Vienna

LECTURES ON TWISTED K-THEORY AND ORIENTIFOLDS
DANIEL S. FREED
This is an improved version of informal supplementary notes for lectures delivered in June, 2012
at the Erwin Schrödinger International Institute for Mathematical Physics as part of a graduate
workshop in the program K-Theory and Quantum Fields. This is not a record of the actual lectures,
but rather is in large part supplementary reading and exercises. These notes will also be the basis
of a formal paper on some of this material, but that will necessarily contain less exposition, so
we are making these informal notes available. We do not make any attempt to assemble even a
representative list of references in these lecture notes; we do so in the published papers. There were
four lectures, but these notes are divided into three lectures as is more natural.
The goal of the lectures is to give an overview of some aspects of ongoing joint work with
Jacques Distler and Greg Moore about topological aspects of superstring theory. The inclusion of
“orientifolds” has especially subtle topological features, and it is these that we focus on. There
are two papers1,2 about this work so far and several more to follow. These theories are known
in the physics literature as “Type II”; the “Type I” theories are a special case of the orientifold
construction, so are included here. The “heterotic” string is not part of this discussion, nor is
“M-theory”, “F-theory”, or other variants of string theory with supersymmetry.
Lecture 1 is a purely mathematical discussion of twistings of K-theory, with a little about twisted
K-theory. The simplest twisting of a cohomology theory is by degree shift, which is not always
considered a twisting. The second simplest is by a double cover, and this sort of twisting exists
for any cohomology theory. These two types of twistings are connected by a nontrivial k-invariant,
and this already justifies including the degree as part of the twisting. For K-theory there is also a
twisting by a “gerbe”, and it is this one which is usually discussed in isolation. In fact, there is a
whole tower of higher twistings, but they do not enter these lectures. These three “lowest” twistings
of K-theory are all connected by nontrivial k-invariants, and we describe a geometric model which
encodes them. It is based on the Donovan-Karoubi paper as well as on work with Mike Hopkins
and Constantin Teleman. We describe the classifying spectrum for these twistings and some maps
relating the real and complex cases.
Lecture 2 is an overview of the fields in the Type II superstring. In fact, we spend much of the
lecture on the oriented bosonic string to illustrate the important ideas of orbifolds and orientifolds.
Both are constructed by “gauging a symmetry”. We interpret fields in field theory as simplicial
sheaves, and in that context the natural quotient construction by a symmetry group is precisely
Date: June 15, 2012.

1J. Distler, D. S. Freed, G. W. Moore, Orientifold précis, in “Mathematical Foundations of Quantum Field
Theory and Perturbative String Theory”, Proceedings of Symposia in Pure Mathematics, 83 (2011), 159–172,
(arXiv:0906.0795).
2J. Distler, D. S. Freed, G. W. Moore, Spin structures and superstrings, in Perspectives in Mathematics
and Physics: Essays dedicated to Isadore Singer’s 85th birthday, Surveys in Differential Geometry, 15 (2010),
(arXiv:1007.4581).
1
2 D. S. FREED
the physicists’ gauging. We emphasize the “B-field”, which is different in the two string theories.
Also noteworthy is the description of spin structures in the superstring; there is a more detailed
discussion in 2 . We also include a general discussion of the notion of a field. Our use of the term
is broader than usual: it includes topological structures such as orientations, for example.
Lecture 3 is a sketch of the anomaly cancellation on the worldsheet of the Type II superstring
(with orientifold). There is an anomaly in the functional integral over the fermionic fields, as usual:
it is the pfaffian of a Dirac operator, which is a section of a Pfaffian line bundle. In this case that
bundle is flat and the line bundle has finite order. So the anomaly is a subtle torsion effect. The
B-field amplitude is also—somewhat surprisingly—anomalous. The theorem states that these two
anomalies cancel. The key ingredient in the cancellation is the twisted spin structure on spacetime.
We remark that even in oriented Type II superstring (no orientifold) there is no anomaly in the
B-field amplitude, and the trivialization of the Pfaffian line bundle uses the spin structure on
spacetime, though we do not know an explicit construction. In fact, what is lacking here, as in
other anomaly arguments, is a categorified index theorem which would construct a trivialization of
the anomaly line bundle, not just prove that one exists, which is what we do here. The entire story
depends very heavily on our choice of Dirac quantization condition for the B-field.
Contents
Lecture 1: Models of twistings 3
Introduction 3
Examples of twistings and twisted K-theory 5
Invertible super algebras, bimodules, and intertwiners 8
The algebraic topology of Alg ˆ 10
The spectrum R and cAlgCˆ 12
Bundles of invertible algebras 15
Twistings of K-theory 18
Twisted vector bundles and Fredholm operators 20
Twistings of KR-theory 22
Differential twistings 24
The classifying spectrum of cAlgRˆ and the transfer map 27
Lecture 2: Fields and superstrings 35
General discussion of fields 35
Quantization of charge 37
The oriented bosonic string 40
Orbifolds in string theory and in geometry 42
Orientifolds of the oriented bosonic string 45
Fields in the oriented Type II superstring 48
The complete Type II superstring 52
Lecture 3: Worldsheet anomalies 55
Digression: a categorified index theorem 56
A twisted R-class on Σ 56
The Pfaffian line 59
q
More twisted R-classes on Σ 61
The anomalous B-field amplitude 65
Putting it all together 65
JUNE, 2012 ESI LECTURES 3
Lecture 1: Models of twistings
In this lecture we give concrete models for twistings of K-theory and its cousins (KO-theory, KR-
theory). The model is based on joint work3 with Michael Hopkins and Constantin Teleman and also
on the original paper of Donovan and Karoubi.4 Isomorphism classes of the twistings we need for the
B-field in superstring theory are classified by a spectrum with three nonzero homotopy groups. The
geometric model we use involves bundles of invertible complex Z{2Z-graded algebras, bimodules,
and intertwiners. It has as special cases many other models which have appeared in the literature
(including 3 , which only uses two homotopy groups). We also give a concrete model for geometric
representatives of twisted K-theory classes. There is a brief discussion of differential twistings as
well. Differential twistings and differential twisted K-theory classes appear5 in our description of
superstring theory, but in subsequent lectures we work in a model-independent manner as nothing
we do depends on a particular model, though we do use our model for some computations. There
is also a notion of a hermitian structure on a twisting, analogous to a hermitian structure on a line
bundle, but we do not describe it here. We end the lecture with several important constructions
and formulas. The arguments we use go back and forth between the explicit models and general
algebraic topology.
Introduction
(1.1) Twisted real cohomology. Twisted versions of cohomology—and the geometric objects which
represent cohomology classes—arise in many situations. For example, on manifolds without an
orientation there is cohomology twisted by the orientation bundle. In topology it appears in the
statement of Poincaré duality. In geometry real cohomology is represented by differential forms,
and forms of the top degree twisted by the orientation bundle are densities, which are the objects
on a manifold which can be integrated. Concretely, if X is a smooth manifold untwisted real
cohomology is computed by the de Rham complex
d d
(1.2) Ω‚ pXq : Ω0 pXq ÝÑ Ω1 pXq ÝÑ ¨ ¨ ¨
Let L Ñ X be a real line bundle with a flat covariant derivative. Then we can consider differential
forms with values in L, which computes twisted cohomology:
d d
(1.3) Ω‚ pX; Lq : Ω0 pX; Lq ÝÑ Ω1 pX; Lq ÝÑ ¨ ¨ ¨
Features to notice: (i) Ω‚ pXq has a ring structure, but Ω‚ pX; Lq does not; (ii) Ω‚ pX; Lq is a rank 1
module over Ω‚ pXq; (iii) the twisting L has a nontrivial automorphism—multiplication by ´1—and
it induces a nontrivial map on Ω‚ pX; Lq and on the twisted cohomology.
3D. S. Freed, M. J. Hopkins, C. Teleman, Loop groups and twisted K-theory I , J. Topology, 4 (2011), 737–798
(arXiv:0711.1906 [math.AT]).
4They do not consider covers by locally equivalent groupoids, so do not realize all twistings with finite dimensional
invertible algebra bundles, only those whose isomorphism class is torsion. In other approaches this is obviated by
generalizing to infinite dimensional algebras.
5As we do not discuss Ramond-Ramond fields except in passing, there is very little about (differential) twisted
K-theory; rather, we focus on the twistings.

4 D. S. FREED
We remark that vector bundles of rank greater than 1 also twist real cohomology, but the kinds
of twistings of K-theory we discuss are rank 1.
(1.4) Twisted vector spaces. K-theory is made from a more primitive source than ordinary coho-
mology: linear algebra. The objects of interest in linear algebra over a point are vector spaces; for
linear algebra over a space X they are vector bundles. Twistings are made by finding a “free rank 1
module” for the notion of a vector space. (For complex K-theory the ground field is C; for real
KO-theory the ground field is R.) As a first example, suppose A is an algebra over the complex
numbers, such as the algebra of n ˆ n matrices. Then in lieu of a complex vector space we can
consider an A-module. Over a space X we would then have a complex vector bundle A Ñ X whose
fibers are complex algebras, and in lieu of vector bundles E Ñ X we consider vector bundles which
are fiberwise A-modules: there is an action map A b E Ñ E which satisfies the usual associativity
property a1 pa2 ¨ eq “ pa1 a2 q ¨ e. We consider only finite rank algebras, but elaborate on this idea to
obtain more general notions of twisting.
More formally: Vector spaces are the objects of a symmetric monoidal category Vect and twisted
vector spaces are the objects of a module category for Vect.
Question: What property of the algebra A tells that the collection of A-modules is free of rank 1
over Vect? Answer: A is invertible in a certain precise sense, and this translates to the classical
notion that A is a central simple algebra.
For ordinary vector spaces over C this leads to nothing new: all invertible algebras are equivalent
to the ground field C and so rank 1 modules over VectC are equivalent to VectC . The situation
is more interesting over R where up to equivalence there is a nontrivial possibility: the algebra H
of quaternions. We work in the richer Z{2Z-graded, or super, world. This is a manifestation of
the basic notions of group completion and stability in K-theory, which in the old literature is
expressed in terms of formal differences E 0 ´ E 1 of vector spaces. Formally, we use the symmetric
monoidal category of Z{2Z-graded vector spaces E “ E 0 ‘ E 1 with the usual tensor product and
the symmetry
E b E 1 ÝÑ E 1 b E
(1.5) 1
e b e1 ÞÝÑ p´1q|e||e | e1 b e,
where |e| P t0, 1u denotes the parity of the homogeneous element e P E. For simplicity we still
use ‘Vect’ to denote this category. Now there are 2 equivalence classes of complex invertible super
algebras and 8 equivalence classes of real invertible super algebras. They are all represented by
Clifford algebras.
(1.6) Clifford algebras. Let pV, Qq be a finite dimensional (ungraded) real vector space V with
a nondegenerate quadratic form Q (equivalently, a nondegenerate symmetric bilinear form). The
Clifford algebra CℓpV, Qq is the free associative real algebra with identity 1 such that v 2 “ Qpvq ¨ 1.
It is naturally Z{2Z-graded. For n P Z“0 set Cℓn “ CℓpRn , Qn q, where Qn is the standard quadratic
form
` ˘
(1.7) Qn pξ 1 , . . . , ξ n q “ ˘ pξ 1 q2 ` ¨ ¨ ¨ ` pξ n q2
with the sign chosen according to the sign of n. By convention Cℓ0 “ R. There are also complex
Clifford algebras CℓC pV, Qq “ CℓpV, Qq b C. The Clifford algebra is invertible in the super sense,
and super modules over a Clifford algebra are a free rank 1 module over super vector spaces. So
this is one possibility for a twisted notion of vector space to make twisted K-theory.
Examples of twistings and twisted K-theory
(1.8) Real vector bundles. Let X be a nice topological space (locally contractible, paracompact,
completely regular), for example, a smooth manifold. Let V Ñ X be a real vector bundle of
finite rank, and assume for simplicity that V is endowed with a positive definite quadratic form Q.
Let CℓpV, Qq Ñ X denote the associated bundle of Clifford algebras. It leads to a twisted notion
of vector bundle: a CℓpV, Qq-twisted vector bundle is a real Z{2Z-graded vector bundle E Ñ X
which fiberwise is a left module over CℓpV, Qq. We can also consider complex Z{2Z-graded vector
bundles, in which case we may as well use the complexification CℓC pV, Qq. When X is a Riemannian
manifold this applies to the tangent bundle V “ T X.
Let VectpXq denote the category of super vector bundles over X and VectCℓpV,Qq pXq the cat-
egory of CℓpV, Qq-twisted vector bundles. The bundle CℓpV, Qq Ñ X is a left CℓpV, Qq-module
by pointwise multiplication—a special twisted vector bundle called the Euler class of V Ñ X. It
may happen that CℓpV, Qq Ñ X is “equivalent” to a bundle of algebras with constant fiber a fixed
Clifford algebra; we define the proper notion of equivalence below. Over the reals we need that the
bundle V Ñ X carry a spin structure; over the complexes V Ñ X must carry a spinc -structure.
We revisit this in (1.92).
(1.9) Twisted equivariant vector bundles. Now suppose G is a compact Lie group and X a nice
G-space. There is a symmetric monoidal category VectG pXq of equivariant super vector bundles
E Ñ X. One way to define a twisted notion is to specify a central extension of G. Over the reals
we take the center to be t˘1u, the multiplicative group of real numbers of unit norm; over the
complexes the center is the circle group T, the multiplicative group of complex numbers of unit
norm. Let
(1.10) 1 ÝÑ t˘1u ÝÑ Gτ ÝÑ G ÝÑ 1
be a central extension in the real case. We use ‘τ ’ to denote the central extension (1.10). Then
a τ -twisted super bundle E Ñ X carries an action of Gτ which covers the action of G on X.
Furthermore, we require that the central element ´1 P t˘1u, which covers the identity map of X,
act on each fiber as scalar multiplication by ´1. Example: Let X “ T with the action of G “ Z{2Z
by a half-turn x ÞÑ ´x for x P T. Then the Möbius line bundle E Ñ X (whose grading is purely
even) carries an action of the central extension
(1.11) 1 ÝÑ t˘1u ÝÑ Z{4Z ÝÑ Z{2Z ÝÑ 1
and so is a twisted equivariant bundle.

6 D. S. FREED
For another kind of twisting we specify a homomorphism
(1.12) ǫ : G ÝÑ Z{2Z
An ǫ-twisted super bundle E “ E 0 ‘ E 1 Ñ X carries an action of G in which g P G is an even

transformation if ǫpgq “ 0 and is an odd transformation if ǫpgq “ 1. Recall that a transformation
is even if it preserves the grading E 0 ‘ E 1 and odd if it reverses it. A group equipped with a
homomorphism (1.12) is called a Z{2Z-graded group.
We have three kinds of twisting in the equivariant case specified by a triple pA, ǫ, τ q and it is
natural to combine them. We use the Koszul sign rule (1.5) strictly. Notice that we can take X “ pt
to obtain twisted notions of the category VectG of Z{2Z-graded representations of G.
(1.13) Examples. Let G “ Z{2Z be nontrivially graded by the identity map ǫ : Z{2Z Ñ Z{2Z.
Take E “ R1|1 the Z{2Z-graded real vector space whose even and odd subspaces are each the
trivial real line R. Define
ˆ ˙ ˆ ˙
0 ´1 0 1
(1.14) x“ , γ“ .
1 0 1 0
Notice that each matrix represents an odd endomorphism of E and they commute in the super
sense of the sign rule (1.5). (This last point is crucial: check it in detail!) We interpret it in two
ways. First: γ generates the Clifford algebra A “ Cℓ1 and x is the action of the generator of the
nontrivial central extension (1.11) of G “ Z{2Z. So E is a real representation of Z{2Z twisted in all
3 ways: it is a module over Cℓ1 , the group Z{2Z is Z{2Z-graded, and there is a nontrivial central
extension. Second interpretation: x generates the Clifford algebra A “ Cℓ´1 and γ is the action
of the generator of G “ Z{2Z. In this interpretation the representation is a module over Cℓ´1 and
Z{2Z is nontrivially Z{2Z-graded, but there is no central extension. We revisit these twisted KO-
classes in (1.142). We remark that the first is the KO-theory Euler class of the sign representation
of the group G “ Z{2Z.
Exercise 1.15. Use the graded tensor product to define powers of both the module and the
twisting. What is the product of the elements described by the two distinct interpretations? What
is the square of the element in the first interpretation? (Very interesting point: the answer to the
latter is in some sense the Bott element u P K 2 pptq.)
(1.16) A complex example with two morals. The previous examples were over the reals. To twist
over C in the equivariant situation we want a central extension by T. If G is a finite dimensional Lie
group every example factors through a central extension by a cyclic subgroup of T. For example,
the central extension
(1.17) 1 ÝÑ T ÝÑ Spincn ÝÑ SOn ÝÑ 1
of the special orthogonal group factors through the extension
(1.18) 1 ÝÑ t˘1u ÝÑ Spinn ÝÑ SOn ÝÑ 1

To obtain something nontrivial we pass to an infinite dimensional example.

Let G be a compact Lie group and fix a principal G-bundle P Ñ S 1 . Let A denote the affine space
of connections on P and G the group of gauge transformations, i.e., the group of automorphisms
of P Ñ S 1 which cover idS 1 . If P “ S 1 ˆ G is the trivial bundle then G “ LG is canonically
the loop group of maps LG “ MappS 1 , Gq. It admits nontrivial central extensions by T, and so
defines a twisted notion of equivariant complex super vector bundle over A. This becomes much
more interesting when we connect the G-space A with finite dimensional geometry. Fix a basepoint
on P ; its projection to S 1 is a basepoint on the base. Then there is a homomorphism G Ñ G
which evaluates a gauge transformation on the fiber containing the basepoint. The holonomy is an
equivariant map A Ñ G, where the group G acts on the space G by conjugation. Below we will
interpret this as a local equivalence of groupoids
(1.19) A{{G ÝÑ G{{G.
This is a global construction of canonical twistings of the equivariant complex K-theory group KG pGq.
Moral 1: To define interesting twistings we must be allowed to pass to locally equivalent

groupoids.
We will not pursue this here, but a positive energy representation of the central extension of the
loop group determines an equivariant family of Fredholm operators parametrized by X. This takes
the place of the finite rank equivariant super vector bundles of the previous examples. See 6 for
details.
Moral 2: To define interesting “bundles” we should allow infinite rank bundles with a Fredholm
operator and not restrict ourselves to finite rank bundles (with the zero operator).
(1.20) Other motivation. The model we present covers all of these examples and many more.
Twisted notions of vector spaces and vector bundles arise in many other circumstances. Here are
two more.
Let
(1.21) 1 ÝÑ G1 ÝÑ G ÝÑ G2 ÝÑ 1
be an extension of (topological) groups and let X be a space of isomorphism classes of represen-

tations of G1 which is stable under the action of G2 . In other words, if ρ : G1 Ñ AutpW q is a
representation and g P G then there is a new representation ρ̃pg 1 q “ ρpgg 1 g ´1 q, g 1 P G1 . Its isomor-
phism class only depends on the image of g in G2 . We ask that if the isomorphism class of ρ lies
in X, then so too does the isomorphism class of ρ̃ for all g P G. In this situation a suitably finite
representation of G determines a vector bundle E Ñ X whose fibers are the multiplicity spaces of
the restriction as a representation of G1 . The group G2 does not in general lift to act on E, but
6D. S. Freed, M. J. Hopkins, C. Teleman, Loop groups and twisted K-theory II , (arXiv:math.AT/0511232).
8 D. S. FREED
rather there is a twisting of the groupoid X{{G2 which is defined and which acts. This is worked
out in 3 and also in 7.
The second situation is a general story in quantum mechanics. The space of pure states in a
quantum mechanical system is the projective space PH of a complex separable Hilbert space H,
and it carries a natural symmetric function p : PH ˆ PH Ñ r0, 1s which encodes “transition proba-
bilities”. A fundamental theorem of Wigner asserts that every symmetry of pPH, pq lifts to either a
unitary or antiunitary transformation of H. If we denote the group of these transformations of H
as Autqtm pHq, then Wigner’s theorem is encoded in the group extension
(1.22) 1 ÝÑ T ÝÑ Autqtm pHq ÝÑ AutpPH, pq ÝÑ 1
The subgroup T of scalar transformations acts trivially on projective space, but it is not central as
scalars do not commute with antiunitary transformations. This is a twisted central extension, and
our first example of a twisting which is relevant to KR-theory.
Invertible super algebras, bimodules, and intertwiners

