pi = Xj, whence L and X must parallel lines X gives rise to one point at infinity;
intersect. (In fact, it is valid to regard the "com- the aggregate of all these points at infinity, cor-
plex point" with position vector pi as the "inter- responding to all these parallel hyperplanes, gives
section" of L and X even in the cases when Pi an 1 system of points at infinity, namely the points
(Xl

is not real, as we shall see later.) Now Pi is real of the null line at infinity L. In fact, to complete
if and only if PJJ' is Hermitian. Since XA and ~A the picture, we must add one further point at in-
are not proportional, we can test the Hermiticity finity, not lying on any finite line, which we call
of PJJ' by taking components with respect to XA , ~A: I. The point I is common to all the lines L at
infinity. The fact that any two null lines at infinity
XAX A'(PAA' - PA'A) = i/-lA,X A' + Z)..A PA = iL"L" , must, for consistency here, be considered to intersect,
~A~'(PAA' - PA'A) = i7JA,~A' + i~ijA = iX"X", follows from (3.7), since if XA = 0 and also ~A = 0,
then xa La = 0 automatically follows. The point I
~AXA'(P.AA' - PA'A} = i7JA,>'A' + i~ApA = ixaL".
then plays the part of the vertex of a null cone at
Thus, the reality of Pi is a consequence of (3.8), infinity the generators of which are the lines at
(3.9), and (3.6), so that in the case of twistors rep- infinity considered above (see Fig. 3).
resenting nonparallel null lines, we can state the The structure of the completed (It compactified")
condition that the lines meet simply as the orthog- Minkowski space arrived at in this way, by adding a
onality condition (3.6) between the twistors. (closed) null cone at infinity, is one which has been
We may further ask what is the geometrical mean- considered by a number of authors. 18 •7 We hence-
forth use the symbol M, here, to refer to the entire
one point at Infinity for
each null hyperplane In MflJ
completed space and not just to the set of finite
{~---, , points. The set of finite points is instead denoted
N~(
"T _.,\
.. .,,:, \I ,
FIG. 3. The con- by M {I} to indicate that the null cone of the
. ;I . struction of the
M= +.-- I I
'I
I
I

til'
•
compact manifold
M.
18 N. H. Kuiper, Ann. Math. 50, 916 (1949); H. Rudberg,
dissertation, University of Uppsala, Uppsa.la, Sweden (1958);
I ( _/ / I
\ I, R. Penrose, in Proceeding8 of the 1965 Conference on RelatirJi8tic
' ... -..... , Theorie8 of Gravitation, Warsaw (Polish Academy of Science,
a'Whole generata corresponds Warsaw, 1965) A. Uhlmann, Acta Phys. Polon. 24, 293 (1963).
to a null direction In MfII Rudberg also mentions the four-valuedness of spinors.

Downloaded 13 Nov 2012 to 139.184.30.132. Redistribution subject to AIP license or copyright; see http://jmp.aip.org/about/rights_and_permissions
 TWISTOR ALGEBRA 351

particular point I has been removed from M. The a general twistor Ra of valence [~], in terms of the
geometry of M is briefly as follows: M is a com- congruence R of lines X whose twistors a (with x
pact nonsingular manifold without boundary which xaXa = 0) satisfy XaRa = O. If RaR" ¢ 0, such
has a well-defined conformal structure everywhere a congruence is referred to here as a Robinson con-
(i.e., a metric defined up to proportionality). It gruence. 20 [If R a R" = 0, then we denote by R the
has a transitive ro 16 group of motions preserving line which is met by all the lines of the congruence
this conformal structure, every point being on an R. On the other hand, if L denotes the line, then
equal footing with every other point. M contains L denotes the corresponding congruence. In gen-
5
00 null" straight" lines (geodesics) each of which eral, we adopt the convention that the operation of
is topologically a circle. Through each point P of applying a bar to any aymbol is its own inverse.
M pass 00 2 of these lines generating the (closed up) From this it follows that the definition of R", given
null cone of P. The removal of anyone of these R a, exactly mirrors the definition (3.5) of La, given
null cones leaves a space with Euclidean topology L", since we could put La = R". Thus,t7 (R") =
which can be assigned a Minkowskian metric (unique (RO, Rt, R2, R 3 ) = (R 2 , R 3 , Ro, Rl)'] If RaR" > 0,
up to dilatations) consistent with the given con- we refer to the Robinson congruence R as right-
formal structure. The entire space M has the topol- handed. If R",R a < 0, it is left-handed.
ogy S3 X SI and can be realized as a real projective Since Robinson congruences occupy such a basic
quadric fourfold with signature (+ + - - - - ). position in the geometry of twistors, it seems worth-
The null lines in M are in one-to-one correspond- while to examine a particular such congruence in
ence with the proportionality classes of twistors some detail here. (In fact, any particular case is
La (Le., "La represents the same line as La, " ¢ 0) completely representative of the general case, since
which satisfy LaL a = 0 and L" ¢ O. Incidence it turns out that any two Robinson congruences
between null lines in M is expressed as orthogonality can be transformed one into the other by a Poincare
(xa La = 0) between the corresponding twistors. transformation, perhaps involving a reflection.)
Choose a real number e and put
IV. ROBINSON CONGRUENCES
(R",) = (e/v'2, 0,1,0) (4.1)
Thus far, the geometrical description of a twistor
given here has been restricted to the case of null so that
twistors (of valence [m. As mentioned earlier, if RaR" = ev'2. (4.2)
on the other hand LaL a ~ 0, then we may think of Let X" be as in (3.1): (X") = (~O, ~\ 1'/0-, 1'/1')' so
L a as representing, in some sense, a kind of "com- that the condition for the line X to belong to the
plexified" null line L. This is, in fact, the way conrgruence R is
we tend to view "geometrically" a general twistor
of valence [~]. However, it is significant that a much (4.3)
more precise realization can be given in terms of Write the position vector Xi of a point on X as
congruences19 of null lines. In order to fix our ideas,
we regard such a congruence as representing, rather,
a twistor of valence m
(e.g., L,,), although this Equation (3.2), with (2.4), gives
makes little difIerence in practice.
Now, any (real) null line L in M can be completely ',In(O 1[
(1'/0-,1'/1') = -(~/v2) ~,~) t-z . -X + iyj.
characterized by the system of all null lines which -x - ~y t+z
meet it. In twistor terms, that is, if we know the
set of all xa satisfying xaLa = 0, xaXa = 0 Using (4.3), we then get _e~o = i~O(-t + z) +
(where LaLa. = 0, La ¢ 0), then we know La up i~I(X+ iy), i.e.,
to proportionality. Thus, while we think of La. as ~o : ~1 = X + iy :t - z + ie. (4.4)
describing the null line L in M, we can think of La
as describing the congruence L of null lines (Le., a Now ~J~J' corresponds to a tangent vector to X,
three-dimensional system19 of lines in M) which so that

[tr- t~I'j dt + dz dx + i dyj.

meet L. In exactly the same way we can represent ex: [

11 The term "congruence" is used to denote a system of

tlr' tl~I' dx - i dy dt - dz
curves (or surfaces, etc.) for which there is just one member
of the system (or at most a discrete number) through a general 20 First investigated by I. Robinson (private communi-
point in the space. cation).

Downloaded 13 Nov 2012 to 139.184.30.132. Redistribution subject to AIP license or copyright; see http://jmp.aip.org/about/rights_and_permissions
352 R. PENROSE

