Topos Methods in the Foundations of Physics∗

arXiv:1004.3564v1 [quant-ph] 20 Apr 2010

Chris J. Isham†
April, 2010

1 Introduction
Over forty years have passed since I first became interested in the problem
of quantum gravity. During that time there have been many diversions and,
perhaps, some advances. Certainly, the naively-optimistic approaches have
long been laid to rest, and the schemes that remain have achieved some degree
of stability. The original ‘canonical’ programme evolved into loop quantum
gravity, which has become one of the two major approaches. The other, of
course, is string theory—a scheme whose roots lie in the old Veneziano model
of hadronic interactions, but whose true value became apparent only after it
had been re-conceived as a theory of quantum gravity.
However, notwithstanding these hard-won developments, there are certain
issues in quantum gravity that transcend any of the current schemes. These
involve deep problems of both a mathematical and a philosophical kind, and
stem from a fundamental paradigm clash between general relativity—the
apotheosis of classical physics—and quantum physics.
In general relativity, space-time ‘itself’ is modelled by a differentiable
manifold M: a set whose elements are interpreted as ‘space-time points’.
The curvature tensor of the pseudo-Riemannian metric on M is then deemed
to represent the gravitational field. As a classical theory, the underlying
philosophical interpretation is realist: both the space-time and its points
truly ‘exist’1 , as does the gravitational field.
∗
Based on joint work with Andreas Döring.
†
c.isham@imperial.ac.uk
1
At least, that would be the view of unreconstructed, space-time substantivalists. How-
ever, even purely within the realm of classical physics this position has often been chal-
lenged, particularly by those who place emphasis on the relational features that are inher-
ent in general relativity.

1
 On the other hand, standard quantum theory employs a background
space-time that is fixed ab initio in regard to both its differential structure
and its metric/curvature. Furthermore, the conventional interpretation is
thoroughly instrumentalist in nature, dealing as it does with counter-factual
statements about what would happen (or, to be more precise, the probability
of what would happen) if a measurement is made of some physical quantity.
In regard to quantum gravity, the immediate question is:

How can such a formalism be applied to space and

time themselves?

Specifically, what could it mean to ‘measure’ properties of space and/or time

if the very act of measurement requires a spatio-temporal background within
which it is made? And how can we meaningfully talk about the ‘probability’
of the results of such measurements? A related question is what meaning, if
any, can be ascribed to quantum superpositions of eigenstates of properties of
space, time, or space-time. Over the years, this issue has been much discussed
by Roger Penrose who concludes that the existing quantum formalism simply
cannot be applied to space and time, and that some new starting point is
needed.
Another twist is provided by the subject of quantum cosmology which
aims at applying quantum theory to the entire universe. There is no prima
facie link between this aspiration and the subject of quantum gravity other
than that we have become accustomed to discussing cosmology using various
simple solutions to the classical equations of general relativity. However, irre-
spective of the link with quantum gravity proper, it is still thought provoking
to consider in general terms what it means to apply quantum theory to the
‘entire’ universe. Of course, this might simply be a stupid thing to do, but
if one does attempt it, then the problem of implementing instrumentalism
becomes manifest: for where is the external observer if the entire universe
is the system? In contexts like these one can see clearly the attractions of
finding a more realist interpretation of quantum theory.
The complexity of such questions is significantly enhanced by the lack of
any experimental data that can be unequivocally identified as pertaining to
the subject of quantum gravity. In normal theoretical physics there is the
(unholy) trinity of (i) real-world data; (ii) mathematical framework; and (iii)
conceptual/philosophical framework. These three factors are closely interre-
lated in any real theory of the natural world: indeed, this tripartite structure
underpins all theoretical physics.
However, in quantum gravity the first factor is largely missing, and this
raises some curious questions. In particular:

2
 • Would we recognise the/a ‘correct’ theory of quantum gravity even if
it was handed to us on a plate? Certainly, the Popperian notion of
refutation is hard to apply with such a sparsity of empirical data.

• What makes any particular idea a ‘good’ one as far as the community is
concerned? Relatedly, how is it that one particular research programme
becomes well established whereas another falls by the wayside or, at
best, gains only a relatively small following?

The second question is not just whimsical, especially for our younger col-
leagues, for it plays a key role in decisions about the award of research grants,
post-doc positions, promotion, and the like.
In practice, many of the past and present research programmes in quan-
tum gravity have been developed by analogy with other theories and the-
oretical structures, particularly standard quantum field theory with gauge
groups. And as for why it is that certain ideas survive and other do not, the
answer is partly that individual scientists indulge in their own philosophical
prejudices, and partly that they like using theoretical tools with which they
are already familiar. This, after all, is the fastest route to writing new papers.
At this point it seems appropriate to mention the ubiquitous, oft-maligned
‘Planck length’. This fundamental unit of length comes from combining
Planck’s constant ~ (the ‘quantum’ in quantum gravity) with Newton’s con-
stant G (the ‘gravity’ in quantum gravity) and the speed of light c (which
always lurks around) in the form
 1/2
G~
LP := (1.1)
c3

which has a value of around 10−35 meters; the corresponding Planck time
(defined as TP := LP /c) has a value or around 10−42 seconds.
The general assumption is that something ‘dramatic’ happens to the na-
ture of space and time at these fundamental scales. Precisely what that
dramatic change might be has been the source of endless speculation and
conjecture. However, there is a fairly wide-spread anticipation that in so far
as spatio-temporal concepts have any meaning at all in the ‘deep’ quantum-
gravity regime, the appropriate mathematical model will not be based on
standard, continuum differential geometry. Indeed, it is not hard to con-
vince oneself that, from a physical perspective, the important ingredient in a
space-time model is not the ‘points’ in that space, but rather the ‘regions’ in
which physical entities can reside. In the context of a topological space, such

3
regions are best modelled by open sets: the closed sets may be too ‘thin’2
to contain a physical entity, and the only physically-meaningful closed sets
are those with a non-empty interior. These reflections lead naturally to the
subject of ‘pointless topology’ and the theory of locales—a natural step along
the road to topos theory.
However, another frequent conjecture is that what we normally call space
and time (or space-time) will only ‘emerge’ from the correct quantum gravity
formalism in some (classical?) limit. Thus a fundamental theory of quantum
gravity may (i) have no intrinsic reference at all to spatio-temporal concepts;
and (ii) be such that some of the spatio-temporal concepts that emerge in
various limits are non-standard and are modelled mathematically with some-
thing other than topology and differential geometry. All this leads naturally
to the main question that lies behind the work reported in this chapter.
Namely:

What is status of (or justification for) using standard quantum

theory when trying to construct a theory of quantum gravity?

It is notable that the main current programmes in quantum gravity do all

use essentially standard quantum theory. However, around twelve years ago
I came to the conclusion that the use of standard quantum theory was fun-
damentally inconsistent, and I stopped working in quantum gravity proper.
Instead, I began studying what, to me, were the central problems in quantum
theory itself: a search that lead quickly to the use of topos theory.
Of the various fundamental issues that arise, I will focus here on just
two of them. The first is the problem mentioned already: viz, applying
the standard instrumentalist interpretation of quantum theory to space and
time. The second is what I claim is a category error in the use, a priori, of
the real and complex numbers in the mathematical formulation of quantum
theory when applied in a quantum gravity context. However, to unpack this
problem it is first necessary to be more precise about the way that the real
(and complex) numbers arise in quantum theory. This is the subject of the
next Section.
2
In a differentiable manifold, closed sets include points and lines whereas physical
entities ‘take up room’.

4
2 The Problem of Using the Real Numbers
in Quantum Gravity
The real numbers arise in theories of physics in three different (but related)
ways: (i) as the values of physical quantities; (ii) as the values of probabilities;
and (iii) as a fundamental ingredient in models of space and time (especially
in those based on differential geometry). All three are of direct concern vis-
a-vis our worries about making unjustified, a priori assumptions in quantum
theory. Let us consider them in turn.
One reason for assuming physical quantities to be real-valued is undoubt-
edly grounded in the remark that, traditionally (i.e., in the pre-digital age),
they are measured with rulers and pointers, or they are defined operationally
in terms of such measurements. However, rulers and pointers are taken to be
classical objects that exist in the physical space of classical physics, and this
space is modelled using the reals. In this sense there is a direct link between
the space in which physical quantities take their values (what we call the
‘quantity-value space’) and the nature of physical space or space-time [7].

Thus assuming physical quantities to be real-valued is problematic

in any theory in which space, or space-time, is not modelled by
a smooth manifold. This is a theoretical-physics analogue of the
philosophers’ category error.

Of course, real numbers also arise as the value space for probabilities via
the relative-frequency interpretation of probability. Thus, in principle, an
experiment is to be repeated a large number, N, times, and the probability
associated with a particular result is defined to be the ratio Ni /N, where Ni
is the number of experiments in which that particular result was obtained.
The rational numbers Ni /N necessarily lie between 0 and 1, and if the limit
N → ∞ is taken—as is appropriate for a hypothetical ‘infinite ensemble’—
real numbers in the closed interval [0, 1] are obtained.
The relative-frequency interpretation of probability is natural in instru-
mentalist theories of physics, but it is inapplicable in the absence of any
classical spatio-temporal background in which the necessary sequence of mea-
surements can be made (as, for example, is the situation in quantum cosmol-
ogy).
In the absence of a relativity-frequency interpretation, the concept of
‘probability’ must be understood in a different way. One possibility involves
the concept of ‘potentiality’, or ‘latency’. In this case there is no compelling

5
reason why the probability-value space should be a subset of the real num-
bers. The minimal requirement on this value-space is that it is an ordered
set, so that one proposition can be said to be more or less probable than
another. However, there is no prima facie reason why this set should be
totally ordered.

In fact, one of our goals is to dispense with probabilities alto-

gether and to replace them with ‘generalised’ truth values for
propositions.

It follows from the above that, for us, a key problem is the way in which
the formalism of standard quantum theory is firmly grounded in the con-
cepts of Newtonian space and time (or the relativistic extensions of these
ideas) which are essentially assumed a priori. The big question is how this
formalism can be modified/generalised/replaced so as to give a framework
for physical theories that is

1. ‘realist’ in some meaning of that word; and

2. not dependent a priori on the real and/or complex numbers.

For example, suppose we are told that there is a background space-time

C, but it is modelled on something other than differential geometry; say, a
causal set. Then what is the quantum formalism that has the same relation
to C as standard quantum theory does to Newtonian space and time?
This question is very non-trivial as the familiar Hilbert-space formalism
is very rigid and does not lend itself to minor ‘fiddling’. What is needed is
something that is radically new and yet which can still embody the basic
principles3 of the quantum formalism, or beyond. To proceed let us return to
first principles and consider what can be said in general about the general
structure of mathematical theories of a physical system.
3
Of course, the question of what precisely are these ‘basic principles’ is much debatable.
I have a fond memory of being in the audience for a seminar by John Wheeler at a
conference on quantum gravity in the early 1970s. John was getting well into the swing of
his usual enthusiastic lecturing style and made some forceful remark about the importance
of the quantum principle. At that point a hand was raised at the back of the lecture room,
and a frail voice asked “What is the quantum principle?”. John Wheeler paused, looked
thoughtfully at his interlocutor, who was Paul Dirac, and answered “Well, to be honest, I
don’t know”. He paused again, and then said “Do you?”. “No” replied Dirac.

6
3 Theories of a Physical System
3.1 The Realism of Classical Physics
“From the range of the basic questions of metaphysics we shall
here ask this one question: What is a thing? The question is
quite old. What remains ever new about it is merely that it must
be asked again and again [8].”

