Choosing a Model for Quantum Matter

A material does not come with a unique Hamiltonian attached to its name. The appropriate model depends on the prepared state, scale, observable, probe, accuracy target, and comparison being attempted. Crystalline silicon can be a multivalley effective-mass problem for low-field mobility, a band-plus-phonon problem for an indirect optical edge, or an atomistic impurity problem near a single dopant. These are not competing answers to one question. They are different retained descriptions for different questions.

This page owns the decision between such candidate descriptions. It does not define the models, fix their notation, or rederive their Hamiltonians. The Quantum Matter Map supplies the system–state–observable problem tuple; From Quantum Mechanics to Materials audits the provenance of a selected reduction; the Model Encyclopedia owns reproducible dossiers for many standard many-body Hamiltonians; the Reference Model Library supplies compact cross-site cards where one is published; and Conventions for Quantum Matter makes the chosen calculation reproducible.

Helpful background. What Is Quantum Matter? distinguishes a model, excitation, phase, and material claim. No page is a hard prerequisite, but those distinctions prevent model selection from becoming label matching.

The Smallest Adequate Model Is Observable-Specific

“Use the most realistic model” is not an operational rule. Restoring every orbital, interaction channel, phonon branch, defect, surface, and environment can make a calculation impossible without making its target prediction more reliable. Conversely, a model that reproduces one dispersion can fail for optical intensities, transport lifetimes, collective modes, or a phase claim.

The useful target is the smallest adequate model: the least elaborate declared description that predicts the requested observable to the requested accuracy throughout a stated window and survives relevant alternatives. This is not necessarily the model with the fewest symbols. A symmetry-adapted multiband model can be safer than a deceptively compact one-band fit because it retains the crossings and matrix elements the observable tests.

Models therefore form a branching set rather than an accuracy ladder. Effective-mass, tight-binding, Hubbard, Heisenberg, Drude, Kubo, and hydrodynamic descriptions answer different questions and are not ordered from “crude” to “exact.” A more microscopic model can improve one observable while introducing poorly constrained parameters for another.

Even “smallest” requires a declared comparison order. Nested candidates can be ordered by retained bands, interaction terms, or resolution, but a band model, a Hubbard model, an Anderson-localization model, and a spin model are generally non-nested. When no single candidate dominates, report a Pareto set of adequate alternatives and name the probe that would distinguish them instead of forcing one winner.

Write a Model-Adequacy Card Before Choosing

Record the following fields for every serious candidate. If two candidates use different inputs or predict different experimental objects, they have not yet been compared on equal terms.

This card is a side-by-side comparison projection of the provenance fields in From Quantum Mechanics to Materials, not a second reduction ledger.

1. System and preparation. State composition, structure, dimensionality, temperature, fields, density or filling, disorder, surfaces, history, and whether the state is equilibrium, driven, or open. Distinguish a finite sample from a thermodynamic target.
2. Target and tolerance. Name the observable, probe geometry, resolution, parameter window, and acceptable discrepancy. “Explain the material” is not a target.
3. Retained degrees of freedom. List bands, orbitals, spins, valleys, phonons, defects, collective fields, reservoirs, and boundaries kept explicitly.
4. Eliminated physics. List remote bands, charge sectors, high-frequency modes, disorder correlations, vertex corrections, environments, and other structures represented only through matched parameters or omitted.
5. Geometry, symmetry, and constraints. Declare lattice or continuum, boundaries, dimensionality, conservation laws, protecting symmetries, and Hilbert-space constraints. Record the finite-size sequence and the order of $L\to\infty$, $T\to0$, $\omega\to0$, and $q\to0$ whenever the claim uses those limits.
6. Parameter provenance. Mark parameters as bare, calculated, screened, downfolded, renormalized, fitted, or independently measured. State the basis and scale at which they apply.
7. State and solver. A Hamiltonian is not a prediction without a state, ensemble or protocol, and a solution method. Separate approximation in the physical model from approximation and numerical error in the solver.
8. Validation set. Reserve observables, temperatures, fields, momenta, or samples not used to fit the parameters. Include sum rules and exact limits when available.
9. Failure and escalation tests. State which discrepancy would restore a band, interaction channel, spatial dimension, nonlocal term, dynamical variable, disorder model, or environment.
10. Canonical owner. Link the model definition, teaching derivation, material application, solver, and reference card instead of reproducing them locally.