We begin by defining the twistings we use for the K-theory of a point, which is to say the twisted
notion of a super vector space.
(1.23) Preliminaries. Let k be a field, which in our application will always be R or C. Let
A “ A0 ‘ A1 be a super algebra.. A homogeneous element z in its center satisfies za “ p´1q|z||a| az
for all homogeneous a P A. The center is itself a super algebra, which is of course commutative
(in the Z{2Z-graded sense). The opposite super algebra Aop to a super algebra A is the same
underlying vector space with product a1 ¨ a2 “ p´1q|a1 ||a2 | a2 a1 on homogeneous elements. All
algebras are assumed unital. Tensor products of super algebras are taken in the graded sense.
Undecorated tensor products are over the ground field. Unless otherwise stated a module is a left
module. An ideal I Ă A in a super algebra is graded if I “ pI X A0 q ‘ pI X A1 q.
Example 1.24. Let S “ S 0 ‘ S 1 be a finite dimensional super vector space over k. Then End S is
a central simple super algebra. Endomorphisms which preserve the grading on S are even, those
which reverse it are odd. A super algebra isomorphic to End S is called a super matrix algebra.
Exercise 1.25. Show that the opposite of the Clifford algebra CℓpV, Qq is CℓpV, ´Qq. In particular,
Cℓn and Cℓń are opposites.
Definition 1.26. A finite dimensional super algebra A over a field k is central simple if its center
is k and its only graded 2-sided ideals are 0 and A.8
Central simple algebras in the super case were investigated by Wall and Deligne. We summarize
some of their main results by recasting them as Theorem 1.28 below.
7D.S. Freed, M. J. Hopkins, C. Teleman, Loop groups and twisted K-theory III , Annals of Math., Volume 174
(2011), pp. 947–1007, (arXiv:math.AT/0312155).
8We also assume there exist a , a P A with a a “ 0 to exclude one-dimensional algebras with the zero product
1 2 1 2
map.
(1.27) 2-category of algebras-bimodules-intertwiners. If A0 , A1 are super algebras, then an pA1 , A0 q-

bimodule is a Z{2Z-graded left A1 b Aop 0 module, a super vector space which is a simultaneous left
module for A1 and right module for A0 . If B, B 1 are pA1 , A0 q-bimodules, then an intertwiner
f : B Ñ B 1 is a linear map which commutes with the actions of A0 and A1 . If α : A0 Ñ A1 is a
homomorphism of super algebras, then A1 is an pA1 , A0 q-bimodule for which pa1 , a0 q acts on a P A1
to yield a1 ¨ a ¨ αpa0 q. In this way bimodules generalize homomorphisms of super algebras.
There is a 2-category Alg “ Algk whose objects are super k-algebras, 1-morphisms are bimodules,
and 2-morphisms are intertwiners. Composition of 1-morphisms is via tensor product: if B is an
pA1 , A0 q-bimodule and B 1 an pA2 , A1 q-bimodule, then B 1 bA1 B is an pA2 , A0 q-bimodule. For any
super algebra A the super vector space A is an pA, Aq-bimodule which represents the identity map
in Alg. Also, Alg is a symmetric monoidal category via tensor product (over k) of super algebras,
bimodules, and intertwiners. Our interest is in the sub-2-category Alg ˆ “ Algkˆ of invertible objects
and morphisms in Alg “ Algk .
An object A P Alg is invertible if it has an inverse under tensor product. Thus there exists
A P Alg and bimodules B : k Ñ A b A1 and B 1 : A b A1 Ñ k which are inverse isomorphisms. Since
1
we are in a 2-category, this means there exist intertwiners f : B 1 bAbA1 B Ñ k and g : BbB 1 Ñ AbA1
which are isomorphisms: f is an isomorphism of k-vector spaces and g is an isomorphism of
pA b A1 , A b A1 q-bimodules.
Theorem 1.28. A super algebra A P Alg is invertible if and only if A is (finite dimensional)
central simple.
Isomorphic algebras in Algkˆ are called Morita equivalent and an invertible bimodule which gives
the isomorphism defines a Morita equivalence. The group of isomorphism classes of objects in Algkˆ
is called the super Brauer group of the field k.
Exercise 1.29. Show that any super matrix algebra is Morita equivalent to the trivial algebra k.
Prove the converse as well.
Theorem 1.30. The super Brauer group of C is cyclic of order 2 and the super Brauer group of R
is cyclic of order 8. In each case it is generated by the Clifford algebra CpL, Qq, where L is a line
and Q a positive definite quadratic form.
In particular, a complex central simple super algebra is either a super matrix algebra or a matrix
algebra tensor the graded algebra C ‘ Cu1 with pu1 q2 “ 1. In the latter case the matrix algebra
can be taken to be even.
For 1-morphisms in Alg invertibility amounts to the following: an pA1 , A0 q-bimodule B is invert-
– –
ible if there exists an pA0 , A1 q-bimodule B 1 and isomorphisms B 1 bA1 B Ý Ñ A0 and B bA0 B 1 ÝÑ A1
as pA0 , A0 q- (respectively pA1 , A1 q-) bimodules. An invertible pk, kq-bimodule is a Z{2Z-graded line,
a 1-dimensional k-vector space which is either even or odd. For any central simple super algebra A
the 1-groupoid of invertible pA, Aq-bimodules is canonically the groupoid of Z{2Z-graded lines: a
Z{2Z-graded line L corresponds to the pA, Aq-bimodule A b L.
(1.31) Morita equivalence and modules. Suppose A0 , A1 are invertible super algebras which are
Morita equivalent, and let B an pA1 , A0 q-bimodule and B 1 an pA0 , A1 q-bimodule be invertible
10 D. S. FREED
bimodules which implement the Morita equivalence. Let A-Mod denote the category of left A-
modules. Then the functors
A0 -Mod ÝÑ A1 -Mod
(1.32)
E ÞÝÑ B bA0 E
and
A1 -Mod ÝÑ A0 -Mod
(1.33)
E ÞÝÑ B 1 bA1 E
are inverse equivalences. In other words, Morita equivalent super algebras define equivalent notions
of twisted vector spaces.
The algebraic topology of Alg ˆ
(1.34) Groupoids and Picard groupoids. Recall that a category C has a classifying space BC. In
this construction morphisms in the category become paths in the classifying space, and so are
invertible up to homotopy: the homotopy inverse of a parametrized path is retreat along the
same trajectory. A groupoid C is a category in which all morphisms are already invertible, so the
classifying space construction does not invert any more. In this case the classifying space BC does
not have any higher homotopy groups, only nonzero π0 , π1 . In the special case that the groupoid is
a group π—that is, there is a single object ˚ and the π “ Autp˚q—then the classifying space has the
single nonzero homotopy group π1 “ π. In other words, the classifying space Bπ is the Eilenberg-
MacLane space Kpπ, 1q. Returning to a general groupoid C, if in addition C has a symmetric
monoidal structure such that every object has an inverse, then C is called a Picard groupoid. In
this case π0 , π1 are abelian groups and the classifying space BC is an infinite loop space. Infinite
loop spaces correspond to spectra, and as for spaces, spectra have a Postnikov decomposition. A
Picard groupoid has only two nonzero homotopy groups, so the Postnikov tower is quite simple:
(1.35) ΣHπ1 BC
k
Hπ0 Σ2 Hπ1
Here Hπ is the Eilenberg-MacLane spectrum which has the single nonzero homotopy group π in
degree 0. Then ΣHπ is the shift which has the nonzero homotopy group in degree 1, the de-looping
of Hπ. (In the special case C has one object with automorphism group the abelian group π, then
BC is the 0-space of the spectrum ΣHπ.) The bottom arrow in (1.35) is called the k-invariant of
the spectrum and it tells how the spectrum is glued together from the Eilenberg-MacLane building
blocks. The k-invariant is a stable cohomology operation H q p´; π0 q Ñ H q`2 p´; π1 q, which is
equivalent to a homomorphism π0 b Z{2Z Ñ π1 . Example: If π0 “ π1 “ Z{2Z, then there is a
single nonzero stable operation Sq 2 : H q p´; Z{2Zq Ñ H q`2 p´; Z{2Zq, the Steenrod square. The k-
invariant can be computed directly from the Picard groupoid C as follows. First, observe that if 1 P C
is the identity object, then for any x P C there is an isomorphism Autp1q – Autp1 b xq – Autpxq
which maps f ÞÑ f b idx . Then the homomorphism π0 b Z{2Z Ñ π1 is the image of the symmetry
(1.36) σx : x b x ÝÑ x b x
under the isomorphism Autpx b xq – Autp1q.
Exercise 1.37. Check that this does indeed define a homomorphism π0 b Z{2Z Ñ π1 .
Twistings, in particular Alg ˆ , form a Picard 2-groupoid. This is a 2-category, which for our
purposes means a category C in which the morphism spaces Cpx, yq are also categories. An ordi-
nary morphism x Ñ y is called a 1-morphism and a morphism in Cpx, yq is called a 2-morphism.
The classifying space BC of a Picard 2-groupoid is an infinite loop space whose only nonzero ho-
motopy groups are π0 , π1 , π2 . There is a filtration whose associated graded consists of two Picard
1-groupoids: the bottom one has the same objects as C and replaces Cpx, yq by π0 Cpx, yq; the top
one is the 1-groupoid Autp1q. These two groupoids do not in general determine C. (In general, there
are extensions when passing from an associated graded to the original filtered object.) However, in
the situation here they are enough to do so; see Lemma 1.53.
(1.38) The 2-groupoid Alg ˆ . We compute the homotopy groups π0 , π1 , π2 for Alg ˆ over the
fields k “ C, R. First, π0 is the super Brauer group of equivalence classes of invertible super
algebras, the computation of which is recalled in Theorem 1.30. Next, π1 , π2 are equal to π0 , π1
of the 1-groupoid Autp1q whose objects are Z{2Z-graded lines. There are two isomorphism classes
of objects—the trivial line k and the odd line Π (generated by an odd element π)—and Πb2 – k.
The tensor unit in Autp1q is the trivial line k and its automorphism group is k ˆ . So we obtain the
following table of nonzero homotopy groups:
AlgCˆ AlgRˆ
π2 Cˆ Rˆ
(1.39)
π1 Z{2Z Z{2Z
π0 Z{2Z Z{8Z
Here Cˆ and Rˆ have the discrete topology. There is a notion of a hermitian structure on algebras
and bimodules, and then a corresponding 2-groupoid of unitary algebras-bimodules-intertwiners
whose π2 is T and Z{2Z in the complex and real cases, respectively.
We are interested in a homotopy classification in which we use the continuous topology on in-
tertwiners. By analogy consider the group Cˆ and maps into it from a nice topological space X.
If Cˆ is discrete there is no notion of equivalence, and the set of continuous maps X Ñ Cˆ
is H 0 pX; Cˆ q. If we use the continuous topology on Cˆ , then homotopy classes form the cohomol-
ogy group H 1 pX; Zq. The Bockstein map H 0 pX; Cˆ q Ñ H 1 pX; Zq connects the two interpretations.
12 D. S. FREED
Similarly, for the topological classification of twistings we use the continuous topology on intertwin-
ers, and so obtain the following table of nonzero homotopy groups, where the initial ‘c’ reminds of
the continuous topology:
cAlgCˆ cAlgRˆ
π3 Z 0
(1.40) π2 0 Z{2Z
π1 Z{2Z Z{2Z
π0 Z{2Z Z{8Z
Remark 1.41. The inclusion of π0 amounts to including the degree in K-theory as part of the
twisting. This is a natural idea and even in ordinary cohomology an alternative way to encode
degree. It was built into our discussion from the beginning in (1.4). From this point of view the
periodicity of K-theory is built-in and arises much as in the classic paper of Atiyah-Bott-Shapiro.
(1.42) k-invariants. Next, we compute the k-invariants between the consecutive groups in (1.40).
We focus on the complex case; the arguments in the real case are similar. The k-invariant on
the bottom is a map HZ{2Z Ñ Σ2 HZ{2Z, and as remarked in (1.34) the only nonzero possibility
is Sq 2 . To compute whether this occurs we take a representative of the nonzero element in π0 , which
is the Clifford algebra A “ CℓC1 with a single odd generator e. The symmetry A b A Ñ A b A is
the involution which fixes 1 b 1, changes the sign of e b e, and exchanges 1 b e Ø e b 1. As stated
in (1.27) it is represented by the vector space A b A as an pA b A, A b Aq-bimodule using the
involution to modify the right multiplication.
Exercise 1.43. Define an isomorphism of this bimodule with the bimodule A b A b Π in which the
left and right A b A actions are by multiplication on A b A tensor the identity on Π. This shows
that the k-invariant is nonzero.
The other k-invariant is the standard one in the category Autp1q of Z{2Z-graded lines. It is the
nonzero map Z{2Z Ñ Cˆ (see (1.39)) as can be seen from the fact that the symmetry ΠbΠ Ñ ΠbΠ
is multiplication by ´1. The corresponding k-invariant in (1.40) is the stable cohomology operation
(1.44) βZ ˝ Sq 2 : H q p´; Z{2Zq ÝÑ H q`3 p´; Zq,
where βZ is the integer Bockstein map.
The spectrum R and cAlgCˆ

Our goal here is to recognize the classifying space of cAlgCˆ , which is the 0-space of a spectrum,
in terms of a more well-known spectrum, namely the spectrum KO. This will enable us to make
topological arguments when we come to superstrings. We use a very small truncation of KO. We
do not have a conceptual explanation of its appearance, except to say that the homotopy groups
in the first column of (1.40) can only be fit together in one nontrivial way, and that occurs in the
KO-spectrum. On the other hand, this description as a truncation of KO is crucial in Lecture 3
because of the Atiyah-Singer index theorem, which relates KO-theory to Dirac operators.
(1.45) A Postnikov truncation of connective ko. Let ko denote the connective KO-theory spec-
trum. It is more primitive than the periodic KO-spectrum: for example, its 0-space is the classifying
space of the symmetric monoidal category VectR of finite dimensional real vector spaces, a fact we
use extensively at the end of this lecture. Its homotopy groups starting with π1 are the Bott song
Z{2Z, Z{2Z, 0, Z, 0, 0, 0, Z, . . . . Furthermore, ko is a commutative ring spectrum (an E 8 ring
spectrum), and we will use the ring structure extensively. There is a truncation procedure which
produces a commutative ring spectrum from a Postnikov truncation. We introduce a central object
in these lectures, the commutative ring spectrum
(1.46) R :“ πď4 ko “ kox0 ¨ ¨ ¨ 4y.
Its homotopy groups (and k-invariants) are those of the bottom part of ko:
R R´1
π4 Z 0
π3 0 Z
(1.47)
π2 Z{2Z 0
π1 Z{2Z Z{2Z
π0 Z Z{2Z
In the second column9 we list the homotopy groups of the connective cover of Σ´1 R. Note that
π´1 Σ´1 R – Z, but it is not detected on spaces or groupoids, so taking the connective cover does
not lose information in our application.
The Postnikov tower for R´1 is
(1.48) Σ3 HZ R´1
i j
ΣHZ{2Z T Σ4 HZ
k
HZ{2Z Σ2 HZ{2Z
for some spectrum T . (In fact, T is the spectrum which represents the Picard groupoid of Z{2Z-
graded real lines, a fact we use in Lecture 3.) We have included the classifying maps for the various
9Our nomenclature ‘R´1 ’ may not be optimal; nonetheless the two possible meanings of ‘R´1 pXq’ for a space or
groupoid X do coincide.
14 D. S. FREED
stages, which are the arrows j, k; their homotopy classes are the k-invariants. The k-invariants can
be computed from the corresponding k-invariants of ko, and we take them as known. In particular,
we have
k „ Sq 2
(1.49)
j ˝ i „ βZ ˝ Sq 2
where as above βZ is the integer Bockstein.

It follows from the Postnikov tower that for any space X there is a short exact sequence of
abelian groups
pt,aq
(1.50) 0 ÝÑ H 3 pX; Zq ÝÑ R´1 pXq ÝÝÝÑ H 0 pX; Z{2Zq ˆ H 1 pX; Z{2Zq ÝÑ 0
It is functorially split as an exact sequence of sets, but not as an exact sequence of abelian groups.
Furthermore, the nonzero map k means that the product structure in the quotient is correct on the
level of abelian groups, but not on the level of cohomology theories.
(1.51) Comparing cAlgCˆ and R´1 . Tables (1.40) and (1.47) show that the classifying spectra
of cAlgCˆ and R´1 have the same homotopy groups. We claim more.
Theorem 1.52. The classifying spectrum of cAlgCˆ and the spectrum R´1 are isomorphic.
This theorem would follow if we produce a map between them which induces an isomorphism on
homotopy groups. Absent that, we check that the k-invariants agree. The k-invariants of the
associated graded to the Postnikov filtration were computed in (1.42), and they agree with the
homotopy classes of the maps k and j ˝ i in (1.48), as listed in (1.49). What we need to know is
that the actual k-invariants, which correspond to the map j, agree. We are lucky(?) in this case:
the extension problem of lifting from the associated graded is trivial.
Lemma 1.53. Let i : ΣHZ{2Z Ñ T denote the map in the Postnikov tower (1.48). Then any map
T Ñ Σ4 HZ is determined (up to homotopy) by its composition with i.
Proof. Fibrations of spectra are also cofibrations, and from (1.48) the cofiber of i is HZ{2Z. Thus
there is an exact sequence of abelian groups
i˚
(1.54) rHZ{2Z, Σ4 HZs ÝÑ rT, Σ4 HZs ÝÝÑ rΣHZ{2Z, Σ4 HZs,
where rS1 , S2 s denotes the group of homotopy classes of maps S1 , S2 of spectra. The first group
` ˘
vanishes, as one can see by computing H q`4 KpZ{2Z, qq; Z{2Z iteratively for small values of q
using the Serre spectral sequence.
Bundles of invertible algebras
(1.55) Review of groupoids. This is a lightening review; we refer to 3 for a more detailed discussion.
To avoid pathologies, as stated in (1.8) all topological spaces are assumed locally contractible,
paracompact, and completely regular. A topological groupoid X consists of a pair of spaces X0 , X1
which form the objects and morphisms (arrows) of a category in which all morphisms are invertible.
The structure maps p0 , p1 : X1 Ñ X0 give the target and source of a morphism; identity morphisms
are identified as the image of a map X0 Ñ X1 ; there is an associative composition map X1 ˆX0
X1 Ñ X1 ; and there is an inversion X1 Ñ X1 . All structure maps are assumed continuous. Let
Xn denote the space whose elements are n composable morphisms. There are maps pi : Xn Ñ
f1 f2 fn
Xn´1 , i “ 0, . . . , n whose value on px0 ÝÑ x1 ÝÑ ¨ ¨ ¨ ÝÑ xn q P Xn replaces the pair fi , fi`1
1 f
with the composition fi`1 fi and omits the node xi . (The map p0 omits x0 ÝÑ and the map pn
fn
omits ÝÑ xn .) For example, composition of two arrows is the map p1 : X2 Ñ X1 . A map of
topological groupoids F : X Ñ Y is a pair of continuous maps Fi : Xi Ñ Yi , i “ 0, 1, which commute
with the structure maps. It is a local equivalence if it is an equivalence of the underlying categories
and has continuous local sections. Topological groupoids related by a chain of local equivalences
are termed weakly equivalent and represent the same underlying topological stack. (There is an
alternative, invariant approach to topological and smooth stacks.) Let X and Y be topological
groupoids. A morphism X Ñ Y of the underlying topological stacks is a local equivalence X̃ Ñ X
and a map of groupoids X̃ Ñ Y . If G is a topological group acting on a space S, then there is a
global quotient groupoid X “ S{{G in which X0 “ S and X1 “ S ˆ G. A topological groupoid X
is a local quotient groupoid if each point in X has a neighborhood which is weakly equivalent to a
global quotient S{{G, where S is a Hausdorff space and G a compact Lie group.
A Lie groupoid X is one for which X0 , X1 are smooth manifolds, all structure maps are smooth,
and the maps p0 , p1 : X1 Ñ X0 are submersions. A Lie groupoid X is a local quotient Lie groupoid
if each point has a neighborhood weakly equivalent to S{{G, where S is a finite dimensional smooth
manifold and G a compact Lie group acting smoothly. A Lie groupoid X is étale if the target and
source maps p0 , p1 : X1 Ñ X0 are local diffeomorphisms. It is proper if p0 ˆ p1 : X1 Ñ X0 ˆ X0 is a
proper map. If X is proper and étale, then the underlying topological stack is called an orbifold or
smooth Deligne-Mumford stack and the representing groupoid an orbifold groupoid. In particular,
the stabilizers are finite groups. We remark that smooth DM stacks may be presented by Lie
groupoids which are not étale—for example, if P Ñ M is a principal G-bundle over a smooth
manifold, then P {{G is locally equivalent to M , a Deligne-Mumford stack which is a manifold.
Orbifolds have a more concrete differential-geometric description as “V-manifolds” in the work of
Satake, Kawasaki, Thurston and others.
Definition 1.56. A vector bundle E Ñ X over a topological groupoid X is a pair E “ pE0 , ψq

consisting of a vector bundle E0 Ñ X0 and an isomorphism ψ : p˚1 E0 Ñ p˚0 E0 on X1 which satisfies
the cocycle constraint
(1.57) ψf2 ˝f1 “ ψf2 ˝ ψf1 .
f
1 f
2
for px0 ÝÑ x1 ÝÑ x 2 q P X2 .
16 D. S. FREED
f
The notation is that the isomorphism ψ at px0 Ý
Ñ x1 q P X1 is ψf : pE0 qx0 Ñ pE0 qx1 . This data
p˜1
determines a groupoid E “ pE0 E1 q where E1 is the pullback p˚1 E0 and p̃0 : E1 Ñ E0 is the
p˜0
ψ
Ñ p˚0 E0 Ñ E0 .
composition p˚1 E0 Ý
(1.58) Bundles of invertible algebras over topological groupoids. We give a sequence of precise
definitions here for the assiduous reader. They have their own internal music—somewhat like a
passacaglia—and once the reader gets the pattern it should be no problem to supply all details.
Definition 1.59. Let X be a topological groupoid. An invertible algebra bundle pA, B, λq over X
is
(i) a fiber bundle A Ñ X0 of central simple super algebras;
(ii) a super vector bundle B Ñ X1 which is an invertible pp˚0 A, p˚1 Aq-bimodule, that is, for
f
px0 Ý
Ñ x1 q P X1 an invertible pAx1 , Ax0 q-bimodule Bf ;
(iii) and an isomorphism of pp˚1 p˚0 A, p˚2 p˚1 Aq-bimodules λ : p˚0 B bp˚ p˚ A p˚2 B Ñ p˚1 B, that is, for
0 2
f1 f2
px0 ÝÑ x1 ÝÑ x2 q P X2 an invertible pAx2 , Ax0 q-intertwiner
(1.60) λf2 ,f1 : Bf2 bAx Bf1 Ñ Bf2 f1 ,

1
f1 f2 f3
such that for px0 ÝÑ x1 ÝÑ x2 ÝÑ x3 q P X3 the diagram
id bλf2 ,f1
(1.61) Bf3 bAx Bf2 bAx Bf1 Bf3 bAx Bf2 f1
2 1 2
λf3 ,f2 bid λf3 ,f2 f1
λf3 ,f2 ,f1