Martin Heidegger

Asking such questions is not a good way for a modern young philosopher
to gain tenure. Plato was not ashamed to do so, and neither was Kant, but
at some point in the last century the question became ‘Wittgensteined’ and
since then it is asked at one’s peril. Fortunately, theoretical physicists are
not confined in this way, and I will address the issue front on. Indeed, why
should a physicist be ashamed of this margaritiferous question? For is it not
what we all strive to answer when we probe the physical world?
Heidegger’s own response makes interesting reading:

“A thing is always something that has such and such properties,

always something that is constituted in such and such a way. This
something is the bearer of the properties; the something, as it
were, that underlies the qualities.”

Let us see how modern physics approaches this fundamental question about
the beingness of Being.
In constructing a theory of any ‘normal’4 branch of physics, the key in-
gredients are the mathematical representations of the following:

1. Space, time, or space-time: the framework within which ‘things’ are

made manifest to us.

2. ‘States’ of the system: the mathematical entities that determine ‘the

way things are’.

3. Physical quantities pertaining to the system.

4. ‘Properties’ of the system: i.e., propositions about the values of physical

quantities. A state assigns truth-values to these propositions.
4
That is, any branch of physics other than quantum gravity!

7
 In the case of standard quantum theory, the mathematical states are
interpreted in a non-realist sense as specifying only what would happen if
certain measurements are made: so the phrase ‘the way things are’ has to be
broadened to include this view. This ontological concept must also apply to
the ‘truth values’ that are intrinsically probabilistic in nature.
In addition, ‘the way things are’ is normally construed as referring to a
specific ‘moment’ of time, but this could be generalised to be compatible
with whatever model of time/space-time is being employed. This could even
be extended to a ‘history theory’, in which case ‘the way things are’ means
the way things are for all moments of time, or whatever is appropriate for
the space-time model being employed.
As theoretical physicists with a philosophical bent, a key issue is how
such a mathematical framework implies, or encompasses, various possible
philosophical positions. In particular, how does it interface with the position
of ‘realism’ ?
This issue is made crystal clear in the case of classical physics, which is
represented in the following way.

1. In Newtonian physics (which will suffice for illustrative purposes) space

is represented by the three-dimensional Euclidean space, R3 , and time
by the one-dimensional space, R.

2. To each system S there is associated a set (actually a symplectic dif-

ferentiable manifold) S of states. At each moment of time, t ∈ R, the
system has a unique state st ∈ S.

3. Any physical quantity, A, is represented by a function Ă : S → R. The

associated interpretation is that if s ∈ S is a state, then the value of
A in that state is just the real number Ă(s). This is the precise sense
in which the philosophical position of (naı̈ve) realism is encoded in the
framework of classical physics.

4. The basic propositions are of the form “A ε ∆” which asserts that the
value of the physical quantity A lies in the (Borel) subset ∆ of the
real numbers. This proposition is represented mathematically by the
subset Ă−1 (∆) ⊆ S—i.e., the collection of all states, s, in S such that
Ă(s) ∈ ∆.

This representation of propositions by subsets of the state space S has a

fundamental implication for the logical structure of classical physics:

8
 The mathematical structure of set theory implies that, of
necessity, the propositions in classical physics have the logical
structure of a Boolean algebra.

We note that classical physics is the paradigmatic implementation of Hei-

degger’s view of a ‘thing’ as the bearer of properties. However, in quantum
theory, the situation is very different. There, the existence of any such real-
ist interpretation is foiled by the famous Kochen-Specker theorem [14]. This
asserts that it is impossible to assign values to all physical quantities at once
if this assignment is to satisfy the consistency condition that the value of a
function of a physical quantity is that function of the value. For example,
the value of ‘energy squared’ is the square of the value of energy.

3.2 A Categorial Generalisation of the Representation

of Physical Quantities
The Kochen-Specker theorem implies that, from Heidegger’s world-view,
there is no ‘way things are’. To cope with this, physicists have histori-
cally fallen back on the instrumentalist interpretation of quantum theory
with which we are all so familiar. However, as I keep emphasising, in the
context of quantum gravity there are good reasons for wanting to achieve a
more realist view, and the central question is how this can be done.
The problem is the great disparity between the mathematical formalism of
classical physics—which is naturally realist—and the formalism of quantum
physics—which is not. Let us summarise these different structures as they
apply to a physical quantity A in a system S with associated propositions
“A ε ∆” where ∆ ⊆ R.

The classical theory of S:

• The state space is a set S.

• A is represented by a function Ă : S → R.

• The propositions “A ε ∆” is represented by the subset Ă−1 (∆) ⊆ S of

S. The collection of all such subsets forms a Boolean lattice.

The quantum theory of S:

• The state space is a Hilbert space H.

9
 • A is represented by a self-adjoint operator, Â, on H.

• The proposition “A ε ∆” is represented by the operator Ê[Â ∈ ∆] that

projects onto the subset ∆ ∩ sp(Â) of the spectrum, sp(Â), of Â. The
collection of all projection operators on H forms a non-distributive lat-
tice.

The ‘non-realism’ of quantum theory is reflected in the fact that propo-

sitions are represented by elements of a non-distributive lattice, whereas in
classical physics a distributive lattice (Boolean algebra) is used.
So how can we find a formalism that goes beyond classical physics and
yet which retains some degree of realist interpretation? One possibility is
to generalise the axioms of classical physics to a category, τ , other than the
category of sets. Such a representation of a physical system would have the
following ingredients:

The τ -category theory of S:

• There are two special objects, Σ, R, in τ known respec-

tively as the state object and quantity-value object.

• A physical quantity, A, is represented by an arrow Ă :

Σ → R in the category τ .

• Propositions about the physical world are represented

by sub-objects of the state object Σ. In standard physics
there must be some way of embedding into this structure
propositions of the form “A ε ∆” where ∆ ⊆ R.

But does such ‘categorification’ work? In particular, is there some category

such that quantum theory can be rewritten in this way?

