Материал: искусственный интеллект

Density Hypercubes, Higher Order

Interference and Hyper-decoherence:

A Categorical Approach

Stefano Gogioso1(B) and Carlo Maria Scandolo2

1 University of Oxford, Oxford, UK

stefano.gogioso@cs.ox.ac.uk

2University of Calgary, Calgary, Canada carlomaria.scandolo@ucalgary.ca

Abstract. In this work, we use the recently introduced double-dilation construction by Zwart and Coecke to construct a new categorical probabilistic theory of density hypercubes. By considering multi-slit experiments, we show that the theory displays higher-order interference of order up to fourth. We also show that the theory possesses hyperdecoherence maps, which can be used to recover quantum theory in the Karoubi envelope.

Keywords: Quantum theory · Categorical probabilistic theories · Higher-order interference · Hyperdecoherence

1 Introduction

Quantum interference is often considered to be one of the fundamental features of quantum theory, responsible for quantum advantage in a number of computational tasks. However, there is a known limit to how much interference quantum theory can exhibit. Sorkin proposed a hierarchy of theories based on the maximum order of interference they exhibit [25, 26], which is quantified by the maximum number of slits on which a theory shows an irreducible interference behaviour. Interference in quantum theory is limited to the second order: the interference pattern of two slits cannot be reduced to the pattern of single slits, but the interference pattern of three slits can be reduced to the pattern arising from pairs of slits and single slits. This limitation has been recently confirmed in various experiments [13, 14, 22–24].

A natural question arises: Why is interference in Nature limited to the second order? Does the presence of higher-order interference create any paradoxical consequences in Nature that conflict with some of the principles we believe to be fundamental? Recent work has shown that higher-order interference—i.e. interference of order higher than the second—is forbidden [1] in physical theories which admit a fundamental level of description where everything is pure and

c Springer Nature Switzerland AG 2019

B. Coecke and A. Lambert-Mogiliansky (Eds.): QI 2018, LNCS 11690, pp. 141–160, 2019. https://doi.org/10.1007/978-3-030-35895-2_10

142 S. Gogioso and C. M. Scandolo

reversible [4, 5]. Further work has ruled out higher-order interference based on thermodynamic considerations [4, 15].

Other literature has instead focused on the analysis of specific feature that theories with higher-order interference would possess, e.g. whether they would provide any advantage in certain computational tasks [16–19]. It was also shown that theories having second-order interference and lacking interference of higher orders are relatively close to quantum theory [2, 21, 27, 28].

Unfortunately, one of the major shortcomings in the study of higher-order interference is the scarcity of concrete models displaying such post-quantum features, so that it has so far been very hard to look for specific examples of paradoxical or counter-intuitive consequences. Two models—density cubes [10] and quartic quantum theory [30]—have been proposed in the past, but are not fully defined operational theories, e.g. because they do not deal with composite systems [18]. This limitation precludes them from being used to study all possible consequences of higher order interference, including potential violation of Tsirelson’s bound.

In this article, we provide the first complete construction of a full-fledged operational theory exhibiting interference up to the fourth order. Our construction is inspired by the double-dilation construction of [29] and the higher-order CPM constructions of [12], and it is carried out in within the framework of categorical probabilistic theories [11]. The resulting theory of ‘density hypercubes’ has composite systems, exhibits higher-order interference and possesses hyperdecoherence maps [18, 20, 30]. Quantum theory, with its second-order interference, is an extension of classical theory: the latter can be recovered by decoherence, which eliminates the second-order interference e ects. Similarly, the theory of density hypercubes, with its thirdand fourth-order interference, is an extension of quantum theory: the latter can now be recovered by hyper-decoherence, which eliminates thirdand fourth-order interference e ects.

The paper is organized as follows. In Sect. 2, we define the categorical probabilistic theory of density hypercubes using the double-dilation construction. In Sect. 3, we define hyper-decoherence maps, and show that quantum theory is recovered in the Karoubi envelope. In Sect. 4, we show that density hypercubes display interference of thirdand fourth-order, but not of fifth-order and above. In Sect. 5, finally, we discuss open questions and future lines of research. Proofs of all results can be found in the Appendix.

2 The Theory of Density Hypercubes

2.1Construction of the Theory

In this section, we define the categorical probabilistic theory of density hypercubes, using a recently introduced construction known as double dilation [29].

The construction is done in two steps: first we define the category DD(fHilb), containing hyper-quantum systems and processes between them, and only in

a second moment we introduce quantum and classical systems, using (hyper-)

decoherence and working in the Karoubi envelope Split(DD(fHilb)).