Bf3 f2 bAx Bf1 Bf 3 f 2 f 1
1
commutes.
The collection of invertible algebra bundles over a fixed topological groupoid X forms a Picard
2-groupoid Alg ˆ pXq. We use the notations AlgCˆ pXq and AlgRˆ pXq when we need to make the
ground field explicit.
Definition 1.62. Let A “ pA, B, λq and A1 “ pA1 , B 1 , λ1 q be invertible algebra bundles over a
topological groupoid X.
(i) A 1-morphism A Ñ A1 is a pair pC, µq consisting of
(a) A fiber bundle C Ñ X0 of invertible pA1 , Aq-bimodules,
f
(b) For px0 Ý
Ñ x1 q P X1 an isomorphism
(1.63) µf : Cx1 bAx1 Bf ÝÑ Bf1 bA1x Cx0

0
f
1 2 f
such that for px0 ÝÑ x1 ÝÑ x2 q P X2 the diagram
(1.64)
µf2 bid id bµf1
Cx2 bAx2 Bf2 bAx1 Bf1 Bf1 2 bA1x Cx1 bAx1 Bf1 Bf1 2 bA1x Bf1 1 bA1x Cx0
1 1 0
id bλf2 ,f1 λ1f bid

2 f1
µf 2 f 1
Cx2 bAx2 Bf2 f1 Bf1 2 f1 bA1x Cx0
0
commutes.
(ii) A 2-morphism pC, µq Ñ pC 1 , µ1 q is an isomorphism
(1.65) ν : C ÝÑ C 1
f
of pA1 , Aq-bimodules on X0 such that for px0 Ý
Ñ x1 q P X1 the diagram
µf
(1.66) Cx1 bAx1 Bf Bf1 bA1x Cx0
0
µbid id bν
µ1f
Cx1 1 bAx1 Bf Bf1 bA1x Cx1 0
0
commutes.
There is a symmetric monoidal structure on Alg ˆ pXq.
Definition 1.67. The tensor product of invertible algebra bundles A “ pA, B, λq and A1 “
pA1 , B 1 , λ1 q over a topological groupoid X is the invertible algebra bundle pA b A1 , B b B 1 , λ b λ1 q,
f f
where the intertwiner pλ b λ1 qf2 ,f1 associated to px0 ÝÑ
1 2
x1 ÝÑ x2 q uses the symmetry of graded
tensor products:
` ˘ ` ˘ (1.5) ` ˘ ` ˘
Bf2 b Bf1 2 bAx bA1x Bf1 b Bf1 1 ÝÝÝÑ Bf2 bAx1 Bf1 b Bf1 2 bA1x Bf1 1
(1.68) 1
λbλ1
ÝÝÝÑ Bf2 f1 b Bf1 2 f1
The symmetry A b A1 Ñ A1 b A is constructed from the symmetry (1.5).
(1.69) Special cases. Specializations of these definitions appear in the literature, as we now illus-
trate.
Example 1.70. Here is the original construction of Donovan and Karoubi in the late 1960s. Sup-
pose the topological groupoid X is a space, so that X1 “ X0 consists only of identity arrows. Then
an invertible algebra bundle over X is a fiber bundle of central simple super algebras. They were in-
troduced by Donovan and Karoubi to twist K-theory and, in particular, to construct a generalized
Thom isomorphism. Subsequently, fiber bundles of infinite dimensional C ˚ -algebras—Dixmier-
Douady bundles—were used by Rosenberg as more general twistings. Below (Definition 1.78) we
allow a space X to be replaced by a weakly equivalent groupoid, and so we realize all isomorphism
classes of infinite dimensional bundles of C ˚ -algebras in our finite dimensional model.
18 D. S. FREED
Example 1.71. Consider the special case in which pA, B, λq P Alg ˆ pXq has A the trivial bundle
with fiber the ground field R or C. Then the bundle B Ñ X1 is a line bundle and λ is an
isomorphism of line bundles; cf. the paragraph preceding (1.31). This is the model explained in
much greater detail in 3 .
Example 1.72. Consider a further special case in which A is trivial and B is the constant line
bundle with fiber k (the ground field) or Π (the trivial odd line). This is equivalent to a function
ǫ : X1 Ñ Z{2Z which encodes the grading, and the existence of λ is in this case a condition, not data:
it asserts that ǫ is a cocycle. In fact, ǫ determines a double cover of the groupoid X; see (1.94).
We remark that double covers twist any cohomology theory.
The next examples, which are commonly used in the literature, do not have Z{2Z-gradings so only
model the spectrum Σ3 HZ, not R´1 . In this case we use the term ‘gerbe’ in place of ‘invertible
algebra bundle’. We will see in the next lecture that gerbes are relevant to the bosonic string,
whereas all of the homotopy groups of R´1 are used in the superstring.
Example 1.73. The next example is another special case of Example 1.71. Here we let B be
possibly nonconstant, but require that it be a purely even line bundle.
Example 1.74. Suppose X is a space and Y Ñ X a surjective submersion. We form the groupoid Ỹ
defined by iterated fiber products Ỹ0 :“ Y , Ỹ1 :“ Y ˆX Y , Ỹ2 :“ Y ˆX Y ˆX Y , etc. Following
Murray, we say a bundle gerbe is a line bundle over Ỹ1 with a product on Ỹ2 which satisfies an
associativity constraint on Ỹ3 . This is an object in Alg ˆ pỸ q of the type in Example 1.71.
Example 1.75. As a special case of Example 1.74 suppose X is a space and tUi uiPI an open cover.
Ů
Then the disjoint union Y :“ Ui surjects onto X and a bundle gerbe reduces to a widely used
i
model for gerbes consisting of a line bundle on each intersection Ui0 X Ui1 and an isomorphism on
each triple intersection Ui0 XUi1 XUi2 which satisfies a cocycle condition on quadruple intersections.
Example 1.76. In another direction if G is a topological group and X “ BG the groupoid with
X0 a point and X1 “ G then an invertible algebra bundle of the type in Example 1.71 is equivalent
to a Z{2Z-graded central extension of G by k ˆ where k “ R or k “ C.
Exercise 1.77. Define the opposite or inverse of an invertible algebra bundle. It is canonical if
you use the opposite algebra, inverse line bundle, etc. Notice the special case of Example 1.73.
What is the effect on the equivalence class in R´1 pXq (or H 3 pX; Zq in the special case)?
Twistings of K-theory
Following §2.3 of 3 , which is motivated by the first moral of (1.16), we make the following
definition.
Definition 1.78. Let X be a topological groupoid. A twisting on X is a pair τ “ pP, Aq consisting

of a local equivalence P Ñ X and an invertible algebra bundle A P Alg ˆ pP q. It is called a
K-twisting in the complex case and a KO-twisting in the real case.
Remark 1.79. It is important that we have the flexibility to “replace” X with a locally equivalent
groupoid P , as foreshadowed in (1.16). For example, if X “ S 3 is the 3-sphere then every invertible
algebra bundle over S 3 is equivalent to a constant invertible algebra bundle in which all fibers are
identified. However, if we replace S 3 by the groupoid formed as in Example 1.75 from the open
cover consisting of two sets Ui “ S 3 ztxi u, i “ 0, 1, for distinct points x0 , x1 P S 3 , then there are
nontrivial complex invertible algebra bundles of the type in Example 1.71 for which the line bundle
on U0 X U1 is topologically nontrivial. These represent nonzero elements of H 3 pS 3 ; Zq.
(1.80) The Picard 2-groupoid of twistings. The twistings in Definition 1.78 are objects in Picard
2-groupoids TwistK pXq and TwistKO pXq. For example, a 1-morphism τ “ pP, Aq Ñ τ 1 “ pP 1 , A1 q
is a 1-morphism p˚ π ˚ τ Ñ p˚ π 1 ˚ τ 1 in Alg ˆ pP 2 q, where in the diagram
(1.81) P2
p
P ˆX P 1
π π1
P P1
the arrow p is a local equivalence. This idea is developed in 3 for the subcategory of twistings in
Example 1.71, and the discussion goes through without change in the general case. One result
is that TwistK pXq and TwistKO pXq are unchanged (up to equivalence) under a local equivalence
X 1 Ñ X of groupoids, so only depend on the stack which underlies X.
The topological classification of twistings is asserted in the following theorem. We will not give
a proof here, but refer to Corollary 2.25 in 3 for the proof in case we do not include the degree as
part of the twisting.
Theorem 1.82. Let X be a local quotient groupoid. Then there is an isomorphism
(1.83) π0 TwistK pXq – R´1 pXq.
We have not yet explained what R´1 pXq means for a groupoid, only for a space. We do so now.
(1.84) Cohomology of local quotient groupoids. Let X be a local quotient groupoid. There is an
associated topological space |X|—the geometric realization or classifying space of X—for which
Xn is the space of n-simplices. If X “ S{{G is a global quotient, then |X| is homotopy equivalent
to the usual Borel construction. Define the cohomology of X with coefficients in an abelian group A
as the ordinary topological cohomology of the geometric realization:
(1.85) H ‚ pX; Aq :“ H ‚ p|X|; Aq.
For a global quotient H ‚ pS{{G; Aq – HG‚ pS; Aq is the usual Borel equivariant cohomology. The
same works for general cohomology theories, and so R´1 pXq “ R´1 p|X|q is the abelian group
of homotopy classes of maps of |X| into the 0-space of the spectrum Σ´1 R. If X is a global
quotient S{G of a nice space S by the action of a compact Lie group G, then R´1 pXq is the Borel
equivariant cohomology of X.
20 D. S. FREED
Twisted vector bundles and Fredholm operators

There is no point to twisting K-theory if we do not have twisted K-theory; so far we only have
the twisting. We first give the definition of a twisted vector bundle for a fixed invertible algebra
bundle; it is a modification of Definition 1.56.
Definition 1.86. Let X be a topological groupoid and A “ pA, B, λq an invertible algebra bundle.
Then an A-twisted vector bundle over X is a pair E “ pE0 , ψq consisting of an A-module E0 Ñ X0
and an isomorphism ψ : B bp˚1 A p˚1 E0 Ñ p˚0 E0 of p˚0 A-modules over X1 . It is required to satisfy a
cocycle relation on X2 analogous to (1.57).
Exercise: Write the exact cocycle relation.
Remark 1.87. Implicit in Definition 1.86 is that the bundle E0 Ñ X0 has finite rank. But such
bundles are not sufficient to represent all twisted K-theory classes, even if we allow general twistings
as in Definition 1.78. Here is a sketch argument in case X is a space. Given an invertible algebra
bundle choose a good open cover (Example 1.75) over which we can trivialize the bundle of algebras.
For simplicity—and because it’s enough to make the point—suppose these algebras are Morita
trivial. Then on the cover we can choose an equivalent invertible algebra bundle for which the
bundle of algebras is trivial, so a twisting of the form in Example 1.71. Now if E is a twisted vector
bundle for that invertible algebra bundle, and its rank is k, then the determinant bundle Det E is
a trivialization of the k th power of the invertible algebra bundle. In other words, the twisting is
necessarily torsion of order k.
(1.88) Fredholm operators. We therefore introduce a more general model which represent elements
of K-theory to realize all twisted classes. Suppose E “ E 0 ‘ E 1 is a Z{2Z-graded Hilbert space.
A bounded linear map t : E 0 Ñ E 1 is Fredholm if it has closed range and finite dimensional kernel
and cokernel. We consider bounded linear operators T : E Ñ E which are odd and skew-adjoint,
` ˚˘
so have the form 0t ´t0 . Then T is Fredholm iff t is. If T is Fredholm, then Ker T is a finite
dimensional super vector space which is equivalent to the pair pE, T q. But in a continuous family
of odd skew-adjoint Fredholm operators the kernels do not form a vector bundle as the rank jumps,
though the (graded) dimension—called the index —is locally constant.
Let Fred be the category of super Hilbert spaces and odd skew-adjoint Fredholms. It, rather
than Vect, gives a model of K-theory which generalizes to the twisted case over a space X. We
first generalize it to a twisted model over a point, as in (1.4). Thus if A is an invertible super
algebra we let FredA denote the category of super Hilbert spaces which are left A-modules and
which carry an odd skew-adjoint Fredholm operator T which (graded) commutes with the A-action.
We remark that we allow the super Hilbert space E to be finite dimensional, and there are many
interesting classes (e.g. the Thom class) which have a nice finite dimensional model. The finite
dimensional model is essentially the “difference construction” in the early K-theory literature.
Remark 1.89. There is a tricky sign in the proper definition of “skew-adjoint” in the super world—
the matrix we wrote above is not really odd skew-adjoint. There is a way around that sign to a
more standard convention and in any case we will not need to worry about this level of detail.
(1.90) Continuous families of Fredholms. To extend Definition 1.86 to twisted families of Fred-
holm operators, there is an important technical point: what is a fiber bundle whose fiber is a (super)
Hilbert space? The issue is the operator topology with which we measure continuity. This is dealt
with nicely in an appendix to the paper of Atiyah-Segal on twisted K-theory. The strong operator
topology, which is essentially the compact-open topology, is the appropriate one to use here, not
the norm topology. Not only does this work theoretically, but important examples are continuous
in the compact-open topology and not in the norm topology. One then needs an adapted model
for Fredholm operators, which ends up being a homotopy equivalent slight thickening of that space
of operators. We defer to all of these references for details. And after all of the preceding detailed
definitions, we leave it to the reader to complete the Exercise: Write carefully the generalization of
Definition 1.86 to twisted families of Fredholm operators, including the correct cocycle relation.
(1.91) Twisted K-theory. Finally, given a twisting (Definition 1.78) we must define twisted K-
theory groups. Again we defer to 3 . We want in the end to obtain a twisted cohomology theory with
all of the exact sequences etc., so we need to do more than just define K-theory groups separately
for each pair pX, τ q of a topological groupoid X and a twisting τ . Rather we define a bundle of
spectra over X and use it to define K-theory groups. The idea is that for a fixed super Hilbert
space there is a spectrum defined by odd skew-adjoint Fredholms, using the standard Clifford
algebras Cℓn to make the shift operators. This was developed in a classic paper of Atiyah-Singer.
Given a twisting τ there exist “locally universal” τ -twisted Hilbert bundle, and we use twisted odd
skew-adjoint Fredholms to form the bundle of spectra we need. The K-theory groups we define are
in the end canonically independent of the choice of the twisted Hilbert bundle. This means that we
can represent K-theory classes by any τ -twisted Hilbert bundle (even finite rank) equipped with
an odd skew-adjoint Fredholm operator.
(1.92) Thom class revisited: an extended exercise. We already gave several examples of twisted
K-theory classes at the beginning of this lecture. The reader should revisit them now. Together we
will only revisit (1.8), and we invite the reader to work out the details of the following as practice
in the formalism.
Let X be a topological groupoid and V Ñ X a real vector bundle of rank n with a positive
definite quadratic form Q. Show that CℓpV, Qq Ñ X is an invertible algebra bundle which is in fact
a bundle of algebras. It is one model for the twisting τ KO pV q determined by V . Here is another.
Define a locally equivalent groupoid P Ñ X which is the principal On -bundle of orthonormal
frames. The group On has an extension
(1.93) 1 ÝÑ Z{2Z ÝÑ Pin´

n ÝÑ On ÝÑ 1
where Pin´ n lies in the Clifford algebra Cℓń . Use this extension to define an invertible algebra
bundle over P of the type in Example 1.71. Construct an isomorphism with the twisting given by
CℓpV, Qq Ñ X. This latter model will play an important role in Lecture 3.
Pull these twistings back by the map π : V Ñ X to obtain twistings over V . For the first
we pull back the opposite Clifford algebras π ˚ CℓpV, ´Qq Ñ V , and consider the vector bundle
π ˚ CℓpV, ´Qq Ñ V as a right CℓpV, ´Qq-module. Define the family of odd skew-adjoint Fredholms
22 D. S. FREED
(in finite rank every operator is Fredholm) whose value at ξ P V is left multiplication by ξ P V Ă
CℓpV, ´Qq. This graded commutes with the left action by CℓpV, `Qq, defined from the right action
of CℓpV, ´Qq using Exercise 1.25. Note that multiplication by ξ is invertible off of the 0-section
˚
of π. The K-theory (or KO-theory) class so defined is the Thom class. It lies in KOπ ClpV,Qq pV qc ,
where the subscript ‘c’ denotes ‘compact support’. Its restriction to the 0-section is the Euler class
defined in (1.8).
Now make a model of the Thom class using the equivalent twisting based on the pin group. You’ll
need to observe that the pullback π ˚ V Ñ P of V Ñ X over P Ñ X is canonically equivalent to
the vector bundle with constant fiber Rn . (Hint: A point of P is an isometry Rn Ñ Vx for some x.)
For this we use the vector bundle over π ˚ V with constant fiber Cℓń , left Clifford multiplication
by ξ P Rn and the right Cℓń -action by multiplication.
Define a spin structure in two equivalent ways. As usual it is a reduction of the On -bundle P Ñ X
to a Spinn -bundle. Show that this trivializes the twisting you defined above in the sense that it is
equivalent to a twisting which shifts the degree by n. Use the trivialization to define an untwisted
Thom class. Use the spin structure to define a vector bundle C Ñ X which is an invertible
` ˘
Cℓn , CℓpV, Qq -bimodule. Compare with Definition 1.62 to see that we obtain an isomorphism of
the twisting CℓpV, Qq Ñ X with the twisting defined by the constant algebra Cℓn , so a degree shift
by `n.
Twistings of KR-theory
Suppose Y is a space with an involution σ : Y Ñ Y . In this situation it is natural to consider
complex vector bundles E Ñ Y with a lift σ̃ : σ ˚ E Ñ E of the involution to a linear isomorphism
on the bundle whose square is the identity. These are simply vector bundles over the groupoid
quotient X “ Y {{σ. But it is also possible to consider lifts which are complex antilinear, or
equivalently a linear isomorphism σ ˚ E Ñ E from the pullback of the complex conjugate bundle E
to E. This is another form of “twisting” of K-theory. We remark that in the previous case twisted
K-theory is a module over K-theory; for this kind of twisting we obtain a module over KO-theory.
This is Atiyah’s KR-theory. Our focus on twistings leads us to regard the KR-theory of Y as a
twisted form of K-theory on the quotient X. From that point of view Y with its involution is
replaced by X with its double cover Y Ñ X. (Recall that double covers have already been used in
Example 1.72 to define more usual twists of K-theory.)
(1.94) Double covers of topological groupoids. A double cover of a local quotient groupoid X is a
morphism X Ñ BZ{2Z, where the groupoid BZ{2Z “ ˚{{pZ{2Zq is the global quotient of a point
by the trivial involution. More concretely, we have the following.
Definition 1.95. Let X be a local quotient groupoid. A double cover of X is a pair pY, φq consisting
of a local equivalence Y Ñ X and a continuous homomorphism φ : Y1 Ñ Z{2Z.
1 f 2 f
The homomorphism property means that for px0 ÝÑ x1 ÝÑ x2 q P Y2 we have φpf2 ˝ f1 q “ φpf2 q `
φpf1 q. The trivial double cover is the map X Ñ BZ{2Z which takes every element in X1 to the
identity arrow in BZ{2Z. Given two double covers of X we can find a single local equivalence
P Ñ X and homomorphisms φ1 , φ2 : P1 Ñ Z{2Z which describe them. The double covers are
f
isomorphic if and only if φ1 ´ φ2 evaluated on px0 Ý Ñ x1 q is χpx1 q ´ χpx2 q for some function
χ : P0 Ñ Z{2Z. See §2.2 in 3 for more details.
If pX̃, φq is a double cover of X we construct a new groupoid Xw as the pullback
(1.96) Xw pt
X BZ{2Z
f
Explicitly, pXw q0 “ X0 ˆ Z{2Z and each arrow px0 Ý Ñ x1 q P X1 gives rise in Xw to two arrows
` ˘
px0 , i0 q Ñ px1 , i1 q P pXw q1 , where i0 “ i1 in Z{2Z if φpf q “ 0 and i0 “ i1 if φpf q “ 1.
A double cover of X is classified up to isomorphism by a class w P H 1 pX; Z{2Zq, where we
recall (1.84) for the definition of cohomology of a local quotient groupoid. In the presentation
above the function φ : X1 Ñ Z{2Z on 1-simplices determines w. We often employ ‘w’ in the
notation for a double cover thus: Xw Ñ X. We also use ‘w’ to denote the double cover itself.
(1.97) w-twisting a twisting. Let V be a complex super vector space. Set
#
φ V, φ “ 0;
(1.98) V “
V, φ “ 1,
where V is the complex conjugate super vector space. The same notation applies to super algebras
and bimodules.
Now suppose that Xw Ñ X is a double cover of topological groupoids, as in Definition 1.95. This
is specified by a homomorphism φ : X1 Ñ Z{2Z, possibly after replacing X by a weakly equivalent
groupoid (which we still denote ‘X’). We use the double cover to twist the notion of invertible
algebra bundles introduced in (1.58).
Definition 1.99. Let Xw Ñ X be a double cover of topological groupoids. A w-twisted complex

invertible algebra bundle A “ pA, B, λq over X is given as in Definition 1.59 with the following
φpf q
modifications. In (ii), Bf is an p Ax1 , Ax0 q-bimodule. In (iii), the intertwiner is a map
φpf1 q
(1.100) λf2 ,f1 : Bf2 bAx1 Bf1 ÝÑ Bf2 f1 .
Finally, in the w-twisted case diagram (1.61) reads
φpf2 f1 q φpf1 q id bλf2 ,f1 φpf2 f1 q

(1.101) Bf3 bAx Bf2 bAx Bf1 Bf3 bAx Bf2 f1
2 1 2
λf3 ,f2 bid λf3 ,f2 f1
φpf1 q λf3 ,f2 ,f1

Bf3 f2 bAx Bf1 Bf 3 f 2 f 1
1
24 D. S. FREED
By way of explanation, if φpf1 q “ 1 then in the tensor product Bf2 bAx Bf1 in (1.100) we have
1
b̄2 b āb1 “ b̄2 ā b b1 for all a P Ax1 , b1 P Bf1 , and b2 P Bf2 .
Remark 1.102. The following special case is illuminating. Suppose Xw is a space with involution σ.
Suppose A Ñ Xw is a complex invertible algebra bundle and the involution σ is lifted to an
isomorphism of algebras σ̃ : σ ˚ A Ñ A such that σ̃σ̃ is the identity. Then σ̃ determines a w-
twisted complex invertible algebra bundle over the quotient groupoid X :“ Xw {{pZ{2Zq. (The
homomorphism φ : X1 Ñ Z{2Z maps non-identity arrows in X to 1 P Z{2Z.)
We leave the necessary modifications of other definitions in the w-twisted case to the reader.
There is a 2-groupoid AlgCˆ pXqw of w-twisted invertible algebra bundles. As in Definition 1.78
we can also form a 2-groupoid TwistK pXqw of “twisted twistings” on X; it is more familiar as a
2-groupoid of twistings of KR-theory on Xw . There is a w-twisted version of Theorem 1.82.
Theorem 1.103. Let X be a local quotient groupoid and Xw Ñ X a double cover. Then there is
an isomorphism
(1.104) π0 TwistK pXqw – Rw´1 pXq,
where Rw pXq is the R-cohomology of X twisted by the double cover Xw Ñ X.
Recall that double covers twist every cohomology theory, so in particular R, and that degree shifts
are also twists. Twistings form a Picard (multi-)groupoid, so can be added using its group law.
Differential twistings
(1.105) General discussion of differential cohomology. Let h be a generalized cohomology theory.

