Monografı́as de la Real Academia de Ciencias de Zaragoza. 29: 37–47, (2006).

Special Holonomy Manifolds in Physics

Luis J. Boya
Departamento de Fı́sica Teórica, Universidad de Zaragoza
E-50009 Zaragoza. Spain
e-mail: [email protected]

Abstract

This is a pedagogical exposition of holonomy groups intended for physicists.

After some pertinent definitions, we focus on special holonomy manifolds, two per
division algebras, and comment upon several cases of interest in physics, associated
with compactification from F -, M - and string theory, on manifolds of 8, 7 and 6
dimensions respectively.

1 Connections and Holonomy groups

Let (E, ∇) be a vector bundle with a connection ∇ : P (M, G) is the principal bundle,
and F supports a representation of G, so E(M, F ) is the associated bundle:

∇ : F ◦→ E → M (1)

∇ allows covariant differentiation of sections, e.g. for the tangent bundle E =

T M, ∇X Y = Z means: covariant derivation of vector field Y along (the flow of) X is
the vector field Z. In the general case ∇X ψ = ψ 0 , where ψ, ψ 0 are sections. ψ : M → E.

Connections allow also parallel transport along paths. E.g., a frame e at a point P
∈ M becomes e0 = g · e also at P, after a loop (closed path) γ from P, where g ∈ G.
Consider all the loops from P and write Hol(∇) := {g}; it is a (sub)group of G, called
the holonomy group of the connection; it was invented by E. Cartan in 1925. For arcwise
connected spaces, which is the case of manifolds, the holonomy group does not depend
(up to equivalence) on the starting point P.

Dedicado a José Cariñena en su 60 aniversario. Querido Pepı́n, parece increı́ble el largo camino
recorrido por los dos en los últimos cuarenta años. Barcelona, Valladolid, Salamanca, Zaragoza... El
jovencito en busca de un padrino de Tesis ha dado paso al respetable catedrático con un brillante historial.
De nada me siento tan orgulloso como de que tú y otros discı́pulos hayais superado tan bien al maestro.
Que sigas ası́, Pepı́n, pues aun esperamos muchos mucho de tı́.

37
 Let Hol0 (∇) be the restriction to contractible loops. Clearly there is an onto map

π1 (M ) → Hol(∇)/Hol0 (∇) (2)

The restricted holonomy group Hol0 (∇) is naturally connected, whereas Hol(∇) needs
not to be. For generic vector bundles the holonomy group is expected to be as large as
the structure group GL(F ). Two important theorems follow; first define

The curvature of a (vector bundle) connection ∇ is the operator on sections (e.g.

vector fields)

R(X, Y ) := [∇X , ∇Y ] − ∇[X,Y ] (3)

The curvature is a local property, the holonomy a global one. But both are related be
the Ambrose-Singer theorem (1953):

”The Lie algebra of the holonomy group is generated by the curvature”.

The other grand result is called the reduction theorem:

” The structure group can be reduced to the holonomy group”.

That is, the total space of the bundle can be restricted by the holonomy loops.

If the curvature is zero, the connection is said flat; the restricted holonomy group
Hol0 (∇) is then {e}. Parallellelizable spaces ( = trivial tangent bundle; they include S 7
and Lie groups) admit flat connections; just define the connection transport as transla-
tions in the (trivializable) tangent bundle.

We shall consider mainly connections in the tangent bundle of a manifold; then there
is another tensor, the torsion, defined as

T (X, Y ) := ∇X Y − ∇Y X − [X, Y ] (4)

Of course, the same space might have several inequivalent connections (e.g. S 3 has the
riemannian Levi-Civita connection, torsionless but curved, and the Lie-group connection
(S 3 ≈ SU (2)), flat but torsionful!).

We shall consider mainly riemannian manifolds (V, g); they enjoy the standard Levi-
Civita connection ∇ = ∇g in the tangent bundle, which is symmetric and isometric:

38
 (symmetric :)T orsion(∇) = 0 = ∇ · g(: isometric) (5)

Let Isom (V, g) be the isometry group of the manifold: ∗ ∈ Isom, means g∗ = g. A
generic riemanifold has no isometries, but the generic holonomy is the structure group,
O(n = dimV) or SO(n). Spaces with maximal isometries have constant curvature; for
example Isom (S n ) = O(n + 1), with constant curvature K > 0.

