
2
developed for entanglement by path identity [19,20]
but also extends to bulk optics or integrated photonics
[12,21]. Integrated photonic chips have made impressive
technological progress recently and offer great potential
for the experimental realization of setups discovered by
pytheus [22–27]. The search for an experiment creating
a target state is formulated as an optimization that max-
imizes the fidelity of a graph. When a solution is found,
the corresponding graph can directly be trannslated to an
experimetnal setup consisting of standard optical compo-
nents. A detailed explanation of digital discovery using
pytheus is provided in the accompanying paper [16].
Here we focus on the scientific consequences of the new
quantum optics concept Halo, the underlying physics,
its connection to modern quantum optics experiments,
and how it can be productively used.
A prominent class of examples for genuine multi-particle
entanglement are the GHZ states [28] [29].
|GHZid
n=1
√d
d
X
i=0 |ii⊗n(1)
where nis the number of particles and dis the dimen-
sion of the GHZ state. |GHZid
nis a generalization of
the original |GHZi2
3state [30]. Fig. 2shows a correla-
tion network for the state |GHZi3
4. An edge of the graph
corresponds to a two-particle correlation as it is intro-
duced by probabilistic photon pair sources, for example
by spontaneous parametric down-conversion (SPDC). A
vertex corresponds to a path to a detector at the end of
the setup. We condition the final quantum state on the
detection of one photon in each of the detectors, which
motivates the usage of perfect matchings in graphs. A
perfect matching is a set of edges by which each vertex
of a graph is covered exactly once. In an event where all
crystals corresponding to the edges in a perfect matching
fire, exactly one photon will enter each detector. Fig. 2
also shows that each perfect matching of the graph can
be understood as a contribution to the created state.
Fig. 2also shows that it is not straightforward to extend
the construction to |GHZi4
4. It is impossible to create
|GHZi4
4using linear optics without the use of additional
resources in the form of ancillary photons [18,21]. This
is a physical limitation that can be explained by graph
theory as follows. There are at most three disjoint per-
fect matchings in a four-vertex graph. This makes it
impossible to create four different GHZ terms without
introducing cross terms with a four-vertex graph.
Including additional resources allows us to go beyond
this limitation. We can use pytheus to search for a
graph corresponding to the state
|ψi=|GHZi4
4⊗ |0000i,(2)
which is the four-dimensional four-particle GHZ state in
a product state with four ancillary particles. For this tar-
get, pytheus discovers minimal solution in Fig. 3(b).
The solution shown has 12 perfect matchings. Four pairs
Figure 2. (a) A graph corresponding to the experimental
setup for the creation of the |GHZi3
4state. A perfect matching
is a set of edges by which each vertex of a graph is covered ex-
actly once. The graph has three perfect matchings (blue, red,
and green). In this case, each perfect matching corresponds
to one term in the target state. (b) shows an incorrect at-
tempt at drawing a graph corresponding to the state |GHZi4
4.
The two additional edges create the wanted fourth term at
the cost of two unwanted cross-terms.
interfere constructively creating the four GHZ terms.
The remaining two pairs each correspond to cross-terms
but interfere destructively.
The physics of HALO – By closer inspection of the
solution for |GHZi4
4(Fig. 3(b)), we see that the graph for
|GHZi3
4(shown in Fig. 2(a)) is included as a subgraph.
The experimental setup for |GHZi3
4can be seen as a basic
setup that is extended by the components corresponding
to the remaining edges of the graph |GHZi4
4. When all
detectors click, the additional components either produce
four correlated particles or none, with all cross-terms de-
structively interfering, imitating a four-particle emitter.
The physical interpretation of the ancillary subgraph is
shown in Fig. 4. This is achieved by an interference
pattern that can be interpreted as an extension of frus-
trated multiphoton interference, an effect described in
[21] and experimentally observed in [31,32]). Building
probabilistic multi-particle emitters is an active area of
research [33,34]. Halo offers a way of emulating cor-
related multi-particle emitters with pair sources in post-
selected experiments, which is a complementary experi-
ment route that can employ physically well-understood
technologies.
Concept Extraction – We give the concept that ap-
pears in the solution for |GHZi4
4a name:
AHalo (Hyperedge Assembly by Linear Optics) is a
subsystem of a linear optics setup, which effectively acts
as a probabilistic multi-photon source.
This definition is not used by the algorithm to produce
solutions, rather we abstract it from the solutions. In the
abstract graph representation, a multi-particle emitter
can be described by a hyperedge (shown in Fig. 3(c)).
A hyperedge is drawn as a shape enclosing the n > 2
vertices it connects. Thus, a Halo-subgraph constructed
from regular edges can also be represented by hyperedges.
Hypergraphs describing quantum experiments involving