The card licenses a bounded statement: “candidate A is adequate for observable $O$ within window $W$ and tolerance $\epsilon$ under controls $C$.” It does not certify every observable of the material or the uniqueness of the model. Budget reduction, parameter, solver, and probe errors separately. Define $\epsilon_{\mathrm{total}}$ as a conservative sum of their bounds and require $\epsilon_{\mathrm{total}}\leq\epsilon_{\mathrm{target}}$. Correlated or unbounded uncertainties cannot be combined in quadrature without an explicit probabilistic model.

A finite system can itself be the target. But a finite-cluster gap, symmetric exact ground state, split edge mode, sharpening peak, or localization-looking plateau cannot by itself establish a thermodynamic transition, spontaneous symmetry breaking, topological phase, or localization. Those claims require the appropriate size sequence, gap or mobility-gap evidence, boundaries, and order of limits.

A Decision Workflow

1. Begin with the measured or computed object

Work backward from conductivity, a spectral peak, a transition temperature, a neutron cross section, a band invariant, or another declared output. Identify the operator to which the probe couples and the resolution it actually has. A transport coefficient and a band velocity are different objects; a measured intensity and a spectral function are different objects.

2. Identify the minimum active structure

Ask which degrees of freedom can change within the energy, momentum, time, and temperature window. Keep every sector required by selection rules, symmetry, conservation, or the target operator. A remote band can be irrelevant to a ground-state energy yet essential to an optical matrix element.

3. Build more than one plausible candidate

For an inverse or mechanism claim, at least one candidate should test the favored mechanism and one should represent a credible alternative. A broad continuum in spectroscopy can arise from fractionalization, disorder, multiparticle decay, or matrix-element and resolution effects. Selecting only the desired model makes the conclusion circular. A routine forward prediction may instead compare a nested convergence sequence or a validated reference model, provided it makes no uniqueness claim.

4. Match inputs without double counting

Compare candidates at the same composition, state, geometry, and observable. Do not combine screened interactions from one active space with a solver that screens the same channels again. Do not fit a relaxation time to the test data and then present agreement with those data as independent validation.

5. Validate held-out predictions

Test held-out observables and limiting cases. Agreement across a dispersion, spectral weight, thermodynamics, and response is stronger than a fit to one curve because the probes constrain different operator combinations. Solver convergence is necessary but cannot repair an inadequate model.

6. Escalate in response to a diagnosed risk

Restore the missing structure indicated by a failure, an unbounded omission, or a known scale or symmetry collision: extra bands for selection rules, nonlocal hopping for dispersion, interactions for spectral weight transfer, phonons for temperature dependence, disorder for sample variation, boundaries for edge response, or an environment for dissipation. Escalation without a diagnosed risk increases cost without defining what was learned; waiting for observed failure is unsafe when an omitted contribution is already comparable to the tolerance.

Candidate Model Families

The families below are entry points, not competing universal theories. Each summary gives the local selection test and then hands the reader to the canonical owner.

Independent-electron and band models

Retains. One-electron states in a periodic or effective potential, with a chosen band or orbital basis. Free-electron, nearly-free-electron, tight-binding, graphene, SSH, Haldane, and Kane–Mele forms are examples with different geometry and symmetry data.

Useful for. Filling, Fermi surfaces, band gaps, velocities, orbital content, Berry geometry, and selected weakly correlated response.

Cannot establish by itself. Exact charged excitations, interaction-driven spectral-weight transfer, scattering lifetimes, or a material topological assignment from a benchmark Hamiltonian alone.

