Low- and high-energy localization landscapes for tight-binding Hamiltonians in two-dimensional lattices
Listen
Radio episode about this paper
Transcript
Introduction to the show: ident: Condensed Matter. The day in condensed matter: materials, phases and the measurements under them.
Kai: I'm Kai.
Mira: And I'm Mira, here with Lev.
Kai: We're looking at the paper: "Low- and high-energy localization landscapes for tight-binding Hamiltonians in two-dimensional lattices".
Title and authors: Kai: So we're looking at this paper, "Low- and high-energy localization landscapes for tight-binding Hamiltonians in two-dimensional lattices," which is about how disorder affects electrons in things like graphene. It tackles a problem because the existing localization landscape theory is mostly for continuous systems, but this work shows how to apply it to discrete settings and almost all known two dimensional lattices <ref:2210.00806#pg1>.
Mira: Right, so the core idea here is extending that continuous theory into the world of tight binding models, which are basically how we model these discrete crystal structures <ref:2210.00806#pg1>. The authors want to show a systematic way to design this localization landscape even for higher dimensions, which is pretty ambitious <ref:2210.00806#pg1>.
Lev: From an engineering standpoint, if you're building hardware, you need to know how these localized states behave under real disorder conditions <ref:2210.00806#pg3>. The paper shows they can accurately predict not just where the waves are trapped, but also the actual energies of those trapped states in both low and high energy regimes <ref:2210.00806#pg1>.
Kai: It's interesting that they use two specific landscapes to do this: a low energy localization landscape, or LL, and a dual localization landscape, or DLL <ref:2210.00806#pg2>. The LL is used for the lower energy states where the effective potential is defined by the reciprocal of some quantity called u <ref:2210.00806#pg3>.
Mira: And then they introduce the dual localization landscape, which gives you information about highly oscillating, high energy wave functions <ref:2210.00806#pg3>. They prove that the reciprocals of both these landscapes act as effective confining potentials in those respective energy regimes <ref:2210.00806#pg3>.
Lev: That dual landscape construction is what I find particularly useful for the high energy part, especially when chiral symmetry isn't present, which we see in some lattices <ref:2210.00806#pg3>. The paper proposes a way to build that dual potential by demodulating the Schrödinger equation around a local maximum of the Laplacian's dispersion relation <ref:2210.00806#pg3>.
Kai: So, what they demonstrate is that for the honeycomb and hexagonal lattices, this approach predicts not just the location but also the energies of localized eigenfu" <ref:2210.00806#pg1> accurately in both low and high energy regimes. The numerical results show a direct proportionality between the lowest eigenvalues and these effective potential minima, with Pearson coefficients close to zero point nine nine in several tests <ref:2210.00806#pg2>.
Mira: That correlation is quite strong, suggesting that the discrete low-energy localization landscape works as well as its continuous counterpart for predicting energy and spatial distribution of localized modes without needing to solve the full eigenvalue equation <ref:2210.00806#pg2>.
Title and authors: Lev: And they extend this robustness across different lattices too; they tested eight types, including three one dimensional and five two dimensional lattices, all showing Pearson coefficients larger than zero point nine six, and even larger than zero point nine eight in most cases <ref:2210.00806#pg3>. They also note that the slopes fall on the one plus d over four limit for weak disorder when the embedding dimension is d <ref:2210.00806#pg3>.
Kai: It's good to hear that it holds up across different lattice types and disorder strengths, because in real materials, you can't just assume everything works the same way <ref:2210.00806#pg3>. But what about constructing this dual potential when chiral symmetry isn't present?
Mira: That’s a key point for the dual landscape construction, right? The paper shows they have proposed a general procedure to build that dual confining potential by "demodulating" the Schrödinger equation around a local maximum of the dispersion relation of the Laplacian <ref:2210.00806#pg3>. This means it relies on the specific band structure of whatever lattice you're looking at <ref:2210.00806#pg3>.
Lev: From an implementation view, building that dual potential when chiral symmetry isn't present is the harder part, I think <ref:2210.00806#pg3>. They show this by defining an envelope phi of an eigenmode psi whose wave vector is close to k max by writing psi n equal e i k max dot r n phi n <ref:2210.00806#pg3>.
Kai: And when they inject that envelope into the original equation, they get a "demodulated" equation that reads something like a1 a2 gamma M k b1 b2 gamma M k gamma <ref:2210.00806#pg3>. The comparison between the deepest wells of one over u prime and the locations of those high-energy states shows that there's a direct match <ref:2210.00806#pg3>.
Mira: So, to summarize this part, they show that for the hexagonal lattice specifically, the high energy theory is directly deduced from the low energy one due to chiral symmetry <ref:2210.00806#pg3>. That's a nice piece of theoretical grounding <ref:2210.00806#pg3>.
Lev: And they also noted that for lattices with symmetric band structure, like the one dimensional dimer chain or the two dimensional Lieb lattice, you can deploy this spectrum inversion strategy <ref:2210.00806#pg3>. That’s a nice generalization beyond just the honeycomb and hexagonal cases <ref:2210.00806#pg3>.
Kai: So, to wrap up on the results, the discrete low energy localization landscape performs as well as its continuous counterpart in predicting localization regions and corresponding energies, which is good for understanding density of states <ref:2210.00806#pg2>. And they also said the high energy dual landscape provides a good estimate of the integrated density of states for both ranges of energy <ref:2210.00806#pg7>.
Title and authors: Mira: But they are careful to state that the construction of that dual confining potential that acts on the upper part of the spectrum is still not completely universal, because they haven't yet extended it to situations like what happens with the Kagome lattice <ref:2210.00806#pg4>. They also mentioned that for the Kagome lattice, defining k max is a challenge because its density of states isn't symmetric and high energy states lie on a flat band <ref:2210.00806#pg4>.
