Coecke (Cat Physics)
Coecke (Cat Physics)
Why?
Why would a physicist care about category theory, why would he want to know
about it, why would he want to show off with it? There could be many reasons. For example, you might find John Baezs webside one of the coolest in the
world. Or you might be fascinated by Chris Ishams and Lee Smolins ideas on
the use of topos theory in Quantum Gravity. Also the connections between knot
theory, braided categories, and sophisticated mathematical physics such as quantum groups and topological quantum field theory might lure you. Or, if you are
also into pure mathematics, you might just appreciate category theory due to
its incredible unifying power of mathematical structures and constructions. But
there is a far more on-the-nose reason which is never mentioned. Namely,
a category is the exact mathematical structure of practicing physics!
What do I mean here by a practicing physics? Consider a physical system of type
A (e.g. a qubit, or two qubits, or an electron, or classical measurement data) and
perform an operation f on it (e.g. perform a measurement on it) which results
in a system possibly of a different type B (e.g. the system together with classical
data which encodes the measurement outcome, or, just classical data in the case
that the measurement destroyed the system). So typically we have
A
f B
where A is the initial type of the system, B is the resulting type, and f is the
operation. One can perform an operation
B
g C
after f since the resulting type B of f is also the initial type of g, and we write
g f for the consecutive application of these two operations. Clearly we have
(h g) f = h (g f ) since putting the brackets merely adds the superficial data
of conceiving two operations as one. If we further set
A
1A A
f gC D
inherited from the operations on the individual systems, then we pass from ordinary categories to a particular case of the 2-dimensional variant of categories
called monoidal categories. (a concept introduced by Jean Benabou in 1963 in
[8]) We will define these monoidal categories in Section 5.
What?
For the same operational reasons as discussed above (and which extend to the
far more compelling case of monoidal categories as we shall see below), category
theory could be expected to play an important role in other fields where operations/processes play a central role e.g. Programing (programs as morphisms)
and Logic & Proof Theory (proofs as morphisms), and indeed, in the theoretical
counterparts to these fields category theory has become quite common practice
cf. the many available textbooks and even undergraduate courses [1].
LOGIC & PROOF THEORY
Propositions
Proofs
PROGRAMMING
Data Types
Programs
PHYSICS
Physical System
Physical Operation
Where?
They truly are everywhere! But thats exactly where people start to get confused. (if you are not up for a storm of data just skip this section and go to
the next one) We consider some examples from mathematics. A group G is a
category with a single object in which every morphism is an isomorphism:
and
f f 1 = 1B .
Quantum?
We can also consider two distinct categories which both have sets as objects,
but one with functions as morphisms denoted by Set and one with relations
as morphisms denoted by Rel. While you might think that since both have
sets as objects they are quite similar, nothing is less true! As a matter of fact,
Rel much more resembles the category of finite dimensional Hilbert spaces and
linear maps FdHilb than it resembles Set, and here things really start to get
interesting. For example, category theory is able to detect the fact that both
vector spaces and relations admit a matrix calculus, respectively over the field K
and over the semiring of booleans B.1 While technically this involves some more
1 A semiring is a ring in without additive inverses. For a matrix calculus it indeed suffices
to be able to add and to multiply scalars, while no substraction is needed.
sophisticated concepts, we are already able to show that both Rel and FdHilb
admit a notion of superposition while Set doesnt. We expose this through the
categorical notion of element i.e. a notion of element which exposes itself at
the level of morphisms. First note that for any set X we have a bijection, i.e.,
categorically, an isomorphism
X {} X ,
where {} is just some singleton set, so we can expect {} to play a special role
both in Set and Rel. Similarly, in finite dimensional Hilbert spaces we have
H C H,
so we expect the one-dimensional Hilbert space C to play a special role in FdHilb.
And indeed, in Set we can define Xs elements as the functions
fx : {} X :: 7 x
since in this way each element x X arises as f () for the function fx : 7 x.