Navigate after choosing. Carry the description provenance, state and filling, gap or crossing question, target observable, and validity window into the Band Theory and Electronic Structure gateway. Use Band Theory Overview for the detailed band-to-many-body dictionary and Tight-Binding Models for an orbital-model construction.

Escalate when. Self-energy, vertex, phonon, disorder, excitonic, or multiorbital effects are comparable with the target resolution.

Envelope, Dirac, and other continuum reductions

Retains. A small neighborhood of band extrema, nodes, valleys, surfaces, or interfaces, represented by effective masses, $k\cdot p$ coefficients, or Dirac-like matrices.

Useful for. Long-wavelength carriers, confinement, valley physics, low energy boundary modes, and device scales much larger than the lattice.

Cannot establish by itself. A local, unregularized Dirac or $k\cdot p$ expansion cannot establish global Brillouin-zone topology; a properly regularized continuum theory sometimes can. Atomistic intervalley processes and fundamental relativistic particle physics also lie outside the generic local expansion.

Escalate when. The occupied window reaches remote bands, atomistic boundaries matter, symmetry-allowed corrections compete with the retained terms, or symmetry-related valleys or nodes required by the claim were omitted. See Effective Mass and Graphene and Dirac Materials.

Lattice vibrations and electron–phonon models

Retains. Ionic displacements or normal modes, force constants, and any electron–phonon vertices required by the observable.

Useful for. Phonon dispersions, thermal properties, structural stability, linewidths, indirect optical processes, and conventional pairing mechanisms.

Cannot establish by itself. Superconductivity from the existence of phonons, or material transport without occupations, scattering channels, and geometry.

Escalate when. Anharmonicity, nonadiabatic dynamics, neglected branch mixing, near-degenerate modes, branches entering the probe window, strong coupling, or structural disorder invalidates the harmonic or lowest-order vertex description. Use Phonons for the lattice dynamics; use the Lattice Vibrations and Collective Modes gateway when the target instead requires a planned material anharmonic, electron–phonon, polaron, plasmon, or exciton branch and its current live fallback. Polarons Preview for one substantive electron–phonon handoff; the dedicated material electron–phonon treatment remains forthcoming.

For a magnetic material question, enter Magnetism and Spin Systems before selecting the precise impurity, exchange, ordered-phase, excitation, spin-transport, or texture owner below.

Interacting fermion, boson, impurity, and Kondo families

Retains. Selected orbitals or sites with explicit interactions, filling, hybridization, and bath structure. Hubbard, multiorbital, Anderson-impurity, Kondo, and Bose–Hubbard descriptions occupy different retained spaces. An Anderson-impurity model and the disorder model behind Anderson localization are distinct despite the shared name.

Useful for. Mott physics, local moments, spectral-weight transfer, screening, mixed valence, low-energy impurity spin dynamics, and interacting lattice-boson coherence or number localization.

Cannot establish by itself. That a named compound is one-band, that an Anderson impurity is always a spin-only Kondo problem, or that strong correlation implies fractionalization.

Escalate when. Charge-transfer states, ligand orbitals, nonlocal interactions, dynamical screening, or lattice coherence enters the target. For lattice bosons, declare the ensemble or filling, trap or inhomogeneity, boundary conditions, and finite-size sequence before inferring a phase. Use Hubbard Physics in Materials, the generic Hubbard Model, and the Bose–Hubbard Model or Kondo Effect for their distinct jobs.

Spin-only models

Retains. Local spin representations, exchange tensors, anisotropy, fields, and lattice or bond geometry after charge fluctuations have been removed. Heisenberg, Ising, XXZ, and bond-directional Kitaev-type models probe different symmetry and coupling structures.

Useful for. Magnetic order, spin waves, frustration, low-energy spin response, and controlled strong-coupling sectors.

Cannot establish by itself. Charge dynamics, optical transitions across a Mott gap, itinerant magnetism, or the microscopic provenance of every exchange constant.

Escalate when. Charge fluctuations, orbital dynamics, itinerancy, spin–phonon coupling, or longer-range interactions enter the window. See Heisenberg Model, Exchange Interactions, and Quantum Spin Liquids. The spin-liquid and fractionalization branch remains owned by Strong Correlations and Emergence.

Pairing, coherence, and order-parameter descriptions

After selecting this model family, use Superfluidity and Superconductivity to choose the material phase, response, defect, interface, or topology branch. This page retains ownership of the model-family decision.

Retains. Pairing channels, quasiparticle sectors, phase and amplitude fields, or phenomenological order parameters. The reduced BCS Hamiltonian, Bogoliubov–de Gennes representation, and Ginzburg–Landau theory are related but not interchangeable.

The reduced BCS model is a number-conserving finite pairing Hamiltonian; BCS mean field is a broken-$U(1)$ saddle. BdG is a Nambu-doubled quadratic eigenproblem whose $\pm E$ partners must not be double counted; a real spectrum alone proves neither self-consistency nor stability. Ginzburg–Landau theory is a long-wavelength phenomenology controlled near the relevant transition or in another declared gradient regime.

Useful for. Gap structure, coherent quasiparticles, vortices, interfaces, phase stiffness, and electromagnetic response in their respective regimes.

Cannot establish by itself. A microscopic pairing mechanism, bulk superconductivity from a tunneling gap, or a topological superconductor from a minimal Kitaev-chain spectrum.

Escalate when. Fluctuations, strong coupling, competing orders, number projection, disorder, or self-consistent material bands matter. Route to BCS Theory, Ginzburg–Landau Theory, or Topological Superconductors according to the claim. For the distinct formal objects, use the Reduced BCS Model dossier and Bogoliubov Quasiparticles.

Disorder, transport, and mesoscopic descriptions

Use Transport, Response, and Optics to select the bulk, kinetic, correlation, coherent-terminal, optical, noise, or nonlinear branch before comparing specialist formulations.

Retains. Random potentials or ensembles, scattering rates, semiclassical distributions, coherent channels, contacts, reservoirs, or response kernels. Anderson localization, Drude, Boltzmann, Landauer–Büttiker, and Kubo descriptions are regime-dependent formulations rather than one accuracy sequence.

Drude and Boltzmann describe incoherent quasiparticle transport under their collision closures; Landauer describes coherent finite terminals; Kubo gives linear response for a specified Hamiltonian and state; Anderson localization retains coherent multiple scattering in a random potential. These are not interchangeable names for one model.

Useful for. Diffusion, localization, mobility, finite-device conductance, and linear response when the associated coherence and collision assumptions hold.

Cannot establish by itself. A microscopic scattering mechanism from a fitted lifetime, a bulk conductivity from a contact-limited device, or localization from high resistance alone.

Escalate when. The chosen formulation omits the dominant interference, dephasing, interactions, inelastic collisions, hydrodynamic flow, disorder, nonlinearity, contacts, finite-size scale, or finite-frequency process. Declare the order of size, time, frequency, and dephasing limits. Anderson localization is distinct from the Anderson impurity model, many-body localization, and a Mott insulator. Compare Drude Theory, Boltzmann Transport, Conductance Quantization, and Anderson Localization by regime, not prestige.

Canonical Handoffs After Selection

Once the candidate family is named, use the narrow owner rather than treating this guide as its definition:

• Ideal Fermi Gas owns the free-gas reference; Nearly Free Electrons and Tight-Binding Models own crystalline weak-potential and orbital reductions.
• Graphene and Dirac Materials owns the material Dirac example; the SSH card owns the compact one-dimensional benchmark.
• Haldane claims route to Chern Numbers in Band Theory and Topological Phase Transitions. Kane–Mele claims route through Spin–Orbit Coupling in Solids to Topological Insulators.
• Hubbard, Bose–Hubbard, Heisenberg, and transverse-field Ising pages own their generic models; Kondo Effect owns the material screening route.
• The Reduced BCS Model dossier and Bogoliubov Quasiparticles distinguish the finite pairing Hamiltonian from Nambu quasiparticle machinery.
• The Kitaev-chain card owns its compact BdG benchmark. Kitaev-honeycomb applications route through Quantum Spin Liquids and Emergent Gauge Fields.