Lev: That limitation with the Kagome lattice is something we need to watch for when we think about applying this theory to more complex, non-honeycomb structures <ref:2210.00806#pg4>. It sets a clear boundary for where this specific methodology currently works <ref:2210.00806#pg3>.
Kai: So, if you boil it down for someone who just wants the big picture, they can see that this paper provides a systematic way to design localization landscapes for almost any known 2D lattice <ref:2210.00806#pg1>. It's a powerful tool because it accurately predicts where things are localized and their energies in both low and high energy regimes <ref:2210.00806#pg1>.
Mira: The implication for the world is that we have a more accessible form of information from a Hamiltonian, allowing us to study disorder effects in two dimensional materials with much greater detail than before <ref:2210.00806#pg1>. It moves the localization landscape theory out of the purely continuous realm and into the discrete realm for tight binding models <ref:2210.00806#pg1>.
Lev: And for hardware, it means if you're simulating disorder in superconducting circuits or graphene, you can use this framework to predict state locations and energies without having to solve the full many body problem every time <ref:2210.00806#pg3>. The overhead is manageable because the theory gives you a way to map the problem onto these effective potentials <ref:2210.00806#pg3>.
Kai: It really shows how useful this approach can be for investigating the role of disorder in these materials <ref:2210.00806#pg1>. It’s a promising tool for understanding transport and magnetic properties in graphene and other 2D materials <ref:2210.00806#pg1>.
Mira: Exactly, it's about taking a theory that was strong in the continuous setting and proving it works in the discrete world for almost all known lattices <ref:2210.00806#pg1>. It’s showing that these localization concepts aren't just limited to simple geometries <ref:2210.00806#pg3>.
Lev: So, we've covered the main results of "Low- and high-energy localization landscapes for tight-binding Hamiltonians in two-dimensional lattices." It’s a lot of heavy lifting to get this discrete framework working across different lattice types <ref:2210.00806#pg3>.
Kai: It’s definitely a solid piece of work, showing how powerful these localization concepts can be when applied systematically to tight binding models <ref:2210.00806#pg1>. We'll keep an eye on how it develops from here.
The paper's summary: Kai: So, to wrap up the main points from that paper, it shows that we can take the localization landscape theory, which is usually for continuous systems, and make it work for discrete tight binding models on almost any two dimensional lattice <ref:2210.00806#pg1>.
Mira: Exactly. They use two things to map this out: a low energy landscape and a dual localization landscape <ref:2210.00806#pg3>. The first one tells you where the waves are trapped at low energies, and the second one tells you about the highly oscillating high energy states <ref:2210.00806#pg3>.
Lev: The big deal there is that they can predict not just where those states are, but also their exact energies in both the low and high energy regimes without having to solve the full complicated eigenvalue equation <ref:2210.00806#pg2>. That kind of accuracy is hard to get when you're building real hardware for error correction <ref:2210.00806#pg3>.
Kai: And they did this for the honeycomb and hexagonal lattices, and the results were really consistent, with those Pearson coefficients hitting nearly zero point nine nine in several tests <ref:2210.00806#pg2>. That means the discrete model is performing just as well as its continuous counterpart for predicting energy density of states <ref:2210.00806#pg7>.
Mira: It's also pretty robust; they tested eight different types of lattices, including three one dimensional and five two dimensional ones, and the correlation stayed above zero point nine six in most cases <ref:2210.00806#pg3>. That tells us this isn't just a fluke for the honeycomb structure <ref:2210.00806#pg1>.
Lev: But they also pointed out that building that dual landscape for the high energy part when chiral symmetry is missing, which happens in some lattices, requires a specific trick called demodulation around a local maximum of the Laplacian's dispersion relation <ref:2210.00806#pg3>. That sounds like it adds complexity to implementation on real chips <ref:2210.00806#pg3>.
Kai: So, the core message is that this framework gives us a systematic way to design these localization landscapes even for higher dimensions <ref:2210.00806#pg1>. It’s a promising tool for understanding how disorder affects transport and magnetic properties in two dimensional materials <ref:2210.00806#pg1>.
Mira: And they also showed that the low energy landscape is valid for any lattice, no matter the dimension, and it doesn't even need nearest neighbor coupling <ref:2210.00806#pg3>. That broad applicability is a strong point for the theory <ref:2210.00806#pg1>.
Lev: The paper also flagged that the dual landscape construction isn't completely universal yet, specifically noting challenges with lattices like the Kagome lattice where the density of states isn't symmetric and high energy states sit on a flat band <ref:2210.00806#pg4>. That’s a clear limitation for applying this method to more complex geometries right now <ref:2210.00806#pg3>.
The paper's improvements: Tom: So, we've talked about how this paper extends localization theory to discrete systems, focusing on those two energy landscapes <ref:2210.00806#pg1>. Now let's look at what they suggest improving this approach looks like <ref:2210.00806#pg3>.
Kai: What they suggest is that the dual localization landscape construction, that part for the high energy states, needs to be more general <ref:2210.00806#pg3>. They propose a method to build it by "demodulating" the Schrödinger equation around a local maximum of the Laplacian's dispersion relation <ref:2210.00806#pg3>.
Mira: That makes sense, because right now it’s specific to certain lattice types, like the honeycomb and hexagonal ones <ref:2210.00806#pg3>. The authors are trying to make that procedure work for any lattice structure <ref:2210.00806#pg3>.
Lev: From a hardware side, making something general is good, but the implementation overhead matters <ref:2210.00806#pg3>. If we want to run this on a real quantum system, that demodulation process adds computation time <ref:2210.00806#pg3>.