Then h b R is (noncanonically) isomorphic to ordinary Eilenberg-MacLane cohomology HR with
coefficients in the Z-graded vector space R‚ “ h‚ pptq b R. (We ignore possible finiteness issues
which are satisfied in our applications.) On a smooth manifold X we can compute HpX; Rq‚ using
smooth differential forms with values in R. (The Z-grading adds the differential form degree and the
degree in R.) Furthermore, there is a “marriage” of h and these differential forms into a differential
cohomology theory q h. This is an active area of development, and we freely use variations (such as
equivariant differential theories) which are perhaps not fully in place.
Rather than delve into any details, we just paint a picture10 of q hq pXq for an integer q and a
smooth manifold X. The main point is that q hq pXq is a (usually infinite-dimensional) abelian Lie
group. An abelian Lie group only has nonzero homotopy groups π0 , π1 , and these are computed by
topological cohomology:
π0 q
hq pXq – hq pXq
(1.106)
π1 q
hq pXq – hq´1 pXq{torsion
10Thisis the correct picture for the examples h “ HZ, R, KO, K that enter these lectures. We could also take
h “ HR, for example, in which case the topology we would introduce on q h‚ pXq would not satisfy (1.106) or other
aspects of our picture.
The identity component of q hq pXq carries a free action of the finite dimensional torus hq´1 pXqbR{Z,
which sits inside as a subgroup, and the quotient is the space of exact differential forms dΩpX; Rqq´1 .
The same torus acts freely on every other component, and the quotient is naturally an affine
translate of dΩpX; Rqq´1 in the space of closed forms in ΩpX; Rqq . The map to differential forms
is a kind of curvature:
(1.107) curv : q
hq pXq ÝÑ ΩpX; Rqqclosed .
The image of (1.107) is a union of affine translates of the subspace ΩpX; Rqqexact Ă ΩpX; Rqqclosed ,
and the image of this union of subspaces under the de Rham map
(1.108) ΩpX; Rqqclosed ÝÑ HpX; Rqq
is a full lattice, which is the image of hq pXq Ñ HpX; Rqq . The fiber of (1.107) over a differential
form ω P ΩpX; Rqqclosed is a torsor for the group hq´1 pX; R{Zq of flat elements in q
hq pXq. Note that
the flat elements hq´1 pX; R{Zq form an abelian Lie group whose identity component is the torus
hq´1 pXq b R{Z.
(1.109) Differential R-theory. Following this general discussion there are differential cohomol-
ogy groups R q‚ pXq attached to a manifold X. In our discussion of superstring theory we en-
counter R q´1 pXq and, for a double cover Xw Ñ X a twisted version R qw´1 pXq. There is a concrete
differential geometric model for these differential classes which we now sketch.
Let X be a smooth manifold and P Ñ X a locally equivalent Lie groupoid (see (1.55)). Let A “
pA, B, λq be an invertible algebra bundle over P , as in Definition 1.59. First, there is contractible
space of “hermitian structures” on A. It consists of a “positive ˚-structure” on the bundle of
algebras A Ñ X0 and a compatible positive hermitian structure on the bimodule B Ñ X1 . We
will not spell out these notions here, but mention that in case A is trivial (Example 1.71), then it
carries a canonical positive ˚-structure and a positive hermitian structure on the Z{2Z-graded line
bundle B Ñ X1 is the usual notion of a positive definite hermitian metric. We fix such a hermitian
structure. Without it we would have complex differential forms in what follows, instead of the real
forms we need.
The differential geometric datum is then a pair pB, ∇q consisting of a real 2-form11 B P Ω2 pP0 q
and a covariant derivative ∇ on B Ñ P1 which is compatible with the bimodule structure and the
hermitian structure. Since the endomorphisms compatible with those structures are sections of a
real line bundle over P1 , the curvature F∇ is a real 2-form on P1 . We demand that pB, ∇q satisfies
the constraint
i
(1.110) p˚1 B ´ p˚0 B “ F∇
2π
on P1 . There is a global closed 3-form H “ dB, and since by (1.110) it satisfies p˚1 H “ p˚0 H it drops
to a global 3-form on the smooth manifold X. This is the curvature of the differential twisting.
11Sorry for the notation clash, but this is the ‘B’ of ‘B-field’.
26 D. S. FREED
q´1 pXq.
A differential twisting has an equivalence class in R
If Xw Ñ X is a double cover, specified by a homomorphism φ : P1 Ñ Z{2Z as in Definition 1.95,
then a differential KR-twisting is the same sort of data on a w-twisted invertible algebra bundle
over P (Definition 1.99), but now (1.110) is modified to:
i
(1.111) p´1qφ p˚1 B ´ p˚0 B “ F∇ .
2π
The curvature is now a twisted differential form H P Ωw`3 pXq.

qw´1 pXq.
A differential KR-twisting has an equivalence class in R
(1.112) Reminder about ordinary orientations. Let V be a real vector space. The set of orienta-
tions of V is a canonical Z{2Z-torsor attached to V , and may be defined as
` ˘
(1.113) opV q :“ π0 Det V zt0u ,
where Det V is the determinant line of V , its highest exterior power. An orientation of V is a
choice of element of opV q.
Let M be a smooth manifold. There is a canonical orientation double cover M x Ñ M whose
fiber at m P M is opTm M q. It represents the first Stiefel-Whitney class w1 pM q. An orientation
x Ñ M . If o is an orientation, then the opposite orientation ó is the section
of M is a section of M
obtained by applying the deck transformation to o.
(1.114) Integration in differential cohomology theories. Let π : X Ñ Y be a fiber bundle of smooth

manifolds with fiber compact manifolds of dimension n. Let h be a multiplicative cohomology
theory12 (based on a commutative ring spectrum). Then there is a notion of h-orientation, and if
π is h-oriented there is an “integration” map
(1.115) π˚ : hq pXq ÝÑ hqń pY q.
An h-orientation induces an ordinary orientation and so an integration of differential forms
(1.116) π˚ : Ωq pX; Rq ÝÑ Ωqń pY ; Rq.
There is a notion of a differential h-orientation, introduced in work of Hopkins-Singer, which is

used to define an integration map
(1.117) π˚ : q
hq pXq ÝÑ q
hqń pY q
in the differential theory which is “compatible” with (1.115) and (1.116); see (1.120) below.
12There is a version of this discussion for a module theory over a ring theory.
We mention three cases of (1.117). For h “ HZ ordinary Eilenberg-MacLane cohomology an

HZ-orientation is a usual orientation and there is no extra data necessary to define a differential
HZ-orientation. For h “ R an R-orientation is a spin structure, just as it is for ko-theory. Again
there is no additional data necessary to integrate in differential R-theory. Finally, for h “ KO a
KO-orientation is again a spin structure, only now there is additional data in a differential KO-
orientation. The most efficient is a Riemannian structure on π : X Ñ Y , which is a pair pg, Hq
consisting of a metric g on the relative tangent bundle T pX{Y q Ñ X and a horizontal distribution,
which is a splitting of the short exact sequence
(1.118) 0 ÝÑ T pX{Y q ÝÑ T X ÝÑ π ˚ T Y ÝÑ 0
of vector bundles over X. This pair determines a Levi-Civita covariant derivative on T pX{Y q Ñ X,
so a curvature and a closed differential form Âpg, Hq P Ω‚ pXq. It enters into the compatibility
between the integrations (1.117) and (1.116). Namely, if
(1.119) } q pXq Ñ ΩpX; Rqq

ω : KO
is the curvature map, then

` ˘ q
(1.120) qq “ π˚ Âpg, Hq ^ ωpq
ωpπ˚ x xq , x } pXq.
q P KO
} and R.
There is a similar compatibility for integration in HZ q
The classifying spectrum of cAlgRˆ and the transfer map

We conclude this lecture with more about the algebraic topology of twistings. Some of what we
sketch here will be useful—even crucial—in the application to superstring theory. Theorem 1.52
locates AlgCˆ in algebraic topology, as a certain truncation of ko. We now investigate the Picard 2-
groupoid cAlgRˆ of real invertible algebras. Here there is not an “off-the-shelf” spectrum to compare
to, so we simply give its classifying spectrum a name: ‘r’. (Despite the choice of letter, this is not
a ring spectrum.) Nonetheless, r does receive a map from ko via the Clifford algebra construction,
and is in fact a ko-module (which factors down to an R-module). Real algebras are the “fixed
points” of complex algebras under complex conjugation, and this leads to another realization in
algebraic topology, also connected to ko and its truncation R. Of course, the involution by complex
conjugation is used in defining w-twisted twistings of K-theory; see Definition 1.99. We also define a
quadratic map—a “transfer map”—from complex algebras to real algebras. This is only interesting
when considered equivariantly for complex conjugation. We prove a formula for this quadratic map
which is crucial in proving the anomaly cancellation in Lecture 3. We remark that this transfer
map is the effect on twistings of a norm map on twisted K-theory classes, and that norm map is
part of the definition of Ramond-Ramond currents, which are not part of these lectures. We also
revisit the Thom twistings attached to a real vector bundle and give formulas for them in algebraic
topology.
I thank Matt Ando and David Gepner for valuable discussions about some of this material.
28 D. S. FREED
(1.121) The spectrum r. Let r denote the classifying spectrum of the Picard 2-groupoid AlgRˆ .
We computed its homotopy groups in (1.40). The computation of the k-invariants is similar to the
computation (1.42) in the complex case. The k-invariants are nonzero, but we won’t need them
for any computations, so we don’t pursue the matter further. Analogous to Theorem 1.82 we have
the following.
Theorem 1.122. Let X be a local quotient groupoid. Then there is an isomorphism
(1.123) π0 TwistKO pXq – r0 pXq.
(1.124) The Clifford map. The Clifford algebra of a real vector space is the map
(1.125) c : VectR ÝÑ AlgRˆ .
which assigns to a real vector space V the Clifford algebra CℓpV, Qq for any positive definite qua-
dratic form Q on V . The set of positive definite inner products on V is contractible, so to make the
construction we can form a category equivalent to VectR whose objects are real vector spaces with
positive definite inner product. Direct sum provides a monoidal structure on VectR and there is
an obvious symmetry. The map c is a homomorphism of symmetric monoidal Picard 2-groupoids:
(1.126) CℓpV1 ‘ V2 , Q1 ‘ Q2 q – CℓpV1 , Q1 q b CℓpV2 , Q2 q.
Therefore it induces a map of the classifying spectra. As mentioned earlier, connective ko is the
classifying spectrum of VectR , so we obtain
(1.127) c : ko ÝÑ r.
We claim that c factors to a map
(1.128) ko R
c
c̄
r
This follows easily since the restriction of c to the homotopy fiber of ko Ñ R is null: there is no
possible nonzero map on homotopy groups. We easily compute the effect of c̄ on homotopy groups:
R ÝÑ r
π4 Z ÝÑ 0
π3 0 ÝÑ 0
(1.129)
π2 Z{2Z ÝÑ Z{2Z
π1 Z{2Z ÝÑ Z{2Z
π0 Z ÝÑ Z{8Z
The map on π0 is surjective and the maps on π1 , π2 are isomorphisms.

(1.130) Complexification. There is an obvious complexification map
(1.131) bC : AlgRˆ ÝÑ AlgCˆ
and it induces a map of spectra
(1.132) bC : r ÝÑ R´1 .
c b
C
The composite ko ÝÑ r ÝÝÑ R´1 attaches the complex Clifford algebra to a real vector space.
Remark 1.133. We do not believe this factors through k, the connective K-theory spectrum which
is the classifying spectrum of VectC , since the space of nondegenerate quadratic forms on a complex
vector space is not contractible.
(1.134) The universal double cover. Define (see also (1.94))
(1.135) BZ{2Z “ pt {{pZ{2Zq,
the groupoid with a single object and automorphism group Z{2Z. There is a canonical double cover
(1.136) π0 : pt ÝÑ BZ{2Z.
It is the universal double cover in the sense that any double cover, say π : Xw Ñ X, is classified by
a unique map
(1.137) Xw pt
π π0
X BZ{2Z
Let w0 denote the double cover (1.136).
(1.138) Complex conjugation. Complex conjugation is the antilinear map C Ñ C which takes
(1.139) z ÞÑ z̄.
The fixed points are R Ă C. What we need is a categorical version. For example, there is an
involution VectC Ñ VectC which takes a complex vector space W to the complex conjugate
vector space W and a linear map to the complex conjugate linear map. (It contains the previous
complex conjugation by restricting to EndpCq “ VectC pC, Cq.) Whereas (1.139) is an involution on
the set C—that is a condition on a map—to say that complex conjugation is an involution on the
category VectC is to give extra data which then satisfies a condition. We do not spell that out here.
30 D. S. FREED
Similarly, we can say that the “fixed points” of that involution on VectC is the category VectR ,
but here ‘fixed points’ must be understood as ‘homotopy fixed points’. So to give a fixed point is
to give a pair pW, Jq consisting of W P VectC and an isomorphism J : W Ñ W which satisfies a
condition relative to the involution which amounts to JJ “ idW .
Our interest here is in complex conjugation on the Picard 2-groupoid AlgCˆ . It maps an invertible
complex algebra A to its complex conjugate algebra A, and similarly for invertible bimodules and
intertwiners. The homotopy fixed point category is AlgRˆ .
In terms of algebraic topology the involution of complex conjugation defines a fibering
(1.140) E ÝÑ BZ{2Z
with fiber R´1 , and the spectrum r is the spectrum ΓpBZ{2Z, Eq of global sections of (1.140). Now
(1.140) is the twisting of R-cohomology from the universal double cover (1.136), and so this gives
a new interpretation of the cohomology theory r in terms of a twisted version of R:
(1.141) r‚ pXq – Rw0 `‚´1 pX ˆ BZ{2Zq.
Since twisted R-cohomology is an R-module, we deduce that r is an R-module, and by pullback a

ko-module as well.
Note that the pullback of the bundle (1.140) of spectra under (1.136) is simply the spectrum R´1 ,
so pullback by π0 is a model for complexification (1.132).
(1.142) Two universal twisted KO-theory classes. We introduce
(1.143) θ P Rw0 ´1 pBZ{2Zq.
It, and the class χ to be introduced shortly, was already defined in (1.13) in terms of our concrete
model. Namely, the twisting w0 of BZ{2Z is defined by the nontrivial grading of the group Z{2Z:
the identity homomorphism Z{2Z Ñ Z{2Z. Twisted ko-classes are then represented by real super
vector spaces with an action of Z{2Z compatible with the grading. The matrix γ in (1.14) is odd
and squares to the identity, so defines such an action on the vector space R1|1 . The degree shift ´1
is implemented by a Cℓ´1 -module structure which (graded) commutes with the group action, and
the generator of the Clifford action is the matrix x in (1.14). This defines a class in kow0 ´1 pBZ{2Zq,
and we let (1.143) be its image under the map ko Ñ R.
Similarly, we define the class
(1.144) χ P R´w0 `1 pBZ{2Zq.
The twisting ´w0 of BZ{2Z is defined by the nontrivial Z{2Z-graded central extension of Z{2Z,
where the grading is as before and the central extension is
(1.145) 1 ÝÑ Z{2Z ÝÑ Z{4Z ÝÑ Z{2Z ÝÑ 1

We simply switch the role of the matrices in (1.14): now x is the action of the generator of Z{2Z
and γ the action of the generator of Cℓ1 . Again this is a class in twisted ko and we define (1.144)
as its image under ko Ñ R.
We need two facts about θ and χ. First, the product θχ is a class in R0 pBZ{2Zq, which is
the image of a class in ko0 pBZ{2Zq, the spectrum associated to (untwisted) representations of the
group Z{2Z. We claim that
(1.146) θχ “ ρ,
where ρ is the image of the regular representation of Z{2Z. Second, the complexification of θ is a
class
(1.147) η P R´1 pptq
which is represented by the Clifford module R1|1 for the Clifford algebra Cℓ´1 . Again, the Clifford
module represents in the first instance a class in ko´1 pptq and we use the map ko Ñ R.
Exercise 1.148. Verify these two facts using the explicit models. For the first you’ll need to
use a Morita trivialization of the Clifford algebra Cℓ1,1 . For the second you’ll need to use that
complexification is the map π0˚ , as explained at the end of (1.138).
(1.149) A commutative diagram. The following shows a relationship between two ways of passing
from a class in ko to a class in r.
Lemma 1.150. The diagram
(1.151) ko R
c̄
c θ
(1.141)
r ΓpBZ{2Z, Eq
commutes.
Proof. Since all spectra in (1.151) are ko-modules, it suffices to verify the diagram for the multi-
plicative unit 1 P ko0 pptq. I do not see a proof using the explicit models, since we don’t have a
model which sees constructs complex invertible super algebras (R´1 ) from Cl´1 -modules (ko´1 ).
Absent that we use the following topological computation, proved in §3 of 2 . We can then define θ
as the generator of this cyclic group which makes the diagram commute.
Theorem 1.152. The group Rw0 ´1 pBZ{2Zq – Rw0 ´2 pBZ{2Z; R{Zq is cyclic of order 8. Further-
more, the pullback of a generator θ under π0 : pt Ñ BZ{2Z is η.
This group has an interpretation as the group of degrees in KO-theory, and the Z{8Z is that of
Bott periodicity. The last statement in Theorem 1.152 was observed around (1.147).
32 D. S. FREED
(1.153) A quadratic map on twistings. According to Theorem 1.103 the group of isomorphism
classes of twistings of KR-theory on Xw , also known as w-twisted twistings, is Rw´1 pXq. For any
orbifold X the group of isomorphism classes of twistings of KO-theory on X, as a special case of the
same theorem, is Rw0 ´1 pX ˆ BZ{2Zq, where the double cover (1.136) is pulled back via projection
onto the second factor. Define
(1.154) ℜ : Rw´1 pXq ÝÑ Rw0 ´1 pX ˆ BZ{2Zq
as pushforward under the map
(1.155) p : X ÝÑ X ˆ BZ{2Z,
where p is the identity onto the first factor and the bottom arrow in (1.137) on the second. Note
–
that (1.137) provides an isomorphism p˚ w0 Ý Ñ w, and this is used to define the pushforward (1.154).
This pushforward may be regarded as a transfer map in equivariant cohomology.
ℜ maps KR-twistings to KO-twistings. It has a geometric definition in the model of Lecture 1.
Roughly, if β is a KR-twisting, then ℜpβq is a real lift of β̄ ` β. In terms of an invertible algebra
bundle pA, B, λq, we form pA b A, B b B, λ b λq. The “real lift” is accomplished via a Morita
equivalence, which is canonical if we introduce a hermitian structure.
According to (1.141) we can also view ℜ as a map
(1.156) ℜ : Rw´1 pXq ÝÑ r0 pXq.
The following lemma plays a crucial role at the end of Lecture 3.
Lemma 1.157. The map ℜ : Rw´1 pXq Ñ r0 pXq is multiplication by q ˚ χ P R´w`1 pXq followed
by c̄ : R0 pXq Ñ r0 pXq.
Here χ P R´w0 `1 pBZ{2Zq is the class defined in (1.144).
Proof. We must show that the two maps (1.154) and multiplication by q ˚ χ are equal on the category
of spaces X equipped with a double cover. There is an equivalent statement about Z{2Z-equivariant
cohomology on the double cover. Now to prove that the two maps are equal it suffices13 to check
two universal cases: (i) X “ BZ{2Z with the nontrivial double cover π0 in (1.136), and (ii) the
lift to X “ pt. In the universal cases it suffices to check on a generator, since both are R-module
maps. For (i) the generator is θ P Rw0 ´1 pBZ{2Zq, as defined in (1.142). Now for X “ BZ{2Z the
map (1.155) is the diagonal ∆ : BZ{2Z Ñ BZ{2Z ˆ BZ{2Z and (1.154) is the pushforward
(1.158) ∆˚ : Rw0 ´1 pBZ{2Zq ÝÑ Rw0 ´1 pBZ{2Z ˆ BZ{2Zq,

13This
follows from a general theorem in equivariant homotopy theory. We used the nonequivariant version of this
theorem in the proof of Lemma 1.150: two maps of spectra are homotopic if they induce the same map on homotopy
groups. In the equivariant case we need to check this on all fixed point spectra, and we need to be careful to use the
correct notion of fixed points.
where the twisting w0 in the codomain is pulled back from the second factor. Using (1.141) we
identify this as the pushforward
(1.159) pπ0 q˚ : r0 pptq ÝÑ r0 pBZ{2Zq,
and this pushforward on the ko-module r is induced by the pushforward
(1.160) pπ0 q˚ : ko0 pptq ÝÑ ko0 pBZ{2Zq,
on ko. The generator 1 P ko0 pptq is represented by the trivial real line R, and its image under pπ0 q˚
is represented by the regular representation ρ of Z{2Z; this is the standard description as the
induced representation of the trivial representation. Now the desired equality follows from (1.146)
and tracing back to (1.158).
For (ii) we prove the equality of ℜ : R´1 pptq Ñ r0 pptq and multiplication by the lift of χ to R1 pptq.
Since R1 pptq “ 0, the latter map is the zero map. As for the former, we check that the action
on homotopy groups (see (1.40)) is trivial. Note the map takes an complex invertible algebra A
to A b A, which is Morita trivial. This proves the map is zero on π0 . For π1 we observe that if L is
a complex super line, then L b L is even, so the induced map on π1 is zero. The maps on π2 , π3
are obviously zero.
(1.161) Thom twistings revisited. The canonical twisting τ h pV q in the cohomology theory h as-
sociated to a real vector bundle V Ñ X is described in (2.44), where its role in integration is
emphasized. Particular models for K-theory are given in (1.92). Here we simply state formulas
for the isomorphism classes of the various twistings. They all have flat differential lifts which we
don’t put into the notation in this section.
Suppose Xw Ñ X is a double cover with classifying map
(1.162) Xw pt
q
X BZ{2Z
Then for the K-theory and KR-theory Thom twistings we have

“ K ‰
τ pV q “ η rV s P R´1 pXq
(1.163) “ KR ‰
τ pV q “ q ˚ pθq rV s P Rw´1 pXq
As a special case of the second equation we have

“ KO ‰
(1.164) τ pV q “ θ rV s P Rw0 ´1 pX ˆ BZ{2Zq,
which, by Lemma 1.150 is equivalent to

“ KO ‰
(1.165) τ pV q “ cpV q P r0 pXq.
34 D. S. FREED
Exercise 1.166. Prove (1.163)–(1.165). Here are a few hints. It is easiest to begin with (1.165)
and observe that the real Clifford bundle, which is the map c, is a model for the Thom twisting;
see (1.92). Then (1.164) follows from (1.151). The K-theory twisting is gotten by complexification,
for which we use the second part of Exercise 1.148.
Lecture 2: Fields and superstrings
In the remaining lectures we turn to geometric structures in superstring theory. This lecture
contains definitions; the next contains a theorem. We execute the traditional three steps in applied
mathematics: (i) model a system external to core mathematics—in this case a physical system—in
mathematical terms; (ii) prove theorems about the mathematical model; and (iii) draw conclusions
about the external system from the mathematical theorems. Today’s lecture is part of Step (i);
tomorrow’s is an example of Step (ii). We do not discuss the physics of Step (iii) here, except to
say that the theorem in Lecture 3 is a consistency check on the physical system. There are other
physical consequences of our work which we do not discuss here. One attraction of this particular
application of mathematics, as with many others, is that it suggests problems and developments in
core mathematics. Here what it suggests are ideas in a rich mix of homotopy theory, differential
geometry, and global analysis. For example, the modern developments in twisted K-theory were
directly inspired by this physical system.
More specifically, we work with a “semi-classical” model for strings in terms of fields, which
are classical objects that belong to differential geometry, though this example presses us to bring
in homotopy theory as well. There are a few key “quantum” ideas which enter also—Dirac’s
quantization of charge, the fermionic functional integral—and they shape our considerations. The
main point of this lecture is to pin down the topological aspects of fields in superstring theory. In
fact, here we only describe the definition of the “fields”, for the most part not the “action” with the
notable exception of the B-field amplitude on the worldsheet. The theorems we prove from these
definitions, including the one in Lecture 3, provide evidence that our mathematical model for the
fields is “correct”. Usual considerations in physics pin down the local field content; the topological
subtleties involve more refined physical and mathematical considerations.
Superstring theory is studied in its usual perturbative formulation as a 2-dimensional field theory
on “string worldsheets”. In this theory spacetime is an external 10-dimensional manifold X. There
is a low energy approximation which is a 10-dimensional field theory formulated in terms of fields
on X. The topological nature of the fields is the same across the two theories, and it is striking
how tightly constrained the system is and how well the same data works in the 2-dimensional and
10-dimensional theories. The “orientifold” construction provides the best testing ground and the
deepest agreement between the two theories.
We begin with a general discussion of the notion of a field, leading to a definition in terms of
simplicial sheaves. A key construction in this lecture is what physicists call “gauging a symmetry”.
In the context of simplicial sheaves this is the natural quotient construction, at least for finite
groups of symmetries. This gives the field content of the gauged theory; the action of the gauged
theory requires additional data. We only discuss one term in the action—the “B-field amplitude”
in the 2-dimensional theory—and the lecture ends with a puzzle: There is no natural extension of
this action to orientifolds. The resolution of the mystery is in Lecture 3.
General discussion of fields

Physicists speak in terms of “scalar” fields, “gauge fields”, etc. Here we give a mathematical
framework in which to think about the notion of a field. We take the notion more broadly—and
36 D. S. FREED
more abstractly—than is usual. For example, we consider topological structures such as orientations
and spin structures to be fields. One motivation for this discussion is our description below in (2.39)
of the orientifold construction as a gauging of the orientation-reversal symmetry. The reader can
safely skip it and get more quickly to the less structural discussions in subsequent sections.
(2.1) Examples of fields. Let’s begin by listing some types of fields by telling what they are on a
smooth manifold M :
(i) a scalar field with values in a fixed manifold Z is a map M Ñ Z, and if Z “ R is is called
a real scalar field ;
(ii) a metric (‘gravitational field’ in physics-speak) is a Riemannian14 metric on M ;
(iii) an orientation is. . . well. . . an orientation on M ; similarly for a spin structure;
(iv) given a complex spinor representation of Spinn , a spinor field on an n-dimensional manifold
with a Riemannian metric and spin structure is a section of the bundle associated to the
given representation;
(v) for a fixed Lie group G a G-connection (‘gauge field’ in physics-speak) is a pair pP, Θq of a
principal bundle P Ñ M and a connection Θ on P ;
(vi) the term ‘B-field ’ has many meanings; in Type II superstring theory it is a twisting of
K-theory on M (Definition 1.78).
(2.2) Categories of manifolds. We want the notion of a field F to be valid not just on a single
manifold M , but rather on a collection of manifolds, and fields must pullback under a collection of
maps between the manifolds. In other words, we will define a category Man of manifolds and a
target category C such that a field is a homomorphism 15
(2.3) F : Manop ÝÑ C.
For example, a real scalar field is the homomorphism F : Manop Ñ Set which assigns the set FpM q “
C 8 pM q to each smooth manifold M . We would then say a real scalar field on M is a particular
element in FpM q.
How should we define Man? For scalar fields (i) we can take Man to be the category of all
smooth manifolds and all smooth maps between them. However, this is too big for (ii): metrics do
not pull back under arbitrary maps. To accommodate metrics, then, we can still take the category
of all smooth manifolds, but restrict to immersions. Now (iii) forces us to rule out immersions
which are not local diffeomorphisms, so we are reduced to the latter. The spinor field (iv) depends
on a particular spinor representation, so requires us to further fix a dimension n and consider only
n-manifolds.
Definition 2.4. For each integer n ě 0 define Mann as the category whose objects are smooth
n-manifolds and morphisms are local diffeomorphisms.
14We work in the “Wick rotated” framework; we could instead use Lorentz metrics.
15a (contravariant) functor in traditional terminology
(2.5) The codomain. For many of the examples in (2.2) we can take the codomain C in (2.3) to
be Set: on a fixed manifold M scalar fields, metrics, orientations, and spinor fields form a set.
But spin structures and G-connections have automorphisms, and we need a structure which tracks
them. In these examples we can take C “ Gpd to be the category of groupoids. The morphisms
of groupoid-valued fields are called “gauge transformations”. But for (vi) we need to go further:
twistings of K-theory on M form a 2-groupoid—see (1.80). Though it is not the subject of these
lectures, we remark that the “C-field” in M-theory takes values in a 3-groupoid. To accommodate
all of these examples, and to have a flexible mathematical framework with plenty of foundational
work in the literature, it is convenient to take C “ Set∆ , the category of simplicial sets. A general
field in an n-dimensional field theory, then, is a homomorphism
(2.6) F : Manop
n ÝÑ Set∆ .
This is sometimes termed a simplicial presheaf on the category Mann . Fields in physics satisfy a
locality property, encoded in mathematics by the sheaf property, which is crucial, but which we do
not spell out here.
Definition 2.7. A field, or collection of fields, in an n-dimensional field theory is a simplicial sheaf
F : Manop
n Ñ Set∆ .
Remark 2.8. In a given field theory we define F to be the collection of all fields in the theory.
Notice that if a theory has, say, a scalar field and a metric, then F “ Fscalar ˆ Fmetric is a product.
But if the theory has a metric, spin structure, and spinor field, then F is not the product of three
factors since a spinor field cannot be defined without first having a metric and spin structure.
So in general F is an iterated fibration of the individual fields. Furthermore, the fields may be
divided into background and dynamical fields (also called fluctuating fields in a quantum theory).
For example, in the theory of a scalar field on a Riemannian manifold we might consider the metric
as background and the scalar field as dynamical. Fields pertinent to the intrinsic geometry of
the manifold—metrics, orientations, spin structures—are background fields in non-gravitational
theories but are dynamical in theories of gravity. String theory is a theory of gravity. Thus, for
example, in the superstring one sums over the spin structures on the worldsheet.
Quantization of charge
(2.9) Classical electromagnetism. We briefly recall the setup for Maxwell’s theory of the elec-
tromagnetic field. We work on a four-dimensional spacetime of the form M 4 “ E1 ˆ N 3 , where
pN 3 , gN q is a Riemannian manifold. We endow M with the Lorentz metric dt2 ´ gN , where t is
a (time) coordinate on E1 and the speed of light is set to unity. Minkowski spacetime is the
case N “ E3 .
38 D. S. FREED
Classical electromagnetism involves four time-dependent fields:
Eptq P Ω1 pN q electric field

Bptq P Ω2 pN q magnetic field
(2.10)
ρE ptq P Ω3c pN q electric charge density
JE ptq P Ω2c pN q electric current
Here Ωc denotes differential forms of compact support. Traditional texts identify E, B, JE with
vector fields and ρE with a function.16 But the differential form language is more convenient and
leads to a better geometric picture. The classical Maxwell equations are
BB
dB “ 0 ` dE “ 0
(2.11) Bt
BE
d ˚N E “ ρE ˚N ´ d ˚N B “ JE
Bt
We reformulate these equations using differential forms on M with its Lorentz metric and corre-
sponding Hodge ˚ operator as follows. Set
F “ B ´ dt ^ E P Ω2 pM q
(2.12)
jE “ ρE ` dt ^ JE P Ω3 pM q.
The electric current jE has compact spatial support. Maxwell’s equations (2.11) are equivalent to
the pair of equations
dF “ 0
(2.13)
d ˚ F “ jE .
As a consequence of the second equation we have
(2.14) djE “ 0.
The de Rham cohomology class of ρE in Hc3 pN ; Rq is the electric charge; (2.14) implies that it is
independent of time.
(2.15) Charges in quantum theory. To write a quantum mechanical theory which incorporates
electromagnetism—for example, the nonrelativistic Schrödinger equation for a charged particle
moving in a background electromagnetic field—the gauge potential A, and not just the electro-
magnetic field F “ dA, appears. This assertion has an experimental basis, due to Aharanov and
Bohm. Furthermore, it is an empirical fact that nobody has written a quantum theory in terms
of F alone. Accepting the necessity of the gauge potential, the quantization of charge is based
16Thisassumes that M is oriented. If not, then ρE , JE are forms twisted by the orientation bundle. The vector
field which corresponds to B is also twisted.
on: (i) the existence of a system in which magnetic current jB and electric current jE are both
nonzero, and (ii) the particular coupling of A to the electric current in the quantum theory. The
text in this discussion, beginning with (2.9), is taken almost verbatim from the leisurely discussion
in 17 to which I refer the reader for a continuation. The upshot is that in a quantum theory the
electric and magnetic charges—the de Rham cohomology classes of jE and jB restricted to a time
slice—are required to live in a full lattice in the de Rham cohomology vector space. There are
further considerations which prompt a refinement of the lattice to an abelian group of charges,
which may include torsion charges. Finally, that abelian group should depend locally on space, so
it is reasonable to postulate that it is a cohomology group in some generalized cohomology theory.
For Maxwell electromagnetism (2.9) the “correct” cohomology theory is ordinary Eilenberg-
MacLane cohomology HZ. So the de Rham class of ρE in Hc3 pN ; Rq lies in the image of Hc3 pN ; Zq,
and in fact there is a cohomology class Q P H 3 pM ; Zq (with compact spatial support) which is
compatible with jE in the sense that the de Rham class of jE is the image of Q under H 3 pM ; Zq Ñ
H 3 pM ; Rq. Recalling the discussion in (1.105) it is natural to assume a refinement q E P Hq 3 pM q
which fits into the diagram:
(2.16) q
E jE
Q QR
The left map gives the component in the differential cohomology group (see (1.106)), the top map
is the curvature, the bottom map is bZ R, and the right map is the de Rham cohomology class.
The right column is classical; the left column is the quantum refinement taking into account Dirac
charge quantization. There is a diagram analogous to (2.16)—with all degrees reduced by 1—for
the gauge field itself. The upper right corner of that diagram is the field strength F and the lower
left corner is sometimes called the flux. The upper left corner is an object in H q 2 pM q, which can
be taken to be a principal T-bundle with connection. This is the usual geometric model for the
Maxwell gauge field.
Perhaps better is to think directly in terms of the picture sketched in (1.105). We then lift the
classical current jE , which is a closed differential form, to its fiber of the map (1.107). The fiber is
a torsor for H 2 pM ; R{Zq and this tells the extra information in the quantum theory—beyond the
restriction that the de Rham class of jE lie in a full lattice.
(2.17) Which cohomology theory? The Dirac argument is taken to apply to all “generalized abelian
gauge fields” in quantum field theories. The field is locally a differential form, but globally has
integrality encoded in a generalized cohomology theory h, and the art is in finding the correct
cohomology theory. Then there are diagrams (2.16) for both the current and gauge field. What
physical and mathematical considerations go into the choice of generalized cohomology theory h
and a particular degree in that theory? There are several possible:
(i) First and foremost, h b R in the given degree must duplicate the known local field content.
17D. S. Freed, K-theory in quantum field theory, Current Developments in Mathematics 2001, International Press,
Somerville, MA, pp. 41-87 (arXiv:math-ph/0206031).
40 D. S. FREED
(ii) There may be torsion charges not detected in the classical formulation with differential
forms, and the theory may contain charged solitonic objects which exhibit the torsion
charges.
(iii) Anomaly cancellation, a condition for the theory to be consistent, is sensitive to the choice
of h. This occurs in the “Green-Schwarz mechanism”, which occurs in several contexts. Here
some form of K-theory is involved as an anomaly from the currents must cancel against an
anomaly computed using the Atiyah-Singer index theorem.
(iv) There may be an equivariant version, e.g. on orientifolds, and then an appropriate equivari-
ant version of h comes into play and must exist with the correct field content after tensoring
with R.
(v) There may be special geometric features in the system.
I refer to 18 for a more detailed discussion of charge quantization for generalized gauge fields and
for many examples.
(2.18) Generalized abelian gauge fields in superstring theory. There are two generalized abelian
gauge fields in the Type II superstring: the “B-field” and the “Ramond-Ramond field”. We will
argue that the former is a differential twisting of K-theory and the latter an object representing a
twisted differential K-theory class. All of the considerations listed in (2.17) are relevant. Point (i)
is always a consideration. In the Type I superstring, for example, there are solitonic objects with
torsion charges, as first identified by Witten, who used (ii) to argue for the correctness of K-theory
to quantize the Ramond-Ramond charges. Lecture 3 is an illustration of (iii) on the worldsheet.
There are also anomaly cancellation arguments for the 10-dimensional spacetime theory, including
the original argument of Green-Schwarz in the Type I theory. Consideration (iv) applies very neatly
to Ramond-Ramond fields on orbifolds. There we use Atiyah-Segal equivariant K-theory, and the
tensor product with R has contributions from twisted sectors (via the localization theorem in
equivariant K-theory) which matches perturbative computations in string theory. Finally, (v) also
applies: the coupling of the B-field and Ramond-Ramond field seen in terms of differential forms
is manifest by using the quantum B-field (with integrality in R-theory) to twist K-theory and so
the notion of a Ramond-Ramond field.
The oriented bosonic string
(2.19) Two sets of fields. Quantum field theory has a fixed dimension of spacetime and fields are
local objects (Definition 2.7) on manifolds of the given dimension. String theory is confusing at
first as there is, in addition to spacetime, a worldsheet of dimension 2. We consistently use the
letter ‘X’ to denote spacetime and ‘Σ’ to denote the worldsheet. Both are smooth manifolds; Σ has
dimension 2 and the dimension of X depends on the particular string theory. We discuss two cases.
For the oriented bosonic string dim X “ 26 and for the superstring dim X “ 10. These dimensions
arise out of standard nontopological considerations and we do not discuss them further here. We
do remark that the number 10 (reduced mod 8) plays a crucial role in Lecture 3; the argument
18D.
S. Freed, Dirac charge quantization and generalized differential cohomology, Surv. Differ. Geom. VII, 2000,
129–194 (arXiv:hep-th/0011220).
there would not work if the dimension of spacetime were 11. We always take Σ to be compact. In
these lectures it will be closed in the sense that it is a manifold without boundary. Worldsheets
with boundary are very important in string theory—they are “open strings”—but we will not have
time to deal with them in these lectures.
For each of string theories there is a set of fields F26 or F10 on spacetime and a set of fields F2 rXs
on the worldsheet. The order is important. A fixed choice of spacetime X and of fields on X—an
element in F26 or 10 pXq—is used as external data to define F2 rXs and is indicated by ‘rXs’ in the
notation. This external data is analogous to both the choice of gauge group in a gauge theory and
to a choice of coupling constants in any theory. On the other hand, F26 or 10 does not depend on
the worldsheet at all.
(2.20) F26 . There are three fields on a spacetime in oriented bosonic string theory, and they are
independent in the sense that F26 is the Cartesian product of three sheaves. In other words, each of
the three fields may be defined without defining the other two. The fields in F26 pXq are (see (2.1)):
(i) a metric;
(ii) a real scalar field, called the “dilaton”;
q a “gerbe with connection”, which on a manifold X has an equivalence class
(iii) a “B-field” β,
q 3
in H pXq.
For the purposes of these lectures we take a gerbe on X to have a geometric model given
by pP ; L, λq, which is a restricted kind of twisting of K-theory. Here, as in Definition 1.78, P is a
topological groupoid equipped with a local equivalence P Ñ X. (Some typical examples appear in
Example 1.74 and Example 1.75.) The pair pL, λq is a special complex invertible algebra bundle
over P in which the bundle of algebras over P0 is the trivial bundle with fiber C and the line bundle
L Ñ P1 is purely even; it was mentioned in Example 1.73. The connection data in this model is a
pair pB, ∇q, as described in (1.109). This case is simpler than the general invertible algebra bundle
described there as ∇ is a covariant derivative on a line bundle.
Remark 2.21. These three fields appear in the superstring as well, but then the B-field is there
an arbitrary twisting of K-theory. The use of the phrase ‘restricted kind of twisting of K-theory’
to describe the oriented bosonic string B-field is pure convenience: there is no K-theory in the
oriented bosonic string and the B-field doesn’t twist anything. The metric and dilaton do not play
any role in these lectures, which focus on topological aspects of strings.
(2.22) F2 rXs. As mentioned in (2.19) to define the worldsheet fields F2 rXs we fix an oriented
string background, which consists of a smooth 26-dimensional manifold X and a choice of fields
in F26 pXq. Let βq denote the chosen B-field. There are three worldsheet fields for the oriented
bosonic string, and they are independent in the sense described in (2.20). The fields are:
(i) an orientation o;
(ii) a metric;
(iii) a scalar field φ with values in X.
On a worldsheet Σ the scalar field is a smooth map φ : Σ Ñ X. If X were a true spacetime
with a Lorentz metric, φ would encode the spacetime motion of a string. Here φ is an analog
42 D. S. FREED
in Riemannian field theory. We remark that one can consider bosonic string theory without an
orientation, in which case there is no B-field.
Note that F2 rXs is set-valued.
(2.23) The B-field amplitude in the oriented bosonic string. For the most part we do not discuss
the worldsheet action for the string and we do not discuss at all the spacetime action which ap-
proximates string theory in terms of low energy fields. The one exception is the B-field amplitude
in the worldsheet action. For the oriented bosonic string it is straightforward. First, recall that in
classical field theory on a manifold Y the action is a function S : FpY q Ñ R on the set of fields. (In
case FpY q is a simplicial set, as it is here because of the presence of the B-field, we require that
S factor through a function on π0 FpY q.) In the functional integral formulation of quantum field
theory, it is the exponential eiS or e´S which is relevant, depending on the signature of the metric.
We call this the exponentiated action. It may happen that only this exponential is well-defined,
and often in those cases the exponential is the same independent of the signature of the metric. In
this case we can view S as a function into R{2πZ. (We will move the ‘2π’ to e2πiS , so take S to
have values in R{Z.)
Now suppose φ : Σ Ñ X is a scalar field as in (2.22)(iii). Then φ˚ βq is a geometric object on Σ
which represents a class in Hq 3 pΣq. We can use the orientation o on Σ to define an integration map,
as in (1.117) and the discussion at the beginning of the paragraph which follows it. Denote that
integration as
ż
(2.24) π˚ “ q 3 pΣq ÝÑ H
:H q 1 pptq – R{Z.
pΣ,oq
This last isomorphism follows from the second isomorphism in (1.106), but of course the reader
will need to understand more about differential cohomology to truly understand it. In any case we
define the R{Z-valued B-field amplitude as
ż
(2.25) Spφ, oq “ q
φ˚ β.
pΣ,oq
Note that the integral only depends on the equivalence class of βq in Hq 3 pXq.
There is no interesting topology in this expression, but the analogous expressions for orientifolds,
superstrings, and superstring orientifolds are more interesting from that point of view.
Orbifolds in string theory and in geometry
(2.26) Gauging a symmetry. We explain the general idea in field theory of gauging, working in
the general framework of Definition 2.7. We gauge the symmetry of a finite, or discrete, group Γ.
(To gauge a Lie group of symmetries, replace Galois covers with connections.)
Let F : Manopn Ñ Set∆ be a collection of fields in an n-dimensional field theory. Let Γ be a
finite group, and suppose that Γ acts of F. Before indicating what ‘Γ acts on F’ means, let’s give
an example. Let Y be a 26-dimensional smooth manifold equipped with a Γ-action. Use it as an
external background to define the fields F2 rY s in the oriented bosonic string, as in (2.22). (We need
to also choose a metric, dilaton, and B-field on Y which are Γ-invariant, but they are not essential
to this discussion.) Then for each 2-manifold Σ, the group Γ acts on the set of fields F2 rY spΣq,
and the action commutes with pullback by local diffeomorphisms (in fact all smooth maps) of 2-
manifolds. The action on the orientation and metric on Σ is trivial; the action of γ P Γ on φ : Σ Ñ Y
yields fγ ˝ φ : Σ Ñ Y , where fγ : Y Ñ Y is the action of γ on Y .
The general definition should be clear, except that for groupoid- or multi-groupoid valued fields
there is more data to specify. We do not give details here.
Given a Γ-action on F, define the sheaf of gauged fields
(2.27) FΓ pΣq “ tpP, Φq : P Ñ Σ is a principal Γ-bundle, Φ P FpP q is Γ-invariantu.
A principal Γ-bundle is also called a Galois covering space with Galois group (group of deck trans-
formations) equal to Γ. The Γ-invariance of Φ means
(2.28) Rγ˚ Φ “ γ ¨ Φ, for all γ P Γ,
where on the left we pull back by the map Rγ : P Ñ P and on the right we use the Γ-action
on FpP q. Note that F Ă FΓ as the gauged fields with trivial Γ-bundle.
Remark 2.29. In this context FΓ is the natural quotient of F by the symmetry Γ. To see this
suppose first that Y is a manifold and Γ acts as a group of symmetries on Y . The natural quotient
construction in the world of simplicial sets is the groupoid Y {{Γ, which as a simplicial set is Y0 “ Y ,
Y1 “ Y ˆ Γ, etc; see (1.55). Now Y corresponds to a sheaf F whose value on a (test) manifold Z is
the set of smooth maps Z Ñ Y . The groupoid Y {{Γ corresponds to a simplicial sheaf whose value
on a (test) manifold M is the set of smooth maps M Ñ Y {{Γ, and this is a pair consisting of a
principal Γ-bundle P Ñ M and a Γ-equivariant map P Ñ Y .
Exercise 2.30. Work out FΓ for the worldsheet example given above. You should see that φ is now
a section of a fiber bundle over Σ with fiber Y . Also, check that for fixed P Ñ Σ its automorphism
group acts on the space of Γ-invariant fields in FpP q. What does that say in the special case when
P is the trivial bundle?
Principal Γ-bundles form a groupoid, and FΓ contains the information of isomorphisms of bun-
dles. So FΓ pΣq naturally breaks up as a union over the isomorphism classes of bundles P Ñ Σ.
These isomorphism classes are called twisted sectors, except for the isomorphism class of the trivial
bundle, which is called the untwisted sector.
Remark 2.31. We have only discussed the fields here, not the action. We can also do that in a
formal framework—there is a bordism category of n-manifolds equipped with a field and an action
is an invertible field theory on that bordism category, and now we can see what it means to extend
the action—but we will not carry that out here. We just point out that to extend the action is to
provide more data, whereas to construct the fields in the gauged theory there is no further data
required.
44 D. S. FREED
(2.32) The usual meaning of twisted sectors. In the case n “ 2 of the string worldsheet theory,
consider a cylinder Σ “ r0, 1s ˆ S 1 , which models the propagation of a single string. Then twisted
sectors are labeled by isomorphism classes of Γ-bundles over the cylinder, which is the same as
isomorphism classes over S 1 , and these are labeled by conjugacy classes in Γ.
Exercise 2.33. In the situation of Exercise 2.30 suppose we fix a basepoint in P Ñ S 1 . Define
the holonomy as an element γ P Γ. Show that Γ-invariant maps φ : P Ñ Y correspond to maps
φ̃ : R Ñ Y such that φ̃pt ` 1q “ fγ ˝ φ̃ptq. What happens as we change the basepoint (in the same
fiber of P Ñ S 1 ?)
(2.34) Orbifolds in geometry. Suppose as above a finite group Γ acts on a smooth 26-manifold Y .
Points of Y connected by elements of Γ represent the same points of spacetime—Γ is a gauge
symmetry—so it is natural to take spacetime as the quotient Y {{Γ. We keep track of isotropy
subgroups, due to non-identity elements γ P Γ and y P Y with fγ pyq “ y. Now an old construction
in differential geometry of Satake, called the ‘orbifold’ by Thurston, does exactly that. Furthermore,
we can admit as spacetimes orbifolds X which are not global quotients by finite groups, thus
widening the collection of models introduced in the previous paragraph. Orbifolds are presented
by a particular class of groupoids, as was explained in (1.55): each point has a neighborhood
weakly equivalent to Y {{Γ for a smooth manifold Y and a finite group Γ. Of course, a special
case is the global quotient X “ Y {{Γ. A worldsheet is then a map φ : Σ Ñ X of orbifolds, and
the infinite-dimensional orbifold of such maps is precisely F2 rXspΣq, once we add a metric and
orientation.
The upshot of this paragraph is that for a scalar field with values in a Γ-manifold Y , the orbifold
quotient implements the gauging (2.26).
And the upshot for oriented bosonic string theory is that we allow spacetime X to be a smooth
26-dimensional orbifold (in the sense of Satake-Thurston). But then we must extend F26 in (2.20)
to orbifolds. This is straightforward for the metric and dilaton. But for the B-field we need some
discussion, as anticipated in (2.17)(iv). We turn to that now.
(2.35) Equivariant cohomology. There are many extensions of a given cohomology theory h to
an equivariant cohomology theory for spaces Y with the action of a compact Lie group G. The
simplest is the Borel construction. It attaches to pY, Gq the space YG “ EG ˆG Y , where EG is
a contractible space with a free G-action. Then one defines the Borel equivariant h-cohomology
as hG pY q :“ hpYG q. This is not a new cohomology theory, but rather the nonequivariant theory
applied to the Borel construction, a functor from G-spaces to spaces. That functor generalizes
to orbifolds which are not necessarily global quotients—the functor is geometric realization—and
so leads to a notion of “Borel cohomology” theories on orbifolds, which we described in greater
generality in (1.84). But usually h has other extensions to an equivariant theory. For example,
the Atiyah-Segal geometric version of equivariant K-theory, defined in terms of equivariant vector
bundles, is more delicate: Borel equivariant K-theory appears as a certain completion. The Atiyah-
Segal theory is extended to orbifolds, in fact to “local quotient groupoids”, in 3 .
We use generalized differential cohomology on orbifolds without further comment. There are
papers which develop it in the case of a global quotient, but as far as I know there is work to be
done in the general case.
(2.36) The B-field on orbifolds. We posit the following generalization of (2.20)(iii) to allow for
X a smooth orbifold.
(iii) the B-field βq is a gerbe with connection on the orbifold X.
Recall that our geometric model for a gerbe with connection, as discussed in (2.20), already makes
sense for an orbifold (and indeed in much greater generality). We use the classification result
Theorem 1.82, restricted to gerbes rather than more general twistings of K-theory, to conclude
that the topological classification of B-fields is by the cohomology group H 3 p|X|; Zq, where as
in (1.84) |X| is the geometric realization of X.
The B-field amplitude (2.25) is unchanged when spacetime X is allowed to be an orbifold; the
worldsheet Σ is still a smooth manifold, not an orbifold, and there is nothing new to say to define
the integral.
Orientifolds of the oriented bosonic string
(2.37) A bigger version of worldsheet fields. In the context of the general discussion of fields at
the beginning of the lecture, culminating in Definition 2.7, observe that a constant (simplicial)
sheaf is trivially a field. In that spirit we now include the spacetime fields, heretofore viewed as
external to F2 rXs, as part of the worldsheet fields. We do this to encode the action of orientation
reversal in string theory, which acts simultaneously on worldsheet and spacetime fields. Thus for a
fixed 26-dimensional orbifold19 X define F Ă2 rXs to include the fields:
(ii) a metric;
(iii) a scalar field φ with values in X;
(iv) a metric on X;
(v) a real scalar field on X;
(vi) a gerbe with connection βq on X.
Ă2 rXs is a homomorphism Man2 Ñ Set∆ . If
The fields (iv)–(vi) are constant in the sense that F
Σ P Man2 is a smooth 2-manifold, then the metric (ii) in F Ă2 rXspΣq is a metric on Σ, so depends
on Σ, as do the fields (i) and (iii).
(2.38) The involution on F Ă2 . Let Y be an orbifold and σ : Y Ñ Y be an involution. The corre-

sponding parity involution on FĂ2 rY s has the following action on the fields enumerated in (2.37):
(i) o ÞÑ ó (the opposite orientation o is defined in (1.112));
(ii) the metric on the 2-manifold is fixed;
(iii) φ ÞÑ σ ˝ φ;
(iv) the metric on Y is pulled back by σ;
(v) the scalar field on Y is pulled back by σ;
(vi) the B-field βq transforms as βq ÞÑ ´σ ˚ β.
q
19The reader can first think through the case when X is a smooth manifold. Allowing X to be an orbifold combines
the orbifold construction above with the orientifold construction.
46 D. S. FREED
The motivation for the minus sign in (vi) is the B-field amplitude (2.25), which is then preserved
by the involution (since both o and βq change sign). The construction of the opposite B-field in the
model of Lecture 1 is hinted at in Exercise 1.77.
(2.39) Gauging the involution. Now apply the general gauging construction in (2.26) to (2.38).
So on Σ P Man2 there is a new field P Ñ Σ which is an arbitrary double cover. (We are gauging
an action of the group Z{2Z, and a double cover is a principal Z{2Z-bundle.) Then we need each
of the fields in (2.37) on P and require them to be invariant under simultaneously pulling back by
the deck transformation of P Ñ Σ and executing the involution (2.38).
The key observation is that if P is an oriented surface, any map P Ñ Σ has a canonical lift
P ÑΣ p to the orientation double cover. Here, since P Ñ Σ is itself a double cover, this map is
an orientation-preserving diffeomorphism. In other words, the field P Ñ Σ is not arbitrary but
must be the orientation double cover and the orientation o is the canonical orientation on Σ. p So
the orientifold of the oriented bosonic string does not have an orientation field : it is an unoriented
string theory.
The metric (ii) in the gauged theory is simply a metric on Σ. More interesting is the gauged
p Ñ Y . The metric (iv)
field φ, for which Exercise 2.30 is relevant. Namely, it is an equivariant map Σ
and dilaton (v) descend to a metric and dilaton on the quotient Y {{σ of Y by the involution σ.
The B-field βq also descends to the quotient, but it is twisted by the double cover Y Ñ Y {{σ, due
to the minus sign in the transformation law.
(2.40) Bosonic orientifold background. We recast the gauged fields (2.39) into new definitions
of F26 and F2 which account for orientifolds. We put hats for the orientifold construction. Note
that the original definitions in (2.20) and (2.22) are the special case when the orientifold double
cover is trivial.
The fields in Fp26 pXq are:
(i) a double cover Xw Ñ X, called the orientifold double cover ;
(ii) a metric;
(iii) a real scalar field, the dilaton;
(iv) a q called the B-field.
w-twisted gerbe with connection β,
The B-field in the bosonic orientifold is a special case of a w-twisted twisting of K-theory, as
defined towards the end of Lecture 1 (as twistings of KR-theory). The equivalence class of βq lies
q w`3 pXq. The orientifold double cover is unramified; it is an ordinary double cover in the sense
in H
of orbifolds, as reviewed in (1.94).
(2.41) Bosonic worldsheet fields for the orientifold. Fix a bosonic orientifold background, which
p
means a smooth 26-manifold X and a set of fields in FpXq. Then the worldsheet fields in that
background form a sheaf Fp2 rXs : Man2 Ñ Set whose fields on a 2-manifold Σ are:
(i) a metric on Σ;
(ii) a map φ : Σ Ñ X;
(iii) an isomorphism ν : φ˚ w Ñ w1 pΣq.
Recall that w1 pΣq is represented by the orientation double cover, so concretely the field in (iii) is
an isomorphism of double covers
ν p
(2.42) φ ˚ Xw Σ
Exercise 2.43. Check that if the double cover Xw Ñ X in the bosonic orientifold background is
trivial, then (iii) reduces to the orientation o in (2.22).
(2.44) Integration of densities. Let M be a smooth manifold of dimension n. Then a density on M

is an element of Ωw1 pM q`n pM q, an n-form twisted by the orientation double cover M x Ñ M . One
xq, one which changes sign under pullback by
representation is as an anti-invariant element of Ωn pM
the deck transformation. Densities can be integrated without any choice of orientation.
There is an analogous story for any cochain theory which represents ordinary cohomology. Thus,
if say M is compact there is an integration map H w1 pM q`n pM q Ñ Z. Similar integration maps exist
for fiber bundles. There are analogous integration maps for a general multiplicative cohomology
theory h, but in that case w1 is replaced by the obstruction to h-orientation. More precisely, a
real vector bundle—here T M Ñ M —gives rise to an associated twisting τ h pM q of hpM q, and the
h
integration is a map hτ pM q`q pM q Ñ hq pptq.
These twisted integrations combine to give a twisted integration on differential cohomology.
(2.45) The B-field amplitude in the bosonic orientifold. Analogous to (2.24), but without the
orientation, we have an integration
ż
(2.46) q w1 pΣq`3 pΣq ÝÑ H
:H q 1 pptq – R{Z.
Σ
Analogous to (2.25) the B-field amplitude is defined to be
ż
(2.47) Spφ, νq “ q
νφ˚ β,
Σ
where the νφ˚ is the composition
˚
(2.48) q w`3 pXq ÝφÝÑ H
H q φ˚ w`3 pΣq ÝνÑ H
q w1 pΣq`3 pΣq.
As with (2.25) the integral (2.47) depends only on the equivalence class of βq in H
q w`3 pXq.
48 D. S. FREED
Fields in the oriented Type II superstring

We deploy the term ‘Type II superstring’ when we include the orientifold construction and use
‘oriented Type II superstring’ when the orientifold data is absent (or viewed as present and trivial).
Some of the structures we describe here are spelled out in greater detail in 2 .
We begin with a technical point.
Remark 2.49. The worldsheet of the perturbative superstring is more properly treated as a super-
manifold, which leads to a more integrated description of some of the fields. Then the moduli space
of super surfaces with conformal structure is itself a supermanifold, and the functional integral of
the perturbative superstring becomes an integral over that supermanifold. That complex super-
manifold does not in general admit a holomorphic splitting, whereas it does admit a C 8 splitting,
as does any supermanifold. We implicitly use one in our description of the fields and in our treat-
ment of the fermions in Lecture 3. We believe that the issues of supergeometry are irrelevant for
our topological considerations.
(2.50) Review of spin structures. As a preliminary we quickly review spin structures. Recall that
the intrinsic geometry of a smooth n-manifold M is encoded in its principal GLn R-bundle of frames
BpM q Ñ M . A point of BpM q is a linear isomorphism Rn Ñ Tm M for some m P M . Choose a
Riemannian metric on M , equivalently, a reduction to an On -bundle of frames BO pM q Ñ M . The
spin group
(2.51) ρ : Spinn ÝÑ On
is the double cover of the index two subgroup SOn Ă On . A spin structure on M is a principal
Spinn -bundle BSpin Ñ M together with an isomorphism of the associated On -bundle with BO pM q.
It induces an orientation on M via the cover Spinn Ñ SOn . The space of Riemannian metrics
is contractible, so a spin structure is a topological choice and can alternatively be described in
terms of a double cover of an index two subgroup of GLn R. An isomorphism of spin structures is
a map BSpin Ñ BSpin1 such that the induced map on On -bundles commutes with the isomorphisms
to BO pM q. The opposite spin structure to BSpin Ñ M is the complement of BSpin in the principal
Pinń -bundle associated to the inclusion
20
Spinn ãÑ Pin´ n . If M admits spin structures, then the
collection of spin structures forms a groupoid whose set of equivalence classes SpM q is a torsor for
H 0 pM ; Z{2ZqˆH 1 pM ; Z{2Zq; the action of a function δ : π0 M Ñ Z{2Z in H 0 pM ; Z{2Zq sends a spin
structure to its opposite on components where δ “ 1 is the nonzero element. The automorphism
group of any spin structure is isomorphic to H 0 pM ; Z{2Zq; a function δ : π0 M Ñ Z{2Z acts by the
central element of Spinn on components where δ “ 1. The manifold M admits spin structures if
and only if the Stiefel-Whitney classes w1 pM q, w2 pM q vanish.
(2.52) Spacetime fields for the oriented Type II superstring. We defer the orientifold construction
but do allow spacetime X to be an orbifold. In the super case there are new fields. Most importantly,
the B-field has a different geometric structure than in the oriented bosonic superstring (2.20).
s : Man
Spacetime in the superstring is 10-dimensional, as mentioned in (2.19). Let F10 10 Ñ Set∆
20Recall that Pin˘

n sits in the Clifford algebra Cℓn . Either sign can be used to construct the opposite spin structure.
denote the sheaf of fields in the Type II superstring. As in (2.34) we allow the domain Man10
to be replaced by the category of 10-dimensional orbifolds and local diffeomorphisms. The fields
s pXq are:
in F10
(i) a metric on X;
(ii) a real scalar field on X;
(iii) a differential twisting βq of K-theory on X (see (1.109));
(iv) a spin structure κ on X;
(v) a Ramond-Ramond field;
(vi) fermionic fields.
The local information in the B-field (iv) is the same as in the B-field in the oriented bosonic
string: it is a closed 3-form H P Ω3 pXq. But the extra global torsion nonnegative homotopy groups
in the cohomology theory Σ´1 R, as opposed to the theory Σ3 HZ, carries relevant information about
q it has an equivalence
the superstring. Let β denote the (nondifferential) twisting which underlies β;
class in R´1 pXq. From the short exact sequence (1.50) we deduce classes
(2.53) q P H 0 pX; Z{2Zq,

t “ tpβq q P H 1 pX; Z{2Zq
a “ apβq
which are topological invariants of the B-field. The class t is the type of the theory. In the usual
nomenclature
t “ 0 is the Type IIB superstring; and

(2.54)
t “ 1 is the Type IIA superstring.
(If X is not connected, then there is a type—A or B—on each component.) One interpretation of
the class a is to define a second spin structure κ ` a on X. Then we consider κ as the “left-moving”
spin structure and κà as the “right-moving” spin structure. These correlate to two spin structures
on the worldsheet, which we define below.
Remark 2.55. We could, therefore, organize the data differently for the Type II superstring. But
that different organization would not generalize to the orientifold.
We do not discuss the Ramond-Ramond field in detail in these lectures, so we only make a few
comments here. One salient point is that the Ramond-Ramond field is self-dual. This means its
quantization is treated differently from that of the other fields in the theory. As with all self-dual
fields we focus on the current rather than the gauge field; see (2.9) for a reminder about currents
and gauge fields in the more familiar context of Maxwell electromagnetism. The Ramond-Ramond
current represents an element of R qβqpXq, twisted differential K-theory group on an orbifold. Here
the relevant equivariant version of K-theory is the Atiyah-Segal theory based on equivariant vector
bundles, or equivariant families of Fredholm operators.
Also, we do not treat the fermionic fields on spacetime in these lectures, though it is interesting
to fit them (and the action) into our picture of the other fields.
Remark 2.56. When we use spacetime fields as fixed external data for the worldsheet theory, we
set the Ramond-Ramond field and fermionic fields to zero.
50 D. S. FREED
(2.57) Worldsheet fields for the oriented Type II superstring. Fix a Type II spacetime background,
by which we mean a smooth 10-dimensional orbifold X and a set of fields in F10 s pXq. As just
remarked, we assume that the Ramond-Ramond field and fermionic fields vanish. The worldsheet
fields F2s rXspΣq on a 2-manifold Σ are:
(ii) a spin structure αℓ which refines the orientation o, and a spin structure αr which refines
the opposite orientation ó;
(iii) a metric;
(iv) a scalar field φ with values in X;
(v) spinor field ψℓ , ψr on Σ with coefficients in φ˚ pT Xq;
(vi) spinor fields χℓ , χr on Σ with coefficients in T ˚ Σ.
The fields (i), (iii), and (iv) are as in the oriented bosonic string (2.22). The fields (ii), (v), (vi)
are new. We emphasize: the spin structures αℓ , αr are independent of each other.
Recall that a spin structure is a trivialization of the first two Stiefel-Whitney classes w1 , w2 , which
detect the bottom two stages of the Postnikov tower for the classifying space BO. An orientation
is a trivialization of w1 , or the bottom stage of the Postnikov tower. It is in that sense that a spin
structure can refine an orientation. See (2.50) for a more concrete description. Physicists speak in
terms of left-movers (ℓ) and right-movers (r), nomenclature which derives from the wave equation
on two-dimensional Minkowski spacetime. There is one spin structure for each orientation. When
we come to orientifolds there is no global orientation, as we have already seen in (2.39), but locally
there are still two spin structures which refine opposite orientations.
The spinor fields ψℓ , χℓ are associated to the spin structure αℓ . The spinor field ψℓ has pos-
itive chirality and χℓ has negative chirality. The spinor fields ψr , χr are associated to the spin
structure αr . The spinor field ψr has positive chirality and χr has negative chirality. (These last
chiralities are measured with respect to ó, the underlying orientation of αr .)
(2.58) The B-field amplitude in the oriented Type II superstring. Let X be a 10-manifold—a su-
perstring spacetime—and βq a B-field on X as defined in (2.52). We define the oriented superstring
B-field amplitude, which only depends on the equivalence class of βq in R q´1 pXq. To do so we
replace (2.24) with a pushforward in differential R-theory; see (1.114). The main point is that the
cohomology theory R is Spin-oriented, that is, there is a pushforward in topological R-theory on
spin manifolds. It is the Postnikov truncation of the pushforward in ko-theory defined from the
spin structure (which by the Atiyah-Singer index theorem has an interpretation as an index of a
Dirac operator). In fact, because we are in sufficiently low dimensions we can identify it exactly
with the pushforward in ko, a fact which is useful in the proof of the Theorem 2.66 below. The
pushforward we need is
ż
(2.59) :Rq´1 pΣq ÝÑ R q´3 pptq – R{Z
pΣ,αℓ q
in differential R-theory defined using the spin structure αℓ on Σ. Then the B-field amplitude is
ż
(2.60) Spφ, αℓ q “ q
φ˚ β.
pΣ,αℓ q
(2.61) Special B-field amplitudes. There is an isomorphism
(2.62) q´1 pptq – R´1 pptq – Z{2Z.