Combining this with (4.4) we get the differential I) at infinity. It then becomes topologically a three-
equations for the line X dimensional sphere S3 (of which the hyperplane
t = T may be regarded as the stereographic pro-
dt + dz : dx - i dy = dx +i dy : dt - dz jection). The vector field on S3 is everywhere non-
=x + iy :t - z + i~. (4.5) singular and nowhere vanishing. The circles con-
stitute an example of what is known as the Hopf
The solution of this is
fibring of S3 (one example being" Clifford parallels"
t - z + i~ = (x + iy)a, (4.6)
on S3).
Note that all the circles in the hyperplane thread
(x - iy) - (t + z)a == {3,
through the particular (smallest) circle of radius
where a and {3 are complex constants defining the I~I and centre z = T, X = Y = 0 given when e = o.
particular line X of R, for consistency satisfying If ~ is small, this circle describes, as T increases, a
path approximating that of a null line. If ~ is zero,
1m (i3a) = ~. (4.7)
the path is exactly the null line z = t, x = Y = O.
In order to get a more visualizable picture of the For small ~, we may think of the lines of the Robins~m
situation, let us consider the intersection with a congruence as defining an approximate null line,
spacelike hyperplane t = T (T const). Each null line but the lines twist around one another and never
X meets this hyperplane in one point (T, x, y, z). quite meet. The twisting has a positive or negative
The direction of X there can be represented by screw sence according as Ra is right- or left-handed.
considering the projection into the hyperplane of [In the limit ~ -7 0, the circles (4.8) all touch the
the tangent vector to X at this point. These pro- z axis at z = T. The tangents to these circles are
jected tangent vectors are then tangents to a series orthogonal to the spheres touching the (x, y) plane
of curves in the hyperplane which should give at z = T, these spheres being the intersections of
some kind of a picture of the structure of the Robin- z = T with the null cones with vertices on z = t, x =
son congruence. To find the differential equations y = O. The lines of the congruence are then just
of these curves, we simply replace t by T in (4.5) the generators of these null cones in this case, as
and dt by ds = (dx 2 + dy2 + dz 2)'. Then we get expected.]
(x + iy)(ds - dz) = (T - Z + i~)(dx +i dy). V. THE ASSOCIATED SPINOR FIELD
OF A TWISTOR
The solutions of this equation are given (apart
from the spurious x + iy = const) by We saw above that the tangent vectors to the
lines of the Robinson congruence constituted a field
x2 + y2 + (z - T)2 - 2~(x sin <p
of null vectors which was regular and nonvanishing
+ y cos <p) tan e = (4.8) everywhere on M. The general Robinson congruence
R can be written compactly in spinor terms in a
Z - T = (x cos <p - y sin <p) tan e, way which exhibits this (and other) facts very
where e and <p are constants defining the different simply. This leads to an interpretation of twistors
curves. These curves are evidently circles, being in terms of certain spinor fields.
intersections of spheres with planes. They twist Put Ra = La, so we can use the notation of Secs.
around one another in such a way that every pair II and III. The condition xa La = 0, for X to belong
of circles is linked. The twisting has a positive screw to the congruence L is then, by (3.7), (3.2),
sense if ~ > 0, i.e., if Ra is right-handed. 21 They lie
on the set of coaxial tori"2 obtained by eliminating
o= ~Ai1A + 7]A,~A' = ~A(i1A - iXAA,~A'),

cp between the two equations. [These tori are the

from which it follows that ~A is proportional to the
rotations about the z axis of a system of coaxial expression in the bracket. Since X defines xa only
circles in the (x, z) plane.] up to proportionality, we are at liberty to choose
From the point of view of the completed space- the scale factor for ~A so that
time M, we should regard the hyperplane t = T as
being completed (conformally) by a point (namely (5.1)

21 The screw sense arising here depends, of course, on the We can think of (5.1) as defining a spinor field,
"handedness" of the choice of matrices in (2.4). since ~A is a function of Xi, the associated null
22 I am grateful to J. Terryl and J. E. Reeve for this obser-
vation concerning the Hopf fibring. vectors Vi ~ ( t ' being tangents to the lines of

Downloaded 13 Nov 2012 to 139.184.30.132. Redistribution subject to AIP license or copyright; see http://jmp.aip.org/about/rights_and_permissions
 TWISTOR ALGEBRA 353

the congruence L. Since XAXAA,X A' is real, it follows fact, there is one line of L totally at infinity, namely
that that corresponding to the direction of XA.) The
quantity (5.6) is also invariant23 under conformal
Re (XA~A) = Re (XApA) = tuL a , (5.2)
transformation (~A of weight t), so we can refer to
so that ~A cannot vanish unless L a is null. The field of La L a as the conformal invariant of the spinor field
null directions defined by (5.1), i.e., by the Robinson ~A or of the twistor La.
congruence L, is thus well defined and regular Some subtleties remain, however, concerning the
throughout M {l). The regularity at infinity also extension of the field ~A across infinity. The pos-
follows from the conformal transformation prop- sibility of defining two-valued (spinor) fields in
erties of (5.1) as we are to see shortly. M{l) is well known. But here we require a spinor
Now write field defined globally on M. The multiple connected-
BB ness of M is such as to cause a slight difficulty in
V BB' == BjBx ', (5.3)
this respect. But in any case, we can think, instead,
so that V BB,XAA' = EABEA'B" Then we have, from in terms of ~A~B or the associated bivector [cf.,
(5.1), (2.7)], since these are not two-valued. Then we
might expect to find a representation of a twistor
(5.4)
up to sign. However, it turns out [cf., (7.25)] that
Thus, the field ~A satisfies the differential equation twistors of odd total valence are four-valued under
the conformal group. Thus, any such conformally
(5.5)
invariant representation of a twistor (of odd va-
Conversely, from this equation we can reconstruct lence), cannot distinguish the twistor from i times
the form of the expression (5.1). For (5.5) implies the twistor. (This is analogous to the fact that
that for some spinor iX B" the derivative V BB'~A has spinors of odd valence cannot be completely rep-
the form given in (5.4). Furthermore, this XB ' must resented in terms of tensors, in a Lorentz covariant
be constant because Vg'V~' ~A = 0 [as follows from way, whereby the spinors are distinguished from
(5.5), since skew-symmetry in B, A and in C, A their negatives.) In the present case, when we extend
implies that the expression vanishes]. Integration the field ~A across infinity, we find that the field
of (5.4) leads straight to (5.1), where ilA is another on the two sides of the null cone of I cannot be ex-
constant. actly matched, but ~A on one side must be matched
We now have the important result that there is a with i~A (or _i~A) on the other side. [We can see
one-to-one relation between twistors of valence m
and this explicitly in terms of the inversion of (7.18),
spinor fields satisfying (5.5). This gives us a more for which the conformal factor is n = xfx;/2a2 •
complete representation of a twistor than just as The null cone of I is transformed from the null
a Robinson congruence, since the factor of propor- cone, n = 0, of the origin, across which n changes
tionality of La is now also represented. Note that sign. The transformed ~A field picks up the factor
La L a is also expressible simply in terms of ~A as n- 1.] Thus, the spinor field (5.1), when defined over
the whole of M, must be thought of as four-valued;
LaL a = 1m (~B'V~'~A)' (5.6)
if ~A is one value, then i~A' -~A' -i~A are the other
But since we are interested in twistors in relation three.
to the whole of M, rather than just M {l), we should Since spinors can be represented geometrically
also ask how the field ~A is to be defined at infinity. (up to sign), this implies a geometrical realization
Essentially all that is required here is to verify of any twistor of valence up to a multiple of m
that (5.5) is invariant under conformal transforma- 1, i, -1, or -i. The phase of ~A defines, as we have
tion [e.g., under inversions, since then I becomes a seen [cf., (2.7)], a half-plane element tangent to
finite point,7 cf., (7.18)]. This invariance is achieved23 the light cone; - ~A defines the same half-plane
by specifying that ~A transform as a conformal element; but i~A and -i~A define the opposite half-
density of weight t. Then the field equation (5.5) is plane element. Thus, a complete unoriented null
also satisfied at infinity, in the sense that it holds plane element is defined at each point by the phase
with respect to some metric which is regular on of La. As we follow one of the lines X belonging to
the null cone of I. This, in particular, defines the the Robinson congruence L, we find that this plane
Robinson congruence L regularly at infinity. (In element rotates about X in the opposite direction
23 See Ref. 7, Eqs. (10.1), (10.7), and (10.8) for the
from the neighboring lines of the congruence and
relevant fonnulas. twice as fast. Owing to the way connections are

Downloaded 13 Nov 2012 to 139.184.30.132. Redistribution subject to AIP license or copyright; see http://jmp.aip.org/about/rights_and_permissions
 354 R. PENROSE