Kai: I mean, if we can automate the design of these potentials based on the lattice structure itself rather than doing it by hand, that would be huge for experimental work <ref:2210.00806#pg1>. It moves us away from just modeling a specific geometry to modeling a class of geometries <ref:2210.00806#pg3>.
Mira: And they are also hinting that for certain lattices, you can use a spectrum inversion strategy, which is what they call it <ref:2210.00806#pg3>. This works well for symmetric band structures like the one dimensional dimer chain or the two dimensional Lieb lattice <ref:2210.00806#pg3>.
Lev: That spectrum inversion sounds like it could simplify things on hardware because it relies more on existing band structure knowledge rather than complex, time-consuming calculations <ref:2210.00806#pg3>. It reduces the computational burden for finding those high energy states <ref:2210.00806#pg3>.
Kai: So, they're pushing toward a more systematic design framework that works across different lattice types and even higher dimensions <ref:2210.00806#pg1>. It shows the localization concepts aren't just limited to simple geometries <ref:2210.00806#pg3>.
Mira: Exactly, it’s showing that these localization concepts are more generalizable than we thought when applied to tight binding models <ref:2210.00806#pg1>. But the paper also flagged that the Kagome lattice is still a hurdle because its density of states isn't symmetric, which complicates things there <ref:2210.00806#pg4>.
Lev: That Kagome issue means we can't just apply this improved method to every structure immediately <ref:2210.00806#pg3>. It sets a clear boundary for where this specific theoretical tool is ready for deployment on real hardware <ref:2210.00806#pg3>.
Conclusion: Tom: So we're wrapping up this talk on "Low- and high-energy localization landscapes for tight-binding Hamiltonians in two-dimensional lattices" <ref:2210.00806#pg1>. It really boils down to showing that we have a way to use continuous theory tools for discrete materials <ref:2210.00806#pg1>.
Kai: Yeah, the main thing is that this framework lets us predict both the location and the energy of localized states in both low and high energy regimes without solving the whole eigenvalue problem <ref:2210.00806#pg2>. That's a big win for experimentalists working with disorder <ref:2210.00806#pg3>.
Mira: I agree, it’s the systematic extension of that continuous theory into the discrete tight binding world that’s important <ref:2210.00806#pg1>. It validates using these localization landscapes for material properties like transport and magnetism <ref:2210.00806#pg1>.
Lev: From an error correction standpoint, the fact that they give us a good estimate of the integrated density of states across both energy ranges is useful <ref:2210.00806#pg7>. But we still have that challenge with building the dual potential for chiral symmetry and non-honeycomb lattices <ref:2210.00806#pg3>.
Kai: Exactly, the caveat there is that if you're working on a system without that symmetry, constructing the dual potential needs to be more involved <ref:2210.00806#pg3>. It’s not as straightforward as it is for the hexagonal lattice <ref:2210.00806#pg3>.
Mira: So, while the low energy landscape is pretty robust across all lattices and dimensions, that high energy dual landscape construction is still where the theoretical work needs to continue <ref:2210.00806#pg4>.
Lev: For real hardware, if we want to use this for error correction on a specific material like a Kagome lattice, we need to figure out how much overhead that demodulation process adds <ref:2210.00806#pg3>. It’s about where the computation really lands <ref:2210.00806#pg3>.
Kai: So, in short, this paper shows the power of applying localization landscape theory to discrete tight binding models for almost any two dimensional lattice <ref:2210.00806#pg1>. It’s a solid tool for understanding disorder effects <ref:2210.00806#pg1>.
Mira: It moves the localization landscape theory out of just the continuous realm and into discrete systems for tight binding models <ref:2210.00806#pg1>. It shows it works without being restricted to nearest neighbor coupling <ref:2210.00806#pg3>.
Lev: That broad applicability is good, but we still have the Kagome lattice case as a known limitation for this specific high energy technique <ref:2210.00806#pg4>.
Kai: So, to wrap up on "Low- and high-energy localization landscapes for tight-binding Hamiltonians in two-dimensional lattices," it’s a powerful systematic way to design these potentials <ref:2210.00806#pg1>. It accurately predicts location and energy in both regimes <ref:2210.00806#pg1>.
Mira: And the implication is that we have a more accessible form of information from a Hamiltonian to study disorder effects in 2D materials <ref:2210.00806#pg1>. It’s moving this theory into the discrete realm for tight binding models <ref:2210.00806#pg3>.
Lev: For practical applications, it means we can use this framework to predict state locations and energies without having to solve the full many body problem every time <ref:2210.00806#pg3>. The overhead is manageable if you use the effective potentials <ref:2210.00806#pg3>.
Kai: So, this paper gives us a way to investigate transport and magnetic properties in graphene and other 2D materials <ref:2210.00806#pg1>. It’s definitely a promising tool for understanding these materials <ref:2210.00806#pg1>.
Luis A. Razo-López, Geoffroy J. Aubry, Marcel Filoche, Fabrice Mortessagne
Université Côte d’Azur · Laboratoire de Physique de la Matière Condenséee, École Polytechnique, Institut Polytechnique de Paris, Institut Langevin, ESPCI, PSL University
cond-mat.dis-nn, cond-mat.mes-hall
Submitted: 2022-10-03
Updated: 2026-11-18
Comments: 3 figures in the main text ; 8 figures in the supplemental material
Journal ref: Phys. Rev. Research 5, 023102 (2023)
DOI: 10.1103/PhysRevResearch.5.023102
License: http://creativecommons.org/licenses/by/4.0/
Importance score: 86/100
The gist: Localization of electronic wave functions in modern two-dimensional (2D) materials such as graphene can impact drastically their transport and magnetic properties The recent localization landscape…
Key concepts
- Tight-Binding Model
- This is a mathematical model used to describe how electrons behave in a lattice structure, like atoms arranged in a pattern. It uses a Hamiltonian that accounts for the energy of an electron at each site and its interaction (coupling) with neighboring sites. Disorder is introduced by making the on-site energies random.
