Материал: искусственный интеллект

146 S. Gogioso and C. M. Scandolo

For example, the component ρ0321 of a 4+-dimensional system will fall into the 1st class from the left above, the component ρ0122 will fall into the 2nd class, the component ρ0003 into the 8th class, the component ρ0011 into the 12th class and the component ρ0000 into the 15th class.

Then we look at the individual orbits of components in each class under the symmetry. Classes with components having orbits of order 4 are shown below: each orbit contributes a single independent complex value to the tensor, i.e. two independent real values, and each component class is annotated by the total number of independent real values contributed in dimension d. Just as we did above, we are using colours to denote values in {1, ..., d}: the geometric action of Z2 × Z2 on the coloured vertices/edges of the squares exactly mirrors the algebraic action of Z2 × Z2 on the components in the di erent classes.

(12) Classes with components having orbits of order 2 and 1 are shown below, each component class annotated by the total number of independent real values contributed in dimension d. Each orbit in the first, second and fourth classes contributes a single independent real value, because each component is stabilised by (at least) one self-adjoining symmetry; each orbit in the third class contributes instead two independent real values, because the components are only stabilised by a self-transposing symmetry.

Adding up the contributions from all orbit classes, we see that the states of d-dimensional density hypercubes form a convex cone of real dimension 12 (d4 − 3d3 + 7d2 − 3d) within the (2d4)-dimensional real vector space of complex fourthorder tensors.

2.3Normalisation and Causality

The “forest” discarding maps DD(H) := CPM( H) in DD(fHilb) (i.e. the

doubled versions of the discarding maps of CPM(fHilb)) form an environment structure [9, 11], and we say that a map of density hypercubes is normalised

if the corresponding CP map is trace preserving (with normalised states as a special case):

(14)

Normalised maps of density hypercubes form a sub-SMC of DD(fHilb), which we refer to as the normalised sub-category. Sub-normalised maps of density hyper-

cubes can be defined analogously by requiring the corresponding CP map to be

trace non-increasing: they also form a sub-SMC of DD(fHilb), which we refer to as the sub-normalised sub-category.

Despite the presence of several kinds of discarding maps, the following results shows that the sub-normalised sub-category is causal [3], or equivalently that that the normalised sub-category is terminal [6, 7].

Proposition 1. The process theory DD(fHilb) is causal, in the following sense: for every object DD(H), the only e ect DD(H) → R+ in DD(fHilb) which yields the scalar 1 on all normalised states of DD(H) is the “forest” discarding map of density hypercubes DD(H).

3 Decoherence and Hyper-decoherence

So far, we have constructed a symmetric monoidal category, which is enriched in convex cones and comes equipped with an environment structure provid-

ing a notion of normalization. The final ingredients necessary for the definition of the categorical probabilistic theory of density hypercubes is the demonstra-

tion that classical systems and quantum systems arise in the Karoubi envelope of DD(fHilb) by choosing some suitable family of decoherence and hyperdecoherence maps.

3.1Decoherence to Classical Theory

Consider a finite-dimensional Hilbert space H and a classical structure on it, associated with some orthonormal basis (|ψx )x X . We define the -decoherence

map dec on the density hypercube DD(H) to be the following morphism in DD(fHilb):

The dec map defined above is idempotent, so it can be used to define classical systems via the Karoubi envelope construction—in the same way as ordinary

148 S. Gogioso and C. M. Scandolo

decoherence maps gives rise to classical systems in quantum theory. It should be noted that decoherence maps defined this way are sub-normalised but not normalised, so that the hyperquantum-to-classical transition in the theory of density hypercubes is not deterministic; we defer further discussion of this point to the next sub-section on hyper-decoherence.