In physics both groups, isometry and holonomy, are important; for example, in the
Kaluza-Klein (de)construction, the gauge groups in the mundane space V4 come from
the isometry group U (1) of the compactification space S 1 : that is why electromagnetism
unfies with gravitation with a circle as fifth dimension, so in this case V5 = S 1 × V4 : the
original Kaluza construction, 1919.

However, in the supersymmetric situation, it is the holonomy group of the compactifi-

ation space which fixes the number of supersymmetries; for example, Calabi-Yau 3-folds
(CY3 , real dimension 6) are favoured for the dim 10 → 4 compactation of the Heterotic
Exceptional string, because the holonomy group, SU (3), allows just N = 1 Susy in our
mundane, 4D space, as we want.

Isometries measure, of course, symmetries, whereas holonomy measures distance (ob-

struction) from flatness; no apparent relation exists, except opposite genericity (as stated
above).

Simple examples in D = 2

The sphere S 2 has isometry O(3), the torus T 2 has U (1)2 ; other genus g > 1 surfaces
have no isometries. In the nonorientable cases, RP 2 and Klein bottle are the only ones
with isometries.

As for holonomy, the 2-Torus T 2 is the only CY1 among surfaces, because is a group
manifold, hence there is a connection with Hol = {e} and SU (1) ≡ {e}. The other sur-
faces with genus 6= 0 have Hol = U (1) = SO(2) (if orientable) and O(2) (if not).

Simple examples in D = 4

The ”round” sphere S 4 has O(5) as isometry, and a connection with SO(4) holonomy.
The 4-Torus T 4 is flat, with isometry U (1)4 . Intermediate is the topologically unique K3
(complex) surface (see later), which is a Calabi-Yau2 space, with dimR = 4, with SU (2)

39
holonomy but no isometries. As for CP 2 , it has U (3) as isometry group, and U (2) for
holonomy; in fact, CP 2 ≈ SU (3)/U (2).

As introductory material, the first book on modern differential geometry is still the
best [1].

Besides the original invention by E. Cartan (who did it in order to construct all sym-
metric spaces ca. 1925/26), and a short revival in the fifties (Berger, Lichnerowicz), the
study of holonomy languished until resuscitation in the mid-eighties, in part by imposi-
tion of physics (as in so many other mathematical questions!). Then Bryant, Salamon
and mainly Dominic Joyce (see book [2]) revitalized greatly the subject.

Finally, let us note that the holonomy groups come to the world with a particular ac-
tion (representation) in the tangent space, so one should properly speak of the holonomy
representation.

2 Special and Exceptional Holonomy Manifolds

What groups can appear as holonomy groups Hol(g) ⊂ O(n) of riemanifolds (V n , g)?
The issue was set and solved by M. Berger in 1955. To state precisely the problem, sup-
pose Hol(·) acts irreducibly in the tangent space, and symmetric spaces G/H are excluded
(because all are known (Cartan) and for them the subgroup H is the holonomy group).
Berger found all possible candidate groups with these prescriptions by a hard case-by-case
method.

Berger´s solution is best understood (Simons, 1962) as the search for transitive groups
over spheres: with two exceptions, these are the special holonomy groups.

The generic case is the orthogonal group acting trans on the sphere, O(n)◦ → S n−1 ,
with isotopy O(n − 1), that is S n−1 = O(n)/O(n − 1). The cases of trans action on
spheres coincident with special holonomy manifolds are (Berger´s list):

R O(n) or SO(n) acting on S n−1

C U (n) or SU (n) acting on S 2n−1
(6)
H Sp(n) · Sp(1) or Sp(n) acting on S 4n−1
O Spin(7) on S 7 or G2 on S 6

40
 We exhibit the association with the four division algebras R, C, H, y O, which is
obvious and remarkable. Recall also that the homology of compact simple Lie groups is
given by that of the product of odd-dimensional spheres, see e.g. [3]. Then the real and
complex cases are clear, for example SU (3) ≈ S 3 × S 5 , as homology sphere product, so we
have S 5 = SU (3)/SU (2). Sp(n) for us is the compact form of the Cn Cartan Lie algebra.
Also there is a ”nonunimodular” form

Sp(n) · Sp(1) := Sp(n) ×/2 Sp(1) (7)

As for the octonion cases, recall dim Spin(7) = 8, type (+1, real); in some sense which
we do not elaborate, it could be said that Spin(7) ”is” Oct(1), and G2 , defined as Aut(O),
is the ”unimodular” form, G2 ≈ SOct(1).