Analogously in FdHilb we define H s elements as linear maps
f|i : C H :: 1 7 |i
since by linearity f|i (1) = |i determines the linear map f|i completely. By
analogy in Rel Xs elements are relations
{}
- X,
but since relations are multi-valued this means that the elements do not correspond with the elements of X but with the subsets Y X, and one can think of
these subsets as superpositions of the singletons. Indeed, setting
Yi := X iff i Y
and
Yi := iff i 6 Y ,
both in FdHilb and Rel we can decompose elements over some notion of bases
respectively as
X
[
|i =
i | i i
Y =
Yi {i} .
iX
iX
Hence the sum becomes a union and the C-valued coefficients become Booleanvalued since {, X} B, the Booleans. In other words, we can think of the subsets
of a set, i.e. the elements in Rel, as being embedded in some vector space:
6
{2}
{1,2}
{1,2,3}
{2,3}
{3}
{1}
{1,3}
Very crucial in all this is the fact that we considered the cartesian product in
Rel and the tensor product in FdHilb, while both categories allow to combine
their objects in many different other ways (e.g. the direct sum of Hilbert spaces).
This shows that it is essential to consider these additional operations as a genuine
part of the structure, introducing monoidal structure.
Which?
(A
fB, C
(A, B)
7
AB
f g
gD)
7
AC
BD
which is bifunctorioral, and comes together with left & right unit natural
isos, a symmetry natural iso and an associativity natural iso.
So it remains to explain what bifunctoriality and those natural isos stand for.
To this means we depict morphisms (i.e. physical processes) as square boxes,
and we label the inputs and outputs of these boxes by types which tell on which
kind of system these boxes act cf. one qubit, n-qubits, classical data etc. Sequential composition (in time) is depicted by connecting matching outputs and
inputs of these boxes by lines, and parallel composition (cf. tensor) by locating boxes side by side. E.g. 1A : A A, f : A B, g f for g : B C,
1A 1B : A B A B, f 1C , f g for f : B D and g : C E, and
(f g) h for h : A B C respectively depict as:
C
B
A
B
A
g
C
h
A
We now show that the requirements bifunctoriality and existence and naturality
for some special isomorphisms with respect to the operation combining systems
are physically so evidently true that they almost seem redundant. (but as we will
see further they do have major implications)
Bifunctoriality. In the graphical language bifunctoriality stands for:
g
f
Bifunctoriality has a very clear conceptual interpretation: If we apply an operation f to one system and an operation g to another system, then the order in
which we apply them doesnt matter. Hence bifunctoriality expresses some notion of locality but still allows for the quantum type of non-locality. The above
pictorial equation can also be written down in term of a commutative diagram:
A1 A2
f 1A2
1B1 g
1 A1 g
?
A1 B2
B1 A2
?
- B1 B2
f 1B2
which expresses that both paths yield the same result. Actually, taking on a
relativistic spirit, (1 g) (f 1) = (f 1) (1 g) expresses that what is at the
left and at the right of the tensor does not temporally compare (cf. are space-like
separated) so we can denote them both without any harm by f g, and hence
assume the slightly more general condition
(g1 g2 ) (f1 f2 ) = (g1 f1 ) (g2 f2 )
from which it easily follows that
(1 g) (f 1) = (1 f ) (g 1) = (f 1) (1 g) = (f 1) (1 g) .
This stronger condition was already implicitly present in the picture calculus
since the latter explicitly ignores the brackets:
g1
g2
f1
f2
i.e. it doesnt matter if we either first consider the sequential composition or the
parallel composition. We read this as: since f1 is causally before g1 and f2 is
causally before g2 , the pair (f1 , f2 ) is causally before (g1 , g2 ) and vice versa, but
we do not assume any a priori space-like correlations along the tensor. Finally
in addition to the above we also require
1A 1B = 1AB
for the tensor, which is again self-evident from an operational perspective.
Symmetry and associativity natural isomorphisms. One can think of
natural isomorphisms as explicitly witnessed canonical isomorphisms. This is
best seen through an example. Consider the following picture:
which, in operational terms, expresses that if we swap the location of two systems
then we also have to swap the operations we intend to apply on them in order to
get the same result. Diagrammatically it corresponds to commutation of:
A1 A2
f g
B1 B2
B1 ,B2
A1 ,A2
?
A2 A1
?
- B2 B1
gf
8
(f1 , . . . , fn-)
(B1 , . . . , Bn )
B1 ,...,Bn
A1 ,...,An
?
(A1 , . . . , An )
(1)
?
- (B1 , . . . , Bn )
(f1 , . . . , fn )
(A1 A2 ) A3
(B1 B2 ) B3
B1 ,B2 ,B3
A1 ,A2 ,A3
?
- B1 (B2 B3 )
f (g h)
?
A1 (A2 A3 )
that is, in a picture,
i.e. it makes no difference if we either want to conceive the first two systems or
the last two systems as one whole. One can of course always choose to have
(A1 A2 ) A3 = A1 (A2 A3 )
with
Bob
C
Alice
A
A,B,C -
Bob
BC
Unit object and unit natural isomorphisms. Physical operations can destroy a system e.g. measurement of the position of a photon. On the other hand,
one can conceive a preparation procedure as the creation of a system from an
unspecified source. Therefore it is useful to have an object standing for no system,
preparation or state then being of the type I A and destruction being of the
type A I in Diracs notation [14] these respectively are the so-called kets
and bras. Clearly, since I stands for no system we have
AIAIA
and these left & right unit natural isomorphisms obviously should satisfy:
A
A
?
IA
- B
- B
?
AI
?
- IB
1I f
(2)
?
- BI
f 1I
i.e. introducing nothing should not alter the effect of an operation. In other words,
the left & right unit natural isomorphisms allow us to introduce or discard such
an extra object at any time. Such an object also comes with a notion of scalar
i.e. a morphism of type s : I I. In particular do these scalars arise when
post-composing a state with a costate i.e. when we have a bra-ket
:I
- A
- I.
10
- B
( 1B ) B
( 1A ) A
?
CA
?
- CB
1C f
- B
A
?
IA
1A
?
C A
B
?
- IB
1I f
Bif unct.
1B
?
- C B
1C f
union) with the empty set as unit object. Again (FdHilb, ) and (Rel, +) are
very similar categorically, but still quite different from (Set, +).
Bases independency. For the particular case of vector spaces over some field
K, setting Ai = Bi := Vi and taking f, g, ... to be a change of bases for the
corresponding vector space, the general naturality diagram (1) exactly expresses
base independency. Hence in the context of vector spaces natural concepts are
always bases independent concepts.
Coherence. We want the different natural isomorphisms introduced above to
coexist peacefully and for that reason we need to require some coherence conditions e.g. I,A A = A and I = I . We will not spell them out explicitly here.
The general theory of coherence in categories is highly non-trivial as a branch
of developing category theory (as opposed to using category theory). The reason
we mention these coherence conditions here is that the axiomatic algebra of categorical quantumness (see Section 11), somewhat surprisingly, first appeared in
the context of coherence theory [21, 23].
Braided categories. One coherence condition for a symmetric monoidal ten1
sor is A,B
= B,A i.e. B,A A,B = 1AB , which depicts as:
=
In braided monoidal categories this is not true anymore, giving rise to braided
structure for B,A A,B :
We refer to the web pages by John Baez and the books by Louis Kauffman for
prose on this body of mathematics research.
How much?
So far nothing quantitative seems to have been going on here. Not true! Given a
category C we will call A := C(I, A) the state space of system A and S := C(I, A)
the scalar monoid. The scalar monoid in (FdHilb, ) is isomorphic to C since
12
I
1I t
t
?
I
?
II
st
?
- I
?
?
?
I I ====== I I ====== I I
1I t
s 1I
?
I
- I
s 1I
?
II
I I ====== I I ====== I I
t
?
- I
This is quite a surprising result. From the very evident operationally motivated
assumptions on compoundness we obtain something as strong as a requirement of
commutation. This for example implies that if we would want to vary quantum
theory by changing the underlying field of the vector space we need to stay
commutative, excluding quaternionic quantum mechanics [16]. But there is much
more. The left-hand-side of the above diagram expresses
st=I
st
II
II
I.
AI
f s
BI
s
s
f
s
13
One can also show that states and costates satisfy a similar property (e.g. [12])
=A
IA
BI
The above introduced notions of bifunctoriality and natural iso are instances
of the key categorical concepts called functor and natural transformation. In
Eilenberg-Mac Lane functors were introduced as morphisms between categories
while natural transformations were introduced as morphisms between functors.
Definition. A functor F : C D is a structure preserving map of
categories i.e. it maps an object A to an object F A, and a morphism
f
Ff
A - B to a morphism F A - F B, and satisfies
F (g f ) = F g F f
and
F (1A ) = 1F A .
Given a category define a new category C C which has pairs (A, B) as objects, pairs (f, g) as morphisms, pairs (1A , 1B ) as identities and with composition
pairwise defined. Hence a functor F : C C C satisfies
F (g1 f1 , g2 f2 ) = F (g1 g2 ) (f1 f2 ) .
Setting F (, ) := it follows that a tensor is indeed a functor, by bifunctoriality. Another example is a group homomorphism which turns out to be
a functor of groups since functoriality implies preservation of inverses:
a1 a = e = a a1
F (a1 a) = F e = F (a a1 )
F (a1 ) F a = e = F a F (a1 )
(F a)1 = F (a1 ) .
14
Ff -
FB
A
?
GA
Gf
?
- GB
While this general definition might be non-intuitive, there are some conceptually
highly significant examples of it. A natural diagonal expresses the process of
copying. It consists of the family {A : A A A}A which again for operational
reasons obviously has to satisfy
A
A
?
AA
- B
B
?
- BB
f f
1 7 |0i + |1i
- CC
{}
- {0, 1}
{(0, (0, 0)), (1, (1, 1))}
{(, (, ))}
?
{(, )} = {} {}
?
- {0, 1} {0, 1}
x 7 f (x)
- Y
x 7 (x, x)
?
X X
?
- Y Y
and this is a consequence of the high-level fact that in Set the cartesian product
is a true product in the categorical sense.
16
p1
h
?
AB
A
p2
- B
p2 : C(C, A B) C(C, B)
p1 [f, g] = f
p2 [f, g] = g .
The three required equalities essentially say that pairing and unpairing are each
other inverse as meta-operations i.e. they allow each operation of type C AB
to be transformed in a pair of operations of respective types C A and C B
and vice versa. If one has such a product structure than one always has a natural
diagonal and provide a notion of copying. Moreover, also the projections are
natural and can be interpreted as a natural notion of deleting. (cf. the no-deleting
theorem in quantum mechanics [30])
17
f g
C D
p1
p1
?
A
?
- C
Specifying the idea of pairing and unpairing for states we have that the information encoded in any bipartite state
:IAB
can be equivalently encoded in the pair
1 = p1 : I A
and
2 = p2 : I B
C
6
which are such that for every pair operations f : A C and g : B C there
exists a unique operation h such that we have commutation of:
- AB
q1
q2
18
Enriched categories
We will not get in detail on this mathematically highly non-trivial subject and
refer the reader for an easy-going introduction to [11]; key historical references are
the inspired [26] and the not so easy [22]. Here we just want to mention the existence of this particular way of adding more structure to categories, since we will
encounter a simple example of it below. Consider the so-called Hilbert-Schmidt
correspondence for finite dimensional Hilbert spaces i.e. given two Hilbert spaces
H1 and H2 there is a natural isomorphism in FdHilb3
H1 H2 FdHilb(H1 , H2 )
(3)
between the tensor product of H1 (i.e. the dual of H1 ) and H2 and Hilbert space of
linear maps between H1 and H2 . In particular do we have that FdHilb(H1 , H2 )
is itself a Hilbert space. Note also that there exists a linear map
FdHilb(H1 , H2 ) FdHilb(H2 , H3 ) FdHilb(H1 , H3 ) :: (f, g) 7 g f
3 Surprisingly
enough, in much of the quantum mechanical literature (e.g. [5, 36]) one does
not encounter this natural correspondence but rather an un-natural one namely H1 H2
FdHilb(H1 , H2 ) which is merely a bijection between sets and which is of course is bases
dependent. The same is the case for many other notions used in the quantum physics literature.
Life could be made so much easier if physicist would learn about the benefits of naturality.
19
due to the universal property of the Hilbert space tensor product i.e. for each
triple H1 , H2 , H3 there exists a particular morphism in FdHilb which internalizes composition of linear maps. Hence we have a situation where the morphismsets of a category C are themselves structured as objects in (possibly another)
category (D, ) in such a manner that composition in C, i.e.
: C(A, B) C(B, C) C(A, C) ,
internalizes in (D, ) as an explicit morphism
cA,B,C : C(A, B) C(B, C) C(A, C) .
Such a category C is called D-enriched or simply a D-category. Each category
is by definition a Set-category. A 2-dimensional category or simply, a 2-category
is defined as a Cat-category. Similarly, a 3-category is a 2-Cat-category, and
a (n + 1)-category is a n-Cat-category, a branch of category which currently
intensively studied, and in particular strongly advertised by John Baez. A particular fragment FdHilb-enrichment (cf. FdHilb is itself FdHilb-enriched) is
enrichment in commutative monoids CMon i.e. linear maps can be added.
Logical closure
Categorical enrichment is not the only way to encode the Hilbert-Schmidt correspondence. From eq.(3) and (H1 H2 ) H1 H2 it follows that
FdHilb(H1H2 , H2 ) (H1H2 )H3 H1(H2H3 ) FdHilb(H1 , H2H3 )
Hence when defining a new connective between Hilbert spaces by setting
H2 H3 := H2 H3 ,
called implication, we obtain
FdHilb(H1 H2 , H2 ) FdHilb(H1 , H2 H3 )
which is a special case of the general situation of monoidal closure:
C(A B, C) C(A, B C)
where we now assume (not necesarily being in a self-enriched context) that the
isomorphism is natural in Set.4 This is precisely the deductive content of general
categorical logic, which states that for each proof
4 Actually we have an example of a so-called adjunction between the two functors B
and B for each object B of the category. While in many category theory books for
very compelling mathematical reasons adjunction will be put forward as the most important
mathematical concept of the whole of category theory, we unfortunately wont have the space
here to develop it, and it would deviate us too much from our story line.
20
comment does not apply to [18] where the whole topos models only a single system.
logical view highly contrasts the Birkhoff-von Neumann proposal in [9] that quantum
logic is a weak logic in which we can do less than classical logic. In fact, we can do more!
6 This
21
10
While till now we have focussed on the tensor product of Hilbert spaces, in this
section we show how the direct sum of Hilbert spaces carries the matrix calculus.
If an object is both terminal and initial we call it a zero-object and in that case
there is a unique zero-maps between any two objects:
0A,B
- B
0
and these zero-maps are moreover closed under composition:
0A,B -
0B,C -
B
6
0
Assume that in addition to this we have a situation of what we roughly describe
as coinciding products and coproducts, and which we will denote by .
Since in this case we both have diagonal and a codiagonal , for each pair
f, g : A B we can define the following sum:
f + g := A
- AA
- AA
f g
- A
f (g1 + g2 ) = (f g1 ) + (f g2 ) .
One verifies that we obtain CMon-enrichment. But we also both have projections
and injections so for each morphism
f :AB C D
we can write down a matrix
(fij )ij =
p1 f q1
p1 f q2
p2 f q1
p2 f q2
and it turns out that we obtain a full-blown matrix calculus in which we can add
and multiply in the usual linear-algebraic fashion. The exact notion which captures the above situation is that of a biproduct. We give two alternative equivalent
definitions.
22
p2 p1
- AB
B
q1
q2
with
pi qj = ij
then C has biproducts.
qi pi = 1AB
Each such biproduct category admits an additive and multiplicative matrix calculus, and each category with numbers as objects and n m-matrices in a commutative semiring as morphisms yields a biproduct category. In particular (Rel, +)
and (VectK , ) are biproduct categories.
Distributivity. We have now seen that in FdHilb there exist two monoidal
structures, namely the -structure which captures entanglement, and the structure which provides the matrix calculus. But these two are not at all independent since there exists a distributivity natural isomorphism:
(A1 A2 ) C
(f1 f2 ) g
- (B1 B2 ) D
DISTA1 ,A2 ,C
?
(A1 C) (A2 C)
DISTB1 ,B2 ,D
?
- (B1 D) (B2 D)
(f1 g) (f2 g)
Such a distributity isomorphism is a very useful tool which for example can be
used to encode classical communication between agents [2]:
(I I) Agent (I Agent) (I Agent) .
However, while the -structure and -structure clearly behave very different,
the first is in the case of finite dimensional objects derivable from the second.