4 The Use of Topos Theory

4.1 The Nature of a Topos
The simple answer to the question of categorification is ‘no’, not in general.
The sub-objects of an object in a general category do not have any logical
structure, and I regard possessing such a structure to be a sine qua non for
the propositions in a physical theory. However, there is a special type of

10
category, known as a topos, in which the sub-objects of any object do have a
logical structure—a remark that underpins our entire research programme.
Broadly speaking, a topos, τ , is a category that behaves in certain critical
respects just like the category, Sets, of sets. In particular, these include the
following:

1. There is an ‘initial’ object, 0τ , and a ‘terminal object’, 1τ . These are

the analogues of, respectively, the empty set, ∅, and the singleton set
{∗}.
A global element (or just element) of an object X is defined to be
an arrow x : 1τ → X. This definition reflects the fact that, in set
theory, any element x of a set X can be associated with a unique
arrow5 x : {∗} → X defined by6 x(∗) := x.
The set of all global elements of an object X is denoted7 ΓX; i.e.,
ΓX := Homτ (1τ , X).

2. One can form ‘products’ and ‘co-products’ of objects8 These are the
analogue of the cartesian product and disjoint union in set theory.

3. In set theory, the collection, B A , of functions f : A → B between sets

A and B is itself a set; i.e., it is an object in the category of sets. In
a topos, there is an analogous operation known as ‘exponentiation’.
This associates to each pair of objects A, B in τ an object B A with the
characteristic property that

Homτ (C, B A ) ≃ Homτ (C × A, B) (4.1)

for all objects C. In set theory, the relevant statement is that a pa-
rameterised family of functions c 7→ fc : A → B, c ∈ C, is equivalent
to a single function F : C × A → B defined by F (c, a) := fc (a) for all
c ∈ C, a ∈ A.

4. Each subset A of a set X is associated with a unique ‘characteristic

function’ χA : X → {0, 1} defined by χA (x) := 1 if x ∈ A and χA (x) = 0
if x 6∈ A. Here, 0 and 1 can be viewed as standing for ‘false’ and
5
The singleton, {∗} is not unique of course. But any two singletons are isomorphic as
sets and it does not matter which one we choose.
6
Note that, hopefully without confusion, we are using the same letter ‘x’ for the element
in X and the associated function from {∗} to X.
7
In general, Homτ (X, Y ) denotes the collection of all arrows in τ from the object X to
the object Y .
8
More generally, there are pull-backs and push-outs.

11
 ‘true’ respectively, so that χA (x) = 1 corresponds to the mathematical
proposition “x ∈ A” being true.
This operation is mirrored in any topos. More precisely, there is a, so-
called, ‘sub-object classifier’, Ωτ , that is the analogue of the set {0, 1}.
Specifically, to each sub-object A of an object X there is associated a
‘characteristic arrow’ χA : X → Ωτ ; conversely, each arrow χ : X → Ωτ
determines a unique sub-object of X. Thus
Sub(X) ≃ Homτ (X, Ωτ ) (4.2)
where Sub(X) denotes the collection of all sub-objects of X.
5. The power set, P X of any set X is defined to be the set of all subsets
of X. Each subset A ⊆ X determines, and is uniquely determined by,
its characteristic function χA : X 7→ {0, 1}. Thus the set P X is in
bijective correspondence with the function space {0, 1}X . Analogously,
in a general topos, τ , we define the power object of an object X, to be
P X := ΩX τ . It follows from (4.1) that

Γ(P X) := Homτ (1τ , P X) = Homτ (1τ , ΩX

τ )
≃ Homτ (1τ × X, Ωτ ) ≃ Homτ (X, Ωτ )
≃ Sub(X) (4.3)
Since Sub(X) is always non-empty (because any object X is always
a sub-object of itself) it follows that, for any X, the object P X is
non-trivial. As a matter of notation, if A is a sub-object of X the
corresponding arrow from 1τ to P X is called the ‘name’ of A and is
denoted pAq : 1τ → P X.
A key result for our purposes is that in any topos, the collection, Sub(X),
of sub-objects of any object X forms a Heyting algebra, as does the set
ΓΩτ := Homτ (1τ , Ωτ ) of global elements of the sub-object classifier Ωτ . A
Heyting algebra is a distributive lattice, h, with top and bottom elements 1
and 0, and such that if α, β ∈ h there exists an element α ⇒ β in h with the
property
γ  (α ⇒ β) if and only if γ ∧ α  β (4.4)
Then the ‘negation’ of any α ∈ h is defined as
¬α := (α ⇒ 0) (4.5)
A Heyting algebra has all the properties of a Boolean algebra except that
the principle of excluded middle may not hold. That is, there may exist α ∈ h
such that α ∨ ¬α ≺ 1, i.e., α ∨ ¬α 6= 1. Equivalently, there may be β ∈ h
such that β ≺ ¬¬β. Of course, in a Boolean algebra we have β = ¬¬β for
all β. We shall return to this feature shortly.

12
4.2 The Mathematics of ‘Neo-realism’
Let us now consider the assignment of truth values to propositions in math-
ematics. In set theory, let K ⊆ X be a subset of some set X, and let x be an
element of X. Then the basic mathematical proposition “x ∈ K” is true if,
and only if, x is an element of the subset K. At the risk of seeming pedantic,
the truth value, [[ x ∈ K ]], of this proposition can be written as

1 if x belongs to K;
[[ x ∈ K ]] := (4.6)
0 otherwise.

In terms of the characteristic function χK : X → {0, 1}, we have

[[ x ∈ K ]] = χK ◦ x (4.7)

where χK ◦ x : {∗} → {0, 1}.9

The reason for writing the truth value (4.6) as (4.7) is that this is the
form that generalises in an arbitrary topos. More precisely, let K ∈ Sub(X)
with characteristic arrow χK : X → Ωτ , and let x be a global element of X
so that x : 1τ → X. Then the ‘generalised’ truth value of the mathematical
proposition “x ∈ K” is defined to be the arrow

[[ x ∈ K ]] := χK ◦ x : 1τ → Ωτ (4.8)

Thus [[ x ∈ K ]] belongs to the Heyting algebra ΓΩτ . The adjective ‘gener-

alised’ refers to the fact that, in a generic topos, the Heyting algebra contains
elements other than just 0 and 1. In this sense, a proposition in a topos can
be only partially true: a concept that seems ideal for application to the fitful
reality of a quantum system.
This brings us to our main contention, which is that it may be profitable
to consider constructing theories of physics in a topos other than the familiar
topos of sets [2, 3, 4, 5, 6, 9, 10, 11, 12]. The propositions in such a theory
admit a ‘neo’-realist interpretation in the sense that there is a ‘way things
are’, but this is specified by generalised truth values that may not be just
true (1) or false (0).
Such a theory has the following ingredients:

1. A physical quantity A is represented by a τ -arrow Ă : Σ → R from the

state object Σ to the quantity-value object R.
9
For the notation to be completely consistent, the left hand side of (4.6) should really
be written as [[ x ∈ K ]](∗).