- Low-Energy Localization Landscape (LL)
- This landscape helps find where states are localized when their energy is low. It uses a quantity derived from the wavefunction, 1/u, which acts like an effective potential well that traps the localized states at lower energy levels in the system.
- Dual Localization Landscape (DLL)
- This landscape predicts the position of highly oscillating, high-energy wave functions. For systems lacking chiral symmetry, it is constructed by 'demodulating' the Schrödinger equation around a local maximum of the Laplacian's dispersion relation to define an effective confining potential for these upper energy states.
Terminology
Summary
Localization of electronic wave functions in modern two-dimensional (2D) materials such as graphene can impact drastically their transport and magnetic properties The recent localization landscape (LL) theory has brought many tools and theoretical results to understand such localization phenomena in the continuous setting, but with very few extensions so far to the discrete realm or to tight-binding Hamiltonians In this paper, we show how this approach can be extended to almost all known 2D lattices, and propose a systematic way of designing LL even for higher dimension We demonstrate in detail how this LL theory works and predicts accurately not only the location, but also the energies of localized eigenfunctions in the low and high energy regimes for the honeycomb and hexagonal lattices, making it a highly promising tool for investigating the role of disorder in these materials
Tight-Binding Model Definition
Tight-binding models are commonly used to study perfect as well as disordered lattices The tightbinding Hamiltonian Hˆ with on-site disorder and nearestneighbor coupling is defined as (Hˆψ)n = −t X m∈hni (ψm − ψn) + (Vn − bn t) ψn (1). In the following, one will assume that t has value 1, thus setting the energy unit, and that Vn = W νn where νn is an i. p. 0 The tightbinding Hamiltonian Hˆ with on-site disorder and nearestneighbor coupling is defined as
(Hˆψ)n = −t
X
m∈hni
(ψm − ψn) + (Vn − bn t) ψn (1)
where ψ ≡ (ψn)n∈[[1,N]] is the wave function defined on
the sites of the lattice (numbered from 1 to N here), Vn is
the on-site potential at site n, −t is the coupling constant
between neighboring sites (assumed to be constant here),
hni indicates the ensemble of nearest neighbors of site n,
and bn is its cardinal. In the following, one will assume
that t has value 1, thus setting the energy unit, and that
Vn = W νn where νn is an i. In the following, one will assume
that t has value 1, thus setting the energy unit, and that
Vn = W νn where νn is an i. random variable with
uniform law in [−0. In the following, one will assume
that t has value 1, thus setting the energy unit, and that
Vn = W νn where νn is an i. random variable with
uniform law in [−0. random variable with
uniform law in [−0. 5], W being therefore the disorder
strength for a given lattice.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Low-Energy Localization Landscape
The low energy LL is obtained by solving the analog of Eq. (2) in the discrete setting, i. p. 0 Aubry,1 Marcel Filoche,2, 3 and Fabrice Mortessagne1, ∗
1Université Côte d’Azur, CNRS, Institut de Physique de Nice (INPHYNI), France
2Laboratoire de Physique de la Matière Condenséee, CNRS,
École Polytechnique, Institut Polytechnique de Paris, 91120 Palaiseau, France
3
Institut Langevin, ESPCI, PSL University, CNRS, Paris, France
(Dated: October 4, 2022)
Localization of electronic wave functions in modern two-dimensional (2D) materials such as
graphene can impact drastically their transport and magnetic properties. The recent localization
landscape (LL) theory has brought many tools and theoretical results to understand such localization phenomena in the continuous setting, but with very few extensions so far to the discrete
realm or to tight-binding Hamiltonians. (2) in the discrete setting, i. , Hˆu = 1
(u ≡ (un)n∈[[1,N]], 1 is a vector filled with 1) with the
same boundary conditions than the eigenvalue problem. (1); (h) Eigenmodes with the four lowest eigenvalues of the inverted Hamiltonian; (i) Inverse of the dual landscape. quantity 1/u—which has the dimension of an energy—
acts as an effective potential confining in its wells the
localized states at low energy [23]. quantity 1/u—which has the dimension of an energy—
acts as an effective potential confining in its wells the
localized states at low energy [23]. Moreover, the energy
of the local fundamental state inside each well was found
to be almost proportional to the value of the potential
1/u at its minimum inside the well [24],
E ≈
1 +
d
4
min(1/u), (3)
where d is the embedding dimension of the system.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
High-Energy Dual Localization Landscape
Another LL, called the dual localization landscape (DLL), gives the position of the envelope of the highly oscillating, high-energy, wave functions. Wang and Zhang have proved mathematically that the reciprocals of these discrete LL and DLL act indeed as effective confining potentials in a tight-binding system at both low- and high-energy regimes, respectively. p. 1 Another LL, called the dual localization landscape (DLL),
gives the position of the envelope of the highly oscillating, high-energy, wave functions. More recently, Wang
and Zhang [26] have proved mathematically that the reciprocals of these discrete LL and DLL act indeed as
effective confining potentials in a tight-binding system at
both low- and high-energy regimes, respectively. More challenging is the construction of the dual confining potential that acts on the upper part of the spectrum. p. 4 It bears the same efficiency than its continuous
counterpart in predicting the localization regions and the
corresponding energies, hence the density of states [28]. More challenging is the construction of the dual confining potential that acts on the upper part of the spectrum. It bears the same efficiency than its continuous