made at infinity, this results in the aggregate of being that they must not all vanish. This co 6 system
plane elements at points of X, effectively constitu- we may think of as constituting a three-dimensional
ting a Mobius strip, which explains why the plane complex projective space which we denote by the
elements cannot be consistently oriented along X. letter C. The" points" of C are just the" complexi-
To see how these plane elements rotate about X, fied" null lines (and the null lines) of M.
consider (~B~B'V BB' )~.4. = ~B'XB'~.4. [see (5.4)]. The In fact, we have two alternative pictures of any
real part of 2i~B'XB' measures the rate of extension given situation, namely the one in terms of M and
of the vector represented by ~.4. and the imaginary the one in terms of C. For example, we may think
part measures the rate of rotation of the half-plane of an object L, with projective coordinates (La)
element about the null direction of ~.4.' By (5.2), either as, say, a" complexified" null line of the real
space-time M (M picture) or simply as a point in a
1m (2i~B'XB') = -2 Re (hAB) = -LaL a,
certain projective three space (C picture). These are
whence the half-plane element rotates in a negative just two different ways of visualizing what is the
or positive sense according as the twistor L a is same physical situation in each case. In order that
right or left handed. (It might appear that the rate the two pictures be completely equivalent, how-
of rotation is constant here but this is misleading ever, we need to be able to interpret, in C, the
since the scaling is different at different points of condition of "reality" of a null line in M, of in-
X.) Note that if La is null, there is no rotation of cidence between null lines in M, and finally of
the half-plane element. We still get a Mobius strip, points in M. In effect, this requires that the conjuga-
however, since a half-twist always occurs, in effect, tion relation La ~ La should have a meaning with
at infinity. regard to the C picture. Now, we have seen that a
To see how the neighboring lines of the congruence twistor La (valence [m
refers to a point L of C; a
rotate about X, consider twistor Ra (valence [m
therefore refers to the dual
concept of a plane R in C, namely the plane of all
~.4.~B'V BB'~.4.
= -i~B'XB'~B' (5.7)
points X for which xaRa = O. (This is, of course,
Here, the real part of -ir''X B , measures the con- a plane in the complex sense. As a real manifold it
vergence of the null lines and the imaginary part is a four-dimensional subset of the six-real-dimen-
measures their rotation about one another.24 The sional manifold C.) The conjugation relation La ~
rotation is in the opposite sense from that of the La therefore describes a point ~ plane correspond-
half-plane elements and half as fast. ence in C, which we may refer to as a Hermitian
The expansion of the null lines and the change correlation of signature (+ + - - ). The signature
in the magnitude of ~.4. are not conformally covariant here refers to the Hermitian form
concepts, whereas the rotations are. Another con-
xaXa = XOX2 + X 1X 3 + X 2XO + X 3 X 1. (6.1)
formally covariant concept for a congruence of null
lines is the shear of these lines. l l •24 In the present We regard this Hermitian correlation as being an
case we have intrinsic part of the geometric structure of C.
The (real) null lines in M are the points of the
~B~.4.V BB'~A = 0, (5.8)
five-real-dimensional subset N (with topology S3 X
which states that the shear vanishes. S2) of C defined by the equation xa Xa = O. Thus,
N is a hypersurface if we regard C as a real six-
VI. THE COMPLEX PROJECTIVE SPACE C dimensional manifold, but it is not a hypersurface
We have seen that the null lines in M form an co 5 in the sense of the complex structure of C. We refer to
system which can be extended to an co 6 system by the subset of C for which xa Xa > 0 holds as C+ and
including" complexified" null lines (the latter being the part for which xa Xa < 0 as C-. The two sets
representable in terms of certain related structures C+ and C- are then disconnected from one another
called Robinson congruences). The members of this and have N as their common boundary (Fig. 4).
co 6 system can be given complex projective co- If L is any point of C, we may regard the plane L
ordinates (La) = (LO, L\ L2, L 3 ). That is, it is as the polar plane of L with respect to N, since
just the three complex ratios LO: L1: L2: L3 which "polarizing" xaXa with U yields xaLa = 0 (or
are significant, the only restriction on L O, '" , L3 La X a = 0), the equation ofthe plane L. The Robin-
son congruence associated with L is now the inter-
24 P. Jordan, J. Ehlers, and R. Sachs, Akad. Wise. Lit.
Mainz, no. 1 (Mainz 2) (1961); E. T. Newman and R. Penrose, section (topology S3) of the plane L with N. (A
J. Math. Phys. 3, 566 (1962). slight inconsistency of notation arises here in that

Downloaded 13 Nov 2012 to 139.184.30.132. Redistribution subject to AIP license or copyright; see http://jmp.aip.org/about/rights_and_permissions
 TWISTOR ALGEBRA 355

it is the Robinson congruence L (\ N that had been

previously labeled L, rather than the entire plane.
The Robinson congruence includes only" real" null
lines of M, by definition. In cases where there is
possibility of confusion, we refer to "the Robinson
congruence L" or" the plane L" as the case may be,
but in general L refers to an entire plane in C.)
When L lies on N, the plane L can be thought of
as the complex tangent plane to N at L. This is also
just the case when L lies on its polar plane.
Now, if we wish to represent a point of M in terms Note that, in the C picture, the condition that a
of the C picture, we may do this using incidence point L lies on a line P, both Land P lying on N, is
properties of null lines in M. Any point P in M can interpreted in the M picture as the condition that
be uniquely represented by an 00 2 system of null the null line L passes through the point P. We have,
lines in M, namely by the generators of the null in fact, a kind of duality correspondence between
cone of P. Let K and L be two null lines in M M and N. To sum up, we have the following cor-
through P. The generators of the null cone of P respondence between the M picture and the subset
are then the null lines common to both K and L N in the C picture:
(i.e., the generators must meet both K and L). In
the C picture this is an 00 2 system25 of lines on N There is a one-to-one relation between the
which lie on the intersection of the two planes K and null lines in M and the points in N. (6.2)
L. This intersection is simply a complex projective
There is a one-to-one relation between the
(straight) line in C. Thus, we have the result that
points in M and the complex lines in N. (6.3)
any point of M is represented, in the C picture, by
a complex projective line which lies entirely on N. The condition for a point to lie on a
(A complex projective line is topologically a sphere null line in M is interpreted, in N, as
8 2 , in terms of its real structure. This agrees with the condition for the corresponding line
the topology of the set of null lines through P, i.e., to pass through the corresponding point. (6.4)
of the null directions at P.) We do not distinguish,
notationally, here, between the point P in M and From (6.4), we see that the condition that two points
the system of null lines in M through P. Then we are in M have a null separation is simply the condition
at liberty to denote also by P the line in the C that their corresponding lines in N should intersect;
picture which represents this point P in the M the condition that two null lines in M should meet
picture. is the condition that the join of the corresponding
Conversely, any line P in the C picture, which points in N should lie entirely in N.
lies entirely on N, represents some point P in the We may ask how we should interpret the lines of
M picture. (In the terminology of projective geom- C which do not lie entirely on N. Lines in a projective
etry, in the C picture, "line" always implies "com- three space form a four-dimensional system, so that
plex projective straight line.") To see this, consider in terms of real dimensions these lines describe
the C picture and let the line P lie entirely on N. an 00 8 system of objects in M. This suggests that
Let K and L be two points on P. Then we have a general line in C represents a complexified point
K"K" = 0, L"L" = 0, and, more generally (K" + in M, where now the straightforward doubling of
(3L")(K" + Pta) = 0 for all complex {3 (the general dimensions suggests that these are complexified
point on the line P having a twistor of the form points in the usual sense, i.e., their position vectors
K" + (3L"). Hence L"K" = O. Thus, in the M are allowed to have imaginary parts. To see that
picture, the null lines Land K must intersect. This this is indeed a consistent interpretation, consider
holds for any pair of null lines belonging to the 00 2 a general line in the C picture as a join of two general
system represented by P. These null lines must points L, X of C. Using the notation of Secs. II
therefore all meet in a point and, in fact, must be and III, we can form the expression (3.10). We may
the generators of the null cone of this point. We thus regard this complex vector Pi as the position vector,
label this point P, and the situation is as before. the M picture, of the complex point of intersection
of the two null lines (" complexified" or otherwise) in
U A symbol co r indicates r dimensionality in the real sense. M, defined by Land X. (Note that in this sense any

Downloaded 13 Nov 2012 to 139.184.30.132. Redistribution subject to AIP license or copyright; see http://jmp.aip.org/about/rights_and_permissions
356 R. PENROSE

two distinct null lines in M have a complex point in properly complexified, the null lines in M become,
common.) in effect, pairs of complex projective planes, namely
Any line P of C falls into one of six classes. If one a plane and one (3 plane for each fully complexi-
l is the imaginary part of the position vector pi fied null line. The "complexification" of the null
for the corresponding complex point P in M, then lines achieved here by the use of twistors amounts
[from (3.10)] to selecting only the a planes to represent the null
lines.
P C C+ <=? qi timelike, future-pointing; (6.5) Much of the earlier discussion could have been
p C C+ U N, P <t N, P <t C+ <=} qi null, carried out in terms of the manifold M*. The em-
future-pointing; (6.6) phasis here, however, has been to try and represent
twistors as far as possible in terms of "real" struc-
P intersects all of C-, N, C+ <=? qi tures in the space-time M. One final condition re-
spacelike, nonzero; (6.7) mains to be represented, however, in order that the
geometry of C can be completely realized in terms
P C C- U N, P <t N, P <t C- <=} qi null,
of M. This is that we must be able to interpret
past-pointing; (6.8) the vanishing of a twistor scalar product in geo-
P C C- <=? qi timelike, past-pointing; (6.9) metrical terms 27 in M.
Consider the condition
peN <=} qi = O. (6.10)
(rR a = O. (6.11)
In cases (6.6) and (6.8), the line P in C touches M. If Q and R are both null lines in M, we have seen
The set (6.5) (in the M picture) is sometimes re- that this is the condition for the lines to intersect.
ferred to as a" future-tube."lo If Q is a null line and R a Robinson congruence, then
There is a representation of lines in projective (6.11) is simply the condition that Q should belong
three-space, well known to geometers, called the to the congruence R. It remains to consider the
Klein (or Grassmann) representation. 26 This is a case when both Q and R are Robinson congruences
four- (complex-) dimensional quadric, the points in M. Let Sand T be two null lines belonging to
of which correspond to the lines of complex pro- the Q congruence and let U and V be null lines
jective three-space. Thus, in our case, we may regard belonging to the R congruence. Suppose the three
the fully complexified version M* of M as the Klein pairs of lines (S, U), (U, T), (T, V) intersect. Then a
representation of the lines in C. We noted in Sec. necessary and sufficient condition for (6.11) to hold
III that M itself was essentially a real ( ++ - - - - ) is that the pair (V, S) also intersect (Fig. 5). To see
quadric fourfold. M* is the complexified version this, consider the points V, U, Q and the planes
of this quadric. There are two systems of planes on S, T, R in the C picture. We have SEQ, T E Q,
M*, called a planes and (3 planes. The a planes cor- whence Q E S, Q E T. Also U E R, V E Rand
respond to points in C and the (3 planes to planes U E S, U E T, VET. Thus, the line UV must be
in C. We may regard M as a submanifold of M*. the intersection of the planes Rand T (Fig. 5).
The a planes which meet M intersect M in the null If (6.11) holds, then Q also lies on this line, whence
lines of M. Similarly with the (3 planes. Thus, when V E S. Conversely if V E S then Q E R. Thus,
in the M picture, the condition for (6.11) to hold
is that the null lines V and S should meet.
This establishes the geometrical equivalence of the
FIG. 5. The condi- M picture with the C picture, whereby the con-
tion for QaRa = 0 is
that, in the M picture, formal geometry of Minkowski space is completely
V and S should meet, represented in twistor terms. The metric geometry
and that in the C
picture, V E S, whence of Minkowski space, on the other hand, requires the
Q ER. introduction of a fixed metric twistor. This is done
in Sec. X.
M-plcture C-p1cture
27 Strictly, we should also show that the concept of a Rob-
inson congruence is "geometrical" in M. An explicit con-
26 See any standard work on classical algebraic geometry, struction in terms of incidence of null lines is given in Sec.IX,
for example, J. A. Todd, Projective and Analytical Geometry but in any case, the geometric (and conformally invariant)
(I. Pitman, London, 1947); J. G. Semple and L. Roth, nature of a Robinson congruence is already implied by Sec.IV
Algebraic Geometry (Clarendon Press, Oxford, England, 1949). and V.

Downloaded 13 Nov 2012 to 139.184.30.132. Redistribution subject to AIP license or copyright; see http://jmp.aip.org/about/rights_and_permissions
 TWISTOR ALGEBRA 357
Vll. TWISTOR TRANSFORMATIONS C- and C+. (The twist orientation of Robinson con-
Although the transformation properties of twis- gruences is reversed under space reflection.) How-
tors have not been given explicitly as yet, they are ever, if we wish to extend this to a space reflection
implicit in the preceding discussion. Any continuous <P of the whole of M* we must, since <P is analytic,
transformation of 111 into itself, which preserves its remove the antiprojective nature of the transforma-
conformal structure (and, hence, the null cone and tion of C and consider, instead, a correlation which
null line structure of 111) must, by (6.4), correspond sends points X of C+[C-] into planes Y for which
to a continuous transformation of N into itself which Y is in C-[C+]. The antiprojective point transforma-
preserves its linearity structure. Furthermore, since tion just considered, which interchanges C+ and C-
the concept of a Robinson congruence is conformally would then correspond to e<p. Similarly, a point
invariant (cf., Sec. V), this transformation of N transformation of C which represents a time reflection
extends uniquely to a transformation of the whole of 111 is antiprojective, here transforming each of
space C. We have seen that the orthogonality rela- C-, C+ into itself. It therefore really represents
tion (6.11) between points and planes in C can be e::l, where ::l describes the (analytic) extension to
stated in terms of (conformally invariant) incidence M* of this time reflection of M. The transformation
properties in 111. Hence, collinearity of points in C of C representing ::l is a correlation sending each
has a conformally invariant interpretation in 111. point X of C+[C-] into a plane Y with Y E C+[C-].
It follows that the conformal transformations of M The transformation <P::l, being the product of two
are represented, in the C picture, by continuous projective correlations, is a projective point trans-
point transformations of C, which preserve its lin- formation of C. It interchanges C- with C+, and
earity structure, and for which the submanifold N is not continuous with the identity unless we widen
is invariant. the group of transformations of M* to include
If we restrict ourselves to conformal transforma- complex conformal transformations. (These would
tions of M which are continuous with the identity, not be point transformations of M, but we may view
then the corresponding transformations of C is them as transformations on the line systems in M
also continuous with the identity. Such transforma- which transform Robinson congruences into one
tions, preserving the linearity structure of C, must another. They are represented, in the C picture, as
therefore be projective point transformations26 of C projective transformations of C which do not pre-
(in the complex sense). If, on the other hand, serve' N.) The transformation <pe::l is an antipro-
we allow transformations of M which involve space jective correlation in C.
or time reflections, then we must also consider Let us examine more explicitly the effect of an
anti-projective transformations of C, that is, trans- allowable transformation on a general twistor A:~:.~·~
formations which combine a complex conjugation of valence [:]. (The indices 01, (3, ... , 8 are r in num-
operation with a projective transformation (so that ber and p, ... , Tare s in number, each ranging over
cross ratios become complex conjugated). We may four values 0, 1, 2, 3.) We may define a general
also consider transformations of C which are not twistor, in terms of twistors of valence [~] and [~],
point transformations but correlations26 (duality cor- in any of the standard ways that" tensors" may be
respondences) in which points of C are mapped to built up from "vectors," e.g., as linear combinations
planes, and planes to points. These would not strictly of outer products, or in terms of multilinear map-
correspond to transformations of M, because of the pings of one-index twistors into the scalars. Alter-
way C has been constructed here, but it is con- natively, we may simply use the transformation
venient to think of them as representing trans- properties to define a twistor A ~:::! of valence [:]
formations of the complexified version M* of M. The (considering for the moment only transformations
natural (analytic) extension to M* of a transforma- continuous with the identity):
tion defined on M sometimes turns out to cor- 7'afl···8 = AKX····tat fl •.. t 8T'" '" T'"r '
.fiP···'T rp"'''' leX "p (7.1)
respond to a correlation in C.
The basic operation e of complex conjugation The matrices (t~), (T~) are inverses 28
of eaeh other
in M*, which leaves the real space M invariant,
(7.2)
corresponds, in the C picture, simply to the
Hermitian correlation given by xa
~ iL. A trans- 28 In fact, the transformations of C would be the same if
we specified only that (T(Ja) be proportional to the inverse of
formation of M equivalent to a space reflection (t(Ja) since the Xa are projective coordinates for C. However,
corresponds, in the C picture, to a point transforma- the stronger requirement (7.2) is adopted here since the factor
of proportionality of a twistor is required when the more
tion of C, which is antiprojective and interchanges complete representation in accordance with Sec. V is used.

Downloaded 13 Nov 2012 to 139.184.30.132. Redistribution subject to AIP license or copyright; see http://jmp.aip.org/about/rights_and_permissions
 358 R. PENROSE

where twistors X a , Y a , of respective valence [~], [~], for which (7.4) does not hold. These correspond to
transform as complex conformal transformations of M*, not pre-
serving M, and the relation between A ~:::: and
(7.3)
A::::a would not be preserved. Such transformations
under a conformal transformation of M continuous will not be discussed here.
with the identity (Le., a projective transformation It will be appropriate to impose one final con-
of C preserving C+). But since Xa is a twistor of dition on the matrix (t p), namely that it should be
valence m,
it follows that unimodular:
T; = l'';, Le., tpl~ = 8~, (7.4) Itpl = 1. (7.6)
where the convention used earlier for twistor indices This amounts to requiring that the Levi-Civita
under complex conjugation [cf., (3.5)] is being em- symbols E a p-y3, E"P-y3 [with fixed components +1,
ployed here, namely that, under complex conjuga- -1,0 according as (a, (3, 'Y, 8) is an even permutation,
tion, upper and lower index positions are inter- an odd permutation, or no permutation of (0, 1,2,3)]
changed and also, the pairs 0, 1 and 2, 3 are inter- shall be twistors rather than just It twistor densities."
changed, i.e.,17 In fact, this condition is necessary if a strict inter-
19 = "ii, l~ = t;, t~ = ~, l~ = ~, pretation of twistors according to the scheme of
Sec. V is to be adhered to. The effect of (7.6) on
l~ = 1:, ... , l; = ~, l: = t:. the twistor algebra is only to enrich it slightly in
Condition (7.4) states that the matrix (t~) is pseudo- that the operation of forming duals of skew-sym-
unitary in the sense that the form (6.1) is left in- metric twistors is now allowable. Note also that
variant, that is, in matrix notation, (7.7)
o 0 1 0 o 0 1 0 since (2, 3, 0, 1) is an even permutation of (0, 1, 2, 3)
000 1 The transformations of the form (7.1) are not
(tp)* 0 0 0 1 (n) (7.5) the only allowable twistor transformations since, as
1 000 1 000 we have seen, the operations e, <P, :>, e<p, e:>, and
o 1 0 0 o 1 0 0 <pe:>, give conformal transformations of M but they
do not correspond to projective point transforma-
where the asterisk denotes Hermitian conjugate in tions of C. On the other hand, <P:> does have the form
the usual sense. 29 (7.1), with (7.2) and (7.6) holding, but where (7.4)
Equation (7.4) ensures that the operation of com- is replaced by
plex conjugation, according to the rules just stated,
is a twistor operation. That is, if A:~:.~·8 is a twistor T; = -lp; (7.8)
of valence [:], then A~P·::.8 is a twistor of valence Thus, for <P:>, we have A::: = - if the total A :::
were, 17 f or exampIe, A- a21••••••oa -- Ao1','2,
[r.] (h a.. . 1 t )
e c.. valence of A::: is odd. (This, again, is assuming
Clearly, the operations of addition, outer multi- that we impose (7.2). Had we chosen t~T: = -8~
plication, contraction (between upper and lower
indices), and index permutation (not mixing upper instead, then A::: = A:::,
but LaR" = -LaRa,
with lower indices) are also twistor operations. [That etc.) The operation e is given by
is to say, they commute with (7.1).] We might also (7.9)
consider complexified twistor transformations (7.1)
while <pe:> is a combination of (7.9) with a trans-
n The representation of twistors in terms of components formation of the form (7.1) satisfying (7.8). The
given in this paper is perhaps the most convenient, but is by operations <P and :> belong to a class of twistor trans-
no means the only one 'possible. It amounts to insisting that
the coordinate basis tWlStors Ea(Oh E~(lh Ea!2)} Ea m are null formations given by
and satisfy Ea(o) E(,)~ == 1 == Ea(l)E(I)", wltn all the other
scalar products vanishing. Another pOSSIble choice would be to Tp···.., -_ (=I).A.>.···.
Ltc/p"" --. 'P''''''U'''UP). ••• u"u-P'P '" u-..,'" ,
require E"(o)Ema == (-1)b8ob (b not summed!) in which case
twistor complex conjugation would take the form Lo ... LO, (7.10)
L1 == - £1, L, == V, La ... - La, instead of (3.5), and the fixed
matrix in (7.5) would become diag(1,-11,-1). (The simple where (Ud) and the transpose of T(U" P) are uni-
connection with spinors (2.13) would be iost, however.) The
only /l8lJential restriction on the way in which the twistors are modular inverse matrices:
represented is that the signature of the form X aX" must be
(++- - ) [ef., (6.1)J. (7.11)

Downloaded 13 Nov 2012 to 139.184.30.132. Redistribution subject to AIP license or copyright; see http://jmp.aip.org/about/rights_and_permissions
 TWISTOR ALGEBRA 359

The minus sign applies when a reversal of space ga = (-~A' a- I 77 A')

orientation is involved and the plus sign when it
since this agrees with (3.2) [and hence with (3.10)].
is the time orientation that is reversed. Equation
Thus, in this case
(7.11) implies that the matrix (u",~) is also pseudo-
unitary in the sense of (7.4) and (7.5), or (7.8).
These transformations give the projective correla- (Ua~) = [ae
o
AB
0].
a-leA'B'
(7.19)
tions of C. The antiprojective point transformations
of C such as e<p and e~, are given by the combination Here a reversal in time orientation (a2 > 0) or space
of (7.10) with (7.9). orientation (a2 < 0) is involved, so we get (7.10)
Let us now examine some twistor transformations rather than (7.1). The space-time reflection
in a little more detail. The translation given by
(7.20)
Zi = Xi + ai, lA = ~A (7.12)
is given if
[using the notation of (3.1), (3.2)] results in
(xa) = (~A, 77..1.' - i~BaBA')' so thaeo
30
whence
(t~) = [EAB 0B,J' (7.21)
(7.13) o EA'

For a space reflection or a time reflection, we must

The Lorentz rotation given, with (b A B) unimodular, introduce a timelike vector Vi at the origin. For
by convenience we normalize Vi to have length V2, so
that
(7.14)
results in (7.22)
iT.. (A The upper sign refers to space reflection and the
(A ) = b B~ B , -b- A , B' 71B') ,
lower sign to time reflection. In spinor terms we
whence have

(t~) = [bo ~
AB JJ' KK'J'J FA' BA' _ B'
;Z = ±x VK VK' i ~ = ~ VB, 77..1. = ±77B,VA
B,J' (7.15)
-b A, with v!'vg, = -o~. Thus,
iT B' A' B
Combining (7.14) with (7.12) and (7.15) with (7.13), Aa = (±VA 77B"VB~),
we get the general restricted Poincare transformation
so that
zJJ' = XKK' bJ Kb J ' K' + a J J', lA = ~BbA B (7.16)
[actually, "inhomogeneous SL(2, C) transforma- (Ua~) = [ 0
A'
(7.23)
V B
tion"lO] represented by
Note that Ua/l is symmetric for a space reflection
and skew-symmetric for a time reflection. Recall,
o
- b- A , B'
J (7.17) also, that Ua/l was skew-symmetric for the inversion
(7.18) [cf., (7.19)]. The significance of this lies in
so that matrices of the form (7.17) give a (two- the fact that there are two distinct kinds of pro-
valued) representation of the Poincar6 group. jective correlation in C which are involutory (i.e.,
The inversion whose squares are the identity), namely polarity
with respect to a quadric (corresponding to Ua/l
;Zi = 2a 2x i (XkXk)-1 (7.18)
symmetric) and null-polarity with respect to a linear
(a' > 0) is given if we put t' = a- 77
I A
', ~A = -a~A' complex26 (ua/l skew-symmetric). The null-polarity
i.e., is distinguished by the fact that any point in C
10 The staggering of the spinor indices is to indicate that, in
lies on the plane into which it is transformed, whence
each case, the left-hand index labels rows and right-hand index any null line in M meets the null line into which it is
labels columns. Also, for notational consistency, -fAB, is used transformed. Thus, in addition to the cases of time
here instead of 8AB' and fA,B' instead of 8A,B', although they
are all numerically equall reflection or inversion just considered, Ud is skew-

Downloaded 13 Nov 2012 to 139.184.30.132. Redistribution subject to AIP license or copyright; see http://jmp.aip.org/about/rights_and_permissions
360 R. PENROSE

symmetric in the case of a space reflection in a set of points representing a Robinson congruence.
plane (that is, in a timelike hyperplane). The The lines of the Robinson congruence are all in-
linear complex involved is, in each case, the system variant under the transformation but not pointwise
of lines in C which are left invariant by the correla- invariant. N oninvolutory transformations with this
tion. Some of these lines lie in N corresponding property also exist. These are intimately related
to the points of M which are invariant under the to twistors and their four-valuedness and so will be
transformation. These points constitute a hyper- described briefly next.
sphere in M which is spacelike [as in (7.18) with Choose a right-handed twistor R a normalized so
2
a > 0, or (7.22) with the lower sign] if a time reversal that
is involved, or timelike [e.g., (7.18) with a2 < 0] if
it is a change of spatial orientation that is involved.
(A hyperplane is to be regarded as a case of a hyper- Define
sphere in M.) Transformations for which U a {3 is
symmetric include, in addition to the space reflection
tee); = eiU(o; - RaR p) + e-3i9RaRp (7.24)
in the origin (or, more correctly, in a timelike line), for each real e. Then tee); satisfies (7.4) and (7.6), so
a reflection in a spacelike line which is accompanied it represents an allowable twistor transformation.
by a time reflection. In these cases, the points of Also
M left invariant by the transformation are rep-
resented by the generators which lie in N of the t(e)"t('P)~ = t(e + 'P)~
quadric (equation: X a X{3u a {3 = 0) defining the polar- so that these transformations form a one-parameter
ity in C. These invariant points in M constitute subgroup of twistor transformations. We can rep-
either a timelike circle (a timelike straight line being resent any line L of the Robinson congruence R, in
one case) or a pair of spacelike circles (a spacelike M, by a twistor La satisfying LaRa = 0 (and
straight line together with a "circle" at infinity U La = 0). Thus,
being one case).
The involutory projective point transformations t(e)"LfJ = ei9 U
of C are called harmonic perspectives. 26 They fall whence each line of R must be left invariant under
into two main classes, depending on whether the the transformation. Furthermore, except for those
invariant points of C constitute two skew lines values
t:
(when = 0) or a point and a plane (t: = ±2, ±2i).
e = tn7r (n = ... , - 2, -1, 0, 1, 2, ... )
The first class is further subdivided according as
the two skew lines both lie in N, both cross N, or for which t(e)" is proportional to there are no 0",
lie one in C+ and one in C-. (No other cases are points of M invariant under the transformation.
possible.) If the lines both lie on N, they correspond (This follows because, in the C picture, the lines
to two special invariant points of M. The spacelike which are left invariant are those which either pass
2-sphere in M of intersection of the null cones of through the point R or lie in the plane R. In neither
these two special points consists also of invarient case can these lines lie on N, since R is not on N
points. Examples of this type of transformation are and R does not touch N.)
the space-time reflection (7.20) in the origin 0 Finally, observe that for e = t7r we have
(whence 0 and I are the special invariant points
and the invariant 2-sphere is at infinity) or a reflec- t(i): = i 0;. (7.25)
tion in a plane together with a time reversal (in
which case the plane is the invariant "2-sphere" This gives the identity transformation on C and
and the special invariant points are both at infinity). therefore also the identity transformation on M.
If the two skew lines both cross N, we get a timelike But it multiplies every twistor of valence [~] by i. Also
2-sphere of invariant points in M (e.g., a timelike (7.25) is continuous with the identity twistor trans-
2-plane, for the case of an ordinary reflection in a formation via (7.24) with 0 :::; e :::; t7r. Hence
line). If neither skew line meets N, there are no twistors are essentially four-valued under conformal
invariant points in M [e.g., the inversion (7.18) transformations of M.
with a2 > 0, followed by the time reflection of (7.22)].
VIII. THE KERR THEOREM
The transformations of the second class are also
interesting in that no points of M are left invariant. Consider the problem of finding all null solutions
The plane R of invariant points of C meets N in a of Maxwell's equations (i.e., with F,jF" = 0 =