The double-dilation category DD(fHilb) is defined to be a symmetric monoidal subcategory of CPM(fHilb) with objects—the density hypercubes—in the form

DD(H) := H H, where H is a finite-dimensional Hilbert space and H := H H is the corresponding doubled system in the CPM category. Even though DD(fHilb) is symmetric monoidal and has its own graphical calculus, in this work we will always use the graphical calculi of CPM(fHilb) and fHilb to talk about density hypercubes. When working in CPM(fHilb), we will use solid black lines for morphisms and calligraphic letters (e.g. H) for objects. When working in fHilb, we will use solid grey lines for morphisms and plain letters (e.g. H) for objects.

The morphisms DD(H) → DD(K) in DD(fHilb) are the CP maps H H → K K taking the following form for a doubled CP map F , some auxiliary systems

E , G and some special commutative †-Frobenius algebra (henceforth known as a classical structure) on G in fHilb:

In the diagram above, F is a doubled CP map H → G K E in CPM(fHilb)—i.e. one in the form F = f f for some f : H → G K E in fHilb—and we have

¯

used F to denote the CP map obtained by inverting the tensor product ordering of inputs and outputs of f (for purely aesthetic reasons). We will always use upper-case letters (e.g. F ) to denote doubled CP maps in CPM(fHilb), lowercase letters to denote the corresponding linear maps in fHilb, and we will always write discarding maps explicitly.

Composition in DD(fHilb) is the same as composition of CP maps, while tensor product is only slightly adjusted to take into account the doubled format of our new morphisms:

G

144 S. Gogioso and C. M. Scandolo

Just as was the case for CP maps, maps of density hypercubes can all be obtained as composition of a “doubled” map and one or two “discarding” maps:

We refer to the discarding map obtained by doubling E as the “forest” and to the discarding map obtained from the classical structure as the “bridge”. The scalars of DD(fHilb) are exactly the scalars R+ of CPM(fHilb), and hence the theory of density hypercubes is probabilistic. It is furthermore convex, because the following “tree-on-a-bridge” e ects can be used to add-up maps of density hypercubes—analogously to the way ordinary discarding maps H can be used to add-up CP maps in CPM(fHilb)—by expanding them in terms of the orthonormal bases |ψx x X associated [8] with the classical structures :

2.2Component Symmetries

States in the theory DD(fHilb) take the form of fourth order tensors, an observation which prompted the choice of “density hypercubes” as a name for the theory. If (|ψx )x X is a choice of orthonormal basis for some finite-dimensional Hilbert space H, the states on DD(H) = H H in DD(fHilb) can be expanded as follows in fHilb:

Recall that density matrices possess a Z2 symmetry given by self-adjointness. This symmetry can be understood in terms of the following action τ : Z2 → Aut(C) of Z2 on the complex numbers:

The components of a density matrix ρ then satisfy the following equation, for every a Z2 (trivial for a = 0, self-adjoint for a = 1):

Instead of a Z2 symmetry, density hypercubes possess a Z2 × Z2 symmetry. This symmetry can be understood in terms of the following action τ : Z2 × Z2 → Aut(C) of Z2 × Z2 on the complex numbers:

The components of a density hypercube ρ satisfy the following equation for every (a, b) Z2 × Z2, where by we have denoted addition in Z2:

We see that the components are related by a trivial symmetry for a = (0, 0), by a self-adjoining symmetry for a = (1, 0) and a = (0, 1), and by a self-transposing symmetry in for a = (1, 1). An alternative way to look at this symmetry is observe that states of density hypercubes can all be expressed as certain sums of doubled states in the following form:

ΦK

For these states, we have the usual self-conjugating Z2 symmetry of density matrices Φ Φ → Φ Φ as well as an independent self-transposing Z2 symmetry

Φ Φ → Φ Φ, which taken together give the same Z2 × Z2 symmetry described above in terms of components.

In order to visualise the Z2 × Z2 symmetry action, we divide the components ρx00x01x10x11 of a d-dimensional density hypercube ρ into 15 classes, depending on which indices x00, x01, x10, x11 have same/distinct values chosen from the set {1, ..., d}. We arrange the indices on a square: index 00 is on the top left corner, 10 acts as reflection about the vertical mid-line, 01 acts as reflection about the horizontal mid-line and 11 acts as 180o rotation about the centre. We use colours as names for index values in {1, ..., d}, with distinct colours denoting distinct values.

(11)