There are two more cases of trans actions on spheres

Sp(n) · U (1) := Sp(n) ×/2 U (1) acting in S 4n−1 (8)

and

Spin(9) acting in S 15 (9)

which, however, do not give rise to new holonomy groups. Spin(9) acts trans in S 15
as Spin(9) ≈ S 3 × S 7 × S 11 × S 15 . In fact, S 15 = Spin(9)/Spin(7), equivalent, in some
sense, to S 15 ≈ ”Oct(2)”/”Oct(1)”. Spin(9) was really in Berger´s list, but the only space
found was OP 2 (Moufang or octonionic plane), which is a symmetric space.

The sphere S 7 of unit octonions is singularized because there are four groups with
trans actions, O(8), U (4), Sp(2) and Spin(7) = ”Oct(1)”; similar for S 15 , but no more.

Notice the next Spin case, Spin(10): the action is not trans in the higher sphere, to
wit, dim Spin(10) = 16, complex, so Spin(10) acts on S 31 , but the sphere homology
product expansion for O(10) is S 3 × S 7 × S 11 × S 15 .

We expand now on the extant cases:

Over the reals we have the groups O(n), generic holonomy, and SO(n): clearly the
second obtains when the space is orientable and the connection oriented: there is an
obstructoin, the first Stiefel-Whitney class:

41
 V orientable manif old ↔ w1 = 0, w1 ∈ H 1 (V, Z2 ) (10)
Alternatively, the manifold V should have a global volume element (reduction GL(n, R)
← SL(n, R)). In Berger´s classification, he took the manifolds as simply connected, which
are then automatically orientable ( if π1 (V) = 0, all first order (co)homology vanishes,
including w1 ). Hence, O(n) did not appear in his list.

Over the complex we have complex manifolds, with structure group U (n); but a
generic hermitian metric h = g + iω will allow in general a connection with holonomy
SO(2n), as ∇g = 0 only, unless the complex structure J is also preserved: this is the case
of Kähler manifolds, with ∇ω(= dω) = 0, where ω = g(J) is the symplectic form.

The ”unimodular” restriction SU (n) obtains when the associated bundle with group
U (1) = U (n)/SU (n) is trivial, which is measured by the first Chern class:

SU (n) holonomy ↔ 0 = c1 ∈ H 2 (V, Z) (11)

The natural name for these spaces would be ”Special Kähler manifolds”, but had
become known instead as ”Calabi-Yau spaces”, after the conjecture of E. Calabi proven by
S.T. Yau. As a bonus, these spaces have trivial Ricci tensor (Ric = T r Riem, contraction
of the Riemann tensor): define the Ricci 2-form ρ := Ric(J); by the same token as above
J parallel (=covariant constant) implies ρ closed, and it turns out that [ρ] = 2πc1 (V).
Therefore

SU (n) holonomy implies Ricci f lat manif olds, Ric = 0. (12)

In other words, Calabi-Yau spaces are candidates to Einstein spaces, solution of vac-
uum Einstein equations (without cosmological term). Also CY spaces have an holomor-
phic volume element.

Over the quaternions we have quaternionic manifolds, like HP n , with ”nonunimodu-

lar” holonomy Sp(1)·Sp(n), and hyperkähler manifolds, with holonomy Sp(n). As clearly
Sp(n) ⊂ SU (2n), the later are also Ricci flat. In fact, the space HP n = Sp(n + 1)/Sp(1) ·
Sp(n), which a quaternionic space, is not valid as special holonomy manifold, because it
is a symmetric space. By contrast, quaternionic manifolds have not to be even Kähler, as
the example of S 4 = HP 1 shows; quaternionic and hyperkähler manifolds are not easy to
come by.

Manifolds with octonionic holonomy : the two groups: Spin(7) ⊂ SO(8), acting as
holonomy groups on 8-dim manifolds, and G2 ⊂ SO(7), in 7-dim. manifolds, are groups

42
associated to the octonions. Spin(7) ≈ ”Oct(1)” and G2 = Aut(O) ≈ ”SOct(1)”; mani-
folds with these holonomy groups are called exceptional holonomy manifolds. We expect
them to play some role in physcs, as for example the iternal spaces in F − and M −theories
have dimensions 8 and 7 bzw.

It is remarkable that, while O, U and Sp are isometry groups of symmetric regular

positive bilinear forms (O), sesquilinear regular positive forms (U ) and antisymmetric
regular forms (Sp), the octonionic cases obtain from invariance of certain 3−forms (G2 )
and selfdual 4-forms (Spin(7)). This cannot go beyond dimension 8, because for example
 
9
the dimension of 3-forms in 9-dimensions is 3
= 84, whereas dim GL(9, R) = 81.