R
Let ηq denote the generator. It is a universal B-field on any spacetime X: pullback using X Ñ pt.
There is an explicit model for the topological class η P R´1 pptq underlying the differential class ηq.
(Because of the first isomorphism in (2.62) there is no extra information in the differential class.)
Following Atiyah-Bott-Shapiro we use the model for KOpptq in terms of Clifford algebras, as
described at the beginning of Lecture 1, and the fact that R is a Postnikov truncation of ko means
we just need give a model for the generator of KO´1 pptq. This is the real super vector space R1|1
with Clifford generator
ˆ ˙
0 ´1
(2.63) γ“
1 0
This already appeared in Lecture 1 around (1.147).

The B-field amplitude for this universal B-field is independent of φ and is an important function
in the theory of spin 2-manifolds. We explain the statement here, cribbing from 2 and refer to that
reference for proofs and more elaboration. Let pΣ, oq be a closed oriented surface and SpΣ, oq the
set of equivalence classes of spin structures which refine the given orientation. Note SpΣ, oq is a
torsor for H 1 pΣ; Z{2Zq. Let
(2.64) q : SpΣ, oq ÝÑ Z{2Z
be the affine quadratic function which distinguishes even and odd spin structures. It dates back
to Riemann and is the Kervaire invariant in dimension two. The characteristic property of the
quadratic function q is
(2.65) qpαà1 à2 q´qpαà1 q´qpαà2 q`qpαq “ a1 ä2 , α P SpΣ, oq, a1 , a2 P H 1 pΣ; Z{2Zq,
where a1 ¨ a2 P Z{2Z is the mod 2 intersection pairing.
Theorem 2.66. Let ηq be the nonzero universal B-field. For any superstring worldsheet φ : Σ Ñ X,
the B-field amplitude is p´1qqpαℓ q .
This demonstrates that the B-field amplitude (2.60) is sensitive to the worldsheet spin structure.
(2.67) Physics interpretation of Theorem 2.66. The perturbative superstring is a 2-dimensional

supergravity theory. In nongravitational field theories the intrinsic geometry which is present in
the theory is fixed. This includes the underlying manifold, topological structures on its tangent
bundle, and a metric or conformal structure. In the language of Remark 2.8 these are background
fields. But in a gravitational theory they are dynamical, which means they are integrated over in
the path integral formulation. In particular, in the perturbative superstring one sums over the spin
structures.
52 D. S. FREED
Now in path integrals over spaces of fields with many components there are often signs or phases
which are attached to each component. They go by different names: “θ-angles”, “discrete torsion”,
etc. But they are not arbitrary: they must be derived from local computations and obey all of the
gluing laws that non-locally-constant quantities obey. In this case the spin structures αℓ , αr can
be used to distinguish components of field space, and so there is a possibility of phases entering.
In fact, there are signs which enter into the usual formulation of the sum over spin structures,
and the precise choice of those signs governs the distinction between Type IIB and Type IIA.
In our approach these signs are embedded in the B-field amplitude because of our choice of the
cohomology theory R´1 to quantize the B-field charges. Theorem 2.66 expresses the signs used to
go from Type IIB to Type IIA, and the signs agree with those in the traditional approach.
The complete Type II superstring

Now we include the orientifold construction into the Type II superstring. Rather than repeat
the gauging procedure described in (2.39), we simply use the formulation in (2.40) and (2.41),
adapted to the Type II superstring. As mentioned earlier, we use ‘Type II superstring’ to include
the orientifold, and use ‘oriented Type II superstring’ if it is absent.
(2.68) More on spin structures. We mentioned in (2.44) that a real vector bundle V Ñ X de-
termines a twisting τ h pV q of any multiplicative cohomology theory h. It includes the rank of the
vector bundle and is an ingredient in the general Thom isomorphism theorem. An h-orientation
(see (1.114)) is a trivialization of τ h pV ´ rank V q, where rank V : X Ñ Z is the rank. Such a
trivialization is an isomorphism 0 Ñ τ h pV ´ rank V q in the (multi-)groupoid of h-twistings. For
h “ KO this is a spin structure. A twisted notion of spin structure enters into (2.69) below.
xs 10 pXq,
(2.69) General Type II background. Let X be a 10-dimensional orbifold. The fields in F
the spacetime fields including the orientifold, are:
a double cover Xw Ñ X, called the orientifold double cover ;
(i)
(ii)
a metric on X;
(iii)
a real scalar field on X;
a w-twisted differential twisting βq of X (see (1.109));
(iv)
a “twisted spin structure” κ : ℜpβq Ñ τ KO pT X ´ 10q;
(v)
(vi)
a Ramond-Ramond field;
(vii)
fermionic fields.
The equivalence class of βq lies in R
qw´1 pXq. The Ramond-Ramond current is now also twisted by w
and represents a twisted differential KR-theory class on Xw . A concrete geometric model for the
twisted spin structure (v) is given in the last section of 2 . There is a topological constraint forced
by the existence of a twisted spin structure:
q
w1 pXq “ tpβqw
(2.70)
q ` tpβqw
w2 pXq “ apβqw q 2
q apβq
where tpβq, q are defined in (2.53). These equations generalize the topological constraints
w1 pXq “ 0, w2 pXq “ 0 imposed by an ordinary spin structure.
(2.71) The Type I superstring. There is a special case of the Type II superstring which is important
in string theory: the Type I superstring. (The nomenclature derives from that in supergravity
theories. For example the ‘I’ and ‘II’ reflect the amount of supersymmetry present in these theories.)
In this case Xw “ Y is a 10-dimensional orbifold with trivial involution; the quotient is X “
Y ˆ BZ{2Z, where BZ{2Z is defined in (1.135). The B-field reduces to an object representing a
q 3 pY q Ñ R
class in H 2 pY ; R{Zq Ă H q´1 pY q; it is a good exercise to see why this is so. The twisted
spin structure reduces to an ordinary spin structure on Y .
(2.72) General Type II superstring worldsheet fields. Fix a Type II background, which means a
smooth 10-dimensional orbifold X and a set of fields in F xs pXq. Then the worldsheet fields in that
background form a sheaf F xs 2 rXs : Man2 Ñ Set whose fields on a 2-manifold Σ are:
p Ñ Σ;
p: Σ
(i) a spin structure α on the total space of the orientation double cover π
(ii) a metric on Σ;
(iii) a map φ : Σ Ñ X;
(iv) an isomorphism ν : φ˚ w Ñ w1 pΣq;
(v) p with coefficients in π
a positive chirality spinor field ψ on Σ p˚ φ˚ pT Xq;
(vi) p with coefficients in T ˚ Σ
and a negative chirality spinor field χ on Σ p (the gravitino).
The spin structure and the spinor fields are the same locally as in the oriented Type II superstring,
but the absence of a global orientation makes the description in terms of the orientation double
cover natural. As far as we know, this description of the spin structure does not appear in the
string theory literature, even for Type I. In particular, we might have thought that the unoriented
superstring would have a pin structure, but this is not the case. The local picture is as in the oriented
case, and in that case the two worldsheet spin structures αℓ , αr are independent; see (2.57). Note
that a pin structure would restrict α since then it would be isomorphic to its pullback by the deck
transformation of Σ p Ñ Σ, which would be analogous to requiring in the oriented case that αr be
the opposite spin structure to αℓ .
Remark 2.73. To illustrate, suppose that the superstring orientifold worldsheet Σ is diffeomorphic
to a 2-dimensional torus. Even though Σ is orientable, the fields (2.72) do not include an orien-
tation. The field α is equivalent to a pair of spin structures α1 , α2 on Σ with opposite underlying
orientations. Up to isomorphism there are 4 choices for each of α1 , α2 , so 16 possibilities in total.
Of those 4 refine uniquely to pin´ structures on Σ.
Exercise 2.74. Check that if the double cover Xw Ñ X in the Type II background is trivial,
then (2.72) reduces to (2.57). The first step is Exercise 2.43. Then you’ll need to reconcile the
description of the spinor fields.
(2.75) B-field amplitude in the general Type II superstring. Now we come to a puzzle, which
we won’t resolve until the next lecture. How do we combine (2.59) and (2.46) to integrate the
pullback φ˚ βq of the B-field? Well, using the isomorphism ν the pullback νφ˚ βq is computed by the
54 D. S. FREED
composition (compare (2.48))
˚
(2.76) qw´1 pXq ÝφÝÑ R
R qφ˚ w´1 pΣq ÝνÑ R
qw1 pΣq´1 pΣq.
But the cohomology theory R is oriented for spin manifolds, and w1 -twisted classes are not “den-
sities” in the sense of (2.44); pw1 , w2 q-twisted classes are densities. We might try to use the spin
structure α on the worldsheet, but it does not move us to densities. Conclusion: There is no obvious
combination of the data which produces a quantity we can integrate in differential R-theory.
(2.77) Pin structures. While not directly relevant to the physics, we can consider the case when
Σ has a pin´ structure. (There are two types of pin structures: pin` and pin´ . Any 2-manifold
admits pin´ structures, but not every 2-manifold admits pin` structures.) Recall from (1.92) that
the tangent bundle T Σ Ñ Σ has an associated KO-twisting τ KO pΣq “ τ R pΣq. The pin´ structure
provides an isomorphism of that twisting with the twisting defined by the orientation double cover:
–
Ñ w1 pσq. We sketch that in the exercise below. Thus the pin´ structure determines a
τ R pΣq Ý
pushforward, or integration, of w1 pΣq-twisted R-theory classes, so in particular a definition of the
q see (2.76).
integral of νφ˚ pβq;
As in (2.61) we can evaluate these amplitudes for a universal B-field θq which is the differential
refinement of θ in Theorem 1.152. (Note that R qw0 ´1 pBZ{2Zq – Rw0 ´1 pBZ{2Zq, so there is no
additional information in the differential refinement.) In §4 of 2 we prove that the resulting integral
is a Z{8Z-valued quadratic function on the H 1 pΣ; Z{2Zq-torsor of pin´ structures. It is the Kervaire
´
invariant of pin´ surfaces, and it induces an isomorphism ΩPin 2 Ñ Z{8Z, where the domain is the
´
bordism group of pin surfaces. This provides a (twisted) R-theory, or KO-theory, interpretation
of this Kervaire invariant.
Lecture 3: Worldsheet anomalies
In this lecture we work exclusively with the worldsheet theory. Thus fix a smooth 10-dimensional
orbifold X and a set of background fields in F xs 10 pXq, as listed in (2.69). The only fields which
play a role are the orientifold double cover Xw Ñ X, the B-field β, q and the twisted spin struc-
KO x s
ture κ : ℜpβq Ñ τ pT X ´ 10q. The worldsheet fields F 2 rXs are listed in (2.72). Let Σ be a
closed 2-manifold. We use all of the worldsheet fields on Σ..
There are two quantities in the effective exponentiated action on which we focus. The first is
obtained by integrating out the two spinor fields ψ, χ. The fermionic path integral over these fields
may be treated formally, and the result is the pfaffian of a Dirac operator on Σ. p It is not a number,
but rather an element of a line,21 the Pfaffian line LPfaff . The second quantity is the exponential
of the B-field amplitude, which was not defined in (2.75). We will define it in this lecture as an
element of another line, which we call the B-field line LB , which depends on the B-field. What we
would like to assert is that the data in the theory gives a trivialization of the tensor product of the
Pfaffian line and the B-field line. Furthermore, it is the twisted spin structure κ, pulled back via φ,
which gives the trivialization. We do not have a direct construction of a geometric trivialization, so
instead prove the weaker statement that a trivialization exists. The stronger statement would be
an example of a “categorified index theorem”; the statement we prove is a geometric index theorem
whose proof leans heavily on results of Atiyah-Patodi-Singer.
The reader may rightly object that any two complex lines are isomorphic—no argument from
the author there—so to get a meaningful statement we work in families of surfaces over a variable
base S. Then LPfaff Ñ S, LB Ñ S are flat hermitian line bundles (in fact of order 2, though that
is not directly evident for LPfaff ), so their isomorphism classes are elements of H 1 pS; R{Zq. Our
main result is
Theorem 3.1. The flat line bundle LPfaff b LB Ñ S is trivializable.
The line bundle LPfaff Ñ S is defined analytically (see §3 of 22), whereas LB Ñ S is defined
purely topologically. Note that the isomorphism class of a flat line bundle is determined23 by the
holonomies around all loops in S, so Theorem 3.1 is an equality between sets of numbers. To
compute these numbers we take the base of the family of surfaces to be the circle S 1 , and so the
total space is a 3-manifold: N 3 Ñ S 1 . Then the holonomy of LPfaff around the base S 1 is computed
by an Atiyah-Patodi-Singer η-invariant. We use a geometric index theorem—a version of the “flat
index theorem” in the 3rd APS paper—to relate it to the holonomy of LB .
We can say this all a bit more nicely in the differential theory, though not every statement has
been proved. (The main theorem in the PhD thesis of Kevin Klonoff locates the η-invariant in
differential K-theory, but does not prove the refinement in KO-theory in dimension 3 that we
need here.) To do so we need a differential version of the the real24 index theorem for families,
21The Pfaffian line and the B-field line below should be regarded as Z{2Z-graded, though the gradings are even
for both.
22D. S. Freed, On determinant line bundles, in “Mathematical Aspects of String Theory” (ed. S. T. Yau), World
Scientific Publishing, 1987.

23This is true for any line bundle with connection; the flatness means the holonomy depends only on the homology
class of the loop.

24There are differential version of the index theorem in complex K-theory, for example in D. S. Freed, J. Lott,An
index theorem in differential K-theory, Geometry & Topology, 14 (2010), pp. 903–966, (arXiv:0907.3508).
56 D. S. FREED
the topological version of which is in the fifth of the classic Atiyah-Singer series of papers. That
theorem asserts that an analytic index is computed by a pushforward in KO-theory. We use a
truncation of the analytic index—the Pfaffian line bundle—and correspondingly a truncation—the
cohomology theory R—of KO-theory. It is very pretty that these truncations match. This works
nicely because we are dimension 2. In higher dimensions we could also truncate KO-theory to
compute the Pfaffian line bundle, but the truncation would keep more homotopy groups and would
not be as geometric as in this low dimension. This relation to the index theorem, and the fact that
the truncation computes the Pfaffian line bundle, is one of the many pieces of supporting evidence
for our choice (2.52) of Dirac quantization condition on the B-field. Again, we need the differential
version of all this, but only in truncated form for the Pfaffian line bundle.
In this lecture we sketch a proof of Theorem 3.1. Warning: Some details in these notes are not
correct and hopefully a corrected version will appear as a paper in due time.
Digression: a categorified index theorem

There is a much easier analog of Theorem 3.1 in supersymmetric quantum mechanics, where
it is an index theorem on 1-manifolds rather than on 2-manifolds. The pfaffian in this case is
computed by an even simpler truncation of KO-theory: mod 2 cohomology! (We get there by
adding the simplifying hypothesis that spacetime X in that supersymmetric quantum mechanics
theory is oriented.) The topological theorem is that the Pfaffian line bundle is computed by
transgressing w2 pXq. The categorified version is that a trivialization of w2 pXq—a spin structure
on X—induces a trivialization of the Pfaffian line bundle. In more detail: the index formula in this
case has the shape
ż
(3.2) LPfaff – φ˚ w2 pXq.
C
If we interpret w2 pXq as a cohomology class, then the right hand side computes the isomorphism
class of LPfaff . But we claim that the formula makes sense on the level of geometric objects and their
isomorphisms: the formula (3.2) actually computes the Pfaffian line bundle, and we can integrate
– –
a trivialization 0 ÝÑ w2 pXq to a trivialization 0 Ý
Ñ LPfaff .
A version of this claim is explained in §5.2 of 2 , though not precisely as an integration for-
mula (3.2). Nonetheless, it is proved that LPfaff has a canonical trivialization if X is spin. The
proof there is a bit of a poof, and the ambitious reader will enjoy writing out more details; the
argument is very explicit and geometric.
A twisted R-class on Σ
p to define a class
In this section we use the spin structure α on Σ
(3.3) δ P Rτ0 pΣq,
where
(3.4) τ0 “ τ R pΣq ´ 2
is the reduced R-twisting defined by the tangent bundle T Σ Ñ Σ. Since R is a truncation of KO,
we have τ R “ τ KO . Explicit models for τ KO pΣq are given in (1.92). We will also give a lift of δ to
a flat differential class
(3.5) δq P R
qτ0 pΣq.
(3.6) An easy construction. Recall from (2.68) that we can regard the spin structure α on the
p Ñ Σ as an isomorphism
p: Σ
total space of π
α p ´ 2q – π
(3.7) 0 ÝÑ τ KO pT Σ p˚ τ KO pT Σ ´ 2q “ π
p ˚ τ0 .
Now integrate (push forward) (3.7) over the double cover π p. We no longer obtain an isomorphism of
twistings, but rather a twisted KO-theory class. Put differently, we can interpret α as an invertible
˚
p let yα P Rπp ˚ τ0 pΣq
twisted KO-class xα P KOπp τ0 pΣq; p be its truncation to a twisted R-class. Then
define
(3.8) p˚ pyα q.
δ“π
Remark 3.9. Recall from (2.44) there is an integration map

