Euclidean Twistor Unification

Peter Woit
Department of Mathematics, Columbia University
woit@math.columbia.edu
October 14, 2021

Abstract
Taking Euclidean signature space-time with its local Spin(4) = SU (2)L ×
SU (2)R group of space-time symmetries as fundamental, one can consis-
tently gauge the SU (2)R factor to get a chiral spin connection formulation
of general relativity, and the SU (2)L factor to get part of the Standard
Model gauge fields. Reconstructing a Lorentz signature theory requires
introducing a degree of freedom specifying the imaginary time direction,
which will play the role of the Higgs field.
Conformally compactifying R4 to S 4 , one can identify this S 4 with
the quaternionic projective space HP1 , and the tautological H bundle
as the bundle of right-handed spinors. Euclidean twistor geometry is
based on the idea that one should work with the projective twistor space
P T = CP3 , which is a bundle over HP1 with fiber CP1 . A point in
the fiber is a complex line C inside the H specifying the point in HP1
and defines a complex structure identifying the tangent space R4 to HP1
with C2 . The Higgs field specifying the imaginary time direction needs
to be taken to be a field on P T , lying in this C2 . CP3 comes with an
internal U (1) × SU (3) symmetry at each point, providing the rest of the
Standard Model internal symmetries. The transformation properties of a
generation of fermions have a simple expression on P T .
This geometry simply encodes the symmetries and degrees of freedom
that go into the Standard Model and general relativity. Such a theory is
naturally defined on projective twistor space rather than the usual space-
time, so will require further development of a gauge theory and spinor
field quantization formalism in that context.

1 Introduction
Quantum field theories in Minkowski space-time suffer from inherent definitional
problems, which sometimes can be resolved by defining the theory in terms of
an analytic continuation from Euclidean space-time. The change in space-time
signature changes quite a bit the nature of the theory. Minkowski space-time
quantum fields are non-commuting operators satisfying an equation of motion,
with free-field Wightman functions that are distributional boundary values of

1
holomorphic functions and no distinguished time direction. Euclidean space-
time quantum fields commute and satisfy no equation of motion, with free-field
Schwinger functions that are actual functions and an imaginary time direction
specified by the choice of analytic continuation to Minkowski signature.
While in Minkowski space-time one can define the space of states covari-
antly, starting with a Euclidean theory one needs to pick an imaginary time
direction, which gets used to define the state space and a Lorentz-invariant
inner product (using Osterwalder-Schrader reflection in the imaginary time di-
rection). In the path integral formalism this is the direction perpendicular to
the hypersurface used to define states. If one instead starts with a Minkowski
space-time theory and decides to analytically continue to Euclidean space-time,
one finds that there is an infinity of possible Euclidean slices of the complex-
ification to use, with each one characterized by the choice of imaginary time
direction. This can be clearly seen in the twistor formalism (see appendix A)
where the SU (2, 2) conformal symmetry of Minkowski space is determined by
a (2, 2) signature Hermitian form Φ. On projective twistor space P T the null
space of Φ is five-dimensional, projecting down to the τ = 0 S 3 subspace of
compactified Euclidean space S 4 . Different projections correspond to different
choices of an imaginary time parameter τ .
The Lorentz group SL(2, C) is a simple group which acts on the physical
state space of the Minkowski space-time theory, while the Euclidean analog
Spin(4) = SU (2)L × SU (2)R is a product of two simple groups which does not
act on the physical state space. The quite different nature of these two groups
has always made it difficult to understand the relationship between quantum
field theories of spinor fields in Minkowski and Euclidean space. For a detailed
discussion of these issues, see appendix B. Gauging space-time symmetries gives
one approach to quantum gravity theories, where again one finds a problematic
relationship between Minkowski and Euclidean signature theories.
We will argue here that one should take as fundamental four-dimensional
quantum field theory in Euclidean signature, and if one does this, the symmetries
and degrees of freedom of the Standard Model and general relativity appear
very naturally. The connection one gets from gauging the SU (2)R subgroup of
Spin(4) can be used to formulate Einstein’s equations and general relativity.
The SU (2)L subgroup plays the role of the internal symmetry of the weak
interactions, which after gauging gives part of the Standard Model gauge fields.
Starting with a Euclidean signature theory, definition of the state space and
reconstruction of a Minkowski signature theory by analytic continuation require
introducing a degree of freedom that breaks the Spin(4) symmetry by picking
out an imaginary time direction. Having such a new degree of freedom also
allows one to consistently treat SU (2)R symmetry as a space-time symmetry,
SU (2)L as an internal symmetry. This degree of freedom has the correct proper-
ties to get identified with the Higgs field of the Standard Model. For this to work
correctly, one needs to take advantage of another aspect of four-dimensional ge-
ometry by using twistor theory.
Penrose’s 1967 [32] twistor geometry provides a remarkable alternative to
conventional ways of thinking about the geometry of space and time. In the

2
usual description of space-time as a pseudo-Riemannian manifold, the spinor
degree of freedom carried by all matter particles has no simple or natural ex-
planation. Twistor geometry characterizes a point in Minkowski space-time
as a complex 2-plane in C4 , with this C2 providing tautologically the (Weyl)
spinor degree of freedom at the point and an inherent parity-asymmetry. The
C4 is the twistor space T , and it is often convenient to work with its projec-
tive version P T = CP3 , the space of complex lines in T . Conformal symmetry
becomes very simple to understand, with conformal transformations given by
linear transformations of C4 .
Twistor geometry most naturally describes not Minkowski space-time, but
its complexification, as the Grassmanian G2,4 (C) of all complex 2-planes in the
twistor space T . This provides a joint complexification of the Euclidean and
Minkowski signature spinor and twistor geometries, allowing one to see how
they are related by analytic continuation. Focusing on the Euclidean rather than
Minkowski version, it is a remarkable fact that the specific internal symmetry
groups and degrees of freedom of the Standard Model appear naturally, unified
with the space-time degrees of freedom. Besides the two SU (2)s from Spin(4),
since projective twistor space P T can be thought of as
SU (4)
CP3 =
U (1) × SU (3)
there are U (1) and SU (3) internal symmetry groups at each point in projective
twistor space. Lifting the choice of a tangent vector in the imaginary time
direction from Euclidean space-time to P T , the internal U (1) × SU (2) acts
on this degree of freedom in the same way the Standard Model electroweak
symmetry acts on the Higgs field.

2 Four dimensional geometry

2.1 Four dimensional geometry in terms of two by two
matrices
Instead of describing four-dimensional Euclidean space E4 (R4 with the usual
positive definite norm) by a list (x0 , x1 , x2 , x3 ) of four real coordinates, one
can work with a subspace E4 ⊂ M (2, C) of the two by two complex matrices,
identifying
(x0 , x1 , x2 , x3 ) ↔ x = x0 1 − i(x1 σ1 + x2 σ2 + x3 σ3 )
(here σj are the Pauli matrices). The norm-squared is then given by
|x|2 = det x
Instead of describing the group SO(4) of four-dimensional rotations in terms of
orthogonal real four by four matrices, one can now use pairs (gL , gR ) of SU (2)
matrices, acting by
−1
x → gL xgR

3
This action preserves the subspace E4 , as well as the norm |x|. (gL , gR ) and
(−gL , −gR ) give the same rotation, and one finds that the product group

Spin(4) = SU (2)L × SU (2)R

is a double cover of the rotation group SO(4). Dimension four is very special:
it is only in this dimension that rotations are not a simple group, but a product
of two factors.
Instead of using complex matrices, one can use the algebra H of quaternions,
identifying
(x0 , x1 , x2 , x3 ) ↔ x = x0 1 + x1 i + x2 j + x3 k
The norm-squared is
|x|2 = xx
Spin(4) acts as above, except now SU (2) = Sp(1) is the group of unit quater-
nions, and two such groups Sp(1)L and Sp(1)R act indepently by left and right
multiplication. Note that, using either complex matrices or quaternions, the
imaginary time direction is distinguished, since it corresponds to the identity
matrix.
In special relativity one takes space-time to be not E4 , but E3,1 , meaning R4
with the Minkowski norm. Here again one can take E3,1 ⊂ M (2, C), identifying

(x0 , x1 , x2 , x3 ) ↔ x = −i(x0 1 + x1 σ1 + x2 σ2 + x3 σ3 )

with the norm-squared again given by the determinant

|x|2 = det x = −x20 + x21 + x22 + x23

The spin double cover of the group SO(3, 1) of linear transformations preserving
the Minkowski norm is SL(2, C), with group elements acting by

x → (g † )−1 xg −1

Of less relevance to physics is E2,2 ⊂ M (2, C), the subspace of real matrices,
in which case one identifies
 
x0 + x3 x1 + x2
(x0 , x1 , x2 , x3 ) ↔ x =
x1 − x2 x0 − x3

where
|x|2 = det x = x20 − x21 + x22 − x23
The spin double cover of SO(2, 2) is

Spin(2, 2) = SL(2, R)L × SL(2, R)R

given by pairs gL , gR of elements of SL(2, R) acting by

−1
x → gL xgR

4
 If one complexifies (takes complex rather than real linear combinations) the
real vector spaces E4 , E3,1 , E2,2 , in each case one gets the same result, the
complex vector space M (2, C) of all complex two by two matrices. The group
preserving the norm-squared given by the determinant is now SO(4, C) which
has spin double cover
Spin(4, C) = SL(2, C)L × SL(2, C)R
given by pairs gL , gR of elements of SL(2, C) acting by
−1
x → gL xgR
(x is now an arbitrary complex two by two matrix).

2.2 Spinors and twistors

Writing (complexified) four-dimensional vectors as two by two complex matrices
identifies vectors as linear maps from one C2 (which we’ll call SR ) to another
C2 (which we’ll call SL ). SR and SL are the spinor spaces for four-dimensional
geometry, with vectors elements of the space Hom(SR , SL ) of linear maps. In
the case of E2,2 one can use R2 instead of C2 , while for E4 and E3,1 one
needs C2 . The cases E4 and E3,1 are however very different from each other.
For E4 , SL and SR are completely independent spaces, transforming under
Spin(4) by two different SU (2) groups. For E3,1 on the other hand, when
g ∈ Spin(3, 1) = SL(2, C) acts on SR , this determines its action on SL (by
(g † )−1 ).
In the twistor theory approach to four-dimensional geometry (see Appendix
A), points are given by a C2 ⊂ C4 (C4 is the twistor space T ), with the C2
a spinor space SR . Tangent vectors at such a point are linear maps from this
SR to SL which gets identified with the quotient T /SR . The four dimensional
geometry here is a complex geometry, with E4 , E3,1 , E2,2 occuring as real four-
dimensional subspaces. This provides a context in which the spinor space at a
point is tautologically defined, and in which one can study analytic continuation
between the Euclidean geometry E4 and the Minkowski geometry E3,1 .
The twistor picture naturally includes a much large group of symmetries,
the conformal group. One can identify points at infinity in such a way (the con-
formal compactification) that the complex four-dimensional geometry is that of
G2,4 (C), the Grassmannian of all C2 ⊂ C4 . The group SL(4, C) then acts,
with subgroups Spin(5, 1), Spin(4, 2), Spin(3, 3) that act as conformal transfor-
mations on the compactifications of E4 (S 4 ), E3,1 (S 3 × S 1 ) and E2,2 (G2,4 (R))
respectively.

3 Four dimensional Euclidean quantum field the-

ory and gravi-weak unification
There is a long history of attempts to quantize general relativity in Euclidean
space, for a discussion see for instance [18]. Such an attempt runs into both

5
technical problems and conceptual puzzles. Here we’ll propose a somewhat
different context for this problem, which may shed new light on these issues.
There have been various proposals (see e.g. [3] and [30]) for unifying the
weak and gravitational interactions by gauging SU (2) and Lorentz (SL(2, C))
subgroups of the complexified space-time symmetry group Spin(4, C). We will
argue that one should instead work with Euclidean quantum field theory and
the symmetry group Spin(4) = SU (2)L × SU (2)R . One can consistently take
SU (2)L to be an internal symmetry, gauge it and construct the usual Yang-Mills
SU (2) gauge theory responsible for the weak interactions. The SU (2)R will be
a space-time symmetry, and gauging it leads to a conventional version of general
relativity, expressed in terms of a chiral spin connection.
The existence of a non-zero distinguished vector e0 ∈ Hom(SR , SL ) in the
imaginary time direction (necessary for reconstructing a Minkowski space-time
theory) allows one to recover the usual geometry of rotations and spin in the
spatial directions. Identifying an arbitary vector x with e−1
0 x ∈ Hom(SR , SR )
one finds that this transforms under Spin(4) as
e−1 −1 −1 −1 −1 −1
0 x → gR e0 gL gL xgR = gR e0 xgR

If x is in the e0 direction then e−10 x is invariant under this action. If x is a

spatial vector, e−1
0 x is invariant under SU (2)L , and transforms as a usual E3
vector under SU (2)R . SU (2)R is thus the spin double over of the SO(3) group
of spatial rotations, and SR is the usual spin representation in three-dimensional
space.