13
 2. Propositions about S are represented by elements of the Heyting alge-
bra, Sub(Σ), of sub-objects of Σ. If Q is such a proposition, we denote
by δ(Q) the associated sub-object of Σ; the ‘name’ of δ(Q) is the arrow
pδ(Q)q : 1τ → P Σ.
3. The topos analogue of a state is a ‘pseudo-state’ which is a particular
type of sub-object of Σ. Given this pseudo-state, each proposition can
be assigned a truth value in the Heyting algebra ΓΩτ . Equivalently, we
can use ‘truth objects’: see below.
Conceptually, such a theory is neo-realist in the sense that the propositions
and their truth values belong to structures that are ‘almost’ Boolean: in fact
they differ from Boolean algebras only in so far as the principle of excluded
middle may not apply.
Thus a theory expressed in this way ‘looks like’ classical physics ex-
cept that classical physics always employs the topos Sets, whereas other
theories—including, we claim, quantum theory—use a different topos. If the
theory requires a background space-time (or functional equivalent thereof)
this could be represented by another special object, M, in the topos. It
would even be possible to mimic the actions of differential calculus if the
topos is such as to support synthetic differential geometry10 .
The presence of intrinsic logical structures in a topos has another striking
implication. Namely, a topos can be used as a foundation for mathematics
itself, just as set theory is used in the foundations of ‘normal’ (or ‘classical’
mathematics). Thus classical physics is modelled in the topos of sets, and
thereby by standard mathematics. But a theory of physics modelled in a
topos, τ , other than Sets is being represented in an alternative mathematical
universe! The absence of excluded middle means that proofs by contradiction
cannot be used, but apart from that this, so-called, ‘intuitionistic’ logic can
be handled in the same way as classical logic.
A closely-related feature is that each topos has an ‘internal language’
that is functionally similar to the formal language on which set theory is
based. It is this internal language that is used in formulating axioms for the
mathematical universe associated with the topos. The same language is also
used in constructing the neo-realist interpretation of the physical theory.

4.3 The Idea of a Pseudo-state

In discussing the construction of truth values it is important to distinguish
clearly between truth values of mathematical propositions and truth values
10
This approach to calculus is based on the existence of genuine infinitesimals in certain
topoi.

14
of physical propositions. A key step in constructing a physical theory is to
translate the latter into the former. For example, in classical physics, the
physical proposition “A ε ∆” is represented by the subset Ă−1 (∆) of the state
space, S; i.e.,
δ(A ε ∆) := Ă−1 (∆). (4.9)
Then, for any state s ∈ S, the truth value of the physical proposition
“A ε ∆” is defined to be the truth value of the mathematical proposition
“s ∈ δ(A ε ∆)” (or, equivalently, the truth value of “Ă(s) ∈ ∆”).
Given the ideas discussed above, in a topos theory one might expect to
represent a physical state by a global element s : 1τ → Σ of the state object
Σ. The truth value of a proposition, Q, represented by a sub-object δ(Q) of
Σ, would then be defined as the global element

ν(Q; s) := [[ s ∈ δ(Q) ]] = χδ(Q) ◦ s : 1τ → Ωτ (4.10)

of Ωτ . However, in the topos version of quantum theory (see Section 5)

it transpires that Σ has no global elements at all : in fact, this turns out
to be precisely equivalent to the Kochen-Specker theorem! This absence of
global elements of the state object could well be a generic feature of topos-
formulated physics, in which case we cannot use (4.10) and need to proceed
in a different way.
In set theory, there are two mathematical statements that are equivalent
to “s ∈ K”’. These are:

1. K ∈ Ts (4.11)

where the ‘truth object’ T s is defined by11

T s := {J ⊆ S | s ∈ J}; (4.13)

and

2. {s} ⊆ K. (4.14)

Thus, in classical physics, “A ε ∆” is true in a state s if, and only, if (i)

δ(A ε ∆) ∈ T s ; and (ii) {s} ⊆ δ(A ε ∆).
Let us consider in turn the topos analogue of these two options.
11
Note that s can be recovered from T s via
\
{s} := K (4.12)
K∈T s

15
Option 1 (the truth-object option): Note that T s is a collection of
subsets of S: i.e., T s ∈ Sub(P S). The interesting thing about (4.11) is that
it is of the form “x ∈ X” and therefore has an immediate generalisation to
any topos. More precisely, in a general topos τ a truth object, T , would
be a sub-object of P Σ (equivalently, a global element of P (P Σ)) with a
characteristic arrow χT : P Σ → Ωτ . Then the physical proposition Q has
the topos truth value

ν(Q; T ) := [[ δ(Q) ∈ T ]] = χT ◦ pδ(Q)q : 1τ → Ωτ (4.15)

The key remark is that although (4.10) is inapplicable if there are no global
elements of Σ, (4.15) can be used since global elements of a power object
(like P (P Σ)) always exist. Thus, in option 1, the analogue of a classical
state s ∈ S is played by the truth object T ∈ Sub(P Σ).