counterpart in predicting the localization regions and the
corresponding energies, hence the density of states [28]. More challenging is the construction of the dual confining potential that acts on the upper part of the spectrum.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Predictive Accuracy and Generalization
We plot on Fig. 2(a) (resp. p. 2 We
plot on Fig. 2(a) (resp. 2(a) (resp. 2(b)) the lowest (resp. 4
DOS
(e) (f)
W = 0
W = 3
W = −3
-0. (a) The honeycomb lattice; (b) Plot of the random potential Vn/W = νn; (c) Eigenmodes with the four lowest
eigenvalues of a honeycomb lattices with on-site disorder, N = 964 sites and W = 3; (d) Inverse of the localization landscape
calculated for the system as in (c) where the four lowest minima are numbered; (e) Band structure of the honeycomb lattice;
(f) density of state of the honeycomb lattice without and with disorder; (g) Eigenmodes with the four highest eigenvalues of
Eq. highest)
eigenvalues of Eq. (1) versus the local minimum values
of the effective potential 1/u (resp. Each
scatter plots corresponds to 100 realizations of the disordered potential. In both cases, a direct proportionality
is clearly observed for the lowest part of the plots, with
Pearson coefficients of the linear regression close to 0. Note also that in
order to obtain this proportionality (which is more than
a simple linear dependency), one has to choose Vshift so
that the shifted potential has a minimum value close to
zero (see Supplemental Material A). These observations indicate that the discrete lowenergy localization landscape performs as well as its continuous analog in predicting energy and spatial distribution of localized modes without resolving an eigenvalue equation.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Lattice Dependence and Robustness
The proportionality between the minima of 1/u' in the basins and the actual energies still holds. We have performed extensive simulations on 8 types of lattices (3 1D and 5 2D) for various disorder strengths, all exhibiting Pearson coefficients larger than 0. p. 4 5
basins and the actual energies still holds (see Fig S7 in
Supplemental Material). We have performed extensive
simulations on 8 types of lattices (3 1D and 5 2D) for
various disorder strengths, all exhibiting Pearson coefficients larger than 0. We have performed extensive
simulations on 8 types of lattices (3 1D and 5 2D) for
various disorder strengths, all exhibiting Pearson coefficients larger than 0. 96 and even larger than 0. Each
scatter plots corresponds to 100 realizations of the disordered potential. In both cases, a direct proportionality
is clearly observed for the lowest part of the plots, with
Pearson coefficients of the linear regression close to 0. This is indeed observed:
the slopes fall on the (1 + d/4) limit for hIPRi → 0. In
the other limit case, when the disorder is so strong that
an eigenstate is localized on a single site (hIPRi → 1),
the LL is essentially supported locally on the same site.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Dual Landscape Construction
To construct the dual confining potential that acts on the upper part of the spectrum when chiral symmetry is not present, we have proposed a general procedure to build it by “demodulating” the Schrödinger equation around a local maximum of the dispersion relation of the Laplacian. This procedure can be applied to any lattice, leading to a DLL that relies on the specific band structure of the given lattice. p. 3 We will
show in the last part of the letter that when this property cannot be used, there remains a general procedure
which consists in “demodulating” the Schrödinger equation around a local maximum of the dispersion relation
of the Laplacian. This procedure that can be applied to
any lattice, leading to a DLL that relies on the specific
band structure of the given lattice.Where the paper says it
High-Energy State Analysis
For the hexagonal lattice, we define an envelope φ of an eigenmode ψ whose wave vector is close to kmax by ψn = e i kmax· rn φn, where rn is the position of site n. By injecting φ in Eq. p. 3 To
remove the rapidly oscillating part of the high-energy
eigenstates, we define the envelope φ of an eigenmode ψ
whose wave vector is close to ~kmax by ψn = e
i~kmax·~rn φn,
where ~rn is the position of site n (see Supplemental Material E for the derivation of the landscape around a local
maximum of the dispersion relation). By injecting φ in
Eq. By injecting φ in
Eq. (1), we obtain a “demodulated” equation that reads
~a1
~a2
Γ
M
K
~b1
~b2
Γ M K Γ
Path in k-space
−6
−4
−2
0
2
Energy
(a)
(b)
(c)
x
y
ψ
N
ψ
N−1
ψ
N−2
ψ
N−3
x
1
2
3
4
(d) (e) 1/|u
?Where the paper says it
Where the paper says it
Summary of Findings
The discrete lowenergy localization landscape performs as well as its continuous counterpart in predicting localization regions and corresponding energies, hence the density of states. More challenging is the construction of the dual confining potential that acts on the upper part of the spectrum. The low-energy LL is valid for any lattice, in any dimension, and are not restricted to nearest-neighbor coupling. p. 3 (2) at the only site n supporting the
wave function therefore becomes Hun = 1 ≈ Eun, hence
E ≈ 1/un and a slope s ' 1 is expected. The definition and the properties of the low-energy LL
are valid for any lattice, in any dimension, and are not
restricted to nearest-neighbor coupling. (2) at the only site n supporting the
wave function therefore becomes Hun = 1 ≈ Eun, hence
E ≈ 1/un and a slope s ' 1 is expected. The definition and the properties of the low-energy LL
are valid for any lattice, in any dimension, and are not
restricted to nearest-neighbor coupling. These observations indicate that the discrete lowenergy localization landscape performs as well as its continuous analog in predicting energy and spatial distribution of localized modes without resolving an eigenvalue equation. Moreover, the high-energy DLL also exhibits the same properties. These observations indicate that the discrete lowenergy localization landscape performs as well as its continuous analog in predicting energy and spatial distribution of localized modes without resolving an eigenvalue equation. Moreover, the high-energy DLL also exhibits the same properties.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Data Summary
The simulations show that for both range of energy, LL and DLL provide a good estimate of the integrated density of states. The Pearson correlation coefficient is very close to 1 in most cases. p. 2 Each
scatter plots corresponds to 100 realizations of the disordered potential. In both cases, a direct proportionality
is clearly observed for the lowest part of the plots, with
Pearson coefficients of the linear regression close to 0. Each
scatter plots corresponds to 100 realizations of the disordered potential. In both cases, a direct proportionality
is clearly observed for the lowest part of the plots, with
Pearson coefficients of the linear regression close to 0. 98
ρ
(b)
W = 3
W = 6
W = 15
W = 30
W = 60
W = 120
W = 240
FIG. Pearson correlation coefficient as a function of
number of minima taken into account for the honeycomb lattice with different strengths of disorder W; (a) The landscape
prediction; (b) The dual landscape prediction using the symmetry of the bandstructure; The linear regression quality do
not evolve when the number of minima considered is larger
than the actual number of minima of 1/u (1/u
? 98
ρ
(b)
W = 3
W = 6
W = 15
W = 30
W = 60
W = 120
W = 240
FIG. Pearson correlation coefficient as a function of
number of minima taken into account for the honeycomb lattice with different strengths of disorder W; (a) The landscape
prediction; (b) The dual landscape prediction using the symmetry of the bandstructure; The linear regression quality do
not evolve when the number of minima considered is larger
than the actual number of minima of 1/u (1/u
? 98
ρ
(b)
W = 3
W = 6
W = 15
W = 30
W = 60
W = 120
W = 240
FIG. Pearson correlation coefficient as a function of
number of minima taken into account for the honeycomb lattice with different strengths of disorder W; (a) The landscape
prediction; (b) The dual landscape prediction using the symmetry of the bandstructure; The linear regression quality do
not evolve when the number of minima considered is larger
than the actual number of minima of 1/u (1/u
?Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Lattice Specifics
The number of edges of the graph on which this operator is relying is given by the number bn of nearest-neighbor couplings, which itself depends on the elementary motif of each given lattice. The IPR is defined for a given eigenvector ψ(j) = P n ψ (j) n ni as IPRj = P n / P n 4/P n 2. p. 2 The number of edges of the graph on which this operator
is relying is given by the number bn of nearest-neighbor
couplings, which itself depends on the elementary motif of each given lattice. Therefore, we use a less contingent quantity, namely the inverse participation ratio
(IPR) defined for a given eigenvector ψ(j) =
P
n ψ
(j)
n |ni
as IPRj =
P
n
ψ
(j)
n
4
/
P
n
ψ
(j)
n
2
2
. (c)
Dispersion relation (d) Four states with the highest eigenvalues of a hexagonal lattice with on-site disorder, N = 1068 and
W = 3; (e) Effective potential where the 4 deepest minima
are marked. for ~kmax =
4π
3a
xˆ:
e
i
4π
3 φn+~a1 + e
−i
4π
3 φn−~a1 + e
i
2π
3 φn+~a2
+e
−i
2π
3 φn−~a2 + e
i
2π
3 φn+~a1−~a2 + e
−i
2π
3 φn−~a1+~a2
− W νnφn = −Eφn,
(5)
where the band structure has been inverted by changing
the signs of both the couplings and the on-site energies,
and where the notation n + ~a denotes the site reached
from site n by a translation of a vector ~a [see Fig.Where the paper says it
Where the paper says it
Future Directions
The high-energy DLL construction has not yet been completely universal, as it has yet to be extended to situations like the one encountered with the Kagome lattice. Future works should address this situation. p. 4 In this case, the
DOS is not symmetric, and the high energy states lie on
a flat band: the definition of ~kmax remains a challenge. Future works should address this situation.Where the paper says it
Appendix Details
The Hamiltonian is written in a concise way as (Hˆψ)n = −t X m∈hni ψm + (Vn + Vshift)ψn, where Vn is the on-site potential at site n, and t is the coupling constant between neighboring sites. The energy shift Vshift = - min(E) + W/2 where E is the energy of the system without disorder at zero on-site energy. p. 6 5], W being therefore
the disorder strength for a given lattice, and Vshift the
energy shift that avoids negative eigenvalues. Finally
Vshift = − min(E) + W/2 where E is the energy of the
system without disorder at zero on-site energy. 5], W being therefore
the disorder strength for a given lattice, and Vshift the
energy shift that avoids negative eigenvalues. Finally
Vshift = − min(E) + W/2 where E is the energy of the
system without disorder at zero on-site energy.Where the paper says it
Where the paper says it
Table Summary
Table S3 summarizes the simulation details, listing different lattices studied with their properties. p. 7 Appendix C: Exhaustive study of various 1D and
2D lattices and simulations details
In Table. S2, we list the different lattices studied in this
study with their properties. 00
(c) W = 120
−100 0 100
(d) W = 240
E
10
20
30
5
10
20
40
10
20
FIG. 2D histogram showing how the eigenvalues and the
IPR are distributed. 0
0
20
40
60
(a)
Honeycomb
Eigenenergies
Effective potential
10 15 20
0
10
20
30
(b)
Hexagonal
236 238 240 242
2080
2100
2120
(c)
470 475 480 485
1040
1050
1060
(d)
FIG. The counting function computed by [(a) and (b)]
the confining potential, and by [(c) and (d)] the dual confining
potential. 0
0
20
40
60
(a)
Honeycomb
Eigenenergies
Effective potential
10 15 20
0
10
20
30
(b)
Hexagonal
236 238 240 242
2080
2100
2120
(c)
470 475 480 485
1040
1050
1060
(d)
FIG. The counting function computed by [(a) and (b)]
the confining potential, and by [(c) and (d)] the dual confining
potential. 0
0
20
40
60
(a)
Honeycomb
Eigenenergies
Effective potential
10 15 20
0
10
20
30
(b)
Hexagonal
236 238 240 242
2080
2100
2120
(c)
470 475 480 485
1040
1050
1060
(d)
FIG. The counting function computed by [(a) and (b)]
the confining potential, and by [(c) and (d)] the dual confining
potential. 0
0
20
40
60
(a)
Honeycomb
Eigenenergies
Effective potential
10 15 20
0
10
20
30
(b)
Hexagonal
236 238 240 242
2080
2100
2120
(c)
470 475 480 485
1040
1050
1060
(d)
FIG. The counting function computed by [(a) and (b)]
the confining potential, and by [(c) and (d)] the dual confining
potential. 0
0
20
40
60
(a)
Honeycomb
Eigenenergies
Effective potential
10 15 20
0
10
20
30
(b)
Hexagonal
236 238 240 242
2080
2100
2120
(c)
470 475 480 485
1040
1050
1060
(d)
FIG. The counting function computed by [(a) and (b)]
the confining potential, and by [(c) and (d)] the dual confining
potential.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Final Conclusion
Born a decade ago, the localization landscape theory has proven its remarkable efficiency to bring in a more accessible form the information contains in a Hamiltonian. In this letter, we have extended its scope to discrete systems described by a tight-binding Hamiltonian, with a focus on 2D lattices. p. 4 Born a decade ago, the localization landscape theory
has proven its remarkable efficiency to bring in a more
accessible form the information contains in a Hamiltonian [27]. In this letter, we have extended its scope to
discrete systems described by a tight-binding Hamiltonian, with a focus on 2D lattices. In this letter, we have extended its scope to
discrete systems described by a tight-binding Hamiltonian, with a focus on 2D lattices. The low-energy part
of the spectrum is described by a discrete extension of
the effective confining potential defined for continuous
systems.Where the paper says it
Where the paper says it
References
[1] D. Rhodes, S. H. Chae, R. Ribeiro-Palau, and J. p. 4 mortessagne@univ-cotedazur. Ribeiro-Palau, and J. Ribeiro-Palau, and J. Hone,
Disorder in van der Waals heterostructures of 2d materials, Nat. Hone,
Disorder in van der Waals heterostructures of 2d materials, Nat. 18, 541 (2019). 18, 541 (2019). Radisavljevic and A. Radisavljevic and A. Kis, Mobility engineering and
a metal–insulator transition in monolayer MoS2, Nat. Kis, Mobility engineering and
a metal–insulator transition in monolayer MoS2, Nat. 12, 815 (2013). 12, 815 (2013). Anderson, Absence of diffusion in certain random
lattices, Phys. Anderson, Absence of diffusion in certain random
lattices, Phys. 109, 1492 (1958). Aspect, and P. Bouyer, Three-dimensional localization of ultracold atoms in an optical disordered potential,
Nat. Bouyer, Three-dimensional localization of ultracold atoms in an optical disordered potential,
Nat. 8, 398 (2012). Skipetrov, and
B. van Tiggelen, Localization of ultrasound in a threedimensional elastic network, Nat. Froufe-Pérez, and F. Scheffold, Transition from light diffusion to localization in threedimensional amorphous dielectric networks near the band
edge, Nat. Scheffold, and F. Mortessagne, Experimental tuning
of transport regimes in hyperuniform disordered photonic
materials, Phys. Skipetrov, T. Cao, Anderson localization of electromagnetic waves in three dimensions (2022). Magkiriadou, and L. Froufe-Pérez, Transport through amorphous photonic
materials with localization and bandgap-regimes, Phys. Froufe-Pérez, Transport through amorphous photonic
materials with localization and bandgap-regimes, Phys. Rothstein, Gravitational Anderson localization,
Phys. Mayboroda, Universal mechanism for
Anderson and weak localization, Proc. USA 109, 14761 (2012). Filoche, Spectral functions
and localization-landscape theory in speckle potentials,
Phys. A 105, 023314 (2022). Mayboroda, Localization landscape theory
of disorder in semiconductors. Theory and modeling,
Phys. Weisbuch, and
M. Filoche, Universality of fold-encoded localized vibrations in enzymes, Sci. Tworzydło, and C. Beenakker, Localization
landscape for dirac fermions, Phys. Bardarson, L
2
localization landscape for highly excited states, Phys. B 101, 220201
(2020). Montambaux, and F. Mortessagne, Tight-binding couplings in microwave artificial
graphene, Phys. Gustavsson, and W. Oliver, Quantum transport and localization in 1d and 2d tight-binding
lattices, npj Quantum Inf. Zhong, and R. Römer, Disorder
effects in the two-dimensional Lieb lattice and its extensions, Phys. Römer, Disorder
effects in the two-dimensional Lieb lattice and its extensions, Phys. E: Low-Dimens. Zhong, and R. Römer, Localization,
phases, and transitions in three-dimensional extended
Lieb lattices, Phys. Mayboroda,
and M. Filoche, Effective confining potential of quantum
states in disordered media, Phys.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Table S3 Summary
Table S3 lists the different lattices studied in this study with their properties. Table S5 shows the 2D histogram showing how the eigenvalues and the IPR are distributed The counting function in Fig. p. 8 00
(c) W = 120
−100 0 100
(d) W = 240
E
10
20
30
5
10
20
40
10
20
FIG. 2D histogram showing how the eigenvalues and the
IPR are distributed.Where the paper says it
Figure S6 Summary
Figure S6 shows that in every case considered in the paper, the eigenvalues and the minima of the effective confining potential are highly correlated. The results are robust across different lattice types and disorder strengths.
Figure S7 Summary
Figure S7 shows that for both range of energy, LL and DLL provide a good estimate of the integrated density of states. The Pearson correlation coefficient is very close to 1 in most cases.
Appendix Summary
The Hamiltonian is written in a concise way as (Hˆψ)n = −t X m∈hni ψm + (Vn + Vshift)ψn, where Vn is the on-site potential at site n, and t is the coupling constant between neighboring sites. The energy shift Vshift = - min(E) + W/2 where E is the energy of the system without disorder at zero on-site energy. The results are robust across different lattice types and disorder strengths.
Table S1 Summary
Table S1 shows the smallest and largest eigenvalues of the tight binding Hamiltonian without on-site potential and with Vshift = 0. The values shown in Table S1 correspond to the particular case t2 = t1/√8. p. 6 Note that for the 1D chain with 2nd
neighbors coupling the energy is
E = −2 [t1 cos (kxa) + t2 cos (2kxa)] , (S2)
where the corresponding kmax is the solutions of
t1 sin (kxa) = −2t2 sin (2kxa). (S3)
Then, values shown in Table S1 correspond to the particular case t2 = t1/
√
8. Note that for the 1D chain with 2nd
neighbors coupling the energy is
E = −2 [t1 cos (kxa) + t2 cos (2kxa)] , (S2)
where the corresponding kmax is the solutions of
t1 sin (kxa) = −2t2 sin (2kxa). (S3)
Then, values shown in Table S1 correspond to the particular case t2 = t1/
√
8. Smallest and largest eigenvalues of the tight binding Hamiltonian without on-site potential and with Vshift = 0. The couplings are all t = 1, except for 1D dimer chain and
the 1D chain with 2nd neighbor coupling cases where they are
explicitly written. Smallest and largest eigenvalues of the tight binding Hamiltonian without on-site potential and with Vshift = 0. The couplings are all t = 1, except for 1D dimer chain and
the 1D chain with 2nd neighbor coupling cases where they are
explicitly written. Lattice − min(E) max(E)
1D Chain 2 2
1D dimer chain t1 + t2 t1 + t2
1D chain with 2nd neighbors coupling 2 (t1 + t2)
√
2t1
Square 4 4
Lieb √
8
√
8
tts 5 3
Hexagonal 6 3
Honeycomb 3 3
Kagome 4 2
TABLE S1. Smallest and largest eigenvalues of the tight binding Hamiltonian without on-site potential and with Vshift = 0.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Appendix A Summary
For the simulations presented in the paper, we considered the Hamiltonian given by Eq. (1) in the main body of the paper, but written in a more concise way: (Hˆψ)n = −t X m∈hni ψm + (Vn + Vshift)ψn, with Vn is the on-site potential at site n, and t is the coupling constant between neighboring sites. p. 6 1
Low and high-energy localization landscapes for tight-binding Hamiltonians in
2D lattices
Supplemental Material
Appendix A: Energy shift
For the simulations presented in the paper, we considered the Hamiltonian given by Eq. (1) in the main body
of the paper, but written in a more concise way:
(Hˆψ)n = −t
X
m∈{nn}
ψm + (Vn + Vshift)ψn , (S1)
where Vn is the on-site potential at site n, and t is the
coupling constant between neighboring sites. (1) in the main body
of the paper, but written in a more concise way:
(Hˆψ)n = −t
X
m∈{nn}
ψm + (Vn + Vshift)ψn , (S1)
where Vn is the on-site potential at site n, and t is the
coupling constant between neighboring sites. Additionally, Vn = W νn where νn is an i. random variable with uniform law in [−0. 5], W being therefore
the disorder strength for a given lattice, and Vshift the
energy shift that avoids negative eigenvalues. 5], W being therefore
the disorder strength for a given lattice, and Vshift the
energy shift that avoids negative eigenvalues. Finally
Vshift = − min(E) + W/2 where E is the energy of the
system without disorder at zero on-site energy. Finally
Vshift = − min(E) + W/2 where E is the energy of the
system without disorder at zero on-site energy. The considered values are shown in Table S1.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.
Improvements for AI systems
-
Improved AI systems can accurately predict not only the location but also the energies of localized eigenfunctions in both low and high energy regimes for honeycomb and hexagonal lattices by using two discrete localization landscapes: the low energy LL and the dual localization landscape (DLL).
-
The improved system can predict the location of localized modes near both band edges (low and high energy) as well as their energies without solving the Schrödinger equation, achieving Pearson coefficients close to 0.99 in several tests.
-
For systems where chiral symmetry is not present, the improved system can construct a dual confining potential by
demodulating
the Schrödinger equation around a local maximum of the dispersion relation of the Laplacian to obtain an effective DLL.
Abstract
Localization of electronic wave functions in modern two-dimensional (2D) materials such as graphene can impact drastically their transport and magnetic properties. The recent localization landscape (LL) theory has brought many tools and theoretical results to understand such localization phenomena in the continuous setting, but with very few extensions so far to the discrete realm or to tight-binding Hamiltonians. In this paper, we show how this approach can be extended to almost all known 2D lattices, and propose a systematic way of designing LL even for higher dimensions. We demonstrate in detail how this LL theory works and predicts accurately not only the locations, but also the energies of localized eigenfunctions in the low and high energy regimes for the honeycomb and hexagonal lattices, making it a highly promising tool for investigating the role of disorder in these materials.
Related papers
- Topological metal-insulator transitions in one-dimensional non-Hermitian quasicrystals: beyond PT-symmetry
- Microscopic and collective signatures of feature learning in neural networks
- Synchronous and asynchronous Monte Carlo dynamics of mean-field spin glasses with rotationally invariant couplings: what they share and where they differ
- Reentrant Localization Length in a Quasiperiodic Thue-Morse Chain
- Beyond Ground States: Learning Many-Body Hamiltonians for p-Bit Invertible Arithmetic
- Linear equivalence of nonlinear recurrent neural networks