3.1 General relativity in terms of chiral spin connections

The geometry of a Riemannian manifold M of dimension n can be described in
a formalism close to that of gauge theory, by using the principal SO(n) bundle
of orthonormal frames (see for example [25]). On this bundle (or on a spin
double-cover) one has two kinds of 1-forms:
• Spin connection 1-forms ω which take values in the Lie algebra spin(n).
These describe infinitesimal parallel transport of not just vectors, but also
spinors. These are the usual connection 1-forms one has for any principal
G-bundle (here G = SO(n) or Spin(n)).
• Canonical 1-forms e which take values in Rn , and at a point in the frame
bundle give the coordinates of a vector with respect to the orthonormal
frame. These 1-forms are special to frame bundles.
In the Palatini formalism for general relativity in four dimensions, one takes as
fundamental fields e, ω, with an action of the form
Z
ABCD eA ∧ eB ∧ ΩCD (ω) (3.1)
M

where the indices take values 0, 1, 2, 3 and Ω(ω) is the curvature 2-form for the
spin connection ω (like ω, it has values in spin(4)).
The equations of motion are then

6
 • Varying the ω AB gives

deA + ωB
A
∧ eB = 0

This is the torsion-free condition, determining ω in terms of e to be the

Levi-Civita connection.
• Varying the eA gives the Einstein equations (written in terms of the e
instead of the metric).
The decomposition Spin(4) = SU (2)L ×SU (2)R implies very special proper-
ties of the spin connection and Riemannian curvature tensor in four dimensions
(for details of the following, see e.g. [6]). Since

spin(4) = su(2)R ⊕ su(2)L

the spin connection and curvature decompose as

ω = ωR + ωL , Ω = ΩR (ωR ) + ΩL (ωL )

(the curvature of an su(2)R -valued connection is su(2)R -valued, similarly for

su(2)L ). Acting on spinors, ωR and ΩR take values in End(SR ), ωL and ΩL
take values in End(SL )
Two-forms on a four-dimensional Riemannian manifold decompose as

Λ2 (M ) = Λ2+ (M ) ⊕ Λ2− (M )

into ±1 (self-dual and anti-self-dual) eigenspaces of the Hodge star operator.

Using the identification of tangent vectors with linear maps from SR to SL and
the identification of tangent vectors and 1-forms given by the metric, one finds
that when one writes 2-forms in terms of spinors

Λ2+ (M ) ⊂ Hom(SR , SR ), Λ2− (M ) ⊂ Hom(SL , SL )

Taking this into account, the curvature two-form in four-dimensions has a de-
composition

Ω(ω) = Ω+,R (ω) + Ω+,L (ω) + Ω−,R (ω) + Ω−,L (ω)

If ω is torsion-free, then the condition that it be the connection for a solution

to the vacuum Einstein equations is

Ω−,R (ω) = 0

and this implies that one also has

Ω+,L (ω) = 0

Solutions with Ω−,L = 0 will be self-dual (or “half-flat”). It’s a remarkable

aspect of four-dimensional Riemannian geometry that one can express the con-
dition for a connection and curvature to be Einstein just using (ωR , ΩR (ωR )).

7
This leads to the possibility of “chiral” formulations of general relativity that
use only these degrees of freedom of the geometry. In particular, one can take
as the action that of equation 3.1, but replacing Ω by ΩR . The equation of
motion coming from varying the connection then sets ωR as the su(2)R com-
ponent of the torsion-free connection. For an extensive discussion, see [26].
The (ωL , ΩL (ωL )) degrees of freedom make no appearance in such chiral for-
mulations. The torsion-free condition picks out a specific ωL , but if one allows
torsion then ωL can be arbitrary.
In the Hamiltonian rather than covariant formalism, this chiral formulation
of gravity is the one studied by Ashtekar and others (for details see for instance
[4] or [5]), giving the starting point for the loop quantum gravity program.
To get from a four-dimensional Lagrangian to a three-dimensional Hamiltonian
formalism one needs to specify a time coordinate and a three-dimensional hy-
persurface M . One can then restrict four-dimensional right-handed spinor fields
to M , and use the time coordinate to treat them as three-dimensional spinor
fields S. The canonical 1-form θ is also restricted to M , giving an R3 -valued
canonical 1-form, where treating tangent vectors as linear maps on spinors, the
R3 ⊂ End(S).
In the Ashtekar variable Hamiltonian formalism the canonical phase space
variables in the Euclidean case are the same as for SU (2) Yang-Mills theory. In
the Yang-Mills case these are interpreted as the gauge-field and electric field,
in the Ashtekar case as the spin-connection and canonical 1-form. In both
cases there is a Gauss-law constraint, coming from the action of SU (2) gauge
transformations. In the Yang-Mills case the Hamiltonian is the sum of the norm-
squares of the electric and magnetic fields. In the Ashtekar case diffeomorphism
invariance gives four extra constraints, three corresponding to translations in
space directions, and the fourth to translation in the time direction. This last
constraint sets the Hamiltonian to zero (at least when there are no boundaries).
The use of a chiral formalism does not by itself resolve the well-known prob-
lems with quantizing gravity. The different treatment we propose in the next
section for the degrees of freedom of the other chirality and for the imaginary
time component of e may provide some new possibilities to examine.

3.2 Weak interactions

In the previous section we have seen that a gravity theory in four dimensions can
be expressed in terms of the R4 -valued canonical 1-form θ and the su(2)R -valued
connection ωR for right-handed spinor fields. The SU (2) part of the Standard
Model can be constructed using instead the su(2)L -valued connection ωL for
left-handed spinor fields and one component (the imaginary time component)
of the canonical 1-form as Higgs field. These can be given exactly the dynamics
of the Standard Model, using the usual Yang-Mills action and the standard
Higgs field kinetic term and potential.
There are two problems with this:

• The electroweak part of the Standard Model has an extra U (1) gauge

8
 symmetry, not just a SU (2)L gauge symmetry.
• As a 1-form, the Higgs field transforms non-trivially under both SU (2)L
and SU (2)R , but we want it to be invariant under SU (2)R . The extra
U (1) gauge symmetry should combine with SU (2)L to give a U (2), with
the Higgs field transforming as the defining C2 representation.

A solution to these problems can be found by formulating the theory in twistor

space, as described in the next section.

4 Euclidean twistor theory and the Standard

Model
For a detailed summary of twistor geometry, see section A. This geometry plays
several different roles simultaneously that allow a new sort of unification of
fundamental interactions:
• It provides a tautological construction of the C2 right-handed spinor de-
gree of freedom, since a point in complexified space-time is precisely the
spinor C2 , realized as a subspace of twistor space T = C4 .

• This complexified space time has Minkowski and Euclidean real slices,
allowing an understanding of how analytic continuation relates Minkowski
and Euclidean spinor fields.
• A point in projective twistor space defines U (1) and SU (3) groups that
can be gauged, giving the rest of the internal symmetries of the standard
model (besides the SU (2)L and SU (2)R of the previous section).
We will take Euclidean signature space-time as fundamental, so will be studying
just the Euclidean twistors of section A.3.3 and this will be the geometric arena
for unification.

4.1 Projective twistors as the bundle of complex struc-

tures
It is well-known that the study of two-dimensional Riemannian geometry ac-
quires very useful new structure if one identifies E2 = C. To do this one just
needs a complex structure: a J ∈ SO(2) such that J 2 = −1, which will play the
role of multiplication by i. There are two such J, of opposite sign. If one tries
to do the same thing in four dimensions, there is an S 2 worth of J ∈ SO(4)
satisfying J 2 = −1, so an S 2 worth of inequivalent different ways of giving E4
a complex structure. One way to motivate Euclidean twistors is to argue that
one should consider all such complex structures simultaneously. The projective

9
twistor space P T of the fiber bundle A.6

S2 P T = CP3
π

S 4 = HP1

is exactly the bundle of all such complex structures for the manifold S 4 . The
fiber above a point x is the S 2 of complex structures on the tangent space at x.
Unlike S 4 , P T = CP3 is a complex manifold and can be studied using complex
analysis. More generally, if M is any four-manifold with anti-self-dual metric,
its bundle of complex structures (generalizing P T ), will be a complex manifold.
The fibers of this fibration can also be thought of as the CP1 of complex
lines C ⊂ C2 = H in the fiber of the tautological H bundle over HP1 . This
C2 is the spinor space SR at the corresponding point, and tangent vectors are
linear maps from SR to SL . A point in P T is a complex line l ⊂ C4 . This line
also lies in C2 = SR and thus gives the point in the fiber CP1 . Given l, we can
identify tangent vectors on S 4 with

HomC (l, SL )

the complex two dimensional space of complex linear maps from l to  SL 

. More
1
explicitly, if one for instance takes as l the complex line generated by then
0
one gets the identification
   
4 1 x0 − ix3
(x0 , x1 , x2 , x3 ) ∈ R ↔ (x0 1 − i(x1 σ1 + x2 σ2 + x3 σ3 )) =
0 −ix1 + x2

Working on P T instead of S 4 , one can resolve the problems noted with

taking SU (2)L to be the weak interaction gauge symmetry:
• A point in P T is a complex line l, and this defines a U (1) bundle over
P T , with fiber the unit vectors in l. This provides the extra U (1) gauge
symmetry needed for an electroweak theory.
• Tangent vectors to S 4 (lifted to P T ) lie in HomC (l, SL ) and are acted on
by the U (1) (through the action on l) and SU (2)L (through the action on
SL ). Lifted to P T , a vector in the imaginary time direction will transform
under U (1) and SU (2)L as a Higgs field.
As usual in the electroweak theory, the Higgs picks out a specific U (1) subgroup
of the U (2) group, the subgroup that leaves the field vector invariant. The gauge
theory of this U (1) group is that of the theory of electromagnetism.
Given a complex structure J acting on R4 , one can extend its action to the
complexification R4 ⊗ C and decompose this space into eigenspaces of J with
eigenvalues +i and −i related by complex conjugation

R4 ⊗ C = C2 ⊕ C2

10
The spinor spaces can then be constructed in terms of antisymmetric tensors as
1
SR ⊕ SL = Λ∗ C2 ⊗ (Λ2 C2 ) 2 (4.1)
with
1
SR = (Λ0 C2 ⊕ Λ2 C2 ) ⊗ (Λ2 C2 ) 2
and
1
SL = Λ1 C2 ⊗ (Λ2 C2 ) 2
The U (2) ⊂ SO(4) subgroup picked out by the choice of J has a double cover
SU (2) × U (1) ⊂ Spin(4), with the SU (2) identified as SU (2)L and the U (1) a
subgroup of SU (2)R . For much more detail about this construction see chapter
31 of [41].
At a point on P T , one can consider not just the lifted tangent vectors from
S 4 , but also the spinor spaces SR and SL , now expressed as in equation 4.1
using the complex structure given by the point. We have seen that there is
a U (1) group determined by the point on P T , and it can be identified with
the U (1) ⊂ SU (2)R that acts on right-handed spinors. As representations of
SU (2)L and this U (1), spinors lifted to P T are
SL = C20 , SR = C−1 ⊕ C+1
where the subscript is the U (1) weight.

4.2 PT and SU (3) gauge symmetry

So far we have just been using aspects of twistor geometry that at a point l ∈ P T
involve the fiber l ⊂ C4 of the tautological line bundle L over P T , as well as
the fibration A.6 to S 4 . Just as in the case of the Grassmanian Gr2,4 (C) of
section A.1, where one could define not just a tautological bundle S, but also
a quotient bundle S ⊥ , over P T one has not just L, but also a quotient bundle
L⊥ . This quotient bundle will have a complex 3-dimensional fiber at p given by
l⊥ = C4 /l. One can think of P T as
U (4) SU (4) SU (4)
PT = = =
U (1) × U (3) S(U (1) × U (3)) U (3)
where the U (1) factor acts as unitary transformations on the fiber l, while
the U (3) acts as unitary transformations on the fiber C4 /l. The SU (3) ⊂
U (3) subgroup provides the color gauge group of the Standard Model, with
fermion fields taking values in l⊥ giving the quarks. For the U (1) to lie in
S(U (1) × U (3)) ⊂ U (4), if it acts with weight 1 on l, it will act with weight − 13
on l⊥ .
We now have, at each point on P T , a principal bundle with fiber the internal
symmetries of the Standard Model, SU (3) × SU (2)L × U (1). A generation of
Standard Model matter fields has exactly the transformation properties under
this group of the space of linear maps from C4 to itself, thought of as
Hom(l ⊕ l⊥ , SR ⊕ SL ) = (l ⊕ l⊥ )∗ ⊗ (SL ⊕ SR )

11
One could write this space as

(C−1 ⊗ C31 ) ⊗ (C20 ⊕ C−1 ⊕ C+1 )

3

which is
C2−1 ⊕ C−2 ⊕ C0 ⊕ (C3 ⊗ C2 ) 31 + C3− 2 + C34
3 3

Here the subscripts are U (1) weights (weak hypercharge), the C2 are the funda-
mental representation of SU (2)L and the C3 are the fundamental representation
of SU (3). For the first generation, the terms above correspond respectively to
the fundamental particles
   
νe u
, e , (ν ) , ,u ,d
e L R e R d L R R

Note that the sort of unification of internal symmetries taking place here is
quite different than the usual one of GUT theories: there is no attempt to fit all
symmetry groups into a single large Lie group. The degrees of freedom of a single
generation are unified in quite different way than in the usual GUT theories (e.g.
SO(10) GUT theories, where a generation fits into a spinor representation).