Indeed, given a biproduct category BP with an object I such that BP(I, I) is
commutative, define a new category:
the objects are the natural numbers N {I . . . I | n N}
| {z }
n
23
nm
In turns out that we always obtain a compact closed category, something which
can exist independently without the presence of an underlying biproduct structure. We strongly believe that the essence of quantum mechanics does not lie in
its matrix calculus but in the independent structure of the tensor product. Recall
here that unveiling the underlying axiomatic structure of quantum mechanics was
a quest started by the formalisms creator John von Neumann [34], initiated in
[9], but did not lead to a satisfactory ending [31]. The main difficulty was the
axiomatization of the tensor product. We believe that the missing ingredient is
the concept of monoidal categories and its underlying operational significance.
11
Categorical quantumness
In [2, 3] Abramsky and myself striped down the tensor product to its bare categorical bones. The structure which emerged was a slightly refined version of Kellys
compact closure [21] which we called strong compact closure. But the greatest
virtue of this structure is that, as it was the case for monoidal categories, it can
still be captured by a simple graphical calculus, as surveyed in my lecture notes
entitled Kindergarten Quantum Mechanics [13]. What we need to add to the
symmetric monoidal graphical calculus is orientation of the lines connecting the
boxes and an operation adjoint which reverses boxes:
A
B
A
A*
A
A*
24
A*
A
A*
=
which now tells us that we are allowed to yank the black line:
=
It turns out that with this bit of structure we basically capture all the behavioral properties of quantum mechanics, and are even able to define notions such
as inner-product, unitarity, Hilbert-Schmidt inner-product, Hilbert-Schmidt mapstate duality, projection, positivity, measurement, Born-rule (which provides the
probabilities) [2, 3], transposition vs. adjoint, global phase and elimination thereof,
vectorial vs. projective formalism [12], full and partial trace [20], completely positive maps and Jamiolkowski map-state duality [33].7 Inventing the quantum teleportation protocol boils down in this calculus to a trivial application of yanking:
7 While
25
Alice
Bob
Alice
Bob
Alice
Bob
12
Conclusion
26
Acknowledgments
This journey was put together as an invited tutorial on category theory for
quantum informaticians at the Perimeter Institute for Theoretical Physics in
July 2005 and was also presented as an invited talk at the conference Impact of
References
[1] Abramsky, S. (2002 ...) Categories, Proofs and Processes. Course at Oxford University Computing Laboratory. Documentation and lecture notes are
available at web.comlab.ox.ac.uk/oucl/courses/topics05-06/cpp/
[2] Abramsky, S. and Coecke, B. (2004) A categorical semantics of quantum protocols. In: Proceedings of the 19th Annual IEEE Symposium on Logic in Computer Science (LiCS04), IEEE Computer Science Press. An extended & improved version is available at arXiv:quant-ph/0402130
[3] Abramsky, S. and Coecke, B. (2005) Abstract physical traces. Theory and Applications of Categories 14, 111124. Available at http://www.tac.mta.ca/
tac/volumes/14/6/14-06abs.html
[4] Abramsky, S., Gay, S. J. and Nagarajan, R. (1995) Interaction categories and
foundations of typed concurrent programming. In: Deductive Program Design:
Proceedings of the 1994 Marktoberdorf International Summer School, pp. 35
113, NATO ASI Series F, Springer-Verlag.
[5] Arrighi, P. and Patricot, C. (2004) On quantum operations as quantum states.
Annals of Physics 311, 2652. arXiv:quant-ph/0307024
[6] Baez, J. (2004) Quantum quandaries: a category-theoretic perspective. In:
S. French et al. (Eds.) Structural Foundations of Quantum Gravity, Oxford
University Press. arXiv:quant-ph/0404040
[7] Barr, M. (1979) -Autonomous Categories. Lecture Notes in Mathematics
752, Springer-Verlag.
[8] Benabou, J. (1963) Categories avec multiplication. Comptes Rendus des
Seances de lAcademie des Sciences. Paris. 256, 18871890.
[9] Birkhoff, G. and von Neumann, J. (1936) The logic of quantum mechanics.
Annals of Mathematics 37, 823843.
27
[10] Blute, R. and Scott, Ph. (2004) Category theory for linear logicians. In:
Linear Logic in Computer Science, T. Ehrhard, J.-Y. Girard, P. Ruet and
Ph. Scott, pp. 364, London Mathematical Society Lecture Note Series 316,
Cambridge University Press.
[11] Borceux, F. and Stubbe, I. (2000) Short introduction to enriched categories.