Downloaded 13 Nov 2012 to 139.184.30.132. Redistribution subject to AIP license or copyright; see http://jmp.aip.org/about/rights_and_permissions
 TWISTOR ALGEBRA 361

F'iFklt'ikl) in free space. Robinsonl l showed that plex analytic surface S in C. We see that the con-
the problem could be reduced to that of finding all gruence must be shear free. Putting 7JA' = -if\l:AA"
the shear-free null geodetic congruences in the space- as in (3.2), Eq. (8.2) can be used to solve for ~A, up
time [cf., (5.8)], since associated with any null to proportionality, as a function of XAA', i.e., to give
field was a congruence of this type, while conversely, the direction of the line X as a function of a point
given such a congruence, all the corresponding solu- on X with position vector Xi. Euler's condition on
tions of Maxwell's equations could then be defined tp, for it to be homogeneous (degree n), gives
in terms of arbitrary complex functions. The result
applies to curved space-times as well as flat. Weare Xa otp =" ~A otp + 7JA' otp = nip.
oxa O~A 07JA'
concerned here only with space-tim~s which are
(conformally) flat. With this restriction, Robinson's Hence, by (3.2) and (8.2),
result applies also to null zero rest-mass fields of
arbitrary spin (>!). (The result for spin 2 has a ~A(OtpA - iXAA' :.otp ) = 0,
bearing on the construction of null solutions of o~ V7JA'
Einstein's nonlinear field equations. 31 )
whence
The solution of the remainder of the problem for
flat space-time, namely the construction of all the
(8.3)
shear-free null congruences, is given by a remarkable
theorem due to Kerr. 32 The theorem takes the fol-
lowing very natural form when stated in twistor for some K. The derivative of (8.2) with respect to
terms. XAA' [ef., (5.3)] must also vanish

A congruence of null lines in M is shear-free

if and only if it is representable in C as the
° = VBB'tp = otp
O~A
VBB'~A +~V
07JA'
BB
' 7JA'

intersection of N with a complex analytic (8.1)

surface S in C (or as a limiting case of = otp VBB'~A
O~A
such an intersection).
otp ( . '("7BB' /:A ·,.A'("7BB' )
"Complex analytic" means here analytic in the + -:.-
V7JA'
-tXAA'V <; - '1,1; V XAA'
sense of the complex structure of C, where S is a
"surface" in the complex sense (i.e., it has four real = K~A VBB'~A _ i otp ~B,
dimensions). Thus, (8.1) tells us that the shear-free 07JB'
condition [cf., (5.8)] is, effectively, a Cauchy-Rie-
mann type of relation in the C picture. [The reason by (3.2) and (8.3). Assuming K ~ 0, contraction with
~B gives ~B~A VBB' ~A = 0, i.e., the condition (5.8)
for the final parenthetic remark in (8.1) is to enable
certain exceptional cases to be incorporated, for for the null directions defined by ~A to be tangent to
which the congruence does not form an analytic shear-free null straight lines, as required. Finally,
K cannot vanish if (8.2) represents a genuine condition
system even in the real sense. These exceptional
~A, since the left-hand side of (8.3) is simply the
cases appear to occur only when the rotation of
the congruence also vanishes, cf., (5.7).] derivative, with respect to ~A, of tp considered as a
Suppose, first, that a congruence of null lines in function of ~A and x AA , •
M is defined by an equation
The converse result that (8.2) represents essenti-
ally the most general shear-free congruence in M
is somewhat less straightforward because of the
existence of exceptional cases. However, if we assume
where tp is analytic and homogeneous in the four that the congruence is analytic in the real sense-
complex variables xa so that (8.2) defines a com- and any nonanalytic congruence (presumably neces-
sarily rotation-free) can be approximated arbitrarily
311. Robinson and A. Trautman, Proc. Roy. Soc. (London)
closely by analytic ones-then we can see from the
A26S, 463 (1962).
32 R. P. Kerr, private communication; cf. also R. P. Kerr, form of (5.8) that the congruence is determined once
Phys. Rev. Letters 11,238 (1963); R. P. Kerr and A. Schild, the directions of the lines (i.e., of ~A) are known on
in Proceedings of the American Mathematical Society Sym-
posium, April 1964. According to Kerr's original construction, any spacelike 2-sphere F in M. [If ~o / = r, then e
a general shear-free null geodetic congruence is defined by an
analytic relation ¢(." xOO' + '1X10', xOl' + .,xll ') = 0, where
(5.8) becomes or/ox00' = /
ox
10'
-ror
/
ox
11'
= ,or
dx 0o ' + .,dx10 ' = 0 = dx° 1' + .,dxll '. - r- or/
1
OX01'. This defines the propagation of r

Downloaded 13 Nov 2012 to 139.184.30.132. Redistribution subject to AIP license or copyright; see http://jmp.aip.org/about/rights_and_permissions
362 R. PENROSE

off the 2-plane xoo' = Xli' = 0. On this 2-plane- where SaP"'! is symmetric, of valence [~l. In terms of
which from the point of view of M is a 2-sphere- ~A and 71A' = -i~AxAA" (8.4) becomes
t can be given arbitrarily as a function of the com-
plex variable X0 1 ' and its complex conjugate XlO'.] ~A~B ••• ~Dcf>AB"'D = 0, (8.5)
The 2-sphere F is the intersection of two null cones where the spinor field cf> AB"'D is defined by
(vertices A, B) in M and is therefore represented,
in the C picture by the lines lying in N which meet
two skew lines A, B in N. The complexification F*
of F is, in the C picture, the system of all lines in C
meeting both A and B. Now (in the M picture), F the ; .. .'s being constants which are essentially the
is a real environmentlO for F*, so that any complex an- coefficients SaP"'" (The round brackets denote'sym-
alytic function on F extends uniquely to a complex metrization.) The cf>AB"'D is symmetric, satisfies the
analytic function on F*. The null directions of the tI field equation"
congruence, defined at points of F, are represented
in the C picture as an co 2 system of points lying on (8.7)
the lines of the F system (describing a real analytic
surface). These extend uniquely to an co 4 system of by virtue of (8.6) and has a "canonical decomposi-
points S on the lines of the F* system (describing a tion,,7
complex analytic surface). The intersection of S 1 2