Option 2 (the pseudo-state option): We cannot use (4.14) in a literal

way in the topos theory since, if there are no global elements, s, of Σ, trying
to construct an analogue of {s} is meaningless. However, although there are
no global elements of Σ, there may nevertheless be certain sub-objects, w, of
Σ (what we call ‘pseudo-states’) that are, in some sense, as ‘close as we can
get’ to the (non-existent) analogue of the singleton subsets, {s}, of S. We
must then consider the mathematical proposition “w ⊆ δ(Q)” where w and
δ(Q) are both sub-objects of Σ.
As we have seen in (4.8), the truth value of the mathematical proposition
“x ∈ K” is a global element of Ωτ : as such, it may have a value other than
1 (‘true’) or 0 (‘false’). In other words, “x ∈ K” can be only ‘partially true’.
What is important for us is that an analogous situation arises if J, K are
sub-objects of some object X. Namely, a global element, [[ J ⊆ K ]], of Ωτ
can be assigned to the proposition “J ⊆ K”: i.e., there is a precise sense in
which one sub-object of X can be only ‘partially’ a sub-object of another. In
this scenario, a physical proposition Q has the topos truth value

ν(Q; w) := [[ w ⊆ δ(Q) ]] (4.16)

when the pseudo-state is w. The general definition of [[ J ⊆ K ]] is not

important for our present purposes. In the quantum case, the explicit form
is discussed in the companion article by my collaborator Andreas Döring [1].
To summarise what has been said so far, the key ingredients in formulat-
ing a theory of a physical system in a topos τ are the following:

1. There is a ‘state object’, Σ, in τ .

16
 2. To each physical proposition, Q, there is associated a sub-object, δ(Q),
of Σ.

3. The analogue of a classical state is given by either (i) a truth object,

T ; or (ii) a pseudo-state w. The topos truth-value of the proposition
Q is then [[ δ(Q) ∈ T ]], or [[ w ⊆ δ(Q) ]], respectively.

Note that no mention has been made here of the quantity-value object
R. However, in practice, this also is expected to play a key role, not least
in constructing the physical propositions. More precisely, we anticipate that
each physical quantity A will be represented by an arrow Ă : Σ → R, and
then a typical proposition will be of the form “A ε Ξ” where Ξ is some sub-
object of R.
If any ‘normal’ physics is addressed in this way, physical quantities are
expected to be real -valued, and the physical propositions are of the form
“A ε ∆” for some ∆ ⊆ R. In this case, it is necessary to decide what sub-
object of R in the topos corresponds to the external quantity ∆ ⊆ R.
There is subtlety here, however, since although (in any topos we are likely
to consider) there is a precise topos analogue of the real numbers12 , it would
be a mistake to assume that this is necessarily the quantity-value object:
indeed, in our topos version of quantum theory this is definitely not the
case.
The question of relating ∆ ⊆ R to some sub-object of the quantity-value
object R in τ is just one aspect of the more general issue of distinguishing
quantities that are external to the topos, and those that are internal. Thus,
for example, a sub-object Ξ of R in τ is an internal concept, whereas ∆ ⊆ R
is external, referring as it does to something, R, that lies outside τ . Any
reference to a background space, time or space-time would also be exter-
nal. Ultimately perhaps—or, at least, certainly in the context of quantum
cosmology—one would want to have no external quantities at all. However,
in any more ‘normal’ branch of physics it is natural that the propositions
about the system refer to the external (to the theory) world to which the-
oretical physics is meant to apply. And, even in quantum cosmology, the
actual collection of what counts as physical quantities is external to the for-
malism itself.
12
Albeit defined using the analogue of Dedekind cuts, not Cauchy sequences.

17
5 The Topos of Quantum Theory
5.1 The Kochen-Specker theorem and contextuality
To motivate our choice of topos for quantum theory let us return again to the
Kochen-Specker theorem which asserts the impossibility of assigning values
to all physical quantities whilst, at the same time, preserving the functional
relations between them [14].
In a quantum theory, a physical quantity A is represented by a self-adjoint
operator  on the Hilbert space, H, of the system. A ‘valuation’ is defined to
be a real-valued function λ on the set of all bounded, self-adjoint operators,
with the properties that: (i) the value λ(Â) belongs to the spectrum of Â;
and (ii) the functional composition principle (or FUNC for short) holds:

λ(B̂) = h(λ(Â)) (5.1)

for any pair of self-adjoint operators Â, B̂ such that B̂ = h(Â) for some
real-valued function h.
Several important results follow from this definition. For example, if Â1
and Â2 commute, it follows from the spectral theorem that there exists an
operator Ĉ and functions h1 and h2 such that Â1 = h1 (Ĉ) and Â2 = h2 (Ĉ).
It then follows from FUNC that

λ(Â1 + Â2 ) = λ(Â1 ) + λ(Â2 ) (5.2)

and
λ(Â1 Â2 ) = λ(Â1 )λ(Â2 ). (5.3)
The Kochen-Specker theorem says that no valuations exist if dim(H) > 2.
On the other hand, (5.2–5.3) show that, if it existed, a valuation restricted
to a commutative sub-algebra of operators would be just an element of the
spectrum of the algebra, and of course such elements do exist. Thus valu-
ations exist on any commutative sub-algebra13 of operators, but not on the
(non-commutative) algebra, B(H), of all bounded operators. We shall call
such valuations ‘local’.
Within the instrumentalist interpretation of quantum theory, the exis-
tence of local valuations is closely related to the possibility of making ‘simul-
taneous’ measurements on commutating observables. However, the existence
of local valuations also plays a key role in the, so-called, ‘modal’ interpre-
tations in which values are given to the physical quantities that belong to
13
More precisely, since we want to include projection operators, we assume that the
commutative algebras are von Neumann algebras. These algebras are defined over the
complex numbers, so that non self-adjoint operators are included too.