Proposition 2. Let Split(DD(fHilb)) be the Karoubi envelope of DD(fHilb), and write Split(DD(fHilb))K for the full subcategory of Split(DD(fHilb)) spanned by objects in the form (DD(H), dec). There is an R+-linear monoidal equivalence of categories between Split(DD(fHilb))K and the probabilistic theory R+-Mat of classical systems. Furthermore, classical stochastic maps correspond

H

3.2Hyper-decoherence to Quantum Theory

If is a classical structure on a density hypercube DD(H), we define the -

Hyper-decoherence maps are idempotent, and hence we can consider the full subcategory C of the Karoubi envelope Split(DD(fHilb)) spanned by objects in the form (DD(H), hypdec): doing so allows us to prove that the hyperdecoherence maps defined above truly provide the desired “hyper-quantum–to– quantum” decoherence, as considered by [18, 20].

Proposition 3. Let Split(DD(fHilb)) be the Karoubi envelope of DD(fHilb), and write Split(DD(fHilb))Q for the full subcategory of Split(DD(fHilb)) spanned

by objects in the form (DD(H), hypdec). There is an R+-linear monoidal equivalence of categories between Split(DD(fHilb))Q and the probabilistic theory CPM(fHilb) of quantum systems and CP maps between them. Furthermore, trace-preserving CP maps correspond to the maps in Split(DD(fHilb))Q nor-

Taking the double-dilation construction together with the content of Propositions 2 and 3, we come to the following definition of a categorical probabilistic theory [11] of density hypercubes.

Definition 1. The categorical probabilistic theory of density hypercubes DH(fHilb) is defined the be the full sub-SMC of Split(DD(fHilb)) spanned by objects in the following form:

–the density hypercubes (DD(H), idDD(H));

–the quantum systems (DD(H), hypdec), for all classical structures on H;

–the classical systems (DD(H), dec), for all classical structures on H.

The environment structure for the categorical probabilistic theory is given by the

discarding maps DD(H), (DD(H),hypdec ) and (DD(H),dec ) respectively. The classical sub-category for the categorical probabilistic theory is the full sub-

SMC spanned by the classical systems.

The hyper-quantum–to–classical and hyper-quantum–to–quantum decoherence maps of density hypercubes play well together with the quantum–to– classical decoherence map of quantum theory: the decoherence map dec : (DD(H), idDD(H)) → (DD(H), dec) of density hypercubes factors, as one would expect, into the hyper-decoherence map hypdec, from (DD(H), idDD(H)) to (DD(H), hypdec), followed by the decoherence map of quantum systems dec, from (DD(H), hypdec) to (DD(H), dec). From this, it is clear that the reason why hyper-quantum–to–classical transition was sub-normalised is that the hyper-quantum–to–quantum transition itself is sub-normalised (cf. Appendix B).

The sub-normalisation of hyper-decoherence maps is a sign that the theory of density hypercubes presented here is still partially incomplete, and that some suitable extension will need to be researched in the future. What we know for sure is that the current theory does not satisfy the no-restriction condition on e ects, and that an extension in which hyper-decoherence maps are normalised is possible: the additional e ect needed by normalisation exists in CPM(fHilb) and is non-negative on all states of DD(fHilb) (cf. Appendix B). In line with the recent no-go theorem of [20], preliminary considerations seem to indicated that the addition of said e ect would mean that the theory no longer satisfies purification.

150 S. Gogioso and C. M. Scandolo

4 Higher Order Interference

In this section, we will show that the theory of density hypercubes displays thirdand fourth-order interference e ects, broadly inspired by the framework for higher-order interference in GPTs presented by [1, 2, 18]. Because interference has to do with decompositions of the identity map in terms of certain projectors, we begin by introducing a handy graphical notation for keeping track of the various pieces that the identity map is composed of.

The identity map of hyper-quantum systems idDD(H) : DD(H) → DD(H) takes the following explicit form in fHilb, for any orthonormal basis (|ψx )x X

of the Hilbert space H:

In order to denote the pieces in the decomposition corresponding to specific values x00, x01, x10, x11 X of the indices, we adopt the following graphical notation, inspired by the Z2 × Z2 symmetry of the components:

In fact, we will adopt the same colour-based notation for index values which we originally introduced in Sect. 2, so that the following is a decomposition piece involving two distinct index values {•, •} X:

Using the colour-based notation defined above for its pieces, the identity on a 2-dimensional hyper-quantum system (with X = {•, •}) would be fully decomposed as follows: