Spin 2010 (jmf) 54

Lecture 8: Parallel and Killing spinor fields

In this lecture we will characterise manifolds admitting spinor fields satisfying some natural differen-
tial equations. We will first revisit parallel (or covariantly constant) spinor fields, which were already
discussed in §6.3, from the point of view of the holonomy representation. We will then introduce the
notion of a (real) Killing spinor field, as a special case of a “twistor” spinor field.

8.1 Manifolds admitting parallel spinor fields

Recall that a covariantly constant spinor field ψ — that is, one obeying d ∇ ψ = 0 — is invariant under
parallel transport and hence its value at any point m is a spinor which is invariant under (the spin
lift of ) the holonomy group Hol(m). We also learnt that in positive-definite signature, a spin manifold
admits parallel spinor fields only if it is Ricci-flat. This means that if Hol(m) leaves invariant a (nonzero)
spinor, the manifold must be Ricci-flat. As we discussed in the last lecture, there are four Ricci-flat
holonomy representations: SU(n) ⊂ SO(2n), Sp(n) ⊂ SO(4n), G2 ⊂ SO(7) and Spin(7) ⊂ SO(8). Curiously,
as shown by Wang [Wan89], each of these representations preserve a nonzero spinor. His results can
be summarised in the following table. The column labelled “Parallel spinors” lists the dimension of the
space of parallel spinors. In even dimensions, this is further refined according to chirality, in such a way
that (n + , n − ) means that the space of positive (resp. negative) parallel half-spinors has (real) dimension
n + (resp. n − ). Of course, changing the orientation of the manifold interchanges n + and n − .

Table 1: Irreducible, simply-connected manifolds admitting parallel spinors

Holonomy representation Geometry Parallel spinors

SU(2n + 1) ⊂ SO(4n + 2) Calabi–Yau (1, 1)
SU(2n) ⊂ SO(4n) Calabi–Yau (2, 0)
Sp(n) ⊂ SO(4n) hyperkähler (k + 1, 0)
G2 ⊂ SO(7) exceptional 1
Spin(7) ⊂ SO(8) exceptional (1, 0)

We will concentrate on two examples: SU(3) ⊂ SO(6) and G2 ⊂ SO(7).

8.1.1 Calabi–Yau 3-folds

We start with the following lemma.

Lemma 8.1. The spin representation gives an isomorphism Spin(6) ∼

= SU(4).

Proof. First of all we remark that Spin(6) ⊂ C�(6)0 ∼= C(4). Thus we have an injective homomorphism
ι : Spin(6) → GL(4, C), which is the spin representation. Since Spin(6) is compact, its image in GL(4, C)
must lie inside a maximal compact subgroup of GL(4, C): namely, a copy of U(4). Since Spin(6) is simple,
its image must be inside SU(4). Finally, since dim Spin(6) = dim SU(4) = 15, and since both Spin(6) and
SU(4) are connected, ι is an isomorphism.

This means that the spinor representation of Spin(6) is the defining representation of SU(4) on C4 .
A nonzero spinor is a vector ψ ∈ C4 . Without loss of generality we can assume that ψ = (z, 0, 0, 0) for
some 0 �= z ∈ C. It is then clear that the subgroup of SU(4) leaving that vector invariant is an SU(3)
subgroup, which is the image under ι of an SU(3) subgroup of Spin(6). Since −1 ∈ Spin(6) does not
leave ψ invariant, it does not belong to SU(3) whence its image under Ad� : Spin(6) → SO(6) is an SU(3)
subgroup of SO(6). This is precisely the holonomy representation SU(3) ⊂ SO(6) in Berger’s table. The
complex conjugate spinor ψ has the opposite chirality to ψ and is also left invariant by the same SU(3)
subgroup, whence the (1, 1) in the corresponding entry in the table.
Spin 2010 (jmf) 55

8.1.2 Manifolds of G2 holonomy

Let O denote the real division algebra of octonions, obtained from the quaternions by the Cayley–
Dickson doubling construction. It is a normed algebra with a positive-definite inner product B(x, y) =
Re(x y). The octonions are not associative, but they are alternating, which means that the subalgebra
generated by any two elements is associative. In particular, if x, y ∈ O, then x(x y) = x 2 y and (y x)x = y x 2 .
This is equivalent to associator (x y)z − x(y z) being totally skewsymmetric in x, y, z. Consider now the
linear maps defined by � : x �→ �x and r : x �→ r x , where �x and r x are, respectively, left and right multi-
plication by x ∈ O.

Lemma 8.2. The linear maps �, r : ImO → EndR (O) defined above are Clifford.

Proof. We will prove the lemma for � and leave r as an exercise. First of all, notice that the alternating
property of O says that for all x, y, z ∈ O,

(103) x(y z) − (x y)z = −y(xz) + (y x)z ,

whence

�x � y z + � y �x z = x(y z) + y(xz)
= (x y)z + (y x)z by equation (103)
= (x y + y x)z .

But notice that since x, y ∈ ImO, x y + y x ∈ R ⊂ O and is indeed equal to −2B(x, y), whence we conclude
that
�x � y + � y �x = −2B(x, y)1 .

This means that � and r extend to representations of the Clifford algebra C�(ImO) ∼ = C�(7). Indeed,
the isomorphism C�(7) ∼ = R(8)⊕R(8) is the Clifford extension of the Clifford map x �→ (�x , r x ). The spinor
representation of Spin(7) is obtained by restricting either of these two Clifford modules to Spin(7) ⊂
C�(7). This defines a map Spin(7) → GL(8, R) whose image, since Spin(7) is compact and connected, lies
inside SO(8), for some SO(8) subgroup of GL(8, R). Indeed, it is the SO(8) which preserves the octonionic
inner product. This follows form the fact that O is a normed algebra, whence B(x y, x y) = B(x, x)B(y, y),
whence if B(x, x) = 1 then both �x and r x are isometries.
Let ψ be a nonzero spinor, which we may take to correspond to 1 ∈ R ⊂ O. The subgroup of Spin(7)
which fixes ψ is a G2 subgroup of Spin(7) which does not contain −1 and hence projects under Ad �:
Spin(7) → SO(7) to a G2 subgroup of SO(7), which is precisely the holonomy representation G2 ⊂ SO(7).
Any other spinor left invariant by this G2 subgroup is proportional to ψ.

8.1.3 Some comments about indefinite signature

In physical applications it is often necessary to determine the lorentzian (or even higher index) spin
manifolds admitting parallel spinors. There is a classification of lorentzian holonomy groups due to
Leistner and Galaev [GL08], as well as earlier results of Bryant [Bry00a] and myself [FO00]. A lorent-
zian spin n-dimensional manifold admits parallel spinors if its holonomy representation is G � Rn−2 ⊂
SO0 (n −1, 1), where G ⊂ SO(n −2) is one of the riemannian holonomy representations admitting parallel
spinors. The subgroup G � Rn−2 of SO(n −1, 1) is such that G acts on Rn−2 via the holonomy representa-
tion G ⊂ SO(n −2), and the abelian normal subgroup Rn−2 acts as null rotations on Rn−1,1 . The situation
for higher index is much less clear and still the subject of investigation.

8.2 Manifolds admitting (real) Killing spinor fields

On a spin manifold one can define natural equations satisfied by spinor fields other than d ∇ ψ = 0.
In this section we will discuss the Killing spinor equation which is a special case of the twistor spinor
equation, about which we will not say anything beyond its definition.
Spin 2010 (jmf) 56

8.2.1 The Dirac operator

Let E = (e i ) be a local frame and let (e i ) denote the dual frame, so that g (e i , e j ) = δij . The Dirac operator
is the differential operator D acting on a spinor field ψ as
�
(104) Dψ = e i · ∇e i ψ ,
i

where the dot (·) stands for Clifford action. More invariantly, it is defined as the composition of the
following two maps

(105) d∇ cl
C∞ (M, S(M)) −−−−−→ C∞ (M, T ∗ M ⊗ S(M)) −−−−−→ C∞ (M, S(M))
where the first map is the covariant derivative and the second map is the fibrewise Clifford action T ∗ M⊗
S(M) → S(M).
Example 8.3. The original Dirac operator was defined on four-dimensional Minkowski spacetime. Re-
lative to flat coordinates x µ and the associated frame, the Dirac operator takes the form
� � ∂ψ
(106) Dψ = Γµ · ∇ ∂ ψ = Γµ µ ,
µ ∂x µ
µ ∂x
�
where ψ : R3,1 → C4 and Γµ = ν ηµν Γν and Γµ are the 4 × 4 gamma matrices representing the Clifford
action by the frame vectors ∂x∂µ .
Spinors which are annihilated by the Dirac operator are known as harmonic spinors. The origin of
the name is due to the fact that squaring the original Dirac operator, one gets the laplacian:
∂ ∂
(107) D2 ψ = −ηµν ψ = �ψ .
∂x µ ∂x ν
In the general case, squaring the Dirac operator results in a curvature-dependent correction:
s
(108) D2 ψ = ∇∗ ∇ψ + ψ ,
4
where s is the scalar curvature and ∇∗ ∇ is the covariant laplacian.
An immediate corollary of this calculation is the following theorem due to Lichnerowicz.
Theorem 8.4 (Lichnerowicz). If (M, g ) is a compact positive-definite riemannian spin manifold with
s ≥ 0 and s > 0 at at least one point, then (M, g ) admits no nonzero harmonic spinor fields; whereas if
s ≡ 0 then a harmonic spinor field is parallel.
Proof. Indeed, let (−, −) denote the invariant inner product on the spinor bundle, and consider the
integral � � �
� � 1
ψ, D2 ψ = |d ∇ ψ|2 + s|ψ|2 .
M M 4 M
Let Dψ = 0, so that the LHS vanishes. Then if s ≥ 0, the RHS is positive-semidefinite and in particular
we see that d ∇ ψ = 0. This being the case, ψ is determined uniquely by its value at any point, so that in
particular if it vanishes anywhere, it must vanish everywhere. If s > 0 at at least one point, then it s > 0
is a neighbourhood of that point and hence ψ = 0 in a neighbourhood of that point and hence ψ = 0
everywhere.

8.2.2 The Penrose operator and twistor spinor fields

Let W ⊂ T ∗ M ⊗ S(M) denote the subbundle defined as the kernel of the Clifford action T ∗ M ⊗ S(M) →
S(M). Let π : T ∗ M ⊗ S(M) → W denote the projection onto W along S(M). The Penrose operator P :
C∞ (M, S(M)) → C∞ (M, W) is defined as the composition
(109) d∇ π
C∞ (M, S(M)) −−−−−→ C∞ (M, T ∗ M ⊗ S(M)) −−−−−→ C∞ (M, W) .
Explicitly, we can write for all spinor fields ψ and all vector fields X,
(110) PX ψ = ∇X ψ + n1 X · Dψ ,
where n = dim M. Spinor fields in the kernel of the Penrose operator are known as twistor fields.
Spin 2010 (jmf) 57

8.2.3 Killing spinor fields

A Killing spinor field is a special type of twistor field ψ which satisfies the stronger equation

(111) ∇X ψ = λX · ψ ,

for some constant λ ∈ C called the Killing constant. A calculation similar to that in section 6.3 reveals
that the integrability condition for the existence of Killing spinor fields is

(R(X) − 4λ2 (n − 1)X) · ψ = 0 ,

for all vector fields X and where X �→ R(X) is the Ricci operator and n = dim M. In positive-definite signa-
ture, it says that R(X) = 4λ2 (n − 1)X for all vector fields X, or equivalently after taking the inner product
with a second vector field Y, that r (X, Y) = 4λ2 (n − 1)g (X, Y), whence (M, g ) is Einstein. In indefinite sig-
nature this is no longer the case, but we can take the Clifford trace of the above equation to conclude
that
sψ = 4λ2 n(n − 1)ψ ,
whence if ψ is not identically zero, the scalar curvature is constrained in terms of λ: namely, s = 4λ2 n(n−
1). Since the scalar curvature is real, we see that λ2 is real, whence it is either real or pure imaginary.
The nature of the Killing constant gives rise to two different kinds of Killing spinor fields: real and ima-
ginary, respectively. They each have a very different flavour and in the rest of this lecture we will con-
centrate on the real case. Furthermore via a homothety (i.e., rescaling the metric by a constant positive
number) we can further assume that λ = ± 12 . Finally, we will concentrate on positive-definite signa-
ture, whence we will be interested in characterising those positive-definite riemannian spin manifolds
admitting nonzero spinor fields ψ satisfying

(112) ∇X ψ = ± 12 X · ψ ,

for all vector fields X.

Bär’s cone construction [Bär93] will relate such Killing spinor fields to parallel spinor fields in an
auxiliary geometry. To at least demonstrate the plausibility of such a construction, let us first of all
notice that a spinor field obeying equation (112) is actually parallel with respect to the connection DX =
∇X ∓ 12 X·. The connection one-form associated with D is given, relative to a local frame E = (e i ), by
� �
(113) 1
4 ωi j e i e j ∓ 12 θi e i ,
i,j i

where ωi j (X) = g (∇X e i , e j ) and θi (X) = g (X, e i ). But now notice that 14 [e i , e j ] and ∓ 12 e i in C�(n) span an
so(n+1) subalgebra of C�(n), whence the above connection one-form is so(n+1)-valued, which suggests
that it could very well be the spin connection of an (n + 1)-dimensional manifold. This manifold is the
metric cone as we now review.

8.2.4 The cone construction

Let (M, g ) be an n-dimensional riemannian manifold and let M � = R+ × M. We parametrise R+ by r > 0

2 2
� by g� = d r + r g . The riemannian manifold (M, � g�) thus constructed is the
and define a metric g� on M
metric cone of (M, g ). (M, g ) embeds isometrically into (M, � g�) as the submanifold at r = 1. Generically
the metric on M � cannot be extended smoothly to r = 0. The exception occurs when (M, g ) is the round
n-sphere, in which case the cone is Rn+1 \ {0} with the flat euclidean metric, since in that case the flat
metric is clearly regular at the origin and can be extended there.
The cone M � admits a homothetic action by R+ , where e t ∈ R+ acts by rescaling the “radial” coordin-
ate: (r, x) �→ (e t r, x). The conformal Killing vector generating this action is the Euler vector ξ = r1 ∂r
∂
. A
vector field X ∈ X (M) admits a unique lift to M, � also denoted X with a little abuse of notation, such that
it is orthogonal to ξ and such that it maps to X under the natural projection M � → M, sending (r, x) to x.
Let ∇ � denote the Levi-Civita connection on M. �
Spin 2010 (jmf) 58

� be lifts of vector fields on M. Then

Lemma 8.5. Let X, Y ∈ X (M)

� ξξ = ξ ,
∇ � ξX = X ,
∇ �Xξ = X
∇ and � X Y = ∇X Y − g (X, Y)ξ .
∇

Remark 8.6. In fact, a result of Gibbons and Rychenkova [GR98] states that a riemannian manifold is a
metric cone if and only if there exists a vector field ξ such that ∇V ξ = V for all vector fields V, where ∇ is
the Levi-Civita connection.

Now given a local frame E = (e i ) for M, we extend it to a local frame E� = (� ∂

e 0 = ∂r , e�i = r1 e i ) for M.
� The
� �
connection coefficients of ∇ relative to E are given in terms of the connection coefficients of ∇ relative
to E by the following formulae.

Lemma 8.7. Let ω

� ab = g�(∇� � relative to the local frame E�. Then
� e a , e�b ) be the connection 1-form in M

∂
� ab ( ∂r
ω )=0, � 0i (e j ) = δi j
ω and � i j (e k ) = ωi j (e j ) .
ω

Since R+ is contractible, the cone M

� is homotopy equivalent to M, whence if M is spin, so is M. �
�
Furthermore, if M is spin, the embedding (at r = 1) of M into M sets up a bijective correspondence
between the spin structures on M and on M.� From now on we assume that both M and M � are spin,
with corresponding spin structures. Now let ψ �
� be a spinor field on M. Its covariant derivative can be
computed from equation (89) and the previous lemma and one finds

�
∇ ∂ �=
ψ ∂
∂r
�
ψ and �e ψ
∇ k
� + 12 e�0 e�k ψ
� = ∇e k ψ �.
∂r

Therefore a parallel spinor field ψ � on M� satisfies ∂ ψ

∂r
� = 12 e�k e�0 ψ.
� = 0 and ∇e k ψ � The restriction of ψ�
to r = 1 is a spinor field on M which satisfies the second of the above equations. To understand this
equation intrinsically, we recall the isomorphism C�(n) ∼ = C�(n + 1)0 given in Proposition 2.8. In fact,
there are two possible isomorphisms, distinguished by a sign: e i �→ ε� e i e�0 , for ε2 = 1. It is now that
we must make a distinction between even- and odd-dimensional M. Consider the volume element
e 1 · · · e n ∈ C�(n). Its image in C�(n + 1)0 under the above isomorphism is given by
�
e 0 e�1 · · · e�n
−ε� n odd
(114) e 1 · · · e n �→
e�1 · · · e�n n even.

If n = dim M is odd, then there are two inequivalent Clifford modules, each determined by the action
of the volume element e 1 · · · e n in C�(n), which goes over to −ε times the action of the volume element
e�0 e�1 · · · e�n in C�(n + 1). This means that ε can be fixed in order to relate Killing spinor fields on M (with
respect to one choice of Clifford module) to the chirality of the parallel spinor field on M. � Hence the sign
of the Killing constant and the chirality of the parallel spinor field are correlated. On the other hand, if
n is even, then ε is not fixed and for every parallel spinor field on M � we obtain a Killing spinor field on
M with either sign of the Killing constant, simply by making the right choice of ε.

8.2.5 The classification

We have just reduced the problem of which riemannian manifolds admit real Killing spinor fields to
which metric cones admit parallel spinors. We will assume that (M, g ) is complete and admits real
Killing spinor fields. Then since it is Einstein, Myers Theorem [CE75, Theorem 1.26] implies that it
is compact. Then a result of Gallot’s [Gal79, Proposition 3.1] says that if (M, g ) is in addition simply
connected, the cone (M, � g�) is either irreducible or flat. If the latter, (M, g ) is the round sphere; if the
former it is one of the geometries in Table 8.1.
Every geometry in Table 8.1 admits parallel forms, constructed via the holonomy principle from the
invariants under the holonomy representation and indeed constructed out of the parallel spinors. Since
in addition the manifold in question is a cone, and hence we have at our disposal also the Euler vector
field ξ, we can construct a number of geometric structures on the manifold M, which are listed in Table
8.2.5, where N± is the dimension of the space of Killing spinor fields with Killing constant ± 12 .
Spin 2010 (jmf) 59

Table 2: Simply-connected, complete riemannian manifolds with real Killing spinor fields
dim Geometry Cone (N+ , N− )
n round sphere flat (2�n/2� , 2�n/2� )
4k − 1 3-Sasaki hyperkähler (k + 1, 0)
4k − 1 Sasaki–Einstein Calabi–Yau (2, 0)
4k + 1 Sasaki–Einstein Calabi–Yau (1, 1)
6 nearly Kähler G2 (1, 1)
7 weak G2 Spin(7) (1, 0)

For example, if the cone is Calabi–Yau, then we have a parallel complex structure J. The vector field
χ = Jξ is orthogonal to ξ and it is the lift of a vector field on M, which we also denote χ. It is easy to
show that χ is a Killing vector and has unit norm. Its dual one-form θ is (the restriction to r = 1 of) the
contraction of the Euler vector into the Kähler form on the cone. The covariant derivative ∇χ defines a
skewsymmetric endomorphism T of the TM such that T(X) = ∇X χ. The fact that J is parallel means that

(115) (∇X T)(Y) = θ(Y)X − g (X, Y)χ .

The triple (χ, θ, T) defines a Sasakian structure on M, whence M is Sasaki–Einstein.

For another example, consider the case of a G2 -holonomy cone. We have a parallel 3-form φ into
which we contract the Euler vector field ξ to define a 2-form ω ∈ Ω2 (M): ω(X, Y) = φ(ξ, X, Y), evaluated
at r = 1. This defines an endomorphism J of TM by g (J(X), Y) = ω(X, Y). One can show that J is an
orthogonal almost complex structure. It is not parallel, but it satisfies (∇X J)(X) = 0 for all vector fields
X ∈ X (M). This defines a (non-Kähler) nearly Kähler structure on M.
These geometries defined via Killing spinors are presently under very active investigation, largely
due to their rôle in the gauge/gravity correspondence (see, e.g., [AFOHS98, MP99]).