with N then defines the given shear-free congruence cf>AB"'D = ~(Ah •• , ~D)' (8.8)
as required (since the congruence defined by S
where each ~ of (8.8) satisfies (8.5). Conversely,
agrees with the given one on F).
every solution of (8.7) has the form (8.6) and each
The Kerr theorem provides a very convenient
resulting ~ of (8.8) [or (8.5)] defines (one branch of)
means of studying the structure of shear-free null
the corresponding algebraic shear-free null congru-
congruences in M, in general. Only a few results
ence. This generalizes the results of Sec. V (for
are briefly indicated here. For example, the lines
which n = 1) and gives us a representation of an
of a shear-free null congruence along which the rota-
arbitrary symmetric twistor of valence [~], in terms
tion vanishes are represented in C by the points
of a symmetric spinor field satisfying (8.7).
where lines of N touch S. We can also generally
Robinson's construction of the general, null, zero
form the reciprocal 26 S of S with respect to the
rest-mass field from its associated shear-free con-
Hermitian correlation defined by N (i.e., the en-
gruence can also conveniently be represented in
velope of polar planes with respect to N of points
twistor terms: the field can be defined in terms of a
of S). Then S defines a shear-free null congruence
complex function on S. However, such matters are
in M which is, in a sense, "reciprocal" to the original
not entered into here. The twister description of
one. The congruences which are everywhere rotation-
physical fields, generally, is left to a later paper.
free are the ones which are self-reciprocal (although
the individual lines of the congruence do not gen-
IX. GEOMETRICAL APPLICATIONS OF TWISTORS
erally reciprocate to themselves). In this case, S
is a ruled surface 26 in C, with ex> 1 of its generators Many interesting geometrical properties arise from
lying in N. This ex> 1 system corresponds, in M, to a the interplay between the geometric structure of
curve and the lines of the congruence are just the M* and that of C. Some of these result simply from
null lines meeting this curve. 33 This curve in M is the fact that the former is the Klein representation
null if the ruled surface in C is developable. of the latter and are, therefore, essentially classical
Of special interest are the algebraic shear-free null results of algebraic geometry.26 Others, however,
congruences. 34 In this case S is algebraic variety26 take into account the reality structure of M and so
and tp can be given as a homogeneous polynomial have a more direct relevance to the structure of the
a physical world. The natural algebra of the C pic-
tp(X ) == SaP ... IXaX P •.• Xl = 0, (8.4)
ture-namely twistor algebra-can be used to
II The rotation-free, shear-free null congruences which are derive certain geometrical properties of M. A small
not analrtic in the real sense emerge here simply as the system selection is given here.
of null lines meeting a nonanalytic curve in M.
84 Explicit shear-free null congruences of this type have been A linear space of dimension r in C can be rep-
used to generate explicit solutions of Einstein's equations. For resented by a (simple) skew-symmetric twistor of
example, in Kerr's construction of the field of a rotating body,
tp(x II) is quadratic. For details, see Ref. 32. valence [r~l] or by its dual of valence [3~']' (r =

Downloaded 13 Nov 2012 to 139.184.30.132. Redistribution subject to AIP license or copyright; see http://jmp.aip.org/about/rights_and_permissions
 TWISTOR ALGEBRA 363

0, 1, 2). The relation between general skew-sym- terms of the involutory transformation (see Sec.
metric twistors and their duals is VII) which is represented in C by the harmonic
perspective whose axes are the lines P and P in C
(9.1) (assuming P and P are skew, i.e., P allPaP ¢ 0);
8
Ca -iC!l-r e"II"Y8, or in terms of linear systems of Robinson congru-
ences. However, if qi is spacelike, we have a more
where conversely
easily visualizable representation. 36 In this case, the
lines P, P in C each meets N in an <Xl 1 system
(9.2)
of points. In the M picture, these become two <Xl 1
systems of null lines. No two null lines belonging
Because of (7.7) and the symmetry between (9.1) to the same system can intersect, but every null
and (9.2), the operations of complex conjugation line of the P system meets every null line of the
and of forming the dual commute. The meaning P system (since in the C picture, all the points of the
of, for example, Ball, given Ball, is therefore un- line P are conjugate under the Hermitian correla-
ambiguous. We call Ball real if Ball = Ball (whence tion to all the points of the line P.) We can refer to
also Ball = Ball). the P system of lines in M as a null regulus and the
A skew-symmetric twistor pall is simple if it is of P system as its complementary null regulus. 26 Any
the form two nonintersecting null lines in M belong to a
unique null regulus; the null transversals of these
(9.3) two null lines generate the complementary null
whence P a/3, pa/3, Pall are all also simple. (Call"Y and regulus. The null reguli in M thus geometrically
Aa/l')' are necessarily simple.) Equivalent alternative represent the points in M* whose position vectors
conditions for pall to be simple are have spacelike imaginary parts.
A real skew-symmetric twistor B all ( =Ball), which
pallp a"y = 0 or pa/lp a/3 = O. (9.4) is not simple, also has a direct interpretation in M.
As given by (9.3), pall represents, in the C picture, We can, in fact, normalize Ball so that
the line joining two points X and Y. In the M pic- (9.7)
ture, pall represents a complex point P (i.e., point
of M*). Thus, we can use simple skew-symmetric Then, if we put BaP = Ua/l in (7.10) we get an
twistors pall for projective coordinates in M*. involutory projective correlation, in C, of the type
Now, the Hermitian correlation defined by N for which the invariant points in M constitute a
effects the correspondence pa/3 ~ p a/l' This is the hypersphere (ct, Sec. VII). Thus, Ball represents a
complex conjugation operation in M* [cf., (7.9)] so, hypersphere in M which is spacelike or timelike
in terms of our coordinates pall, complex conjuga- according as the upper or lower sign occurs in (9.7).
tion in M* is defined by pall ~ pall, the points of M (In the limiting case when Ball becomes simple, the
being given when paP is real (in the above sense). hypersphere becomes a null cone with vertex B.)
Corresponding to (6.5)-(6.10) we can also express, We have seen that A a represents a point A in C
in terms of pa/3, the space-time nature of the imag- and that A a represents its" polar plane" with respect
inary part q' of the position vector of a complex to N (cf., Sec. VI). A lies onN if and only if A aAa =
point P of M*: 0, which is also the condition for A to touch N. In
terms of the dual twistor A a/ln this condition is
> . . >
all-
P 'P a/l < 0 accordmg as q' qi <0, (9.5) (9.8)

and if qiqi ~ 0, then for all Za Let xa, va, za be three (nonzero) null twistors so

Z tllpa"Yp f~ 0 if qi future-pointing, 81 When qi is timelike we may represent P as a parallelism

a /l'l'l (9.6) on M (with torsion; left-handed if PCC+) which is closely

~ 0 if qi past-pointing. related to Clifford parallelism on 8 1. The sets of null directions
which are to be regarded as parallel are those of the Robinson
congruences represeI!ted by the points of the line P in C (i.e.,
[The point of intersection of the plane Z with the by planes through P). A transitive four-parameter group of
line P in C is represented by Zapa"Y and, cf., (6.5), (conformal) motions of M preserves this parallelism, namely
that given by twistor transformations (7.1) for which the line
(6.6), (6.8), and (6.9).] P is left pointwise invariant. This group is readily Been to be
A complex point of M* can be realized in terms of the group of unitary (2 X 2) matices and leads to Uhlmann's
representation (Bee Ref. 18) of the points of M in tertns of
a real structure in M in various ways; for example, in Buch matrices.

Downloaded 13 Nov 2012 to 139.184.30.132. Redistribution subject to AIP license or copyright; see http://jmp.aip.org/about/rights_and_permissions
364 R. PENROSE

just considered (cf., Fig. 6), the graph would be that

of a tetrahedron. 36 Any other polyhedron of triangles
FIG. 6. A geometrical theorem con- would also give rise to a theorem on null lines.
cerning null lines in M.
Finally, let us consider the construction of a Robin-
son congruence R containing three given null lines
U, V, W in general position in M. (In the C picture,
that, in the M picture, X, Y, and Z are three null R is the plane of the three points U, V, W.) We
lines. Put wish to construct the line of R which passes through
a general point 0 in M. Now, a Robinson congruence
(9.9) is characterized by the fact that, if any two null
(Square brackets denote skew-symmetrization.) In lines X, Y belong to the congruence, then so does
the M picture, A is generally a "complexified null every line of the null regulus containing X and Y.
line". But if (9.8) is satisfied, then A is a real null (In the C picture, the join of any two points in a
line in M which meets each of X, Y, and Z. (A "X" = plane also lies on the plane.) We have seen that the
OfollowsfromA 1aP'YX6 ] = 0, etc.) Thus, the condition null regulus containing X and Y is simply the set
for the three null lines X, Y, Z to have a common null of null transversals of any pair of null transversals
transversal is simply (9.9) substit.uted into (9.8), of X and Y. Thus, if we can find X and Y belonging,
i.e., respectively, to the null regulus containing U, V and
{XY} I YZIIZXI = 1, (9.10) to that containing U, W such that the null regulus
where containing X and Y also contains a line Z through
IXYI = -(Xay,,)(yp.xpf i = {YXI- 1 o (see Fig. 7), the construction is complete. What
is required, in fact, is to show that there is a null
etc. (since X a, y a, and za are null). We assume no line Q through 0 which meets a line (namely X)
two of X, Y, Z meet so that (9.10) is well defined. of the U, V regulus and a distinct line (namely
If X, Y, and Z have two common transversals, then Y) of the U, W regulus; i.e., the triplets U, V, Q
(9.9) vanishes and X, Y, and Z belong to a null and U, W, Q must each have a null transversal.
regulus. In fact, there is, in general, a unique such line Q
It is possible to derive a host of geometrical theo- through O. For, putting Qa = sa + rT a, where T
rems concerning incidence of null lines in M, from and S are null lines through 0, with T meeting U,
the condition (9.10). For example, if four null lines and substituting in (9.10), we get two simultaneous
in M, of which no two meet, are such that there is a linear equations in rand f. The condition for these
null transversal to each of three different selections equations to have a unique solution reduces to
of triplets of the lines, then there is also a null {UVI I VTI {TWI I WUI :;'" 1 which, for general
transversal to the remaining triplet. (The configura- positions of 0, is indeed satisfied. 37
tion is that depicted in Fig. 6-except for the case
when there is a single common transversal to all four X. THE METRIC TWISTOR
lines.) Some theorems of this general type can be If we wish to use twistors to describe the Minkow-
generated by means of a diagrammatic notation: a ski metric st.ructure of M (i.e., Mill), rather than
graph made up of triangles whose vertices represent j list its conformal structure, we may do this by
null lines in M, can be used. The triangles represent introducing a metric twistor r
P which represents
t.riples of lines wit.h a common null transversal. Now, the point I of M, according to the scheme of Sec.
if anyone of t.hese triangles represents a circuit in IX. Thus, r P is simple, skew-symmetric, and real
the graph linearly dependent on the circuits given (in the sense of Sec. IX):
by other triangles, then, because of the form of
laPI,,'Y = 0, (10.1)
(9.10), we have a geometrical theorem. For the case
laP = laP (10.2)