4.3 Relating theories on P T and space-time

The unification proposal being made here involves certain changes in the nature
of fundamental degrees of freedom with respect to the conventional description
of the Standard Model and general relativity. In particular, the U (1) and SU (3)
principal bundles on which gauge theory connections are to be defined live on
the projective twistor space P T , not Euclidean space-time S 4 . A full definition
of a theory on P T will not be given here and remains to be done, but several
indications of how this might work are as follows.
Considering first free massless matter fields, without coupling to gauge fields
or the Higgs field, note that the single-particle state space H1 for a spinor field
can be identified with the initial data at t = 0 for a solution to the Weyl equation.
This has the disadvantage of obscuring the Poincaré group action on H1 , but
in the twistor formalism one can identify this t = 0 subspace of compactified
Minkowski space (an S 3 ) with an equator in compactified Euclidean space S 4
4
that divides the space into upper (τ > 0) and lower (τ < 0) hemispheres S+ and
4 4
S− . In the coordinates for S of equation A.5, setting τ = x0 = 0 corresponds
to the condition that the real part of the numerator vanish, so

s⊥ ⊥ ⊥ ⊥
1 s1 + s2 s2 + s1 s1 + s2 s2 = 0

Note that (by equation A.3), this is exactly the condition

Φ(s, s) = 0

12
that describes the five-dimensional subspace N = P T0 of P T which contains
the complex lines corresponding to Minkowski space. One has the fibration

CP1 N = P T0 P T = CP3
π

S3 S 4 = HP1

as well as
CP1 P T± P T = CP3
π

4
S± S 4 = HP1
One can study the analytic continuation of solutions of the massless Dirac
4
equation between S+ and compactified Minkowski space (where they are hyper-
fuctions), by using the Euclidean and Minkowski Penrose transforms to relate
both to holomorphic objects on P T . The single-particle state space then will be
given by holomorphic cohomology classes on P T+ , the part of P T that projects
to the upper hemisphere of S 4 . The relevant case of the Penrose transform (see
section A.2) is an identification of the cohomology group H 1 (P T+ , O(−3)) with
4
solutions of the Weyl equation on S+ . To get the value of a solution at a point
from the cohomology class, one restricts the class to the fiber CP1 and uses the
isomorphism
H 1 (CP1 , O(−3)) = C2
For an alternate point of view on this, [14] has a discussion of the relation of
the Dirac operator on a manifold such as S 4 to the Dolbeault operator on the
the projective twistor space.
4
Introducing gauge fields on S+ , for anti-self-dual connections on a vector
4
bundle on S+ , the Penrose-Ward correspondence gives an identification with
a holomorphic bundle F on P T+ , and solutions of the Weyl equation coupled
to that connection are given by cohomology classes H 1 (P T+ , O(F (−3))). This
however only applies to gauge fields satisfying the anti-self-duality equations
rather than the more general Yang-Mills equations. The idea of studying quan-
tum Yang-Mills theory on twistor space has attracted attention over the years,
going back for instance to work of Nair [29] in 1988. This works best in a
formalism based on expanding about the anti-self-dual theory, as studied by
Chalmers-Siegel [12]. In 2003 Witten [38] made major advances in calculating
Yang-Mills amplitudes using twistor space, and this led to an active ongoing
program of studying such amplitudes that exploits twistor space ideas. For
more of the literature relating supersymmetric Yang-Mills theories and quan-
tum field theories on projective twistor space, see for instance [9], [13] and the
review article [1].
In the usual formalism, states for the full theory will be functionals of fields
(spinor fields, gauge fields, vierbeins, Higgs field) on the t = 0 subspace of
Minkowski space-time. In a Euclidean twistor formalism, fields will live on the

13
five-dimensional space P T0 fibered over the τ = 0 S 3 subspace of Euclidean
space-time, with fiber CP1 . One needs to somehow exploit the holomorphic
structure of CP1 to relate fields on P T0 and on S 3 . The usual Penrose-Ward
correspondence relates bundles on space-time and their pull-backs to projective
twistor space P T . The unification proposal here inherently involves U (1) and
SU (3) principal bundles which live on P T and are not pull-backs from space-
time, but vary along the CP1 fiber.

5 Open problems and speculative remarks

The unification proposal discussed here is still missing some crucial aspects.
Most critically, it is unclear what the origin of generations might be. This issue
is crucial for any hope of understanding where fermion masses and mixing angles
come from. It is possible that the fundamental theory involves not just the
usual twistor geometry of P T , but should be formulated on the seven-sphere
S 7 , which is a circle bundle over P T . S 7 is a remarkably unusual geometric
structure, exhibiting a wide range of different symmetry groups, since one has

S 7 = Spin(8)/Spin(7) = Spin(7)/G2 = Spin(6)/SU (3) = Spin(5)/Sp(1)

as well as algebraic structures arising from identifying S 7 with the unit octo-
nions. Our discussion has exploited the last two geometries on S 7 , not the
first two. The chiral formulation of general relativity may have an interesting
formulation using the geometry of S 7 , see [26] and [20].
The construction of a generation of fermion fields using four-by-four com-
plex matrices with the geometrical interpretation of Hom(l ⊕ l⊥ , SR ⊕ SL ) may
perhaps be understood as a lift from S 4 of the Clifford algebra bundle on S 4 ,
which also is a bundle of four-by-four complex matrices. It is well-known that
if one attempts to formulate a lattice gauge theory version of fermions in four
Euclidean dimensions, one has problems avoiding multiple copies of spinor repre-
sentations. The simplest constuction uses Kogut-Susskind fermions, which can
be thought of as taking values in the exterior algebra of differential form (this
is the associated graded algebra for the Clifford algebra). In the usual Kogut-
Susskind formalism one is getting multiple copies of the space-time spinors of
both chiralities. The lift of this to P T here would be something different, cor-
responding to 16 copies of a Weyl spinor, with the Weyl spin degree of freedom
coming from the fiber CP 1 at a point, not the Clifford or exterior algebra.
For another point of view on spaces of solutions of massless field equations
and the geometry of P T , note that this is an example of the geometric quanti-
zation construction of representations of the group SU (2, 2). P T+ is an orbit of
SU (2, 2) acting on P T , providing an important example of the case of “minimal”
orbits, for which geometric quantization runs into difficult technical problems
due to the lack of an appropriate invariant polarization. For a 1982 history of
work on this specific case, see appendix A of [33]. It may be that the analytic
continuation to the Euclidean space-time perspective will give new insight into
these problems.

14
 The anti-self-duality equations can be formulated as the vanishing of a mo-
ment map, for an old speculative discussion of the significance of the path in-
tegral over gauge fields in this context see [40]. N = 2 and N = 4 super
Yang-Mills give topological quantum field theories (see [37]), with the feature
that a “twisting” of the space-time symmetry into an internal symmetry plays a
crucial role. One might speculate that these theories have some relation to the
hypothetical theory on P T described here. Twisted N = 4 super Yang-Mills
theories have remarkable properties, including providing a quantum field theory
version of geometric Langlands [23], with the geometric Langlands program in
recent years having given evidence for a dramatic unified perspective relating
number theory, geometry and representation theory (and, potentially, physics).

6 Conclusions
The main conclusion of this work is that twistor geometry provides a compelling
picture of fundamental physics, integrating internal and space-time symmetries,
as long as one treats together its Euclidean and Minkowski aspects, related
through the projective twistor space PT. The Euclidean aspect is crucial for
understanding the origin of the Standard Model internal symmetries and the
breaking of electroweak symmetry, which is inherent in the Euclidean space-time
definition of physical states.
This picture has many attractive aspects:

• Spinors are tautological objects (a point in space-time is a space of Weyl

spinors), rather than complicated objects that must be separately intro-
duced in the usual geometrical formalism.
• Analytic continuation between Minkowski and Euclidean space-time can
be naturally performed, since twistor geometry provides their joint com-
plexification.
• Exactly the internal symmetries of the Standard Model occur.
• Electroweak symmetry breaking has a novel origin in the breaking of Eu-
clidean SO(4) symmetry necessary to define a physical theory.

• The intricate transformation properties of a generation of Standard Model

fermions correspond to a simple construction.
• One gets a new chiral formulation of gravity, unified with the Standard
Model.

• Conformal symmetry is built into the picture in a fundamental way.

Much work remains to be done to explicitly construct and understand a full
theory defined on P T that would correspond to the Standard Model and general
relativity, with the expected three generations of matter fields. Such a theory
might allow understanding of currently unexplained features of the Standard

15
Model, as well as possibly making testable predictions that differ from those of
the Standard Model. In particular, the framework proposed is fundamentally
chiral as a theory of gravity, not just in the electroweak sector, and this may
have observable implications.

Appendices
A Twistor geometry
Twistor geometry is a 1967 proposal [32] due to Roger Penrose for a very dif-
ferent way of formulating four-dimensional space-time geometry. For a detailed
expository treatment of the subject, see [35] (for a version aimed at physicists
and applications in amplitude calculations, see [2]). Fundamental to twistor ge-
ometry is the twistor space T = C4 , as well as its projective version, the space
P T = CP3 of complex lines in T .

A.1 Compactified and complexified space-time

The relation of twistor space to conventional space-time is that complexified
and compactified space-time is identified with the Grassmanian M = G2,4 (C)
of complex two-dimensional linear subspaces in T . A space-time point is thus
a C2 in C4 which tautologically provides the spinor degree of freedom at that
point. The spinor bundle S is the tautological two-dimensional complex vector
bundle over M whose fiber Sm at a point m ∈ M is the C2 that defines the
point.
The group SL(4, C) acts on T and transitively on the spaces P T and M
of its complex subspaces. Points in the Grassmanian M can be represented as
elements
ω = (v1 ⊗ v2 − v2 ⊗ v1 ) ∈ Λ2 (C4 )
by taking two vectors v1 , v2 spanning the subspace. Λ2 (C4 ) is six conplex di-
mensional and scalar multiples of ω gives the same point in M , so ω identifies
M with a subspace of P (Λ2 (C4 )) = CP5 . Such ω satisfy the equation

ω∧ω =0 (A.1)

which identifies (the “Klein correspondence”) M with a submanifold of CP5

given by a non-degenerate quadratic form. Twistors are spinors in six dimen-
sions, with the action of SL(4, C) on Λ2 (C4 ) = C6 preserving the quadratic
form A.1, and giving the spin double cover homomorphism

SL(4, C) = Spin(6, C) → SO(6, C)

To get the tangent bundle of M , one needs not just the spinor bundle S,
but also another two complex-dimensional vector bundle, the quotient bundle

16
S ⊥ with fiber Sm
⊥
= C4 /Sm . Then the tangent bundle is

T M = Hom(S, S ⊥ ) = S ∗ ⊗ S ⊥

with the tangent space Tm M a four complex dimensional vector space given by
⊥ ⊥
Hom(Sm , Sm ), the linear maps from Sm to Sm .
A choice of coordinate chart on M is given by picking a point m ∈ M and
⊥
identifying Sm with a complex two plane transverse to Sm . The point m will be
⊥
the origin of our coordinate system, so we will denote Sm by S0 and Sm by S0⊥ .
⊥
Now T = S0 ⊕ S0 and one can choose basis elements e1 , e2 ∈ S0 , e3 , e4 ∈ S0⊥
for T . The coordinate of the two-plane spanned by the columns of
 
1 0
 0 1 
 
z01 z01 
z10 z11

will be the 2 by 2 complex matrix

 
z z01
Z = 01
z10 z11

This coordinate chart does not include all of M , since it misses those points
in M corresponding to complex two-planes that are not transverse to S0⊥ . Our
interest however will ultimately be not in the global structure of M , but in its
local structure near the chosen point m, which we will study using the 2 by
2 complex matrix Z as coordinates. When we discuss M we will sometimes
not distinguish between M and its local version as a complex four-dimensional
vector space with origin of coordinates at m.
Writing elements of T as  
s1
 s2 
 ⊥
s1 
s⊥2

an element of T will be in the complex two plane with coordinate Z when

 ⊥  
s1 s1
= Z (A.2)
s⊥
2 s2

This incidence equation characterizes in coordinates the relation between lines

(elements of P T ) and planes (elements of M ) in twistor space T . We’ll some-
times also write this as
s⊥ = Zs
An SL(4, C) determinant 1 matrix
 
A B
C D

17
acts on T by
As + Bs⊥
   
s
→
s⊥ Cs + Ds⊥
On lines in the plane Z this is
     
s As + BZs (A + BZ)s
→ =
Zs Cs + DZs (C + DZ)(A + BZ)−1 (A + BZ)s

so the corresponding action on M will be given by

Z → (C + DZ)(A + BZ)−1

Since Λ2 (S0 ) = Λ2 (S0⊥ ) = C, S0 and S0⊥ have (up to scalars) unique choices
S0 and S0⊥ of non-degenerate antisymmetric bilinear forms, and corresponding
choices of SL(2, C) ⊂ GL(2, C) acting on S0 and S0⊥ . These give (again, up to
scalars), a unique choice of a non-degenerate symmetric form on Hom(S0 , S0⊥ ),
such that
hZ, Zi = det Z
The subgroup

Spin(4, C) = SL(2, C) × SL(2, C) ⊂ SL(4, C)

of matrices of the form  

A 0
0 D
with
det A = det D = 1
acts on M in coordinates by

Z → DZA−1

preserving hZ, Zi.

Besides the spaces P T and M of complex lines and planes in T , it is also
useful to consider the correspondence space whose elements are complex lines
inside a complex plane in T . This space can also be thought of as P (S), the
projective spinor bundle over M . There is a diagram of maps

P (S)
µ ν

PT M

where ν is the projection map for the bundle P (S) and µ is the identification
of a complex line in S as a complex line in T . µ and ν give a correspondence
between geometric objects in P T and M . One can easily see that µ(ν −1 (m)) is
the complex projective line in P T corresponding to a point m ∈ M (a complex

18
two plane in T is a complex projective line in P T ). In the other direction,
ν(µ−1 ) takes a point p in P T to α(p), a copy of CP 2 in M , called the “α-plane”
corresponding to p.
In our chosen coordinate chart, this diagram of maps is given by

(Z, s) ∈ P (S)
µ
ν
 
s
∈ PT Z∈M
Zs

The incidence equation A.2 relating P T and M implies that an α-plane is a null
plane in the metric discussed above. Given two points Z1 , Z2 in M corresponding
to the same point in P T , their difference satisfies

s⊥ = (Z1 − Z2 )s = 0

Z1 − Z2 is not an invertible matrix, so has determinant 0 and is a null vector.

A.2 The Penrose-Ward transform

The Penrose transform relates solutions of conformally-invariant wave equations
on M to sheaf cohomology groups, identifying
• Solutions to a helicity k
2 holomorphic massless wave equation on U .
• The sheaf cohomology group

H 1 (U
b , O(−k − 2))

Here U ⊂ M and U
b ⊂ P T are open sets related by the twistor correspondence,
i.e.
b = µ(ν −1 (U ))
U
We will be interested in cases where U and U b are orbits in M and P T for a
real form of SL(4, C). Here O(−k − 2) is the sheaf of holomorphic sections of
the line bundle L⊗(−k−2) where L is the tautological line bundle over P T . For
a detailed discussion, see for instance chapter 7 of [35].
The Penrose-Ward transform is a generalization of the above, introducing
a coupling to gauge fields. One aspect of this is the Ward correspondence, an
isomorphism between
• Holomorphic anti-self-dual GL(n, C) connections A on U ⊂ M .

• Holomorphic rank n vector bundles E over U

b ⊂ PT.

19
Here “anti-self-dual” means the curvature of the connection satisfies
∗FA = −FA
where ∗ is the Hodge dual. There are some restrictions on the open set U , and
E needs to be trivial on the complex projective lines corresponding to points
m ∈ U.
In one direction, the above isomorphism is due to the fact that the curva-
ture FA is anti-self-dual exactly when the connection A is integrable on the
intersection of an α-plane with U . One can then construct the fiber Ep of E
at p as the covariantly constant sections of the bundle with connection on the
corresponding α-plane in M . In the other direction, one can construct a vector
bundle E e on U by taking as fiber at m ∈ U the holomorphic sections of E
on the corresponding complex projective line in P T . Parallel transport in this
vector bundle can be defined using the fact that two points m1 , m2 in U on the
same α-plane correspond to intersecting projective lines in P T . For details, see
chapter 8 of [35] and chapter 10 of [28].
Given an anti-self-dual gauge field as above, the Penrose transform can be
generalized to a Penrose-Ward transform, relating
• Solutions to a helicity k holomorphic massless wave equation on U , coupled
to a vector bundle E e with anti-self-dual connection A.

• The sheaf cohomology group

H 1 (U
b , O(E)(−k − 2))

For more about this generalization, see [16].

A.3 Twistor geometry and real forms

So far we have only considered complex twistor geometry, in which the relation
to space-time geometry is that M is a complexified version of a four real dimen-
sional space-time. From the point of view of group symmetry, the Lie algebra
of SL(4, C) is the complexification
sl(4, C) = g ⊗ C
for several different real Lie algebras g, which are the real forms of sl(4, C).
To organize the possibilities, recall that SL(4, C) is Spin(6, C), the spin group
for orthogonal linear transformations in six complex dimensions, so sl(4, C) =
so(6, C). If one instead considers orthogonal linear transformations in six real
dimensions, there are different possible signatures of the inner product to con-
sider, all of which become equivalent after complexification. This corresponds
to the possible real forms
g = so(3, 3), so(4, 2), so(5, 1), and so(6)
which we will discuss (there’s another real form, su(3, 1), which we won’t con-
sider). For more about real methods in twistor theory, see [42].

20
A.3.1 Spin(3, 3) = SL(4, R)
The simplest way to get a real version of twistor geometry is to take the discus-
sion of section A and replace complex numbers by real numbers. Equivalently,
one can look at subspaces invariant under the usual conjugation, given by the
map σ
s1
   
s1
 s2   s2 
σ
s⊥
 =  ⊥
 s1 
1
s⊥2 s⊥
2

which acts not just on T but on P T and M . The fixed point set of the action
on M is M 2,2 = G2,4 (R), the Grassmanian of real two-planes in R4 . As a
manifold, G2,4 (R) is S 2 × S 2 , quotiented by a Z2 . M 2,2 is acted on by the
group Spin(3, 3) = SL(4, R) of conformal transformations. σ acting on P T
acts on the CP 1 corresponding to a point in M 2,2 with an action whose fixed
points form an equatorial circle.
Coordinates can be chosen as in the complex case, but with everything real.
A point in M 2,2 is given by a real 2 by 2 matrix, which can be written in the
form  
x0 + x3 x1 − x2
Z=
x1 + x2 x0 − x3
for real numbers x0 , x1 , x2 , x3 . M 2,2 is acted on by the group Spin(3, 3) =
SL(4, R) of conformal transformations as in the complex case by

Z → (C + DZ)(A + BZ)−1

with the subgroup of rotations

Z → DZA−1

for A, D ∈ SL(2, R) given by

Spin(2, 2) = SL(2, R) × SL(2, R)

This subgroup preserves

hZ, Zi = det Z = x20 − x23 − x21 + x22

For the Penrose transform in this case, see Atiyah’s account in section 6.5
of [7]. For the Ward correspondence, see section 10.5 of [28].

A.3.2 Spin(4, 2) = SU (2, 2)

The real case of twistor geometry most often studied (a good reference is [35]) is
that where the real space-time is the physical Minkowski space of special relativ-
ity. The conformal compactification of Minkowski space is a real submanifold of
M , denoted here by M 3,1 . It is acted upon transitively by the conformal group

21
Spin(4, 2) = SU (2, 2). This conformal group action on M 3,1 is most naturally
understood using twistor space, as the action on complex planes in T coming
from the action of the real form SU (2, 2) ⊂ SL(4, C) on T .
SU (2, 2) is the subgroup of SL(4, C) preserving a real Hermitian form Φ of
signature (2, 2) on T = C4 . In our coordinates for T , a standard choice for Φ is
given by
   0     0 
s s 0 1 s
Φ , = s s ⊥ = s† (s⊥ )0 + (s⊥ )† s0 (A.3)
s⊥ (s⊥ )0 1 0 (s⊥ )0

Minkowski space is given by complex planes on which Φ = 0, so

   
s s
Φ , = s† (Z + Z † )s = 0
Zs Zs

Thus coordinates of points on Minkowski space are anti-Hermitian matrices Z,

which can be written in the form
 
x0 + x3 x1 − ix2
Z = −i = −i(x0 1 + x · σ)
x1 + ix2 x0 − x3

where σj are the Pauli matrices. The metric is the usual Minkowski metric,
since
hZ, Zi = det Z = −x20 + x21 + x22 + x23
One can identify compactified Minkowski space M 3,1 as a manifold with the Lie
group U (2) which is diffeomorphic to (S 3 × S 1 )/Z2 . The identification of the
tangent space with anti-Hermitian matrices reflects the usual identification of
the tangent space of U (2) at the identity with the Lie algebra of anti-Hermitian
matrices.
SL(4, C) matrices are in SU (2, 2) when they satisfy
 †
A C†
    
0 1 A B 0 1
=
B † D† 1 0 C D 1 0

The Poincaré subgroup P of SU (2, 2) is given by elements of SU (2, 2) of the

form  
A 0
C (A† )−1
where A ∈ SL(2, C) and A† C = −C † A. These act on Minkowski space by

Z → (C + (A† )−1 Z)A−1 = (A† )−1 ZA−1 + CA−1

One can show that CA−1 is anti-Hermitian and gives arbitrary translations on
Minkowski space. The Lorentz subroup is Spin(3, 1) = SL(2, C) acting by

Z → (A† )−1 ZA−1

Here SL(2, C) is acting by the standard representation on S0 , and by the

conjugate-dual representation on S0⊥ .

22
 Note that, for the action of the Lorentz SL(2, C) subgroup, twistors written
as elements of S0 ⊕S0⊥ behave like usual Dirac spinors (direct sums of a standard
SL(2, C) spinor and one in the conjugate-dual representation), with the usual
Dirac adjoint, in which the SL(2, C)-invariant inner product is given by the
signature (2, 2) Hermitian form

hψ1 , ψ2 i = ψ1† γ0 ψ2

Twistors, with their SU (2, 2) conformal group action and incidence relation to
space-time points, are however something different than Dirac spinors.
The SU (2, 2) action on M has six orbits: M++ , M−− , M+0 , M−0 , M00 , where
the subscript indicates the signature of Φ restricted to planes corresponding
to points in the orbit. The last of these is a closed orbit M 3,1 , compactified
Minkowski space. Acting on projective twistor space P T , there are three orbits:
P T+ , P T− , P T0 , where the subscript indicates the sign of Φ restricted to the
line in T corresponding to a point in the orbit. The first two are open orbits
with six real dimensions, the last a closed orbit with five real dimensions. The
points in compactified Minkowski space M00 = M 3,1 correspond to projective
lines in P T that lie in the five dimensional space P T0 . Points in M++ and M−−
correspond to projective lines in P T+ or P T− respectively.
One can construct infinite dimensional irreducible unitary representations of
SU (2, 2) using holomorphic geometry on P T+ or M++ , with the Penrose trans-
form relating the two constructions [15]. For P T+ the closure of the orbit P T+ ,
the Penrose transform identifies the sheaf cohomology groups H 1 (P T+ , O(−k −
2)) for k > 0 with holomorphic solutions to the helicity k2 wave equation on
M++ . Taking boundary values on M 3,1 , these will be real-analytic solutions to
the helicity k2 wave equation on compactified Minkowski space. If one instead
considers the sheaf cohomology H 1 (P T+ , O(−k − 2)) for the open orbit P T+
and takes boundary values on M 3,1 of solutions on M++ , the solutions will be
hyperfunctions, see [36].
The Ward correspondence relates holomorphic vector bundles on P T+ with
anti-self-dual GL(n, C) gauge fields on M++ . However, in this Minkowski sig-
nature case, all solutions to the anti-self-duality equations as boundary values
of such gauge fields are complex, so one does not get anti-self-dual gauge fields
for compact gauge groups like SU (n).

A.3.3 Spin(5, 1) = SL(2, H)

Changing from Minkowski space-time signature (3, 1) to Euclidean space-time
signature (4, 0), the compactified space-time M 4 = S 4 is again a real subman-
ifold of M . To understand the conformal group and how twistors work in this
case, it is best to work with quaternions instead of complex numbers, iden-
tifying T = H2 . When working with quaternions, one can often instead use
corresponding complex 2 by 2 matrices, with a standard choice

q = q0 + q1 i + q2 j + q3 k ↔ q0 − i(q1 σ1 + q2 σ2 + q3 σ3 )

23
For more details of the quaternionic geometry that appears here, see [7] or [34]
The relevant conformal group acting on S 4 is Spin(5, 1) = SL(2, H), again
best understood in terms of twistors and the linear action of SL(2, H) on T =
H2 . The group SL(2, H) is the group of quaternionic 2 by 2 matrices satisfying a
single condition that one can think of as setting the determinant to one, although
the usual determinant does not make sense in the quaternionic case. Here one
can interpret the determinant using the isomorphism with complex matrices,
or, at the Lie algebra level, sl(2, H) is the Lie algebra of 2 by 2 quaternionic
matrices with purely imaginary trace.
While one can continue to think of points in S 4 ⊂ M as complex two planes,
one can also identify these complex two planes as quaternionic lines and S 4 as
HP1 , the projective space of quaternionic lines in H2 . The conventional choice
of identification between C2 and H is
 
s
s = 1 ↔ s = s1 + s2 j
s2

One can then think of the quaternionic structure as providing an alternate

notion of conjugation than the usual one, given instead by left multiplying by
j ∈ H. Using jzj = −z one can show that

−s2
   
s1
 s2   s1 
σ
s⊥
 =  ⊥
 −s2  (A.4)
1
⊥
s2 s⊥
1

σ satisfies σ 2 = −1 on T , so σ 2 = 1 on P T . We will see later that while σ has

no fixed points on P T , it does fix complex projective lines.
The same coordinates used in the complex case can be used here, where now
S0⊥ is a quaternionic line transverse to S0 , so coordinates on T are the pair of
quaternions  
s
s⊥
These are also homogeneous coordinates for points on S 4 = HP1 and our choice
of Z ∈ H given by  
s
Zs
as the coordinate in a coordinate system with origin the point with homogeneous
coordinates  
1
0
The point at ∞ will be the one with homogeneous coordinates
 
0
1

24
This is the quaternionic version of the usual sort of choice of coordinates in the
case of S 2 = CP 1 , replacing complex numbers by quaternions. The coordinate
of a point on S 4 with homogeneous coordinates
 
s
s⊥
will be
(s⊥ ⊥
1 + s2 j)(s1 − s1 j) s⊥ s1 + s⊥ ⊥ ⊥
2 s2 + (−s1 s2 + s2 s1 )j
s⊥ s−1 = 2 2
= 1 2 2
(A.5)
|s1 | + |s2 | |s1 | + |s2 |
A coordinate of a point will now be a quaternion Z = x0 + x1 i + x2 j + x3 k
corresponding to the 2 by 2 complex matrix
 
x0 − ix3 −ix1 − x2
Z = x0 1 − ix · σ =
−ix1 + x2 x0 + ix3
The metric is the usual Euclidean metric, since
hZ, Zi = det Z = x20 + x21 + x22 + x23
The conformal group SL(2, H) acts on T = H2 by the matrix
 
A B
C D
where A, B, C, D are now quaternions, satisfying together the determinant 1
condition. These act on the coordinate Z as in the complex case, by
Z → (C + DZ)(A + BZ)−1
The Euclidean group in four dimensions will be the subgroup of elements of the
form  
A 0
C D
such that A and D are independent unit quaternions, thus in the group Sp(1) =
SU (2), and C is an arbitrary quaternion. The Euclidean group acts by
Z → DZA−1 + CA−1
with the spin double cover of the rotational subgroup now Spin(4) = Sp(1) ×
Sp(1). Note that spinors behave quite differently than in Minkowski space:
there are independent unitary SU (2) actions on S0 and S0⊥ rather than a non-
unitary SL(2, C) action on S0 that acts at the same time on S0⊥ by the conjugate
transpose representation.
The projective twistor space P T is fibered over S 4 by complex projective
lines
CP 1 P T = CP 3
π (A.6)

S 4 = HP 1

25
The projection map π is just the map that takes a complex line in T identified
with H2 to the corresponding quaternionic line it generates (multiplying ele-
ments by arbitrary quaternions). In this case the conjugation map σ of A.4 has
no fixed points on P T , but does fix the complex projective line fibers and thus
the points in S 4 ⊂ M . The action of σ on a fiber takes a point on the sphere to
the opposite point, so has no fixed points.
Note that the Euclidean case of twistor geometry is quite different and much
simpler than the Minkowski one. The correspondence space P (S) (here the
complex lines in the quaternionic line specifying a point in M 4 = S 4 ) is just P T
itself, and the twistor correspondence between P T and S 4 is just the projection
π. Unlike the Minkowski case where the real form SU (2, 2) has a non-trivial
orbit structure when acting on P T , in the Euclidean case the action of the real
form SL(2, H) is transitive on P T .
In the Euclidean case, the projective twistor space has another interpreta-
tion, as the bundle of orientation preserving orthogonal complex structures on
S 4 . A complex structure on a real vector space V is a linear map J such that
J 2 = −1, providing a way to give V the structure of a complex vector space
(multiplication by i is multiplication by J). J is orthogonal if it preserves an
inner product on V . While on R2 there is just one orientation-preserving or-
thogonal complex structure, on R4 the possibilities can be parametrized by a
sphere S 2 . The fiber S 2 = CP 1 of A.6 above a point on S 4 can be interpreted
as the space of orientation preserving orthogonal complex structures on the four
real dimensional tangent space to S 4 at that point.
One way of exhibiting these complex structures on R4 is to identify R4 = H
and then note that, for any real numbers x1 , x2 , x3 such that x21 + x22 + x23 = 1,
one gets an orthogonal complex structure on R4 by taking

J = x1 i + x2 j + x3 k

Another way to see this is to note that the rotation group SO(4) acts on orthogo-
nal complex structures, with a U (2) subgroup preserving the complex structure,
so the space of these is SO(4)/U (2), which can be identified with S 2 .
More explicitly, in our choice of coordinates, the projection map is
   
s x0 − ix3 −ix1 − x2
π: ⊥ →Z=
s = Zs −ix1 + x2 x0 + ix3

For any choice of s in the fiber above Z, s⊥ associates to the

four real coordinates
specifying Z an element of C2 . For instance, if s = 1, 0 , the identification of
R4 with C2 is  
x0  
  ↔ x0 − ix3
x1 
x2  −ix1 + x2
x3
The complex structure on R4 one gets is not changed if s gets multiplied by a
complex scalar, so it just depends on the point [s] in the CP 1 fiber.

26
 For another point of view on this, one can see that for each point p ∈ P T ,
the corresponding α-plane ν(µ−1 (p)) in M intersects its conjugate σ(ν(µ−1 (p)))
in exactly one real point, π(p) ∈ M 4 . The corresponding line in P T is the line
determined by the two points p and σ(p). At the same time, this α-plane
provides an identification of the tangent space to M 4 at π(p) with a complex
two plane, the α-plane itself. The CP 1 of α -planes corresponding to a point in
S 4 are the different possible ways of identifying the tangent space at that point
with a complex vector space. The situation in the Minkowski space case is quite
different: there if CP 1 ⊂ P T0 corresponds to a point Z ∈ M 3,1 , each point p
in that CP 1 gives an α-plane intersecting M 3,1 in a null line, and the CP 1 can
be identified with the “celestial sphere” of null lines through Z.
In the Euclidean case , the Penrose transform will identify the sheaf cohomol-
ogy group H 1 (π −1 (U ), O(−k − 2)) for k > 0 with solutions of helicity k2 linear
field equations on an open set U ⊂ S 4 . Unlike in the Minkowski space case, in
Euclidean space there are U (n) bundles E e with connections having non-trivial
anti-self-dual curvature. The Ward correspondence between such connections
and holomorphic bundles E on P T for U = S 4 has been the object of inten-
sive study, see for example Atiyah’s survey [7]. The Penrose-Ward transform
identifies
• Solutions to a field equation on U for sections Γ(S k ⊗ E),
e with covariant
derivative given by an anti-self-dual connection A, where S k is the k’th
symmetric power of the spinor bundle.
• The sheaf cohomology group

H 1 (U
b , O(E)(−k − 2))

b = π −1 U .
where U
For the details of the Penrose-Ward transform in this case, see [21].

A.3.4 Spin(6) = SU (4)

If one picks a positive definite Hermitian inner product on T , this determines a
subgroup SU (4) = Spin(6) that acts on T , and thus on P T, M and P (S). One
has
SU (4) SU (4) SU (4)
PT = , M= , P (S) =
U (3) S(U (2) × U (2)) S(U (1) × U (2))
and the SU (4) action is transitive on these three spaces. There is no four real
dimensional orbit in M that could be interpreted as a real space-time that would
give M after complexification.
In this case the Borel-Weil-Bott theorem relates sheaf-cohomology groups
of equivariant holomorphic vector-bundles on P T, M and P (S), giving them
explicitly as certain finite dimensional irreducible representations of SU (4). For
more details of the relation between the Penrose transform and Borel-Weil-
Bott, see [8]. The Borel-Weil-Bott theorem [10] can be recast in terms of index

27
theory, replacing the use of sheaf-cohomology with the Dirac equation [11]. For
a more general discussion of the relation of representation theory and the Dirac
operator, see [22].

B Euclidean quantum fields

But besides this, by freeing
ourselves from the limitation of
the Lorentz group, which has
produced all the well-known
difficulties of quantum field
theory, one has here a possibility
— if this is indeed necessary — of
producing new theories. That is,
one has the possibility of
constructing new theories in the
Euclidean space and then
translating them back into the
Lorentz system to see what they
imply.

J. Schwinger, 1958[47]

Should the Feynman path

integral be well-defined only in
Euclidean space, as axiomaticians
would have it, then there seems
to exist a very real problem when
dealing with Weyl fields as in the
theory of weak interactions or in
its unification with QCD.

P. Ramond, 1981[60]

A certain sense of mystery

surrounds Euclidean fermions.

A. Jaffe and G. Ritter, 2008[75]

That one chirality of Euclidean space-time rotations appears after analytic

continuation to Minkowski space-time as an internal symmetry is the most hard
to believe aspect of the proposed framework for a unified theory outlined in
this paper. One reason for the very long time that has passed since an earlier
embryonic version of this idea (see [39]) is that the author has always found
this hard to believe himself. While the fact that the quantization of Euclidean

28
spinor fields is not straightforward is well-known, Schwinger’s early hope that
this might have important physical significance (see above) does not appear to
have attracted much attention. In this section we’ll outline the basic issue with
Euclidean spinor fields, and argue that common assumptions about analytic
continuation of the space-time symmetry do not hold in this case. This issue
becomes apparent in the simplest possible context of free field theory. There are
also well-known problems when one attempts to construct a non-perturbative
lattice-regularized theory of chiral spinors coupled to gauge fields.
Since Schwinger’s first proposal in 1958[46], over the years it has become
increasingly clear that the quantum field theories governing our best under-
standing of fundamental physics have a much simpler behavior if one takes time
to be a complex variable, and considers the analytic continuation of the theory
to imaginary values of the time parameter. In imaginary time the invariant no-
tion of distance between different points becomes positive, path integrals often
become well-defined rather than formal integrals, field operators commute, and
expectation values of field operators are conventional functions rather than the
boundary values of holomorphic functions found at real time.
While momentum eigenvalues can be arbitrarily positive or negative, energy
eigenvalues go in one direction only, which by convention is that of positive ener-
gies. Having states supported only at non-negative energies implies (by Fourier
transformation) that, as a function of complex time, states can be analytically
continued in one complex half plane, not the other. A quantum theory in Eu-
clidean space has a fundamental asymmetry in the direction of imaginary time,
corresponding to the fundamental asymmetry in energy eigenvalues.
Quantum field theories can be characterized by their n-point Wightman
(Minkowski space-time) or Schwinger (Euclidean space-time) functions, with
the Wightman functions not actual functions, but boundary values of analytic
continuations of the Schwinger functions. For free field theories these are all
determined by the 2-point functions W2 or S2 . The Wightman function W2 is
Poincaré-covariant, while the Schwinger function S2 is Euclidean-covariant.
This simple relation between the Minkowski and Euclidean space-time free
field theories masks a much more subtle relationship at the level of fields, states
and group actions on these. In both cases one can construct fields and a
Fock space built out of a single-particle state space carrying a representation
of the space-time symmetry group. For the Minkowski theory, fields are non-
commuting operators obeying an equation of motion and the single-particle state
space is an irreducible unitary representation of the Poincaré group.
The Euclidean theory is quite different. Euclidean fields commute and do
not obey an equation of motion. The Euclidean single-particle state space is
a unitary representation of the Euclidean group, but far from irreducible. It
describes not physical states, but instead all possible trajectories in the space
of physical states (parametrized by imaginary time). The Euclidean state space
and the Euclidean fields are not in any sense analytic continuations of the cor-
responding Minkowski space constructions. For a general theory encompassing
the relation between the Euclidean group and Poincaré group representations,
see [64].

29
 One can recover the physical Minkowski theory from the Euclidean theory,
but to do so one must break the Euclidean symmetry by choosing an imaginary
time direction. In the following sections we will outline the relation between the
Minkowski and Euclidean theories for the cases of the harmonic oscillator, the
free scalar field theory, and the free chiral spinor field theory.

B.1 The harmonic oscillator

The two-point Schwinger function for the one-dimensional quantum harmonic
oscillator of frequency ω (ω > 0) is
1
S2 (τ ) = e−ω|τ |
(2π)(2ω)

with Fourier transform

Z ∞
f2 (s) = √1
S eisτ S2 (τ )dτ =
1 1
2π −∞ (2π)3/2 s2 + ω 2

In the complex z = t + iτ plane, S2 can be analytically continued to the upper

half plane as
1
eiωz
(2π)(2ω)
and to the lower half plane as
1
e−iωz
(2π)(2ω)

The Wightman functions are the analytic continuations to the t (real z) axis,
so come in two varieties:
1
W2− (t) = lim eiω(t+i)
→0+ (2π)(2ω)

and
1
W2+ (t) = lim+ e−iω(t−i)
→0 (2π)(2ω)
The conventional interpretation of W2± is not as functions, but as distribu-
tions, given as the boundary values of holomorphic functions. Alternatively (see
appendix C), one can interpret W2± (t) as the lower and upper half-plane holo-
morphic functions defining a hyperfunction. Like distributions, hyperfunctions
can be thought of a elements of a dual space to a space of well-behaved test
functions, in this case a space of real analytic functions. The Fourier transform
of W2 is then the hyperfunction
i 1
W
f2 (E) =
(2π)3/2 ω 2 − E 2

30
which is a sum of terms W f ± (E) supported at ω > 0 and −ω < 0. Note that
2
−iEt
the convention that e has positive energy means that Fourier transforms of
positive energy functions are holomorphic for τ < 0.
The physical state space of the harmonic oscillator is determined by the
single-particle state space H1 = C. H1 is the state space for a single quantum,
it can be thought of as the space of positive energy solutions to the equation of
motion
d2
( 2 + ω 2 )φ = 0 (B.1)
dt
H1 can also be constructed using W2+ , by defining
Z ∞Z ∞
(f, g) = f (t2 )W2+ (t2 − t1 )g(t1 )dt1 dt2
−∞ −∞
∞ ∞
e−iω(t2 −t1 )
Z Z
= f (t2 ) g(t1 )dt1 dt2
−∞ −∞ (2π)(2ω)
1
= fe(ω)e
g (ω)
2ω
for f, g functions in S(R) and taking the space of equivalence classes

H1 = [f ] ∈ {f ∈ S(R)}/{(f, f ) = 0}

One can identify such equivalence classes as

1
[f ] = √ fe(ω)
2ω
H1 is C with standard Hermitian inner product
1 ˜
h[f ], [g]i = f (ω)g̃(ω)
2ω
Note that it doesn’t matter whether one takes real or complex valued functions
f , in either case one gets the same quotient complex vector space H1 .
Given H1 and the inner product h·, ·i, the full state space H is an inner
product space given by the Fock space construction, with
∞
M
H = S ∗ (H1 ) = S k (H1 )
k=0

In this case the symmetrized tensor product S k (H1 ) of k copies of H1 = C is

just again C, the states with k-quanta. A creation operator a† (f ) (for f real)
acts by symmetrized tensor product with [f ] and a(f ) is the adjoint operator.
One can define an operator
b ) = a(f ) + a† (f )
φ(f

and then
h0|φ(f
b )φ(g)|0i
b = h[f ], [g]i

31
φ(t)
b should be interpreted as an operator-valued distribution, writing
Z ∞
φ(f
b )= φ(t)f
b (t)dt
−∞

φb satisfies the equation of motion B.1.