The Model-to-Volume Cross-Link Index owns the fuller dossier–teaching–reference–benchmark–application routing map.

Minimal Topological Models Are Benchmarks, Not Material Proof

SSH, Haldane, Kane–Mele, Kitaev-chain, and Kitaev-honeycomb models isolate important mechanisms: chiral or sublattice structure, Chern response, time-reversal-protected band topology, superconducting boundary modes, or fractionalized spin dynamics. Their value is precisely that they remove most material detail.

The benchmark qualifications are model specific. The SSH winding statement requires chiral symmetry and a declared unit-cell or origin convention; a boundary-state claim additionally requires the termination. “Haldane model” here means the Chern-insulator model, not the spin-1 Haldane chain. The opposite-Chern spin-block picture of Kane–Mele requires conserved $s_z$; with generic Rashba mixing the relevant test is $\mathbb Z_2$. The Kitaev chain is a mean-field BdG toy: one needs a bulk gap and parity structure, and a zero mode alone is not proof of a material phase. The honeycomb model’s exact $\mathbb Z_2$ gauge structure is an ideal-model property, and non-Abelian anyons occur only in the appropriate gapped time-reversal-broken regime.

Using one as a material model requires a separate match of degrees of freedom, dimension, symmetry, filling, gap or mobility gap, boundary, interactions, disorder, and the probe operator. A shared edge-like spectrum is not enough. The Reference Model Library identifies the compact model cards where they exist. After a topology-capable model family survives the adequacy tests, enter Topological Quantum Matter to choose the appropriate material branch; Topology in Quantum Matter and Chern Numbers in Band Theory own the material phase and invariant tests.

Worked Audit: One Correlated Material, Three Questions

Consider a transition-metal compound with narrow partially filled $d$ bands, an optical gap-like feature, and low-temperature antiferromagnetic response.

Question 1: What dispersion reaches the Fermi level? A multiorbital band or quasiparticle model is the first candidate because momentum-resolved dispersion and orbital matrix elements are the target. A static one-band Hubbard model may be too small if ligand weight or several crystal-field levels are visible.

Question 2: Is the insulating response interaction driven? The candidate set must compare a validated independent-electron band account with Hubbard/charge-transfer and disorder/localization alternatives. Test compressibility or addition spectra, optical onset and spectral-weight transfer, momentum structure, temperature dependence, and disorder scaling. A band model can support a band-insulator mechanism; it cannot establish an interaction-driven mechanism. An optical suppression alone is not a charge gap because selection rules, excitons, disorder, and matrix elements can shift or suppress the onset. A spin-only model cannot answer the charge question.

Question 3: What organizes the lowest magnetic excitations? If a charge gap separates the magnetic window and local moments are established, a spin Hamiltonian can become the smallest adequate model. Its exchange constants must reproduce momentum- and polarization-resolved magnetic spectra. Failure near the charge scale, strong damping, or orbital excitations requires the orbital model again.

The three models can all be appropriate without being equivalent. The Hubbard-to-Heisenberg arrow discards charge excitations and dresses observables; it is not an identity of full Hilbert spaces. In the simplest repulsive single-band convention, the leading exchange $J=4t^2/U$ for real nearest-neighbor hopping requires $U\gg |t|$, one particle per site or doping negligible for the target, and a target below the charge gap. At appreciable doping the natural first handoff is a projected itinerant model such as the t–J Model Preview, not a pure Heisenberg model. Higher-order hopping, multiorbital structure, ligands, and spin–orbit coupling generate additional terms. A finite-cluster gap or ordering signal does not establish a thermodynamic Mott or magnetic phase.

