phonon theory

Phonon theory describes collective atomic motion in condensed matter by expanding the potential energy around a reference structure, finding its normal modes, and quantizing those modes. In the harmonic limit, each mode labeled by wave vector $\mathbf q$ and branch $\nu$ behaves as an independent quantum oscillator of energy $\hbar\omega_{\mathbf q\nu}(n+1/2)$. Force constants and atomic masses determine frequencies and polarization vectors; anharmonic force constants create thermal expansion, finite lifetimes, frequency shifts, and heat resistance. A complete phonon model must declare the structure, boundary conditions, force model, electrostatics, dimensionality, quantum statistics, approximation order, and observable. ```svg ``` **A phonon is a quantized normal mode rather than one atom vibrating.** Every atom participating in a mode moves with an amplitude and phase specified by a polarization eigenvector. A phonon occupation adds one quantum to that collective oscillator. Localized pictures are useful wave-packet constructions made from many wave vectors, but a perfect-crystal phonon eigenstate is extended. Confusing phonons with literal point particles hides coherence, polarization, and interference. **The reference structure must be a stationary point of the potential energy.** Expand atomic displacements $u_{l\kappa\alpha}$ about equilibrium positions where first derivatives vanish. If residual forces remain, a linear term drives motion and the harmonic matrix does not describe oscillations about a true stationary configuration. Tight structural relaxation, stress control, magnetic state, charge state, and electronic convergence are prerequisites. A high-symmetry saddle may intentionally have imaginary modes, but it should not be mislabeled an equilibrium phase. **Second-order force constants are the Hessian of potential energy.** $\Phi_{\alpha\beta}(l\kappa,l'\kappa')=\partial^2U/(\partial u_{l\kappa\alpha}\partial u_{l'\kappa'\beta})=-\partial F_{l'\kappa'\beta}/\partial u_{l\kappa\alpha}$. They couple a displacement of one atom and direction to the force response elsewhere. Translational symmetry reduces them to cell differences. Their range and symmetry encode bonding, electrostatics, and the chosen electronic or empirical potential surface. **Mass weighting converts the force-constant problem into a Hermitian eigenproblem.** The dynamical matrix divides force-constant blocks by $\sqrt{M_\kappa M_{\kappa'}}$ and Fourier transforms over lattice vectors. Diagonalizing $D(\mathbf q)$ gives eigenvalues $\omega^2_{\mathbf q\nu}$ and normalized eigenvectors. Phonopy's official formulation makes this mass weighting and phase convention explicit. Changing isotope mass shifts frequencies without changing a Born–Oppenheimer force-constant surface to leading order. **Phase conventions change eigenvectors but not physical frequencies.** Dynamical matrices may place basis-atom position phases inside or outside the Fourier factor. Eigenvectors then differ by unitary, wave-vector-dependent phases. Scattering matrix elements, group velocities, structure factors, and interpolations must use one convention consistently. Comparing raw complex eigenvectors across codes without aligning phases and degenerate subspaces can suggest disagreement where observables agree. **There are three acoustic branches in an ordinary three-dimensional crystal.** Uniform translation costs no energy, giving frequencies approaching zero at $\Gamma$ for one longitudinal and two transverse acoustic branches in a bulk elastic medium. Their small-$q$ slopes are sound velocities related to elastic constants and density. In lower-dimensional membranes, flexural branches can be quadratic rather than linear. A missing or gapped acoustic mode often signals broken translational invariance, external pinning, or numerical error. **Optical branches arise when the primitive cell contains more than one atom.** A cell with $s$ atoms has $3s$ branches in three dimensions: three acoustic and $3s-3$ optical. Near $\Gamma$, optical modes involve relative motion of basis atoms and can couple to light or electric fields depending on symmetry and polarity. “Optical” names historical spectroscopy, not a guarantee that every branch is infrared or Raman active. **Polarization vectors carry atomic, directional, and phase information.** Longitudinal and transverse labels are exact only along symmetry directions or in isotropic limits. Away from them, modes can be mixed. Degenerate eigenvectors are not individually unique; any unitary rotation within their subspace is valid. Track subspaces using overlaps, symmetry irreducible representations, or parallel transport rather than sorting solely by frequency near crossings. ```svg ``` **A monatomic one-dimensional chain provides the canonical dispersion test.