One can use the Schwinger function S2 to set up a Euclidean (imaginary
time τ ) Fock space, taking E1 to be the space of real-valued functions in S(R)
with inner product
Z ∞Z ∞
(f, g)E1 = f (τ2 )S2 (τ2 − τ1 )g(τ1 )dτ1 dτ2
−∞ −∞
∞ ∞
e−ω|τ2 −τ1 |
Z Z
= f (τ2 ) g(τ1 )dτ1 dτ2
−∞ −∞ (2π)(2ω)

The Fock space will be

E = S ∗ (E1 ⊗ C)
based on the complexification of E1 , with operators a†E (f ), aE (f ), φbE (f ), φbE (τ )
defined for f ∈ E1 . Expectation values of products of fields φbE (f ) for such
real-valued f can be given a probabilistic interpretation (see for instance [19]).
Note that the imaginary time state space and operators are of a quite dif-
ferent nature than those for real time. The operators φbE (τ ) do not satisfy an
equation of motion, and commute for all τ . They describe not the annihila-
tion and creation of a single quantum, but an arbitrary path in imaginary time
of a configuration-space observable. The state space is much larger than the
real-time state space, with E1 infinite dimensional as opposed to H1 = C.
One way to reconstruct the physical real-time theory from the Euclidean
theory is to consider the fixed τ subspace of E1 ⊗C of complex functions localized
at τ0 . Here f (τ ) = aδ(τ − τ0 ) for a ∈ C and one defines a Hermitian inner
product on E1 ⊗ C by
Z ∞Z ∞
(f, g)E1 ⊗C = f (τ2 )S2 (τ2 − τ1 )g(τ1 )dτ1 dτ2
−∞ −∞

While elements of E1 satisfy no differential equation and have no dynamics,

one does have an action of time translations on E1 , with translation by τ0 taking
aδ(τ ) to aδ(τ − τ0 ). Since the inner product satisfies

(aδ(τ ), bδ(τ − τ0 )) = abe−ω|τ0 |

one sees that one can define a Hamiltonian operator generating imaginary time
translations on these states by taking H to be multiplication by ω. The imagi-
nary time translation operator e−τ0 ω can be analytically continued from τ0 > 0
to real time t as
U (t) = e−itω

32
 Another way to reconstruct the real-time theory is the Osterwalder-Schrader
method, which begins by picking out the subspace E1+ ⊂ E1 ⊗ C of functions
supported on τ < 0. Defining a time reflection operator on E1 by
Θf (τ ) = f (−τ )
one can define
(f, g)OS = (Θf, g)E1 ⊗C
The physical H1 can then be recovered as
{f ∈ E1+ }
H1 =
{(f, f )OS = 0}
Note that for f, g ∈ E1+ one has (since f, g are supported for τ < 0)
Z ∞Z ∞
e−ω|τ2 −τ1 |
(f, g)OS = f (−τ2 ) g(τ1 )dτ1 dτ2
−∞ −∞ (2π)(2ω)
Z ∞Z ∞
eω(τ2 +τ1 )
= f (τ2 ) g(τ1 )dτ1 dτ2
−∞ −∞ (2π)(2ω)
Z ∞ Z ∞
1
= f (τ2 )eωτ2 dτ2 g(τ1 )eωτ1 dτ1
(2π)(2ω) −∞ −∞
1 e
= f (−iω)e g (−iω)
2ω
This gives a map
f ∈ E1+ → [f ] ∈ H1
similar to that of the real-time case
Z ∞
1 1
f → [f ] = √ fe(−iω) = √ √ eωτ f (τ )dτ
2ω 2ω 2π −∞

B.2 Relativistic scalar fields

The theory of a mass m free real scalar field in 3 + 1 dimensions can be treated
as a straightforward generalization of the above discussion of the harmonic oscil-
lator, treating time in the same way, spatial dimensions with the usual Fourier
transform. Defining p
ωp = |p|2 + m2
the Fourier transform of the Schwinger function is
1 1
S
f2 (s, p) =
(2π)3 s2 + ωp2
and the Schwinger function itself is
Z
1 1 1
S2 (τ, x) = ei(τ s+x·p) dsd3 p
(2π)2 R4 (2π)3 s2 + ωp2
m p
= p K1 (m τ 2 + |x|2 )
3 2
(2π) τ + |x| 2

33
where K1 is a modified Bessel function. This has an analytic continuation to
the z = t + iτ plane, with branch cuts on the t axis from |x| to ∞ and −|x| to
−∞.
τ

t
−|x| |x|

The Wightman function W2+ (t, x) will be defined as the limit of the analytic
continuation of S2 as one approaches the t-axis from negative values of τ . This
will be analytic for spacelike t < |x|, but will approach a branch cut for timelike
t > |x|. The Fourier transform of W2± will be, as a hyperfunction (in the
time-energy coordinate)
1 i
W
f2 (p) =
(2π)3 ωp2 − E 2
or, as a distribution, the delta-function distribution

f + (p) = 1
W 2 θ(E)δ(E 2 − ωp2 )
(2π)2
on the positive energy mass shell E = +ωp . Here W2+ (x) is
Z
1 1 −iωp t ip·x 3
W2+ (t, x) = 4
e e d p
(2π) R3 2ωp
As in the harmonic oscillator case, one can use it to reconstruct the single
particle state space H1 , defining
Z Z
(f, g) = f (x)W2+ (x − y)g(y)d4 xd4 y
R4 R4
4 4
for f, g ∈ S(R ) (R is Minkowski space), and equivalence classes
H1 = [f ] ∈ {f ∈ S(R4 )}/{(f, f ) = 0}
The inner product on H1 is given by
Z
h[f ], [g]i = θ(E)δ(E 2 − ωp2 )fe(p)e
g (p)d4 p
R4

34
where p = (E, p) and θ is the Heaviside step function. Elements [f ] of H1 can
be represented by functions fe on R3 of the form

fe(p) = fe(ωp , p)

In this representation, H1 has the Lorentz-invariant Hermitian inner product

d3 p
Z
h[f ], [g]i = fe(p)e
g (p)
R3 2ωp

Using the Fock space construction (as in the harmonic oscillator case, where
H1 = C), the full physical state space is
∞
M
H = S ∗ (H1 ) = S k (H1 )
k=0

with creation operators a† (f ) acting by symmetrized tensor product with [f ].

a(f ) is the adjoint operator and one can define field operators by

b ) = a(f ) + a† (f )
φ(f

Writing these distributions as φ(t,

b x), one recovers the usual description of
Wightman functions as

W2+ (x − y) = h0|φ(x)
b φ(y)|0i
b

The operators φ(x)

b satisfy the equation of motion
 2 
∂ 2
− ∆ + m φb = 0
∂t2

and φ(x),
b φ(y)
b commute for x and y space-like separated, but not for time-like
separations (due to the branch cuts described above).
The Euclidean (imaginary time) theory has the Fock space

E = S ∗ (E1 ⊗ C)

where E1 is the space of real-valued functions in S(R4 ) (now R4 is Euclidean

space) with inner product
Z Z
(f, g)E1 = f (x)S2 (x − y)g(y)d4 xd4 y
R4 R4

This Fock space comes with operators a†E (f ), aE (f ), φbE (f ), φbE (x) defined for
f ∈ E1 . Expectation values of products of fields φb( f ) for such real-valued f can
be given a probabilistic interpretation in terms of a Gaussian measure on the
distribution space S 0 (R4 ) (for details, see [19]).

35
 As in the harmonic oscillator case, there are two ways to recover the real
time theory from the Euclidean theory. In the first, one takes H1 ⊂ E1 to be the
functions on Euclidean space-time localized at a specific value of τ , say τ = 0,
of the form
1
f (τ, x) = δ(τ )F (x)
2π
Evaluating the inner product for these, one finds

d3 p
Z
(f, g)E1 = Fe(p)G(p)
e
R3 2ωp

which is the usual Lorentz-invariant inner product. The rotation group SO(3)
of spatial rotations acts on this τ = 0 subspace of E1 and this action passes to
an action on the physical H1 . Time translations act on E1 and one can use the
infinitesimal action of such translations to define the Hamiltonian operator on
H1 .
To recover the physical state space from the Euclidean theory by the Osterwalder-
Schrader method, one has to start by picking an imaginary time direction in the
Euclidean space R4 , with coordinate τ . One can then restrict to the subspace
E1+ ⊂ E1 of functions supported on τ < 0. Defining a time reflection operator
on E1 by
Θf (τ, x) = f (−τ, x)
one can define
(f, g)OS = (Θf, g)E1
The physical H1 can be recovered as

{f ∈ E1+ }
H1 =
{(f, f )OS = 0}
In both the Euclidean and Minkowski space-time formalisms one has a uni-
tary representation of the space-time symmetry groups (the Euclidean group
E(4) and the Poincaré group P respectively) on the spaces E1 , H1 and the cor-
responding Fock spaces. In the Minkowski space-time case this is an irreducible
representation, while in the Euclidean case it is far from irreducible, and the
representations in the two cases are not in any sense analytic continuations of
each other.
The spatial Euclidean group E(3) is in both E(4) and P , and the two meth-
ods for passing from the Euclidean to Minkowski space theory preserve this
group action. For translations in the remaining direction, one can fairly readily
define the Hamiltonian operator using the semi-group of positive imaginary time
translations in Euclidean space, then multiply by i and show that this generates
real time translations in Minkowski space-time.
More delicate is the question of what happens for group transformations in
other directions in SO(3, 1) (the boosts) and SO(4). In the Minkowski theory,
boosts act on H1 , preserving the inner product, so one has a unitary action
of the Poincaré group on H1 and from this on the full state space (the Fock

36
space). But while elements of SO(4) not in the spatial SO(3) act on E1 pre-
serving (·, ·)E1 , they do not preserve the positive time subspace E1+ and do not
commute with the time reflection operator Θ. One can construct operators on
E1+ giving infinitesimal generators corresponding to directions in the Lie algebra
complementary to the Lie algebra of SO(3), and then show that these can be
analytically continued and exponentiated to give the action of boosts on H1 .
That this can be done was first shown by Klein and Landau in 1982 (by a not
completely straight-forward argument, see [24]).

B.3 Spinor fields

While scalar field theories and pure gauge theories have well-understood and
straightforward formulations in Euclidean space-time, the question of how to
define spinor quantum field theories in Euclidean space-time has always been
(see the quote from Jaffe and Ritter above) much more problematic. At the
end of this paper one can find a fairly complete bibliography of attempts to
address this question over the years, none of which provide a fully satisfactory
answer. Schwinger’s earliest work argued that in Euclidean space a doubling of
the spinor degrees of freedom was necessary, and a version of Euclidean spinor
fields due to Osterwalder and Schrader [51] that includes such a doubling has
been the conventionally accepted best solution to the definitional problem.
We’ll consider the theory of a chiral (Weyl) spinor field in Minkowski space,
and then see what problems arise when one tries to find a corresponding Eu-
clidean field theory. It is well-known (see the quote at the beginning of this
section from [60]) that a problem arises immediately if one tries to write down
a Euclidean path integral for such a theory: there is no way to write an SO(4)
invariant Lagrangian just using one chirality.
The equation of motion for a right-handed Weyl spinor is
 
∂
+ σ · ∇ ψ(t, x) = 0
∂t
or, in energy-momentum space

(E − σ · p)ψ(E,
e p) = 0 (B.2)

Since one has

(E + σ · p)(E − σ · p) = E 2 − |p|2
solutions in energy-momentum space will also satisfy

(E 2 − |p|2 )ψ(E,
e p) = 0

and be supported on the light-cone E = ±|p|.

The momentum space Wightman function for the Weyl spinor theory will
be the hyperfunction
−i 1 −i E + σ · p
W
f2 (E, p) = =
(2π)3 E − σ · p (2π)3 E 2 − |p|2

37
or equivalently the distribution

+ 1
W
g
2 (E, p) = θ(E)(E + σ · p)δ(E 2 − |p|2 )
(2π)2

This is matrix-valued, and on solutions to B.2 gives the inner product

Z
hψe1 , ψe2 i = ψe1† (E, p)(E + σ · p)ψe2 (E, p)θ(E)δ(E 2 − |p|2 )dEd3 p
R4
d3 p
Z
= ψe1† (p)(|p| + σ · p)ψe2 (p)
3 2|p|
ZR
= ψe1† (p)ψe2 (p)d3 p
R3

Here ψ(p)
e = ψ(|p|,
e p).
The last expression is manifestly invariant under spatial (Spin(3)) rotations,
but not Lorentz (Spin(3, 1) = SL(2, C)) transformations. One can see Lorentz
invariance using the first expression, since for Ω ∈ SL(2, C) one has

(Ω† )−1 (E + σ · p)Ω−1 = E 0 + σ · p0

where E 0 , p0 are the Lorentz-transformed energy-momenta

(E 0 , p0 ) = Λ−1 · (E, p)

(Λ ∈ SO(3, 1) corresponds to Ω ∈ Spin(3, 1) in the spin double cover).

