From Models to Measured Observables

The same material can support several phases, and the same Hamiltonian can make different predictions for different states and measurements. To organize a quantum-matter problem, connect the sample, the model, the state, the observable, and the scales. An unexplained discrepancy often comes from one of these missing connections.

Required background. Quantum Scales and Collective Behavior distinguishes a material, a phase, an excitation, and a measured response. For a study sequence rather than a problem to analyze, use Condensed Matter Roadmap.

Five parts of a material problem

Part	What to specify	Why it matters
Sample	Composition, structure, thickness, contacts, disorder, applied fields	Two samples with the same nominal composition can have different carrier densities or scattering
Model	Degrees of freedom, Hamiltonian, parameters, boundaries	A spin model and an electronic model retain different excitations
State	Filling, temperature, preparation, and relevant history	The Hamiltonian alone does not specify occupations or order
Observable	The measured quantity and its probe coupling	A mode can be visible in one polarization and absent in another
Scales	Energy, length, time, momentum, and resolution	A low-energy approximation may fail at the frequency of the experiment

For an equilibrium sample, a thermal ensemble may specify the state. For a quench, a pulse experiment, or a hysteretic magnet, the preparation history can be essential. Boundaries and contacts matter especially in small devices, where terminal conductance need not be described by a local bulk resistivity.

The forward problem predicts an observable from a specified model and state. The inverse problem asks which models and parameter values are consistent with measured data. The inverse problem often has several answers: one spectrum rarely determines an entire Hamiltonian.

Follow the observable through the description

Microscopic electrons and nuclei can be reorganized into bands, local moments, lattice modes, and collective fields. Choose the description that retains what your observable can excite or resolve.

For long-wavelength sound, an elastic displacement field may be enough. To predict a core-level spectrum, atomic multiplets and their couplings can be essential. To predict a quantum point contact’s conductance, transmitting channels and reservoirs are natural variables. Choosing a Model compares these choices in more detail.

A change of basis and an approximation are different operations. Rewriting a complete selected band subspace in a Wannier basis preserves its information. Discarding bands, distant hoppings, or interaction terms removes information and needs a separate justification. Wannier functions explains the localization conditions; effective Hamiltonians explains how eliminated states affect retained ones.

Energies alone are often insufficient. Two calculations can give similar dispersions but different eigenvectors, and therefore different optical matrix elements or Berry curvature. When comparing with an experiment, carry the relevant observable through the approximation as well as the Hamiltonian.

Compare scales before choosing a method

Ratio	Physical comparison	What it helps decide
$T/T_F$	Temperature relative to the Fermi scale	Whether a simple carrier gas is degenerate
$U/W$	Local interaction relative to bandwidth	Whether weak-interaction reasoning is plausible for the selected orbitals
$\gamma_E/E_*$	Intrinsic energy half width relative to an excitation scale	Whether a narrow-mode description is useful
$\ell/L$	Elastic mean free path relative to device size	Whether elastic scattering is rare or frequent across the device
$\ell_\phi/L$	Phase-coherence length relative to device size	Whether interference across the device can survive
$\xi/L$	Correlation length relative to sample size	Whether finite-size effects are likely to dominate
$\Delta/(k_BT)$	A gap relative to thermal energy	Whether thermal activation across the gap is suppressed

These ratios organize approximations, not universal phase boundaries. For example, $U/W$ omits filling, orbital degeneracy, Hund coupling, and screening. In metals, $k_F\ell$ measures the mean free path in units of inverse Fermi wave vector; because $\lambda_F=2\pi/k_F$, it equals $2\pi\ell/\lambda_F$.

Example: coherence without ballistic motion. Suppose a multichannel wire has $L=1\,\mu\mathrm m$, elastic mean free path $\ell=0.05\,\mu\mathrm m$, and phase-coherence length $\ell_\phi=3\,\mu\mathrm m$. Assume weak disorder ($k_F\ell\gg1$) and a length well below its localization length, so the wire remains in the metallic regime. Electrons then undergo many elastic scattering events and their motion is approximately diffusive, while phase memory can survive across the sample. Weak localization and universal conductance fluctuations exploit precisely this distinction, as reviewed by Beenakker and van Houten. Elastic scattering need not erase phase information. Without the metallic-regime assumption, $\ell\ll L$ alone does not establish diffusion: coherent interference can instead produce strong localization.

A gap has to be attached to an excitation

An optical absorption edge, a single-particle gap, a spin gap, and a transport activation energy are different quantities. A useful comparison starts by identifying the excitation created by the probe.

For a semiconductor optical edge, determine whether the transition is direct or phonon assisted. Include the valence and conduction bands, dipole selection rules, and electron–hole attraction. An exciton can produce absorption below the unbound electron–hole threshold; an indirect transition involves phonon energy and occupation factors. Thus an absorption onset is not automatically the bare difference between two band eigenvalues. See band gaps and excitons.

Likewise, a resistance increase at a commensurate filling in a moiré system does not determine whether the mechanism is a Mott state, charge order, localization, or another gap-producing process. Compressibility, spectroscopy, symmetry-sensitive probes, and temperature dependence can help distinguish them.