In fact, the algebraic definition of Spin(7) is the isotopy group of certain class of self-
dual 4-forms in R4 : it preserves also orientation and euclidean metric, so Spin(7) ⊂ SO(8).
 
8
Notice the selfdual form is not generic, as 82 − 4
/2 = 29 > dim Spin(7) = 21; the special
selfdual four-form is called the Cayley form in the math literature. In any case Spin(7)
covers S 7 with isotopy G2 : 21 − 7 = 14.

The algebraic definition of G2 is this: the stability group of the generic 3-form in R7
as vector space: dim GL(7, R) − dim ∧3 R7 = dim G2 : 49 − 35 = 14. Of course, the
original characterization of G2 as Aut(O) by Cartan is related to this: a 3-form becomes
a T21 tensor through a metric, and this is indicative of an algebra, i.e. a bilinear map
R7 × R7 → R7 , given by the octonionic product (and restriction to the imaginary part).
There is also a sense in which for each division algebra there is a normal form and am
unimodular form:

Reals O(n) ≈ S 0 × S 3 × S 7 ... SO(n) ≈ S 3 × S 7 ...

R Generic Orientable, w1 = 0

Complex U (n) ≈ S 1 × S 3 × S 5 ... SU (n) ≈ S 3 × S 5 ...

C Kähler Calabi − Y au, c1 = 0
(13)
3 3 7 3 7
Quaternions Sp(n) · Sp(1) ≈ S × S × S ... Sp(n) ≈ S × S ...
H Quaternionic Hyperkähler

Octonions Spin(7) ≈ S 3 × S 7 × S 11 G2 ≈ S 3 × S 11
O dim 8 dim 7
NOTES 1). Today there are compact examples of all cases of special holonomy mani-

43
folds: big advances were made recently by Joyce [2], Salamon and others.

2). In dim 4, a remarkable case is the K3 manifold or Kummer surface (the name is
due to A. Weil, 1953, for Kummer, Kähler and Kodaira). It is the only CY manifold in
dimension four; it can be easily constructed (R4 → T 4 → orbif old → blow up; [4] ).
For a long time was the paradigmatic example of SU (n = 2) special holonomy.

3). Notice a generic complex n-manifold would have SO(2n) holonomy inspite the
structure group being U (n)!

4). The Calabi conjecture, proved by Yau, indicates the relation of the Ricci form
with the Kähler structure.

5). In the 80s a big industry, led by Phil Candelas in Austin ( [5]), was to find CY3
manifolds for string compactifications. Mirror symmetry was discovered in this context;
see later.

6). Except G2 , all special holonomy spaces are even dimensional.

7). One can show that G2 and Spin(7) holonomies are Ricci flat.

8). Although special holonomy representations are irreducible in the vector case, there
might be p-forms which split under the holonomy subgroup. For example, for G2 , 3-forms
split as 35 = 1 + 7 + 27; the 7 irrep is justly the octonion product, and the 1 the invariant
3-form. As for Spin(7), a self-4-form splits as 35 = 1 + 7 + 27: it includes the invariant
4-form.

3 Cases in Physics: dimensions 6, 7 and 8

In 1983, just after the first studies in eleven dimensional supergravity (11-dim SuGra),
Duff and Pope realized that it is the holonomy of the compactified space which deter-
mines the number of surviving Susy symmetries down to 4 dimensions. For spinor fields,
as S 7 = Spin(7)/G2 , 7-manifolds with exceptional G2 holonomy would have a surviving
spinor, hence N = 1 Susy down to 4-D. But after the String Revolution, 1984/85, the
descent 10 → 4 took over, and the favourite spaces were CY 3-folds: the heterotic string
has N = 1 supersymmetry in 10−dim., which means N = 4 down to earth; but it will
be 1/4 of these after CY3 compactation: the generic SO(6) holonomy of any (orientable)
dim-6 manifold would become SU (4) = Spin(6) after imposing a (necessary) spin struc-

44
ture, and then if we want one spinor to survive the group descends to SU (3). The 6-dim
manifold has to be orientable, spin, complex, Kähler and Calabi-Yau.

With the advent of M -Theory (1995), P. Townsend resurrected the idea of 7-dim
manifolds with G2 holonomy. One can go even further to Spin(7), the largest exceptional
holonomy group, by considering for example compactifications to 3-dim spaces (which
seems natural; for example, the series of noncompact symmetries of supergravity includes
E7 in 4 dimensions, which is claiming for E8 in three, which is of course the case). An-
other reason is F -Theory, which works in 12 dimensions with (2, 10) signature, and where
Spin(7) (perhaps in a nonpositive form) fits well.

