Take two lattices and put a whole copy of the first at every site of the second. One number then says how strongly the outer lattice couples relative to the inner one. Turn it all the way down and the system is many separate copies of the inner lattice. Turn it all the way up and it is the outer lattice, with everything inner along for the ride. This is about what it is in between.
Nesting of this kind is not a contrivance. Moiré and super-moiré materials, cold atoms in superlattices, DNA-templated arrays and nested photonic networks all have two or more characteristic lengths that matter at once, and behaviour at neither one alone. What this paper adds is the dial: a construction in which the interplay between the scales is a parameter, so the crossover can be walked through rather than happened upon.
A lattice at every site
A site of the nested lattice needs two addresses: where it sits inside its copy, and which copy that is. Written that way the Hamiltonian is a sum of two terms, one acting on each address, weighted so that they always add to one.
with the prefactors normalised so the two weights sum to one and is their ratio. In the spatial basis,
Nothing here makes one lattice inner and the other outer: they enter symmetrically, and the hierarchy has to be put in by hand, by replacing with something that is not the identity — strengthening the links between edges of neighbouring copies relative to the links between their bulks, say. Higher orders nest the same way, , with one energy scale per layer.
Eqs. (1)–(5) · equation numbers throughout are those of arXiv:2511.13831v1
The projector form of the coupling covers moiré systems directly: supercells described by , coupled through with the coupling localised to the supercell interfaces.
Because both lattices have edges and bulks of their own, a state now has a character at each scale independently. The paper writes these as bulkbulk, bulkedge, edgebulk and edgeedge, the base label naming the smallest lattice. A bulkedge state is an edge state of the inner lattice sitting in the bulk of the outer one. That is the breakdown in the title: bulk and edge stop being a dichotomy and become a pair of labels.
One dial
Take both lattices to be anomalous quantum Hall lattices, couple copies at their corners, and turn the dial. At the spectrum is the inner lattice’s, repeated once per copy and degenerate. As the outer hopping comes up the degeneracy lifts; bands split, cross, and at isolated values of collapse to a point — a perfectly flat band — before opening out again. The paper calls the middle of this the cocoon, and defines it precisely: the at which the levels are spread most evenly, where the sum of squared level spacings is least.
What appears in between
At either end of the dial the system is one lattice and has one lattice's worth of structure. In the middle it has more, and the extra is not a mixture of the two ends.
| α → 0 | in the cocoon | α → ∞ | |
|---|---|---|---|
| the spectrum | the inner lattice’s, once per copy, degenerate | degeneracies lifted; bands split, cross and flatten; mini-gaps proliferate | the outer lattice’s, once per site, degenerate |
| topology | a single first Chern number | a second Chern number, and a Chern character that depends on the scale you measure at | a single first Chern number |
| flat bands | whatever the inner lattice has, and no more | perfect flat bands at isolated values of α, some of them embedded in topological gaps | whatever the outer lattice has, and no more |
| states | edge or bulk | edgeedge, edgebulk, bulkedge, bulkbulk — and isolated edge bands detached from any gap | edge or bulk |
Topology that depends on how far away you stand
A Chern number computed band by band cannot see the inner lattice at all: it is an integral over a Brillouin zone, and that zone belongs to the outer periodicity, too small to resolve anything inside a copy. The way around it is to stop integrating over momenta and start averaging over positions, at a chosen resolution.
with the local Chern marker and a normalised Gaussian of spread . The coarse-grained marker is blind to anything smaller than , so is a scale of observation. For a third-order lattice of size , the bulk value flips as is increased. Treating the scales as separate subspaces gives a higher invariant that is simply the product of the per-scale ones, , so several scales with multiply into a large one.
Eqs. (7)–(9)
The point of the product is not bookkeeping. A four-dimensional quantum Hall system carries a second Chern number, and four dimensions is what the cocoon looks like: at the two pairs of coordinates enter on the same footing and there is nothing to distinguish inner from outer. The difference from a topological charge pump is that here the high-dimensional lattice is present all at once, on a flat circuit, rather than being swept out one quasimomentum at a time.
Magic flat bands
Three things can happen to a band as the dial turns. It can deform while keeping its shape; it can touch another band and change its Chern number; or its bandwidth can collapse to zero at one value of and open again. The third is the interesting one, and the paper calls those values magic, by analogy with the magic angles of twisted bilayer graphene.
The analogy is doing work. Landau levels are flat and stay flat as the field changes; so do the flat bands of a Lieb or line-graph lattice as long as the graph is intact, and so do those of an Aharonov–Bohm cage at fixed flux. These have none of that. They exist on a measure-zero set of , the band curvature reverses sign across the flat point, and they do not appear in a single ring, a uniform ring lattice or a simply connected block lattice — the nesting is what makes them. Some are bounded by trivial gaps; others sit inside topological gaps, threaded by edge bands.
In the lattice a perfect flat band sits at near , and the curvature varies slowly enough there that it does not need fine tuning.
There is also a reason to expect them. On a flat band nothing propagates, so the time direction can be dropped, and a four-dimensional theory becomes a four-dimensional Euclidean one — where a chiral anomaly obstructs the flat band unless it cancels. The cancellation condition is a criterion for flatness, of the kind used in Generic Topological Criterion for Flat Bands in Two Dimensions and computed by the method of Path-Integral Approach to Quantum Anomalies in Interacting Models. Nesting, in other words, is a way of building a higher-dimensional anomaly out of two-dimensional parts.
Where it would be built
The proposal is a photonic circuit, in coupled ring resonators of the kind foundries already make at wafer scale. Site rings carry the modes; detuned link rings set the hoppings, one colour of link per scale. A first-order lattice becomes a one and then a one by repeating the same move, and is set by the link rings rather than by growing a new sample.
What you would look at is the drop-port spectrum together with the field pattern: the resonances nest as the order goes up, and the bulk-type and edge-type bands separate in frequency, so the labels above become something you can point at. Above threshold the picture changes again — the several hopping rates are several timescales at once, which is an unusual setting for nonlinear optics.
The flat bands are the part that connects back. Where a band is flat the kinetic energy is gone and whatever interaction is present decides what happens, which is the situation Localizing Transitions via Interaction-Induced Flat Bands is about — here with photons rather than electrons.