18
some specific commuting set. The most famous such interpretation is that
of David Bohm where it is the configuration14 variables in the system that
are regarded as always ‘existing’ (in the sense of possessing values).
The topos-implication of these remarks stems from the following obser-
vations. First, let V, W be a pair of commutative sub-algebras with V ⊆ W .
Then any (local) valuation, λ, on W restricts to give a valuation on the sub-
algebra V . More formally, if ΣW denotes the set of all local valuations on
W , there is a ‘restriction map’ rW V : ΣW → ΣV in which rW V (λ) := λ|V for
all λ ∈ ΣW . It is clear that if U ⊆ V ⊆ W then
rV U (rW V (λ)) = rW U (λ) (5.4)
for all λ ∈ ΣW —i.e., restricting from W to U is the same as going from W
to V and then from V to U.
Note that, if one existed, a valuation, λ, on the non-commutative algebra
of all operators on H would provide an association of a ‘local’ valuation
λV := λ|V to each commutative algebra V such that, for all pairs V, W with
V ⊆ W we have
λW | V = λV (5.5)
The Kochen-Specker theorem asserts there are no such associations V 7→
λV ∈ ΣV if dim H > 2.
To explore this further consider the situation where we have three com-
muting algebras V, W1 , W2 with V ⊆ W1 and V ⊆ W2 and suppose λ1 ∈ ΣW1
and λ2 ∈ ΣW2 are local valuations. If there is some commuting algebra W
so that (i) W1 ⊆ W and W2 ⊆ W ; and (ii) there exists λ ∈ ΣW such that
λ1 = λ|W1 and λ2 = λ|W2 , then (5.4) implies that
λ1 | V = λ2 | V (5.6)
However, suppose now the elements of W1 and W2 do not all commute
with each other: i.e., there is no W such that W1 ⊆ W and W2 ⊆ W . Then
although valuations λ1 ∈ ΣW1 and λ2 ∈ ΣW2 certainly exist, there is no longer
any guarantee that they can be chosen to satisfy the matching condition in
(5.6). Indeed, the Kochen-Specker theorem says precisely that it is impos-
sible to construct a collection of local valuations, λW , for all commutative
sub-algebras W such that all the matching conditions of the form (5.6) are
satisfied.
We note that triples V, W1 , W2 of this type arise when there are non-
commuting self-adjoint operators Â1 , Â2 with a third operator B̂ and func-
tions f1 , f2 : R → R such that B̂ = f1 (Â1 ) = f2 (Â2 ). Of course, [B̂, Â1 ] = 0
14
If taken literally, the word ‘configuration’ implies that the state space of the underlying
classical system must be a cotangent bundle T ∗ Q.

19
and [B̂, Â2 ] = 0 so that, in physical parlance, A1 and B can be given ‘si-
multaneous values’ (or can be measured simultaneously) as can A2 and B.
However, the implication of the discussion above is that the value ascribed to
B (resp. the result of measuring B) depends on whether it is considered to-
gether with A1 , or together with A2 . In other words the value of the physical
quantity B is contextual. This is often considered one of the most important
implications of the Kochen-Specker theorem.
op
5.2 The topos of presheaves SetsV(H)
To see how all this relates to topos theory let us rewrite the above slightly.
Thus, let V(H) denote the collection of all commutative sub-algebras of op-
erators on the Hilbert space H. This is a partially-ordered set with respect to
sub-algebra inclusion. Hence it is also a category whose objects are just the
commutative sub-algebras of B(H); we shall call it the ‘category of contexts’.
We view each commutative algebra as a context with which to view the
quantum system in an essentially classical way in the sense that the physical
quantities in any such algebra can be given consistent values, as in classical
physics. Thus each context is a ‘classical snapshot’, or ‘world-view’, or ‘win-
dow on reality’. In any modal interpretation of quantum theory, only one
context at a time is used15 but our intention is to use the collection of all
contexts in one mega-structure that will capture the entire quantum theory.
To do this, let us consider the association of the spectrum ΣV (the set
of local valuations on V ) to each commutative sub-algebra V . As explained
above, there are restriction maps rW V : ΣW → ΣV for all pairs V, W with
V ⊆ W , and these maps satisfy the conditions in (5.4). In the language of
category theory this means that the operation V 7→ ΣV defines the elements
of a contravariant functor, Σ, from the category V(H) to the category of sets;
equivalently, it is a covariant functor from the opposite category, V(H)op to
Sets.
Now, one of the basic results in topos theory is that for any category C,
the collection of covariant functors F : C op → Sets is a topos, known as the
op
‘topos of presheaves’ over C, and is denoted SetsC . In regard to quantum
theory, our fundamental claim is that the theory can be reformulated so as
op
to look like classical physics, but in the topos SetsV(H) . The object Σ is
known as the ‘spectral presheaf’ and plays a fundamental role in our theory.
In purely mathematical terms this has considerable interest as the foundation
for a type of non-commutative spectral theory; from a physical perspective,
15
In the standard instrumentalist interpretation of quantum theory, a context is selected
by choosing to measure a particular set of commuting observables.

20
we identify it as the state object in the topos. op
The terminal object 1SetsV(H)op in the topos SetsV(H) is the presheaf
that associates to each commutative algebra V a singleton set {∗}V , and the
restriction maps are the obvious ones. It is then easy to see that a global
element λ : 1SetsV(H)op → Σ of the spectral presheaf is an association to each
V of a spectral element λV ∈ ΣV such that, for all pairs V, W with V ⊆ W
we have (5.5). Thus we have the following basic result:

The Kochen-Specker theorem is equivalent to the statement

that the spectral presheaf, Σ, has no global elements.