Worked Audit: A Topological Material Claim

Suppose a fitted honeycomb Hamiltonian and a sample with edge conduction suggest a topological band mechanism. First branch on time reversal and spin structure. A Haldane-like spinless block is a two-band Chern model that breaks time reversal and supports chiral response. Minimal Kane–Mele is spinful and has four bands before filling; it preserves time reversal, has total Chern number zero, and supports a $\mathbb Z_2$/helical phase. Only when $s_z$ is conserved does it decompose into opposite-Chern spin blocks. Rashba mixing removes that decomposition without necessarily destroying the $\mathbb Z_2$ phase.

Candidate models. Begin with the symmetry- and dimension-appropriate band model, but include the actual filling, surfaces, disorder, and coupling to the measurement. Add interactions when they are comparable with the gap or invalidate the band invariant or response, not only when they produce intrinsic topological order. Use a BdG description only when superconducting Nambu redundancy and its particle–hole constraint are appropriate; this is not an ordinary material particle–hole symmetry.

Validation. Check the bulk or mobility gap, invariant in the validated occupied subspace, boundary connectivity, quantized or symmetry-protected response, and robustness to allowed perturbations. Test trivial surface accumulation, band bending, domains, contacts, and inhomogeneity as alternatives.

Escalation. If the minimal model fits an edge spectrum but not the bulk gap, response coefficient, or symmetry dependence, adding decorative terms is not enough. Reopen the orbital basis, interaction model, surface electrostatics, and probe forward model.

Cross-Validation and the Escalation Ladder

Prefer the candidate that passes the declared validation set with the fewest unsupported structures, not the one with the most fashionable vocabulary. Escalate in diagnosed stages:

1. change parameter values only within independently allowed uncertainty;
2. restore a symmetry-allowed term omitted by a controlled truncation;
3. enlarge the retained band, orbital, spin, phonon, boundary, or environment space;
4. replace a static parameter by its momentum, frequency, or state dependence;
5. change the state, ensemble, kinetic closure, or nonequilibrium protocol;
6. replace the model family when its defining variables no longer organize the observable.

Stop when the requested predictions are stable within tolerance across the declared window and held-out tests, and when plausible alternatives are either distinguished or explicitly left unresolved. “The solver converged” and “the curve looks right” are not stopping rules.

Canonical Boundaries

• The Quantum Matter Map owns problem decomposition before candidate selection.
• From Quantum Mechanics to Materials owns provenance and approximation auditing along a selected material pipeline.
• This page owns question-first comparison, adequacy, validation, and escalation among candidate material model families.
• The Model Encyclopedia owns reproducible dossiers once a model family is named.
• The Model-to-Volume Cross-Link Index owns dossier, teaching, reference, benchmark, and application routing.
• Conventions for Quantum Matter owns signs, gauges, normalizations, units, and limit order after a model is selected.
• Computational Many-Body QM owns many-body solver selection, convergence, benchmarks, and reproducibility; other numerical methods remain with their technical owners.
• Computational Quantum Matter owns the material-facing step from a selected model and observable to a routed workflow, validation record, and bounded claim; this page retains model selection.
• Model definitions and derivations remain on their linked teaching pages; material claims must return to probe and evidence pages such as Data Interpretation and Pitfalls.

Common Selection Errors

• Treating a material name as a Hamiltonian or a Hamiltonian name as a complete finite problem.
• Choosing a model before specifying the observable, probe, resolution, and tolerance.
• Assuming a more microscopic or more expensive model is uniformly more predictive.
• Using the same reduced model for spectra, matrix elements, transport, and phase identification without separate validation.
• Calling a Kohn–Sham eigenvalue, tight-binding energy, quasiparticle pole, and measured spectral peak the same “band.”
• Treating fitted parameters as independent tests or mixing incompatible active spaces and screening conventions.
• Confusing a benchmark model with evidence that a material realizes its phase.
• Treating a spin-only reduction as a charge model or a pairing gap as proof of superconducting coherence.
• Reporting solver convergence as model validation.
• Adding terms until one dataset fits without recording predictive penalties, identifiability, or held-out tests.