Static and uniform response need not agree

The order of limits can express a physical difference rather than an algebraic subtlety. Consider a conserved density fluctuation $\delta n$ in an infinite diffusive medium. Let a small field $f$ couple as $H'=-\int f n\,d^dx$, so a static field locally favors $\delta n=\kappa f$, with positive compressibility $\kappa$. A simple constitutive equation is

$$\partial_t\delta n=D\nabla^2(\delta n-\kappa f),\qquad D>0.$$

For Fourier components proportional to $e^{i\mathbf q\cdot\mathbf r-i\omega t}$, it gives

$$\delta n=\chi(\mathbf q,\omega)f,\qquad \chi(\mathbf q,\omega)=\frac{\kappa Dq^2}{Dq^2-i\omega}.$$

At a nonzero wave vector, a sufficiently slow perturbation allows density to redistribute, so $\chi(\mathbf q,0)=\kappa$. At a nonzero frequency, a strictly uniform field cannot change the total particle number of this closed system, so $\chi(\mathbf0,\omega)=0$. Therefore

$$\lim_{q\to0}\lim_{\omega\to0}\chi=\kappa, \qquad \lim_{\omega\to0}\lim_{q\to0}\chi=0.$$

Particle exchange with a reservoir changes the uniform-mode problem. A real measurement must specify that coupling and allow sufficient relaxation time. The general response theory is developed in the Kubo formula; material applications are in transport and response.

Compare predictions with what the instrument records

A useful schematic measurement model is

$$\mathbf y_{\mathrm{measured}} =\mathcal R\mathbf f(\boldsymbol\theta) +\mathbf b+\boldsymbol\epsilon.$$

The ideal prediction $\mathbf f$ depends on physical parameters $\boldsymbol\theta$. The operator $\mathcal R$ represents acceptance and resolution, $\mathbf b$ a background, and $\boldsymbol\epsilon$ noise. Real instruments may require a nonlinear model, but even this simple expression explains two common difficulties.

First, different parameter values may produce nearly indistinguishable data over a restricted measurement window. A fitted number can then be poorly determined even when the curve looks convincing. Second, an excitation can be present but invisible because its matrix element vanishes in the measured geometry. A null result constrains a theory only within the experiment’s sensitivity and coverage.

Use complementary observables to distinguish possibilities. A proposed magnetic order can be tested with diffraction and local-field probes; a candidate superconducting state with electromagnetic and thermodynamic responses; a suggested quasiparticle with its dispersion and linewidth dependence. How Quantum Matter Is Measured and Data Interpretation and Pitfalls develop the practical examples.

Exercises

1. Specify an insulating prediction. A calculation is described only as “the Hubbard model is insulating.” Name the missing information needed to interpret that statement.

Solution

Specify the lattice, hopping, interaction, filling, temperature or preparation, size, and boundary conditions. Identify the observable used: a charge-addition gap, compressibility, Drude weight, and finite-temperature conductivity are related but distinct diagnostics. A material comparison also needs a mapping from its active orbitals and parameters to the model, plus convergence or approximation checks.

2. Basis change and truncation. A selected isolated group of bands is transformed into Wannier orbitals. The Hamiltonian is then restricted to nearest-neighbor hopping and on-site interactions. Which operations can preserve information exactly?

Solution

A unitary Bloch-to-Wannier transformation preserves information within the selected subspace. The existence of exponentially localized or symmetry-respecting Wannier functions has additional topological conditions. Selecting the subspace can already omit relevant remote bands; discarding hopping and interaction terms is another approximation. Compare those omissions with the observable and energy range being studied.

3. Diffusive response. Starting from the density equation above, derive $\chi(\mathbf q,\omega)$ and find the relaxation time for a density mode of wave vector $q\ne0$ after $f$ is switched off.

Solution

Substitution gives $(-i\omega+Dq^2)\delta n=\kappa Dq^2f$, producing the stated susceptibility. With no drive, $\delta n_q(t)=\delta n_q(0)e^{-Dq^2t}$, so the relaxation time is $(Dq^2)^{-1}$. Long-wavelength modes relax increasingly slowly. The $q=0$ mode is conserved and has no finite relaxation time in this closed-system equation.

4. A missing peak. A proposed collective mode is absent in a measured spectrum. Give two checks before concluding that the mode does not exist.

Solution

Check whether the probe operator and polarization couple to the mode, and whether its momentum and energy fall inside the accessible range with adequate sensitivity and resolution. An alternative polarization, momentum scan, or complementary probe can distinguish invisibility from genuine absence.

References

• P. W. Anderson, “More Is Different,” Science 177, 393–396 (1972), doi:10.1126/science.177.4047.393.
• C. W. J. Beenakker and H. van Houten, “Quantum Transport in Semiconductor Nanostructures,” Solid State Physics 44, 1–228 (1991), doi:10.1016/S0081-1947(08)60091-0.
• P. Coleman, Introduction to Many-Body Physics, Cambridge University Press, 2015, doi:10.1017/CBO9781139020916.
• B. Keimer and J. E. Moore, “The physics of quantum materials,” Nature Physics 13, 1045–1055 (2017), doi:10.1038/nphys4302.