** Nearest-neighbor springs of constant $K$ and mass $M$ give $\omega(q)=2\sqrt{K/M}|\sin(qa/2)|$. It is linear near zero, periodic in reciprocal space, and has zero group velocity at the zone boundary. This model verifies phase, mass, units, acoustic behavior, and Fourier interpolation. It also shows why continuum elasticity captures only long wavelengths. **A diatomic chain separates acoustic and optical motion.** Two masses per cell produce two branches with a gap controlled by mass contrast and springs. At long wavelength, the acoustic branch moves atoms mostly in phase, whereas the optical branch has relative motion. The limiting frequencies and eigenvectors expose normalization and atom-order errors. Real crystals add three dimensions, multiple bonds, noncentral forces, and long-range electrostatics, but the branch logic persists. **Group velocity is the wave-packet energy propagation velocity in the harmonic band picture.** $\mathbf v_{\mathbf q\nu}=\nabla_{\mathbf q}\omega_{\mathbf q\nu}$, obtainable from derivatives of the dynamical matrix and eigenvectors away from degeneracies. Phase velocity $\omega/q$ is different. A flat optical band has small group velocity even at high frequency. Heat transport also depends on heat capacity and lifetime, so high group velocity alone does not determine conductivity. **The phonon density of states counts modes per frequency interval.** $g(\omega)=\sum_{\mathbf q\nu}\delta(\omega-\omega_{\mathbf q\nu})$ with normalization chosen per cell, atom, volume, or mole. Van Hove singularities appear where dispersion gradients vanish or topology changes. Partial densities project atomic or directional character but depend on eigenvector normalization. A path dispersion cannot substitute for a full Brillouin-zone mesh in thermodynamics. **Quantization turns each positive harmonic mode into a bosonic oscillator.** The harmonic Hamiltonian is $\sum_{\mathbf q\nu}\hbar\omega_{\mathbf q\nu}(a^\dagger a+1/2)$. Creation and annihilation operators change occupation by one, and modes at $\mathbf q$ and $-\mathbf q$ combine to make real displacements. The Bose–Einstein occupation is $n_B=1/[\exp(\hbar\omega/k_BT)-1]$. Phonon number is not conserved because interactions can create and destroy quanta. **Zero-point motion persists at zero temperature.** Each mode has energy $\hbar\omega/2$ and mean-square displacement even in its ground state. Zero-point energy can shift phase stability, lattice constants, isotope effects, and light-atom structures. Harmonic zero-point motion does not by itself cause thermal resistance because independent modes do not scatter. Nuclear quantum effects beyond harmonic motion require path integrals, vibrational configuration interaction, or other anharmonic treatments. **Classical equipartition is the high-temperature limit of quantum phonon statistics.** When $k_BT\gg\hbar\omega$, a mode's thermal energy approaches $k_BT$ and classical molecular dynamics can sample its occupation. At low temperature, high-frequency modes freeze out. Assigning classical energy to every mode produces the Dulong–Petit heat capacity at all temperatures and misses zero-point motion. Quantum corrections based on a harmonic density of states do not fully repair anharmonic classical dynamics. **Low-temperature acoustic modes produce the Debye $T^3$ heat-capacity law in three dimensions.** The Debye model replaces acoustic dispersions by linear isotropic cones up to a cutoff chosen to count modes. Its density of states scales as $\omega^2$. It describes universal low-frequency behavior but not optical branches, anisotropy, or detailed van Hove structure. In two and one dimensions or for quadratic flexural modes, the power law changes. **Einstein's model captures localized-frequency intuition but not acoustic physics.** Treating all atoms as identical oscillators gives a single frequency and activated low-temperature heat capacity. It historically demonstrates quantum suppression and can approximate narrow optical groups. It has no translational acoustic modes, dispersion, or heat propagation. Multi-Einstein fits can reproduce heat-capacity curves without uniquely identifying microscopic branches. **Harmonic free energy follows directly from the phonon spectrum.** $F_{vib}(T)=\sum_{\mathbf q\nu}[\hbar\omega/2+k_BT\ln(1-e^{-\hbar\omega/k_BT})]$ for stable modes. Entropy and constant-volume heat capacity are temperature derivatives. Imaginary frequencies make this expression ill-defined because the reference is not a harmonic minimum. Numerical integration needs a converged $q$ mesh, especially for low-frequency acoustic contributions. **Quasiharmonic theory adds thermal expansion through volume-dependent frequencies.** Compute electronic energy plus harmonic phonon free energy across volumes, minimize $F(V,T)+PV$, and obtain equilibrium volume, bulk response, and thermal expansion. Mode Grüneisen parameters $\gamma_{\mathbf q\nu}=-\partial\ln\omega/\partial\ln V$ connect frequency shifts to volume. Phonopy documents this finite-volume derivative formulation. QHA neglects explicit same-volume anharmonic frequency shifts and fails near strong instabilities or melting. **Negative thermal expansion comes from weighted mode Grüneisen physics.