In: Current Research in Operational Quantum Logic: Algebras, Categories,
Languages, pp. 167194, B. Coecke, D. J. Moore and A. Wilce, Fundamental
Theories of Physics 111, Kluwer.
[12] Coecke, B. (2005) De-linearizing linearity: Projective quantum axiomatics
from strong compact closure. Electronic Notes in Theoretical Computer Science, Proceedings of the 3rd International Workshop on Quantum Programming Languages. arXiv:quant-ph/0506134
[13] Coecke, B. (2005) Kindergarten quantum mechanics lecture notes. In:
Quantum Theory: Reconsiderations of the Foundations III. A. Khrennikov,
Ed., American Institute of Physics Press. arXiv:quant-ph/0510032
[14] Dirac, P. A. M. (1947) The Principles of Quantum Mechanics (third edition).
Oxford University Press.
[15] Eilenberg, S. and Mac Lane, S. (1945) General theory of natural equivalences.
Transactions of the American Mathematical Society, 58, 231294.
[16] Finkelstein, D. R., Jauch, J. M., Schiminovich, D. and Speiser, D. (1962)
Foundations of quaternion quantum mechanics. Journal of Mathematical
Physics 3, 207220.
[17] Girard, J.-Y. (1987) Linear logic. Theoretical Computer Science 50, 1102.
[18] Isham, C. J. and Butterfield, J. (1998) Topos perspective on the KochenSpecker theorem: Quantum states as generalized valuations. International
Journal of Theoretical Physics 37, 26692733.
[19] Joyal, A. and Street, R. (1991) The Geometry of tensor calculus I. Advances
in Mathematics 88, 55112.
[20] Joyal, A., Street, R. and Verity, D. (1996) Traced monoidal categories. Proceedings of the Cambridge Philosophical Society 119, 447468.
[21] Kelly, G. M. (1972) Many-variable functorial calculus I. In: Coherence in
Categories, pp.66105, G. M. Kelly, M. Laplaza, G. Lewis and S. Mac Lane,
Eds., Lecture Notes in Mathematics 281, Springer.
[22] Kelly, G. M. (1982) Basic concepts of enriched category theory. London
Mathematical Society Lecture Note Series 64. Cambridge University Press.
28
[23] Kelly, G. M. and Laplaza, M. L. (1980) Coherence for compact closed categories. Journal of Pure and Applied Algebra 19, 193213. Transactions of
the American Mathematical Society, 58, 231294.
[24] Lambek, J. (1958) The mathematics of sentence structure. American Mathematics Monthly 65, 154169.
[25] Lambek, J. (2004) What are pregroups? In: Language and Grammar: Studies
in Mathematical Linguistics and Natural Language, C. Casadio, P. J. Scott,
and R. A. G. Seely, CSLI Publications.
[26] Lawvere, F. W. (1974) Metric spaces, generalized logic, and closed categories.
Rendiconti del Seminario Matematico e Fisico di Milano 43, 135166.
[27] Lawvere, F. W. and Schanuel, S. H. (1997) Conceptual mathematics. Cambridge University Press.
[28] Lawvere, F. W. and Rosebrugh, R. (2003) Sets for mathematics. Cambridge
University Press.
[29] Mac Lane, S. (1971) Categories for the Working Mathematician. Springer.
[30] Pati, A. K. and Braunstein, S. L. (2000) Impossibility of deleting an unknown
quantum state. Nature 404, 164165.
[31] Redei, M. (1997) Why John von Neumann did not like the Hilbert space formalism of quantum mechanics (and what he liked instead). Studies in History
and Philosophy of Modern Physics 27, 493510.
[32] Seely, R. A. G. (1998) Linear logic, -autonomous categories and cofree algebras. Contemporary Mathematics 92, 371382.
[33] Selinger, P. (2005) Dagger compact closed categories and completely positive
maps. Electronic Notes in Theoretical Computer Science, Proceedings of the
3rd International Workshop on Quantum Programming Languages.
[34] von Neumann, J. (1932) Mathematische Grundlagen der Quantenmechanik.
Springer-Verlag. English translation (1955): Mathematical Foundations of
Quantum Mechanics. Princeton University Press.
[35] Wootters, W. K. and Zurek, W. (1982) A single quantum cannot be cloned.
Nature 299, 802803.
29
30