Note that the operator E + σ · p is just the momentum space identification
of Minkowski space-time R3,1 with 2 by 2 hermitian matrices:
 
t + x3 x1 − ix2
x = (t, x1 , x2 , x3 ) ↔ M =
x1 + ix2 t − x3

with the Minkowski norm given by −detM . One can identify complexified
Minkowski space-time R3,1 ⊗ C = C4 with all 2 by 2 complex matrices by:
 
t + iτ + z3 z1 − iz2
(t + iτ, z1 , z2 , z3 ) ↔ M =
z1 + iz2 t + iτ − z3

Euclidean space-time R4 will get identified with complex matrices of the form
 
iτ + x3 x1 − ix2
(τ, x1 , x2 , x3 ) ↔ M =
x1 + ix2 iτ − x3

and analytic continuation between Euclidean and Minkowski space takes place
on functions of such matrices.
The group Spin(4, C) = SL(2, C) × SL(2, C) acts on complex matrices by
−1
M → gL M gR

38
preserving the determinant (here gL , gR ∈ SL(2, C)). The subgroup SL(2, C)
† −1
such that gR = (gL ) is the Lorentz group Spin(3, 1) that preserves Minkowski
space-time, the subspace of hermitian matrices. The subgroup

SU (2)L × SU (2)R = Spin(4)

such that gL ∈ SU (2)L and gR ∈ SU (2)R preserves the Euclidean space-time.

If one tries to find a Schwinger function S2 related by analytic continuation
to W2 for the Weyl spinor theory, the factor E + σ · p in the expression for W2
causes two sorts of problems:

• After analytic continuation to Euclidean space-time it takes spinors trans-

forming under SU (2)R to spinors transforming under a different group,
SU (2)L . If the only fields in the theory are right-handed Weyl spinor
fields, the Schwinger function cannot give an invariant inner product.
• After analytic continuation the self-adjoint factor E + σ · p is neither
self-adjoint nor skew-adjoint. This makes it difficult to give S2 an inter-
pretation as inner product for a Euclidean field theory.
The first problem can be addressed by introducing fields of both chiralities,
giving up on having a theory of only one chirality of Weyl spinors. The ad-
jointness problem however still remains. Schwinger and later authors have dealt
with this problem by doubling the number of degrees of freedom. Schwinger’s
argument was that this was necessary in order to have Euclidean transformation
properties that did not distinguish a time direction. The problem also appears
when one tries to find a generalization of the time-reflection operator Θ that
allows reconstruction of the Minkowski theory from the Euclidean theory. The
conventional wisdom has been to follow Osterwalder-Schrader, who deal with
this by doubling the degrees of freedom, using a Θ which interchanges the two
sorts of fields[31]. A fairly complete bibliography of attempts to deal with the
Euclidean quantum spinor field is included at the end of this article.

B.4 Physical states and SO(4) symmetry breaking

It appears to be a fundamental feature of Euclidean quantum field theory that,
although Schwinger functions are SO(4) invariant, recovering a connection to
the physical theory in Minkowski space-time requires breaking SO(4) invari-
ance by a choice of imaginary time direction. In Minkowski space-time there is
a Lorentz-invariant distinction between positive and negative energy, while in
Euclidean space-time the corresponding distinction between positive and nega-
tive imaginary time is not SO(4) invariant. This breaking of SO(4) symmetry is
a sort of spontaneous symmetry breaking with not the lowest energy state, but
the distinction between positive and negative energy necessary for quantization
being responsible for the symmetry breaking.
While the Euclidean Fock space has an SO(4) action, states in it correspond
not to physical states, but to paths in the space of physical states. A choice of

39
imaginary time direction is needed to get physical states, either by restriction to
a constant imaginary time subspace or by restriction to a positive imaginary time
subspace together with use of reflection in imaginary time. The path integral
formalism has the same feature: one can write Schwinger functions as an SO(4)
invariant path integral, but to get states one must choose a hypersurface and
then define states using path integrals with fixed data on the hypersurface.
Needing to double spinor degrees of freedom and not being able to write
down a free chiral spinor theory have always been disconcerting aspects of Eu-
clidean quantum field theory. An alternate interpretation of the problems with
quantizing spinor fields in Euclidean space-time would be that they are a more
severe version of the problem one already sees with relativistic scalars, with
the quantization of such theories requiring the introduction of a new degree of
freedom that picks out an imaginary time direction.

C Hyperfunctions
Wightman functions are conventionally described as tempered distributions on
a Schwartz space of test functions. Such distributions occur as boundary values
of holomorphic functions, and one can instead work with hyperfunctions, which
are spaces of such boundary values. Like distributions, they can be thought of
a elements of a dual space to a space of well-behaved test functions, which will
be real analytic, not just infinitely differentiable. For an enlightening discussion
of hyperfunctions in this context, a good source is chapter 9 of Roger Penrose’s
The Road to Reality [45].

C.1 Hyperfunctions on the circle

In the case of the unit circle, one can generalize the notion of functions by
considering boundary values of holomorphic functions on the open unit disk.
Taking the circle to be the equator of a Riemann sphere, a hyperfunction on
the circle can be defined as a pair of functions, one holomorphic on the open
upper hemisphere, the other holomorphic on the open lower hemisphere, with
pairs equivalent when they differ by a globally holomorphic function.
Boundary values of functions holomorphic on the upper hemisphere corre-
spond to Fourier series with Fourier coefficients satisfying an = 0 for n < 0, those
with an = 0 for n > 0 correspond to boundary values of functions holomorphic
on the lower hemisphere. The global holomorphic functions on the sphere are
just the constants, those with only a0 non-zero. Hyperfunctions allow one to
make sense of a very large class of Fourier series (those with coefficients growing
at less than exponential rate as n → ±∞) as linear functionals on real analytic
test functions (whose coefficients an fall off faster than e−c|n| for some c > 0).
The discrete series representations of the non-compact Lie group SL(2, R)
can naturally be constructed using such hyperfunctions on the circle. The group
SL(2, R) acts on the Riemann sphere, with orbits the upper hemisphere, the
lower hemisphere, and the equator. The discrete series representations are hy-

40
perfunctions on the equator, boundary values of holomorphic sections of a line
bundle on either the upper or lower hemisphere. For more about this, see sec-
tion 10.1 of [8]. For a more general discussion of hyperfunctions on the circle
and their relation to hyperfunctions on R, see the previously mentioned chapter
9 of [45].

C.2 Hyperfunctions on R
Solutions to wave equations are conventionally discussed using the theory of dis-
tributions, since even the simplest plane-wave solutions are delta-functions in
energy-momentum space. Distributions are generalizations of functions that can
be defined as elements of the dual space (linear functionals) of some well-behaved
set of functions, for instance smooth functions of rapid decrease (Schwartz func-
tions) for the case of tempered distributions. The theory of hyperfunctions gives
a further generalization, providing a dual of an even more restricted set of func-
tions, analytic functions. Two references which contain extensive discussions of
the theory of hyperfunctions with applications are [44] and [43].
To motivate the definition of a hyperfunction on R, consider the boundary
values of a holomorphic function Φ+ on the open upper half plane. These give
a generalization of the usual notion of distribution, by considering the linear
functional on analytic functions (satisfying an appropriate growth condition)
on R Z ∞
f → lim+ Φ+ (t + i)f (t)dt
→0 −∞

Usual distributions are often written with a formal integral symbol denoting the
linear functional. In the case of hyperfunctions, this is no longer formal, but
becomes (a limit of) a conventional integral of a holomorphic function in the
complex plane, so contour deformation and residue theorem techniques can be
applied to its evaluation.
It is sometimes more convenient to have a definition involving symmetrically
the upper and lower complex half-planes. The space B(R) of hyperfunctions on
R can be defined as equivalence classes of pairs of functions (Φ+ , Φ− ), where
Φ+ is a holomorphic function on the open upper half-plane, Φ− is a holomorphic
function on the open lower upper half-plane. Pairs (Φ1,+ , Φ1,− ) and (Φ2,+ , Φ2,− )
are equivalent if
Φ2,+ = Φ1,+ + ψ, Φ2,− = Φ1,− + ψ
for some globally holomorphic function ψ. We’ll then write a hyperfunction as

φ = [Φ+ , Φ− ]

The derivative φ0 of a hyperfunction φ is given by taking the complex derivatives

of the pair of holomorphic functions representing it

φ0 = [Φ0+ , Φ0− ]

41
 As a linear functional on analytic functions, the hyperfunction φ is given by
I ∞ Z ∞
f→ φ(t)f (t)dt ≡ lim (Φ+ (t + i) − Φ− (t − i))f (t)dt
−∞ →0+ −∞

We’ll use coordinates t on R, z = t+iτ on C since our interest will be in physical

applications involving functions of time t, as well as their analytic continuations
to imaginary time τ .
One way to get hyperfunctions is by choosing a function Φ(z) on C, holo-
morphic away from the real axis R, and taking

φ = [Φ|U HP , Φ|LHP ]

For example, consider the function

i 1
Φ=
2π z − ω
where ω ∈ R. As a distribution, corresponding hyperfunction will be given by
the limit
 
i 1 1 1 1
φ(t) = lim+ − = lim+
→0 2π t + i − ω t − i − ω →0 π (t − ω)2 + 2

The limit on the right-hand side is well-known as a way to describe the delta
function distribution δ(t − ω) as a limit of functions. Using contour integration
methods one finds that the hyperfunction version of the delta function behaves
as expected since I ∞
i 1
f (t)dt = f (ω)
−∞ 2π t − ω

One would like to define a Fourier transform for hyperfunctions, with the
same sort of definition as an integral in the usual case, so
Z ∞
1
F(φ)(E) = φ(E)
e =√ eiEt φ(t)dt (C.1)
2π −∞
with the inverse Fourier transform defined by
Z ∞
1
F −1 (φ)(t)
e = φ(t) = √ e−iEt φ(E)dE
e
2π −∞
The problem with this though is that the Fourier transform and its inverse don’t
take functions holomorphic on the upper or lower half plane to functions with
the same property.
One can however define a Fourier transform for hyperfunctions (satisfying a
growth condition, called “Fourier hyperfunctions”) by taking advantage of the
fact that for a class of functions f (E) supported on E > 0 (respectively E < 0)
Z ∞
1
√ e−iEz f (E)dE
2π −∞

42
is holomorphic in the lower half (respectively upper half) z plane (since the
exponential falls off there). The decomposition of a hyperfunction φ(t) into
limits of holomorphic functions Φ+ , Φ− on the upper and lower half planes cor-
responds to decomposition of φ(E)
e into hyperfunctions φe− (E), φe+ (E) supported
for negative and positive E respectively. This is similar to what happened for
hyperfunctions on the circle, with Φ+ , Φ− analogous to functions holomorphic
on the upper or lower hemispheres, φe− (E), φe+ (E) analogous to the Fourier co-
efficients for positive or negative n.
For an example, consider the hyperfunction version of a delta function sup-
ported at E = ω, ω > 0:
 
i 1 i 1 1
φ(E) = φ+ (E) =
e e ≡ lim −
2π E − ω 2π →0+ E + i − ω E − i − ω

This has as inverse Fourier transform the hyperfunction

Z ∞
1 i 1 1
φ(t) = √ e−iEt dE = √ e−iωt
2π −∞ 2π E − ω 2π
which has a representation as
1
φ(t) = [0, − √ e−iωz ]
2π
The Fourier transform of this will be
Z ∞
1
φ(E) = √
e eiEt φ(t)dt
2π −∞
Z ∞
1 1
=√ eiEt √ e−iωt dt
2π −∞ 2π
but this needs to be interpreted as a sum of integrals for t negative and t positive
Z 0 Z ∞
1 i(E−i−ω)t
= lim ( e dt + ei(E+i−ω)t dt)
2π →0+ −∞ 0
 
i 1 1
= lim −
2π →0+ E + i − ω E − i − ω

An example that is relevant to the case of the harmonic oscillator is that of

 
1 1 1 1
φ(E)
e = 2 = −
E − ω2 2ω E − ω E + ω
where the first term is a hyperfunction with support only at ω > 0, the second
only at −ω < 0. The inverse Fourier transform is
iπ 1
φ(t) = √ (eiωt − e−iωt )
ω 2π

43
where the first term should be interpreted as the equivalence class
iπ 1
√ [eiωz , 0]
ω 2π
and the second as the equivalence class
iπ 1
√ [0, e−iωz ]
ω 2π

References
[1] Tim Adamo. Twistor actions for gauge theory and gravity. 2013. arXiv:
1308.2820 [hep-th].
[2] Timothy Adamo. “Lectures on twistor theory”. In: Proceedings of the XIII
Modave Summer School in Mathematical Physics. 10-16 September 2017
Modave. 2017, p. 3. arXiv: 1712.02196 [hep-th].
[3] Stephon Alexander, Antonino Marcianò, and Lee Smolin. “Gravitational
origin of the weak interaction’s chirality”. In: Physical Review D 89 (2014),
p. 065017. arXiv: 1212.5246 [hep-th].
[4] Abhay Ashtekar. New Perspectives in Canonical Gravity. 1988.
[5] Abhay Ashtekar. Lectures on Non-Perturbative Canonical Gravity. World
Scientific, 1991. doi: 10.1142/1321.
[6] Michael F. Atiyah, Nigel J. Hitchin, and Isidore M. Singer. “Self-duality
in four-dimensional Riemannian geometry”. In: Proc. R. Soc. Lond. A 362
(1978), pp. 425–461.
[7] Michael F. Atiyah. Geometry of Yang-Mills fields. Lezioni Fermiane. Scuola
Normale Superiore, Pisa, 1979.
[8] Robert J. Baston and Michael G. Eastwood. The Penrose Transform: Its
Interaction With Representation Theory. Oxford University Press, 1989.
[9] Rutger Boels, Lionel Mason, and David Skinner. “Supersymmetric gauge
theories in twistor space”. In: Journal of High Energy Physics 2007 (2007).
arXiv: hep-th/0604040 [hep-th].
[10] Raoul Bott. “Homogeneous Vector Bundles”. In: Annals of Mathematics
66 (1957), pp. 203–248.
[11] Raoul Bott. “The Index Theorem for Homogeneous Differential Opera-
tors”. In: Differential and Combianatorial Topology. Ed. by Stewart Scott
Cairns. Princeton University Press, 1965, pp. 167–186.
[12] G. Chalmers and W. Siegel. “Self-dual sector of QCD amplitudes”. In:
Physical Review D 54 (1996), 7628–7633. issn: 1089-4918. arXiv: hep -
th/9606061 [hep-th].
[13] Kevin J. Costello. Notes on supersymmetric and holomorphic field theories
in dimensions 2 and 4. 2013. arXiv: 1111.4234 [math.QA].

44
[14] Michel Dubois-Violette. “Structures Complexes Au-dessus des Variétés,
Applications”. In: Mathématique et Physique. Ed. by L. Boutet de Montvel,
A. Douady, and J.L. Verdier. Vol. 37. Progress in Mathematics. Birkhaüser,
1983, pp. 1–42.
[15] Michael G. Eastwood, Roger Penrose, and Raymond O. Wells Jr. “Co-
homology and Massless Fields”. In: Communications in Mathematical
Physics 78 (1981), pp. 305–351.
[16] Michael G. Eastwood. “The generalized Penrose-Ward transform”. In:
Proc. R. Soc. Lond. A 370 (1980), pp. 173–191.
[17] Chris Elliott and Philsang Yoo. Quantum Geometric Langlands Cate-
gories from N = 4 Super Yang-Mills Theory. 2020. arXiv: 2008.10988
[math-ph].
[18] G. Gibbons. “Euclidean quantum gravity: The view from 2002”. In: Work-
shop on Conference on the Future of Theoretical Physics and Cosmology
in Honor of Steven Hawking’s 60th Birthday. 2002, pp. 351–372.
[19] James Glimm and Arthur Jaffe. Quantum Physics: A Functional Integral
Point of View. Springer-Verlag, 1981.
[20] Yannick Herfray. “Pure connection formulation, twistors, and the chase
for a twistor action for general relativity”. In: Journal of Mathematical
Physics 58.11 (2017), p. 112505. arXiv: 1610.02343 [hep-th].
[21] N. J. Hitchin. “Linear field equations on self-dual spaces”. In: Math. Proc.
Camb. Phil. Soc. 97 (1984), pp. 165–187.
[22] Jing-Song Huang and Pavle Pandz̆ić. Dirac Operators in Representation
Theory. Birkhäuser, 2006.
[23] Anton Kapustin and Edward Witten. “Electric-Magnetic Duality And
The Geometric Langlands Program”. In: Commun. Num. Theor. Phys.
1 (2007), pp. 1–236. arXiv: hep-th/0604151.
[24] Abel Klein and Lawrence Landau. “From the Euclidean Group to the
Poincaré Group via Osterwalder-Schrader Positivity”. In: Commun. Math.
Phys. 87 (1983), pp. 469–484.
[25] Shoshichi Kobayashi and Katsumi Nomizu. Foundations of Differential
Geometry: Volume I. Interscience Publishers, 1963.
[26] Kirill Krasnov. Formulations of General Relativity. Cambridge University
Press, 2020.
[27] H. Blaine Lawson Jr. and Marie-Louise Michelsohn. Spin geometry. Vol. 38.
Princeton Mathematical Series. Princeton University Press, 1989.
[28] L. J. Mason and N.M.J. Woodhouse. Integrability, Self-duality, and Twistor
Theory. London Mathematical Society Mongraphs. Oxford University Press,
1996.
[29] V.P. Nair. “A current algebra for some gauge theory amplitudes”. In:
Physics Letters B 214 (1988), pp. 215–218.

45
[30] Fabrizio Nesti and Roberto Percacci. “Gravi-weak unification”. In: Jour-
nal of Physics A: Mathematical and Theoretical 41 (2008), p. 075405.
arXiv: 0706.3307 [hep-th].
[31] Konrad Osterwalder and Robert Schrader. “Euclidean Fermi Fields and a
Feynman-Kac formula for Boson-Fermion models”. In: Helvetica Physica
Acta 46 (1973), pp. 277–302.
[32] Roger Penrose. “Twistor Algebra”. In: Journal of Mathematical Physics
8 (1967), pp. 345–366.
[33] John Rawnsley, Wilfried Schmid, and Joseph A. Wolf. “Singular Unitary
Representations and Indefinite Harmonic Theory”. In: Journal of Func-
tional Analysis 51 (1983), pp. 1–114.
[34] Simon Salamon. “Topics in Four-dimensional Riemannian geometry”. In:
Geometry Seminar Luigi Bianchi. Ed. by E. Vesentini. Vol. 1022. Lecture
Notes in Mathematics. Springer, 1983, pp. 33–124.
[35] R. S. Ward and Raymond O. Wells Jr. Twistor geometry and field theory.
Cambridge University Press, 1990.
[36] R. O. Wells Jr. “Hyperfunction Solutions of the Zero-Rest-Mass Field
Equations”. In: Communications in Mathematical Physics 78 (1981), pp. 567–
600.
[37] Edward Witten. “Topological quantum field theory”. In: Communications
in Mathematical Physics 117.3 (1988), pp. 353–386.
[38] Edward Witten. “Perturbative Gauge Theory as a String Theory in Twistor
Space”. In: Communications in Mathematical Physics 252.1-3 (2004), pp. 189–
258. arXiv: hep-th/0312171 [hep-th].
[39] Peter Woit. “Supersymmetric quantum mechanics, spinors and the Stan-
dard Model”. In: Nuclear Physics B303 (1988), pp. 329–342.
[40] Peter Woit. Quantum Field Theory and Representation Theory: A Sketch.
2002. arXiv: hep-th/0206135 [hep-th].
[41] Peter Woit. Quantum theory, groups and representations: An introduction.
Springer, 2017.
[42] N.M.J. Woodhouse. “Real Methods in Twistor Theory”. In: Classical and
Quantum Gravity 2 (1985), pp. 257–291.
[43] Urs Graf. Introduction to hyperfunctions and their integral transforms.
Birkhäuser, 2010.
[44] Isao Imai. Applied hyperfunction theory. Vol. 8. Mathematics and its ap-
plications. Kluwer Academic Publishers, 1992.
[45] Roger Penrose. The Road to Reality. Jonathan Cape, 2004.

46
Euclidean Spinor References
[46] Julian Schwinger. “On the Euclidean structure of relativistic field theory”.
In: Proceedings of the National Academy of Sciences 44.9 (1958), pp. 956–
965.
[47] Julian Schwinger. “Four-dimensional Euclidean formulation of quantum
field theory”. In: 8th International Annual Conference on High Energy
Physics. 1958, pp. 134–140.
[48] Julian Schwinger. “Euclidean Quantum Electrodynamics”. In: Phys. Rev.
115 (3 1959), pp. 721–731.
[49] Tadao Nakano. “Quantum Field Theory in Terms of Euclidean Parame-
ters”. In: Progress of Theoretical Physics 21.2 (1959), pp. 241–259.
[50] K. Osterwalder. “Euclidean fermi fields”. In: International School of Math-
ematical Physics, Ettore Majorana: 1st course: Constructive Quantum
Field Theory. 1973, pp. 326–331.
[51] K. Osterwalder and R. Schrader. “Euclidean fermi fields and a Feynman-
Kac formula for boson-fermion models”. In: Helv. Phys. Acta 46 (1973),
pp. 277–302.
[52] David N. Williams. “Euclidean Fermi fields with a Hermitean Feynman-
Kac-Nelson formula, I”. In: Comm. Math. Phys. 38.1 (1974), pp. 65–80.
[53] David N. Williams. “Euclidean Fermi fields with a Hermitean Feynman-
Kac-Nelson formula, II”. 1974. url: http://www-personal.umich.edu/
~williams/papers/eufermi.pdf.
[54] Edward Nelson. “Markov fields”. In: Proceedings of the International Congress
of Mathematicians (Vancouver, B. C., 1974), Vol. 2. 1975, pp. 395–398.
[55] Jurg Frohlich and Konrad Osterwalder. “Is There a Euclidean Field The-
ory for Fermions?” In: Helv. Phys. Acta 47 (1975), p. 781.
[56] Leonard Gross. “On the formula of Mathews and Salam”. In: Journal of
Functional Analysis 25.2 (1977), pp. 162 –209.
[57] H. Nicolai. “A possible constructive approach to (super − φ3 )4 : (I). Eu-
clidean formulation of the model”. In: Nuclear Physics B 140.2 (1978),
pp. 294 –300. issn: 0550-3213.
[58] Diego de Falco and Francesco Guerra. “On the local structure of the
Euclidean Dirac field”. In: Journal of Mathematical Physics 21.5 (1980),
pp. 1111–1114.
[59] John Palmer. “Euclidean Fermi fields”. In: Journal of Functional Analysis
36.3 (1980), pp. 287 –312.
[60] Pierre Ramond. Field Theory: A Modern Primer. Benjamin/Cummings,
1981.
[61] Shigeaki Nagamachi and Nobumichi Mugibayashi. “Covariance of Eu-
clidean Fermi Fields”. In: Progress of Theoretical Physics 66.3 (1981),
pp. 1061–1077.

47
[62] Bengt Ek. “Parity Conservation and the Euclidean Field Formulation of
Relativistic Quantum Theoriesj”. In: Letters in Mathematical Physics 5
(1981), pp. 149–154.
[63] Bengt Ek. “Euclidean Quantum Fields with Spin on an Indefinite Inner
Product State Space”. In: Publ. RIMS, Kyoto Univ. 18 (1982), pp. 251–
274.
[64] J. Frohlich, K. Osterwalder, and E. Seiler. “On Virtual Representations
of Symmetric Spaces and Their Analytic Continuation”. In: Annals of
Mathematics 118.3 (1983), pp. 461–489.
[65] J. Kupsch. “Measures for Fermionic Integration”. In: Fortschritte der
Physik/Progress of Physics 35.5 (1987), pp. 415–436.
[66] J. Kupsch. “Functional integration for Euclidean Dirac fields”. In: Annales
de l’I.H.P. Physique théorique 50.2 (1989), pp. 143–160.
[67] J. Kupsch and W. D. Thacker. “Euclidean Majorana and Weyl Spinors”.
In: Fortschritte der Physik/Progress of Physics 38.1 (1990), pp. 35–62.
[68] Mayank R. Mehta. “Euclidean continuation of the Dirac fermion”. In:
Phys. Rev. Lett. 65 (16 1990), pp. 1983–1986.
[69] David Borthwick. “Euclidean Majorana fermions, fermionic integrals, and
relative Pfaffians”. In: Journal of Mathematical Physics 34.7 (July 1993),
pp. 2691–2712.
[70] Peter van Nieuwenhuizen and Andrew Waldron. “On Euclidean spinors
and Wick rotations”. In: Physics Letters B 389.1 (1996), 29–36.
[71] P. van Nieuwenhuizen and A. Waldron. “A continuous Wick rotation for
spinor fields and supersymmetry in Euclidean space”. In: Gauge Theories,
Applied Supersymmetry and Quantum Gravity II (1997). eprint: hep -
th/9608174.
[72] Andrew Waldron. “A Wick rotation for spinor fields: the canonical ap-
proach”. In: Physics Letters B 433.3-4 (1998), 369–376. eprint: hep-th/
9702057.
[73] Arthur Mountain. “Wick rotation and supersymmetry”. In: Quantum as-
pects of gauge theories, supersymmetry and unification (Sept. 1, 1999–
Sept. 7, 1999). Paris, 2000. doi: 10.22323/1.004.0036.
[74] V. Wessels and W. N. Polyzou. “Stability of Covariant Relativistic Quan-
tum Theory”. In: Few-Body-Systems 35.1-2 (2004).
[75] Arthur Jaffe and Gordon Ritter. “Reflection positivity and monotonicity”.
In: Journal of Mathematical Physics 49.5 (2008), p. 052301. arXiv: 0705.
0712 [math-ph].
[76] C. Wetterich. “Spinors in euclidean field theory, complex structures and
discrete symmetries”. In: Nuclear Physics B 852.1 (2011), pp. 174 –234.
arXiv: 1002.3556 [hep-th].

48