Are such beasts as CY3 spaces in abundance? Yes, you can produce them in assembly
line, to the point of studying their Hodge numbers statistically! [5]. Another interesting
phenomenological constraint in the “old-fashion” 10 → 4 descent, was the Euler number
χ : it is related, via zeroes of the Dirac operator, to the number of generations, which is
| χ |/2.

As for the extension to F -Theory, we refer the reader to [7]. Besides some attempt to
state the particle content, the theory is rather stagnant at this point (as is M -Theory in
general). For a modern study of special holonomies with Lorentzian metrics, see [8].

4 Relation with Mirror Symmetry

Complex manifolds have a refinement on Betti numbers, as they separate in holo- or

antiholomorphic. The full expression of them is called the the Hodge diamond. For the
previous K3 manifold it is

h0,0 = 1
h1,0 = 0 h0,1 = 0
h2,0 = 1 h1,1 = 20 h0,2 = 1 (14)
h2,1 = 0 h1,2 = 0
h2,2 = 1

with bettis = 1, 0, 22, 0, 1. For Calabi-Yau 3-folds the diamond is bigger, but still
symmetric. A mirror pair X, Y of CY3 are two such spaces with

h1,1 [X] and h2,1 [X] equal h1,2 [Y ] and h1,1 [Y ] (15)

45
 There is no clear reason for this duality, but just another more example of a physics
discovery on pure mathematics. These two numbers measure very different invariants, so
mirror symmetry came up as a big surprise to mathematicians, when many conjectures
by physicists seemed to be true.

Indeed, several of these conjectures were true. It is also true that string theory ”glosses
over” orbifold singularities (i.e. quotienting manifolds by fix-point-action discrete groups),
and the associated (quantum) conformal field theories make perfect sense.

We shall only comment on the interpretation of this mirror symmetry by M. Kont-

sevich [6]. For him, the crux of the matter is a trade between complex and symplectic
geometry. In fact, complex structures preserve an imaginary unit, J ∈ End V, J 2 = −1,
whereas a symplectic manifold enjoys a 2-form ω, whose matrix form is very similar to
J!. In both cases, there is an extra condition:

In the complex case, N (J) = 0, where N stands for the Nihenhuis (obstruction); that
makes up a complex, not only almost-complex, manifold (the Newlander-Nirenberg theo-
rem). In the symplectic case, the 2-form is closed, or, alternatively, the (inverse) Poisson
bracket satisfies Jacobi´s identity. Their isotropy groups, GL(n, C) and Sp(n, R) respec-
tively, are of the same homotopy type, namely the homotopy of the intersection, U (n).
In any case, the relation hidden in Mirror Symmetry is an intrincate one. In the words
of Dijkgraaf: ”Mirror symmetry is the claim that the generating function for certian in-
variants of the symplectic structures on the 2-Torus S 1 × S 1 is a ”nice” function in the
moduli space of complex structures in the same”: the 2-Torus is a self-dual manifold for
Mirror Symmetry (MS).

The general case of MS is best understood in terms of toric varieties, which generalize
projective spaces.

The main lesson of MS for physics seems to be this: certain topology changes (some-
times called “flops”) are compatible with the underlined string theory. That probably
means that the complex-geometric description of strings is too fine... Perhaps it hints
towards a new type of duality.

46
References

[1] S. KOBAYASHI and K. NOMIZU, Foundations of Differential Geometry, 2 Vols. J.

Wiley, New York 1963

[2] D. JOYCE, Compact Manifolds with Special Holonomy. Oxford U.P., 2000

[3] L. J. BOYA, ”The Geometry of Compact Lie Groups”, Rep. Math. Phys. 30 (1991),
149-162.

[4] C. GOMEZ in ”Geometry and Physics” CIEMAT, 1998

[5] P. CANDELAS et al. ”Calabi-Yau manifolds in weighted spaces”, Nucl. Phys. B341
(1990), 383-402.

[6] M. KONSEVICH, talk at the Int. Math. Cong., Zürich 1997

[7] L. J. BOYA, ”Arguments for F-Theory”, Mod. Phys. Lett. A 21 (2006), 287-304.

[8] R. L. BRYANT, ”Pseudo-riemannian metrics with parallel spinor fields and vanishing
Ricci tensor”, arXiv:math.DG/0004073.

47
48