Exercises

Exercise 1: One semiconductor, three models

Choose the first model for (a) low-field bulk mobility near a band minimum, (b) an indirect optical absorption edge, and (c) the local spectrum near an atomically resolved dopant. State one escalation test for each.

Solution

(a) Start with multivalley effective-mass bands plus a Boltzmann or other appropriate transport closure whose scattering channels and carrier density are independently constrained. Escalate for nonparabolicity, confinement, intervalley structure beyond the retained rates, or contact-limited transport.

(b) Retain valence and conduction bands, photon and phonon matrix elements, occupations, and excitonic corrections at the requested resolution. A mobility model omits the optical operator. Escalate when remote bands, excitons, temperature-dependent phonons, or disorder alter the edge.

(c) Use an atomistic impurity Hamiltonian, solved for example with Green functions, plus the surface, screening, and tunneling forward model. Envelope theory can fail at atomic lengths. Escalate the active orbital and electrostatic environment if the spatial pattern or bias dependence is wrong.

Exercise 2: Build an adequacy card

A narrow-band material shows a dispersing spectral peak, a resistivity upturn, and an optical suppression at low temperature. List the minimum card fields before choosing among a band, Hubbard-like, or disorder model. Which single observation would be insufficient to decide?

Solution

Declare composition and structure, temperature and history, dimensionality, filling, disorder controls, each probe and resolution, the exact observables, active orbitals, symmetries, parameter provenance, state and solver, held-out tests, and candidate failure criteria. The resistivity upturn alone is insufficient: localization, a gap, contacts, inhomogeneity, and changing scattering can mimic it. Joint spectral-weight, compressibility or optical, momentum, disorder, and temperature evidence is needed.

Exercise 3: Audit a Hubbard-to-Heisenberg reduction

When is a spin-only model adequate for a half-filled repulsive Hubbard-like material, and which observables does the reduction lose?

Solution

The spin description requires an established low-energy local-moment sector, one particle per active site or doping negligible for the target, $U\gg|t|$ or an analogous charge-scale separation, charge excitations separated from the target window, and exchange parameters matched from the parent model or data. In the simplest repulsive single-band case with real nearest-neighbor hopping, $J=4t^2/U$ is only the leading result. At appreciable doping, route toward a projected itinerant model such as t–J rather than a pure Heisenberg model. The spin model can predict low-energy magnetic correlations and spin excitations. It does not retain charge addition and removal spectra, optical transitions across the charge gap, doublon dynamics, or general doping response. Significant charge fluctuation, ligand physics, orbital dynamics, higher-order exchange, doping, or high-frequency probes require escalation; finite-cluster evidence alone does not establish a bulk phase.

Exercise 4: A benchmark is not a material diagnosis

A two-band model has the same Chern number as a fitted material Hamiltonian, and one sample shows an edge-like conductance feature. What remains before a Chern-material claim is warranted?

Solution

Validate the occupied subspace and filling, bulk or mobility gap, symmetry and dimensionality, invariant convention, boundary connectivity, response coefficient and units, disorder and interaction stability, and the probe and contact forward model. Exclude trivial surface bands, accumulation layers, band bending, inhomogeneity, magnetic domains, and contact artifacts. Matching a benchmark invariant plus one boundary-like signal does not establish the material phase.

References

• A. Altland and B. Simons, Condensed Matter Field Theory, 2nd ed., Cambridge University Press, 2010.
• N. W. Ashcroft and N. D. Mermin, Solid State Physics, Holt, Rinehart and Winston, 1976.
• P. Coleman, Introduction to Many-Body Physics, Cambridge University Press, 2015.
• M. P. Marder, Condensed Matter Physics, 2nd ed., Wiley, 2010.
• R. M. Martin, Electronic Structure: Basic Theory and Practical Methods, Cambridge University Press, 2004.
• P. Phillips, Advanced Solid State Physics, 2nd ed., Cambridge University Press, 2012.