r
,UjwQ
~l;1!t.!tIl..: y Since a real null line in M defines both a point in C and a

Q'<f.&o~v. ~ Construc~i~n
36
-/j
plane in C through this point, the configuration of Fig. 6 is
FIG. 7. of the Robil!Bon represented in the C picture as a pair of mutually inscribed and
T. v, ~~x congr:uence contammg three gIven circmllScribed tetrahedra-a configuration faIniliar to geom-
", null hnes U, V, and W. eters. We may note that the full complexification of a null line
in M leads, in the C picture, strictly to a point in C together
", \k§A with a plane through it. This gives a five-complex-dimensional
"'o:P- z system as we would expect.
87 It only- fails if there is a circle through 0 meeting U, V,
and W which lies on a null cone through U.

Downloaded 13 Nov 2012 to 139.184.30.132. Redistribution subject to AIP license or copyright; see http://jmp.aip.org/about/rights_and_permissions
 TWISTOR ALGEBRA 365

(with r~ = - I~a). The twistor transformations PQ orthogonally. (These twistors are not simple,
which leave r~ invariant give the Poincare group. so they represent 1/ hyperspheres" rather than
(The transformations which leave r~ invariant up points.) We can check the coefficients m (lOA)
to proportionality give the Poincare group plus the using
dilatations. These would be just the conformal trans-
formations of M which leave the point I-and (10.6)
its null cone-invariant.) which is a familiar Poincare invariant expression.
We may also treat de Sitter space-time in a very The derivation of expressions such as (10.6) is
similar way. In this case there is a hypersphere at facilitated if we specialize our twistor structure still
infinity rather than a null cone at infinity. As we further by the introduction of an origin twistor
have seen in Sec. IX, this can be described by a O"~ which represents a particular point 0 (the
real, skew-symmetric twistor which is not simple, 1/ origin") in M. Here, oa~ is to be simple, skew-

i.e., we just drop condition (10.1). Other conformally symmetric, real and normalized as in (10.3)
flat space---times can also be treated. For example,
if we retain (10.1) but drop (10.2) [and replace
(10.3) by its modulus], we can describe the Einstein (10.7)
static universe. The matter is not pursued further p
By putting Qa = oa~ in (lOA) and (10.6), we obtain
here, however.
the twistor expressions for the Lorentz covariant
We can generate Poincare covariant operations
vector operations of linear combination and squared
simply by employing twistor algebra and admitting,
magnitude.
as basic elements of the algebra (in addition to
. t ors I a" ,a", Since the transformation group leaving both r~
a) t h e tWIS
EaPp~, E a~p~ ,Op, I sat'IS f ymg
.
and oa~ invariant is the Lorentz group, we can ex-
(10.1), (10.2). We can represent points P, of M*II}
pect to be able to express spinors, referred to the
by simple skew-symmetric twistors p a " normalized,
origin 0, in terms of twistors. In effect, we carry
for convenience, so that
out the construction given in Secs. II and III for
(10.3) twistors in terms of spinors, but in reverse. This
E M II} if p a{3
gives us a correspondence between spinor indices
(P = pa".) Then the correspondence
and 1/ reduced" twistor indices for which
between P and pa~ is unique. 3s If Qa{3 and Ra~
similarly represent points of M* II I, then an example oafJ +--t EAB , laP +--t EAB,
(10.8)
of a Poincare covariant operation is
OaP +--t EA'B', I"fJ +--t EA'B"

aP
afJ+ bQ"~ + cR"~
The orthogonal idempotents which reduce the twistor
- !lbcQ1~R'Y~ + caR"~P'Y~ + abp.,6Q1~lr". (lOA) space-to a direct sum of two spinor spaces (one
The significance of this particular expression is that unprimed and one primed)-are
if J; = oa 1I p1 +--t o! = lG EBG , (10.9)
a+b+c=l (10.5) J; = OfJ1r1 +--t o!: = EB'G'EA'G"
then (lOA) represents the point in M*1I1 whose We have [from (10.1), (10.2), (10.7), etc.]
position vector is ap; bq; + +
cr;, where pi, l, r;
a -P - a fJ
are the respective position vectors of P, Q, R. [This J ~J 1 = 0 = J pJ 'Yl (10.10)
vector operation is clearly Poincare covariant if
J~J~ = J~,
(10.5) holds and represents a weighted mean.] Ex-
pression (lOA) generalizes to any number of points A general twistor p:~::;6, of valence [:] corre-
in an obvious way. Other less involved expressions sponds to a set of 2r+8 spinors. To obtain such a
than (lOA) can, of course, also be given and define spinor, each index of P:~::;~ is transvected with
Poincare covariant operations in M I I}. For example, either a J~ or a J~. The reSUlting twistor then rep-
p a" +ar" and pa" _ Qa" represent, respectively, resents a spinor with an unprimed index correspond-
a hypersphere center P and the hyperplane bisecting ing to each free J ~ index and a primed index, in the
reverse position, corresponding to each free J~ index.
38 In practice, the X a and ya of the decomposition (9.3)
(which may be specilized if desired) often turn out to be For example,
more convenient coordinates than pafJ. This is of value in
connection with physical fields and will be discussed elsewhere. P ·~~ 9'X JaJ-~J1J-"'JX
IC P
). " (f
+--t nA B'
GR'
s·

Downloaded 13 Nov 2012 to 139.184.30.132. Redistribution subject to AIP license or copyright; see http://jmp.aip.org/about/rights_and_permissions
36~ R. PENROSE

With the different possibilities for J'j and J'j, these where PAB' is given as in (3.10). In the C picture P
give 2 r +. different spinors, which [by (10.10)] to- is the line joining two points X and L; in M*, P is the
gether determine P:!:.~·'. Any spinor operation can (complex) point of intersection of two (" complexi-
thus be mirrored in twistor terms, using J'j and fled") null lines. Thus, according to (3.10), pi rep-
J~. The correspondences of (10.8) and (10.9) are resents the positioo vector, with respect to 0, of
consistent with this. the point P in M*. We can express (10.11) in
The basic relations (2.13) and (3.1) are expressed matrix form as
as
X~J; ~~A, X~ J; ~ 71A',
(P"fJ) = [.EABB -ipAB' 1
'lPA' -iPipiEA'B'
L~J;~>.A, L~J; ~ 1-'04."
Expressions such as (10.6) and (10.4) then follow
If we put at once.

/lP"~ = X"LfJ - L"X~ ACKNOWLEDGMENTS

[cf., (9.3)] with I' chosen so that (10.3) holds, we get The author is greatly indebted particularly to
/I = ~A>.A. Then,
I. Robinson, but also to E. L. Schticking and to
R. P. Kerr, for stimulating ideas in the early stages
P" J~J~ ~ tAB, } of this work.
This work was done partly at the University of
P"'J~j~ ~ -ip~" (10.11)
Texas, where it was supported, in part, by the U. S.
pP(JJ~j! +-+ -tpiP;EA'B', Air Force under Contract, AF-AFOSR-454-63.

Downloaded 13 Nov 2012 to 139.184.30.132. Redistribution subject to AIP license or copyright; see http://jmp.aip.org/about/rights_and_permissions