ż
p
R pT Σ´2q`q
p ÝÑ Rq´2 pptq.
(3.10) : Rτ pΣq
p
Σ
The spin structure α, in the guise of the class yα , is an orientation with which we define integration
of untwisted classes:
ż
p ÝÑ Rq´2 pptq
: Rq pΣq
p
pΣ,αq
(3.11) ż
z ÞÝÑ yα z.
p
Σ
Exercise 3.12. What is the analog of (3.11) in de Rham theory?
(3.13) An explicit model for δ and its differential lift. Let BO pΣq Ñ Σ denote the principal O2 -
bundle of orthonormal frames of Σ. Recall that the orientation double cover Σ p carries a canonical
orientation, and then observe that an orthonormal frame induces an orientation, so there is a map
BO pΣq Ñ Σp which in fact is a principal SO2 -bundle, the oriented orthonormal frame bundle of Σ.p
The spin structure α on Σ p is a principal Spin2 -bundle Br Ñ Σp together with a map Br Ñ BO pΣq
which is a quotient map for the action of the central Z{2Z Ă Spin2 .
Let K Ñ BO pΣq be the even real flat line bundle associated to the double cover Br Ñ BO pΣq.
The spin structure α also determines an odd real flat line bundle ∆ Ñ Σ as follows. Locally
on Σ the Spin2 -action on Br extends in two ways to a Pin´ r
2 -action on B which covers the O2 -action
on BO pΣq. To see this observe that there are two choices for the action of each element in the
58 D. S. FREED
non-identity component of Pin´ 2 , and because we have a Spin2 -action once a choice is made for one
element the choice for every other element is determined. These two local canonical actions define
a global double cover of Σ; let ∆ Ñ Σ be the associated real line bundle, which we take to be odd.
Consider the Z{2Z-graded line bundle
(3.14) δq : K ‘ K∆ ÝÑ BO pΣq,
where we find it more attractive to omit the ‘b’ sign: K∆ “ K b ∆. The definition of ∆ in
terms of local Pin´ ´
2 -actions gives a canonical Pin2 -action on (3.14) which: (i) is compatible with
´
the nontrivial grading Pin2 Ñ Z{2Z in the sense that elements in the identity component Spin2
preserve the grading of K ‘ K∆ and elements in the non-identity component reverse it; and
(ii) has the property that the center Z{2Z Ă Spin2 Ă Pin´ 2 acts by scalar multiplication on the
fibers. (Recall the definition of K.) Recalling the second model of τ KO in (1.92), we see that
(3.14) represents a class δ P Rτ0 pΣq, and we claim it is the same class as defined in (3.6).
Exercise 3.15. Verify this claim. To do so, identify xα in this model.
The following exercise shows explicitly that δ measures the failure of α to be a pin´ structure.
Exercise 3.16. Let Π be the trivial odd real line, and we use the same symbol to denote the
constant line bundle with fiber Π over any space. Verify that a refinement of α to pin´ structure
on Σ is an isomorphism Π Ñ ∆. If it exists, show that the equivalence class of (3.14) is the zero
element in Rτ0 pΣq.
Exercise 3.17. Suppose Σ has an orientation o. Then we can encode α as two spin structures αℓ , αr
on Σ which refine o, ó, respectively, as in (2.57). Prove that in this case the odd line bundle
∆ Ñ Σ represents the difference αℓ ´ αr . The fact that ∆ is odd reflects the different orientations
underlying αℓ , αr . Use the spin structure αℓ to trivialize τ KO pΣq and so identify (3.14) with the
untwisted KO-class on σ represented by the super line bundle R ‘ ∆. This requires an explicit use
of Definition 1.62 and the associated “Morita isomorphism” on twisted vector bundles.
Finally, recall that K, ∆ come with a canonical flat connection, whence so to does (3.14). This
flat connection lifts δ to a differential class δq P R
qτ0 pΣq.
(3.18) The class δ is torsion of order 8. We use the class δq in families of surfaces, so in particular
on a 3-manifold N which fibers over S 1 .
Lemma 3.19. On the 3-manifold N we have 8δq “ 0.
A proof in case Σ is orientable (Exercise 3.17) is buried in §4 of 25; the general case will be a small
modification based on the following universal description of a spin structure on the orientation
double cover.
25D.
S. Freed, E. Witten, Anomalies in string theory with D-branes, Asian J. Math, 3 (1999), 819–851,
(arXiv:hep-th/9907189).
r be
Exercise 3.20. Let i : BSOn Ñ BSOn be the free involution whose quotient is BOn . Let E
defined as the pullback in the diagram
(3.21) r
E B Spinn ˆB Spinn
id î
BSOn BSOn ˆ BSOn
The diagram is compatible with the involution i in the lower left corner and the involution which
exchanges the factors in each entry of the right column. Let E denote the quotient of the induced
r the diagram produces a map E Ñ BOn . Prove that if V Ñ M is a rank n real
involution on E;
vector bundle with classifying map M Ñ BOn , then a spin structure (up to equivalence) on the
double cover of M defined by orientations of V is a lift (up to homotopy) of the classifying map
to E. Construct the universal version of δ.
The Pfaffian line
(3.22) Geometric setup. Let f : M Ñ S be a fiber bundle whose fibers are closed 2-manifolds. The
Riemannian metric on the family is two pieces of data: a metric on the rank 2 relative tangent bundle
T pM {Sq Ñ M and a horizontal distribution on M , which is a complement to T pM {Sq Ă T M . The
fiberwise orientation double cover π p: Mx Ñ M is another fiber bundle f ˝ π p of closed 2-manifolds
over S, and the metric data pulls back to f ˝ π p. The spin structure α is a spin structure on the
relative tangent bundle T pM x{Sq Ñ M x. The field φ is a map φ : M Ñ X. The fermionic functional
integral over the fields ψ, χ in (2.72) is the pfaffian of the family of Dirac operators on the fiber of
f ˝π p Ñ S coupled to the virtual bundle
p: Σ
` ˘
(3.23) p˚ φ˚ pT Xq ´ T ˚ pM {Sq ,
π
which has a covariant derivative from the metrics on X and M {S. (This is a shorthand for the
ratio of the pfaffians for Dirac coupled to each bundle separately.) Note that the bundle (3.23)
is real, and because we are in dimension 2 pmod 8q the Dirac operator is complex skew-adjoint,
which is why there is a pfaffian. The pfaffian is a section of the Pfaffian line bundle LPfaff Ñ S,
which is a hermitian line bundle with a covariant derivative. The theorems in 22 give formulas for
its curvature and holonomy.
Exercise 3.24. Use the curvature formula to prove that LPfaff Ñ S is flat. (Hint: The integrand
p˚ ω of a differential form ω P Ω‚ pM q, since the metric data is pulled
in that formula is the pullback π
back from M , and because the deck transformation of π p reverses orientation, the composition
pq˚ ˝ π
pf ˝ π ˚
p is the zero map.)
(3.25) The holonomy formula. Suppose S “ S 1 , and as before we use ‘N ’ to denote the 3-manifold
which is the total space of f : N Ñ S 1 . Then the holonomy formula asserts that the holonomy is
the exponential of 2πi times
(3.26) p q{2
ξpN pmod 1q,
60 D. S. FREED
where ξpN p q is the Atiyah-Patodi-Singer invariant (roughly half of the η-invariant) on N p . The
extra factor of 2 is because we take the pfaffian, not the determinant. The ξ-invariant is for the
Dirac operator on N , using the bounding spin structure on the base S 1 , and coupled to the virtual
bundle (3.23). Because the curvature vanishes, this invariant is topological, independent of the
metric data.
p {S 1 q Ñ
(3.27) Replacing the cotangent bundle to the surface. The relative cotangent bundle T ˚ pN
1
S is spin, which means that its KO-class on a 3-manifold is equal to its rank, which is 2. Thus
the Pfaffian line bundle, as a flat bundle, is unchanged if we replace (3.23) with
(3.28) p˚ φ˚ pT X ´ 2q.
π
(3.29) APS index theorem. Let’s consider the special case in which the relative tangent bundle
of f : N Ñ S 1 carries an orientation o. That gives a section of πp, and so, combining with the
1
spin structure on the base S , spin structures αℓ , αr on N with opposite underlying orientation.
Then (3.26) reduces to the sum
ξαℓ pN q ξαr pN q
(3.30) ` pmod 1q
2 2
of half ξ-invariants for the two spin structures, this time ξ-invariants on N . That sum is computed
by the flat index theorem26 as a pushforward, or integral, in KO-theory with R{Z coefficients. We
reinterpret it in differential KO-theory. As in Exercise 3.17 the difference of the spin structures
is represented by the odd flat real line bundle ∆ Ñ N . According to Exercise 3.17 the class δq is
represented in this situation by the bundle R ‘ ∆ Ñ N with its flat covariant derivative. The APS
index theorem computes (3.30) as
ż pKOq
}
(3.31) δq ¨ φ˚ rT X ´ 2s,
pN,αℓ q
where the integral is in differential KO and uses the KO-orientation from the spin structure αℓ .
} ´3 pptq – R{Z. Because we are in low dimension, the integral only depends
The integral lands in KO
on the truncation to R-cohomology, and so can be written as
ż pRq
q
(3.32) δq ¨ φ˚ rT X ´ 10s,
pN,αℓ q
where we now interpret δq P R

q0 pN q, which is a flat element, and φ˚ rT X ´ 10s P R0 pN q. Also, we
have used Lemma 3.19 to change ‘2’ to ‘10’.
26adopted to KO-theory
Remark 3.33. Since δq is flat, the product in the integrand does not depend on a differential refine-
ment of φ˚ rT X ´ 2s. We use this property of the product in differential cohomology throughout. It
means that all expressions we write are, in fact, products in topological (generalized) cohomology
with R{Z coefficients.
Finally, we reconfigure (3.32), which is a formula for the holonomy, to a formula for the isomor-
phism class of the flat line bundle LPfaff Ñ S, where now we work with a family f : M Ñ S over
an arbitrary base S. Namely, we claim
ż
(3.34) rLPfaff s “ δq ¨ φ˚ rT X ´ 10s P R
q´2 pSq.
pM {S,αℓ q
Here the integral is in differential R-theory. It remains to note that R´2 has nonzero homotopy
groups π0 “ Z{2Z, π2 – Z with nontrivial k-invariant (see (1.47) and (1.48)), so is represented by
the Picard groupoid of Z{2Z-graded complex lines. Hence elements of R q´2 pSq are represented by
Z{2Z-graded hermitian line bundles with covariant derivative over S. (There is a bit of a mismatch
in the formulas and we might better exponentiate the integral on the right hand side of (3.34)—after
multiplying by 2πi—to interpret it as a line bundle. But we won’t bother in these notes.)
(3.35) An extension to the general case. We can arrive at (3.34) by a more direct route, which
we now employ in the general case. Thus given a family f : M Ñ S with all the data in (2.72) and
no orientation assumption, we presume a truncated differential index theorem which tells that the
Pfaffian line bundle of the family of Dirac operators on Mx Ñ S coupled to (3.28) (with ‘2’ replaced
by ‘10’, as remarked after (3.32)) is
ż
` ˘
(3.36) rLPfaff s “ p˚ φ˚ rT Xs ´ rT ˚ pM {Sqs
π q´2 pSq.
PR
x{S,αq
pM
As in (3.11) we rewrite this as

´ ` ˘¯
˚ ˚ ˚
(3.37) pq˚
rLPfaff s “ pf ˝ π yqα ¨ π
p φ rT Xs ´ rT pM {Sqs ,
where now the integral is written as a pushforward and we use a (flat) differential refinement of yα .
The pushforward does not use any spin structure and is defined because yqα is twisted by the relative
p. Carry out π
tangent bundle of f ˝ π p˚ , use the push-pull formula, use the differential version of
formula (3.8), and make the same substitutions for T ˚ pM {Sq as in (3.28) and (3.32) to derive
` ˘
(3.38) rLPfaff s “ f˚ δq ¨ φ˚ rT X ´ 10s .
This is our final formula for the Pfaffian line bundle.
q
More twisted R-classes on Σ
We introduce some new characters. As in Exercise 3.16 let Π be the trivial odd real line, the
trivial vector space R regarded as odd. We use the same notation for the constant line bundle over
any space.
62 D. S. FREED
(3.39) The class q

ǫ. First, as a variation of (3.14), consider
(3.40) q
ǫ : K ‘ K∆Π Ñ BO pΣq.
This is an even flat vector bundle of rank 2, and it has an action of Pin´2 , but now the action is
purely even. So it represents a twisted class where the twisting only senses the central extension,
not the orientation, and thus the twisting is τ0 ´ w1 pΣq. We denote this class as
(3.41) q qτ0 ´w1 pΣq pΣq.

ǫPR
Note that the “curvature” of q

ǫ, in the sense of differential cohomology, is the rank of (3.40), which
is the constant function 2.
Exercise 3.42. Use the explicit model of the twisting and the symmetric monoidal structure on
twistings to verify that q q
ǫ lives in the twisted R-cohomology group indicated in (3.41).
(3.43) Twice the B-field amplitude. Recalling (2.76) we see that we can integrate q q over Σ,
ǫ ¨νφ˚ pβq
since this product is pτ0 ´1q-twisted. This integral gives a number in R{Z. That would seem to give
a definition of the B-field amplitude, but the problem is that the curvature has an extra factor of 2,
from the curvature of q ǫ, and that means that the curvature of this integral in a family parametrized
q and this is not what we want. (See
by S is the twice the transgression of the 3-form curvature of β,
the computation (3.66) below.) In other words, we have twice the B-field amplitude as a function
or, exponentiating, its square as a function S Ñ T. We want to take a square root of this function,
and the square root is naturally a section of a flat hermitian line bundle of order 2. We give a
differential cohomology version of the construction.
Exercise 3.44. Construct the square root geometrically: Given a function h : M Ñ T on a smooth
manifold, construct a flat hermitian line bundle L Ñ M , a section s, and a trivialization of Lb2
such that sb2 “ h. (Hint: Solve the universal problem M “ T and h “ idT .)
q and its trivialization ζ.

(3.45) The class λ q We easily compute that R1 pptq “ 0, whereas
(3.46) q1 pptq – R
R q0 ppt; R{Zq – R{Z
q to be a representative of 1{2 pmod 1q in the group (3.46). We

and every element is flat. Define λ
can make an explicit model as the vector space
(3.47) q :R‘Π
λ
with Clifford action

ˆ ˙
0 1
(3.48) γ` “ .
1 0
The differential lift is defined by simply specifying the number 1{2 P R{Z – R q1 pptq. The underlying
topological class in R1 pptq vanishes, of course, and that can be seen since there is an extra Clifford
generator
ˆ ˙
0 ´1
(3.49) γ´ “
1 0
which graded commutes with (3.48). (The paper of Atiyah-Bott-Shapiro explains the sense in which
this provides a trivialization of λ.) We claim (3.49) also provides a trivialization ζq of λ
q in the differ-
ential theory without any extra data. This is because differences of differential trivializations form
the group Rq0 pptq, and R q0 pptq – R0 pptq, the latter being the group of differences of trivializations
of the underlying topological element λ.
Remark 3.50. This can probably be said better in a model with superconnections, which would
give some nontrivial data even over pt, but I haven’t yet worked that out. Also, I don’t have time
to write now about what trivializations mean in the differential world. You can work that out in
a general way by starting as follows. Imagine you only know about closed differential forms, not
all differential forms, and you want to “invent” a theory which includes all forms. Then define a
form η of degree q ´ 1 on a smooth manifold M to be a closed q-form ω on the cone CM and
define dη to be the restriction of ω to M Ă CM . We recover an honest form by integrating ω
over the generating line segments of the cone, and Stokes’ theorem tells its differential. Of course,
you’ll want to replace CM by the cylinder r0, 1s ˆ M and use forms which vanish at one end of
the cylinder. Imitate this in a model of differential “cocycles”, or indeed in any geometric model
of cohomology classes.
(3.51) One more class. Finally, we give a representative of the generator of
(3.52) q´w0 `1 pBZ{2Zq – R´w0 `1 pBZ{2Zq – Z{8Z.

R
It already appeared in (1.142). It is the super vector space
(3.53) q:R‘Π
χ
with Clifford generator (3.48) and the generator of Z{2Z lifted to the order four transformation
ˆ ˙
0 ´1
(3.54) .
1 0
q is flat. We remark, as we did earlier, that χ is the Euler class of the sign representation.
Of course, χ
Remark 3.55. Usually the Euler class of an odd rank vector bundle is torsion of order 2. That is
not true for twisted Euler classes, as we see here.
64 D. S. FREED
(3.56) An important isomorphism. Recapitulating, we have the four classes
qPR
λ q1 pptq
`
q´w0 `1 BZ{2Zq
qPR
χ
(3.57)
q qτ0 ´w1 pΣq pΣq
ǫPR
δq P R
qτ0 pΣq
or more accurately the symbols denote geometric representatives of the underlying differential
cohomology classes. Now there is a diagram of double covers
(3.58) p
Σ Xw pt
φ q
Σ X BZ{2Z
The left diagram is the isomorphism ν, which is one of the fields on the worldsheet. More simply:
p.
the composition is the classifying map of the double cover π
We construct an isomorphism
q¨q – q
(3.59) λ ǫ ÝÝÝÝÑ pq ˝ φq˚ pq
χq ¨ δ.
In our model this is an isomorphism of super vector bundles
–
(3.60) pR ‘ ΠqKpR ‘ ∆Πq ÝÝÝÝÑ pR ‘ ΠqKpR ‘ ∆q
over BO pΣq. The isomorphism must be even, commute with the Clifford action of (3.48) on the
first factor, commute with the Pin´ 2 -actions, and it must respect the flat connections. We specify
the isomorphism in terms of bases, working in the fibers over a fixed point of BO pΣq. Let e0 , e1 be
a basis of the first factor R ‘ Π; a0 , b0 a basis of pR ‘ ∆Πq; and c0 , d1 a basis of R ‘ ∆. (So a0 “ c0
and b1 “ Πd1 .) Define (3.60) as the map
e0 a0 ÞÝÑ e0 c0
e0 b0 ÞÝÑ e1 d1
(3.61)
e1 a0 ÞÝÑ e1 c0
e1 b0 ÞÝÑ e0 d1
Exercise 3.62. Check that (3.61) satisfies the requirements listed above: commutation with this
and that.
The anomalous B-field amplitude
(3.63) The B-field line bundle and a section. Working with a family f : M Ñ S as above we
define the B-field line bundle over S as
ż
(3.64) LB “ q¨q
λ q
ǫ ¨ νφ˚ pβq q´2 pSq.
PR
M {S
Note that the integrand is τ0 -twisted, so the integral makes sense. The B-field amplitude is the
“nonflat trivialization”
ż
(3.65) ζq ¨ q q
ǫ ¨ νφ˚ pβq
M {S
of LB . Its covariant derivative, a 1-form on S, is computed as

ż ż ż
q curvpq q “ 1
(3.66) curvpζq ǫq φ˚ pcurv βq ¨ 2 ¨ φ˚ H “ φ˚ H,
Σ{S Σ{S 2 Σ{S
where H P Ωw`3 pXq is the 3-form curvature of the B-field β,q a twisted 3-form on X. This is the
required formula for the B-field amplitude. Note that (3.66) is a closed 1-form, consistent with the
fact that LB Ñ S is flat.
As an important step towards Theorem 3.1 we apply the isomorphism (3.59) to (3.64) to conclude
ż
LB “ χq ¨ δq ¨ νφ˚ pβq
pq ˝ φq˚ pq q
(3.67) M {S
` ˘
“ f˚ δq ¨ φ˚ pq ˚ pχq ¨ βq ,
where we write the integral as a pushforward in the last step and also use the fact that δq is flat, so
the product in differential cohomology only depends on the topological class underlying the second
factor; see Remark 3.33.
Putting it all together
(3.68) The anomaly cancellation. To prove Theorem 3.1, the anomaly cancellation, we must show
that the sum of (3.67) and (3.38) vanishes. We’re off by a sign, so end up showing they’re equal.
(In any case the classes have order two.) This is where the twisted spin structure
–
(3.69) κ : ℜpβq ÝÝÝÝÑ τ KO pT X ´ 10q,
or rather its existence, comes into play. Using Lemma 1.157 and (1.165) we conclude from the
existence of (3.69) that
` ˘
(3.70) c̄ q ˚ pχq ¨ β “ c̄rT X ´ 10s P r0 pXq,
66 D. S. FREED
where rT X ´ 10s P R0 pXq is the isomorphism class of the reduced tangent bundle, defined using
the quotient map ko0 pXq Ñ R0 pXq. Working on a 3-manifold f : N Ñ S 1 , which suffices since the
flat bundles are determined by their holonomies, we conclude
` ˘
(3.71) c̄φ˚ q ˚ pχq ¨ β ´ c̄φ˚ rT X ´ 10s “ 0 P r0 pN q
where φ : N Ñ X. The table (1.129) shows the effect of the map c̄ : R Ñ r on homotopy groups,
and the vanishing in (3.71) and the fact we are on a 3-manifold imply that the class
` ˘
(3.72) φ˚ q ˚ pχq ¨ β ´ φ˚ rT X ´ 10s P R0 pN q
is equal to a multiple of 1 P R0 pN q which is divisible by 8. In fact, it vanishes since the “rank” of

each term is zero. It follows that (3.67) and (3.38) are equal, as desired.

Vienna

Uploaded by

Copyright:

Available Formats

Vienna

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Vienna

Uploaded by

Copyright:

Available Formats

LECTURES ON TWISTED K-THEORY AND ORIENTIFOLDS

Date: June 15, 2012.

Lecture 1: Models of twistings

K-theory; rather, we focus on the twistings.

Examples of twistings and twisted K-theory

(1.11) 1 ÝÑ t˘1u ÝÑ Z{4Z ÝÑ Z{2Z ÝÑ 1

and so is a twisted equivariant bundle.

For another kind of twisting we specify a homomorphism

An ǫ-twisted super bundle E “ E 0 ‘ E 1 Ñ X carries an action of G in which g P G is an even

(1.17) 1 ÝÑ T ÝÑ Spincn ÝÑ SOn ÝÑ 1

of the special orthogonal group factors through the extension

(1.18) 1 ÝÑ t˘1u ÝÑ Spinn ÝÑ SOn ÝÑ 1

To obtain something nontrivial we pass to an infinite dimensional example.

(1.19) A{{G ÝÑ G{{G.

Moral 1: To define interesting twistings we must be allowed to pass to locally equivalent

be an extension of (topological) groups and let X be a space of isomorphism classes of represen-

(1.22) 1 ÝÑ T ÝÑ Autqtm pHq ÝÑ AutpPH, pq ÝÑ 1

Invertible super algebras, bimodules, and intertwiners

(1.27) 2-category of algebras-bimodules-intertwiners. If A0 , A1 are super algebras, then an pA1 , A0 q-

The algebraic topology of Alg ˆ

under the isomorphism Autpx b xq – Autp1q.

(1.44) βZ ˝ Sq 2 : H q p´; Z{2Zq ÝÑ H q`3 p´; Zq,

where βZ is the integer Bockstein map.

The spectrum R and cAlgCˆ

(1.46) R :“ πď4 ko “ kox0 ¨ ¨ ¨ 4y.

where as above βZ is the integer Bockstein.

Bundles of invertible algebras

Definition 1.56. A vector bundle E Ñ X over a topological groupoid X is a pair E “ pE0 , ψq

(1.57) ψf2 ˝f1 “ ψf2 ˝ ψf1 .

(1.60) λf2 ,f1 : Bf2 bAx Bf1 Ñ Bf2 f1 ,

λf3 ,f2 bid λf3 ,f2 f1

λf3 ,f2 ,f1

(1.63) µf : Cx1 bAx1 Bf ÝÑ Bf1 bA1x Cx0

id bλf2 ,f1 λ1f bid

The symmetry A b A1 Ñ A1 b A is constructed from the symmetry (1.5).

Definition 1.78. Let X be a topological groupoid. A twisting on X is a pair τ “ pP, Aq consisting

(1.83) π0 TwistK pXq – R´1 pXq.

(1.85) H ‚ pX; Aq :“ H ‚ p|X|; Aq.

Twisted vector bundles and Fredholm operators

Exercise: Write the exact cocycle relation.

(1.93) 1 ÝÑ Z{2Z ÝÑ Pin´

(1.97) w-twisting a twisting. Let V be a complex super vector space. Set

Definition 1.99. Let Xw Ñ X be a double cover of topological groupoids. A w-twisted complex

Finally, in the w-twisted case diagram (1.61) reads

φpf2 f1 q φpf1 q id bλf2 ,f1 φpf2 f1 q

λf3 ,f2 bid λf3 ,f2 f1

φpf1 q λf3 ,f2 ,f1

(1.104) π0 TwistK pXqw – Rw´1 pXq,

where Rw pXq is the R-cohomology of X twisted by the double cover Xw Ñ X.

(1.105) General discussion of differential cohomology. Let h be a generalized cohomology theory.

(1.108) ΩpX; Rqqclosed ÝÑ HpX; Rqq

The curvature is now a twisted differential form H P Ωw`3 pXq.

(1.114) Integration in differential cohomology theories. Let π : X Ñ Y be a fiber bundle of smooth

(1.115) π˚ : hq pXq ÝÑ hq´n pY q.

An h-orientation induces an ordinary orientation and so an integration of differential forms

(1.116) π˚ : Ωq pX; Rq ÝÑ Ωq´n pY ; Rq.

There is a notion of a differential h-orientation, introduced in work of Hopkins-Singer, which is

We mention three cases of (1.117). For h “ HZ ordinary Eilenberg-MacLane cohomology an

(1.119) } q pXq Ñ ΩpX; Rqq

is the curvature map, then