Of course, identifying the topos and the state object are only the very
first steps in constructing a topos formulation of quantum theory. The next
key step is to associate a sub-object of Σ with each physical proposition. In
quantum theory, propositions are represented by projection operators, and
so what we seek is a map

δ : P(H) → Sub(Σ) (5.7)

where P(H) denotes the lattice of projection operators on H. Thus δ is

a map from a non-distributive quantum op
logic to the (distributive) Heyting
algebra, Sub(Σ), in the topos SetsV(H) .
The precise definition of δ(P̂ ), P̂ ∈ P(H), is described in the companion
article by my collaborator Andreas Döring [1]. Suffice it to say that, at each
context V , it involves approximating P̂ with the ‘closest’ projector to P̂ that
lies in V : an operator that we call ‘daseinisation’ in honour of Heidegger’s
memorable existentialist perspective on ontology.
The next step is to construct the quantity-value object, R. We do this by
applying the Gel’fand spectral transformation in each context V . The most
striking remark about the result is that R is not the real-number object, R,
in the topos, although the latter is a sub-object of the former. One result of
this construction is that we are able to associate an arrow Ă : Σ → R with
each physical quantity A; i.e., with each bounded, self-adjoint operator Â.
This is a type of non-commutative spectral theory.
The final, key, ingredient is to construct the truth objects, or pseudo-
states; in particular, we wish to do this for each vector |ψi in the Hilbert
space H. Again, the details can be found in the chapter by Andreas Döring,
but suffice it to say that the pseudo-state, w |ψi associated with |ψi ∈ H can
written in the simple form

w |ψi = δ( |ψihψ| ) (5.8)

21
In other words, the pseudo-state corresponding to the unit vector |ψi is just
the daseinisation of the projection operator onto |ψi.

6 Conclusions
We have revisited the oft-repeated statement about fundamental incompat-
ibilities between quantum theory and general relativity. In particular, we
have argued that the conventional quantum formalism is inadequate to the
task of quantum gravity both in regard to (i) the non-realist, instrumentalist
interpretation; and (ii) the a priori use of the real and complex numbers.
Our suggestion is to employ a mathematical formalism that ‘looks like’
classical physics (so as to gain some degree of ‘realism’) but in a topos, τ ,
other than the topos of sets. The ingredients in such a theory are:

1. A state-object, Σ, and a quantity-value object, R.

2. A map δ : P → Sub(Σ) from the set of propositions, P, to the Heyting

algebra, Sub(Σ), of sub-objects of Σ.

3. A set of pseudo-states, w, or truth objects, T . Then the truth value

of the physical proposition Q in the pseudo-state w (resp. the truth
object T ) is [[ w ⊆ δ(Q) ]] (resp. [[ δ(Q) ∈ T ]]) which is a global element
of the sub-object classifier, Ωτ , in τ . The set, ΓΩτ of all such global
elements is also a Heyting algebra.

In the case of quantum theory, in our published papers we have shown

in great detail howop this programme goes through with the topos being the
category, SetsV(H) , of presheaves over the category/poset V(H) of commu-
tative sub-algebras of the algebra of all bounded operators on the Hilbert
space H.
The next step in this programme is to experiment with ‘generalisations’
of quantum theory in which the context category is not V(H) but some cat-
egory, C, which has no fundamental link with a Hilbert space, and therefore
with the real and/or complex numbers. In a scheme of this type the intrin-
sic contextuality of quantum theory is kept, but its domain of applicability
could include spatio-temporal concepts that are radically different from those
currently in use.

Acknowledgements. I am most grateful to Andreas Döring for the

stimulating collaboration that we have enjoyed over the last four years.

22
References
[1] A. Döring. The physical interpretation of daseinisation. To appear in
Deep Beauty, ed. Hans Halvorson (2010).
[2] A. Döring, and C.J. Isham. A topos foundation for theories of physics:
I. Formal languages for physics, J. Math. Phys 49, 053515 (2008).
[3] A. Döring, and C.J. Isham. A topos foundation for theories of physics:
II. Daseinisation and the liberation of quantum theory. J. Math. Phys
49, 053516 (2008).
[4] A. Döring, and C.J. Isham. A topos foundation for theories of physics:
III. Quantum theory and the representation of physical quantities with
arrows Ă : Σ → R . J. Math. Phys 49, 053517 (2008).
[5] A. Döring, and C.J. Isham. A topos foundation for theories of physics:
IV. Categories of systems. J. Math. Phys 49, 053518 (2008).
[6] A. Döring, and C.J. Isham. ‘What is a Thing?’: Topos Theory in the
Foundations of Physics. To appear in New Structures in Physics, ed.
B. Coecke, Springer (2010).
[7] C.J. Isham. Some reflections on the status of conventional quantum
theory when applied to quantum gravity. In Proceedings of the Confer-
ence in Honour of Stephen Hawking’s 60’th birthday, ed. G. Gibbons,
Cambridge University Press, Cambridge (2003).
[8] M. Heidegger. What is a Thing? Regenery/Gateway, Indiana (1967).
[9] C.J. Isham and J. Butterfield. A topos perspective on the Kochen-
Specker theorem: I. Quantum states as generalised valuations. Int. J.
Theor. Phys. 37, 2669–2733 (1998).
[10] C.J. Isham and J. Butterfield. A topos perspective on the Kochen-
Specker theorem: II. Conceptual aspects, and classical analogues. Int.
J. Theor. Phys. 38, 827–859 (1999).
[11] C.J. Isham, J. Hamilton and J. Butterfield. A topos perspective on
the Kochen-Specker theorem: III. Von Neumann algebras as the base
category. Int. J. Theor. Phys. 39, 1413-1436 (2000).
[12] C.J. Isham and J. Butterfield. A topos perspective on the Kochen-
Specker theorem: IV. Interval valuations. Int. J. Theor. Phys 41, 613–
639 (2002).

23
[13] C.J. Isham and J. Butterfield. Space-time and the philosophical chal-
lenge of quantum gravity. In Physics Meets Philosophy at the Planck
Scale, eds. C. Callender and N. Huggett, Cambridge University Press,
33–89 (2001).

[14] S. Kochen and E.P. Specker. The problem of hidden variables in quan-
tum mechanics. Journal of Mathematics and Mechanics 17, 59–87
(1967).

[15] L. Smolin. The case for background independence. In The Structural

Foundations of Quantum Gravity, eds. S. French, D. Rickles and J. Sa-
hatsi. Oxford University Press, Oxford (2006).

[16] L. Smolin. Generic predictions of quantum theories of gravity. In Ap-

proaches to Quantum Gravity—Toward a New Understanding of Space,
Time, and Matter, ed. D. Oriti. Cambridge University Press, Cambridge
(2006).

24