** Modes with negative $\gamma$ soften under compression and can drive contraction on heating when their heat-capacity weights dominate. Framework transverse modes and flexural motion are common mechanisms. Averaging Grüneisen parameters without heat-capacity and elastic weighting can give the wrong sign. Temperature can change which modes dominate. ```svg ``` **Imaginary frequency denotes negative curvature under the standard harmonic convention.** A negative dynamical-matrix eigenvalue is often plotted as a negative frequency magnitude. It means the reference structure lowers energy along that mode at harmonic order, not that atoms oscillate with an imaginary clock rate. Follow the eigenvector, distort both signs, relax, and map the energy. Tiny imaginary acoustic values can instead arise from force noise, interpolation, or violated sum rules. **Soft modes connect lattice dynamics to structural phase transitions.** A branch frequency decreasing toward zero signals a weakening restoring force at a particular wave vector and symmetry. Condensing a zone-center soft mode can produce a ferroelectric distortion; a zone-boundary mode enlarges the cell. Anharmonic free energy, strain coupling, disorder, and quantum fluctuations determine the actual transition. Zero-K harmonic instability alone does not give transition temperature. **The acoustic sum rule expresses invariance under rigid translation.** Summing force constants acting on any atom over all partner atoms must vanish. Enforcing the rule restores zero-frequency translations but can redistribute noisy force constants. Rotational invariance adds further constraints, especially important for molecules and low-dimensional flexural dispersions. A correction should be reported and the uncorrected violation used as a force-quality diagnostic. **Crystal symmetry constrains force constants and mode irreducible representations.** Space-group operations relate displaced configurations and tensor components, reducing computation and noise. At high-symmetry wave vectors, little-group irreducible representations label degeneracies and selection rules. Symmetrizing a genuinely distorted, magnetic, or defective structure can erase physics. Use the symmetry of the actual force model, including spin and fields, not merely nominal atomic positions. **Finite-displacement calculations estimate force constants from force differences.** Displace symmetry-inequivalent atoms by small amplitudes in a supercell, compute forces, and fit the linear response. Too small a displacement exposes SCF and force noise; too large includes anharmonicity. Central differences reduce even-order contamination but double calculations. Phonopy's official formulation explicitly relates displacement-induced forces to second-order constants. Check several amplitudes for representative atoms. **The supercell controls real-space force-constant range and reciprocal interpolation.** Periodic images of a displaced atom must be separated enough that truncated interactions are negligible or treated analytically. Covalent short-range materials may converge quickly, while metals, ionic crystals, low-dimensional systems, and soft materials need larger cells. A smooth-looking dispersion from a small supercell can be wrong because Fourier interpolation always produces a curve. Converge frequencies, free energy, and target modes with supercell shape and size. **Density-functional perturbation theory obtains linear response without explicit supercell displacements.** DFPT solves self-consistent first-order electronic response to a phonon perturbation at selected $\mathbf q$, directly producing dynamical matrices. Quantum ESPRESSO's phonon guide documents generic-$q$ frequencies and eigenvectors through DFPT. It is efficient for primitive periodic crystals and polar response, but inherits electronic cutoff, $k$ mesh, smearing, pseudopotential, and functional convergence requirements. **Finite displacement and DFPT should agree in the same physical and numerical limit.** Differences arise from supercell truncation, $q$ sampling, displacement amplitude, force convergence, Fourier convention, nonanalytic corrections, and electronic parameters. Cross-method agreement at selected commensurate $q$ points is a powerful verification. One method is not intrinsically more accurate; each exposes different numerical errors. **Long-range dipole interactions make polar-crystal force constants nonlocal.** Born effective charge tensors couple atomic displacement to macroscopic polarization, and the high-frequency dielectric tensor screens the field. Near $\Gamma$, a direction-dependent nonanalytic dynamical-matrix term produces longitudinal–transverse optical splitting. Omitting it makes LO and TO modes spuriously degenerate; double-counting it after force constants already include an inconsistent long-range treatment is also wrong. **LO–TO splitting depends on the direction of approach to the zone center.** In anisotropic polar crystals, the nonanalytic correction depends on $\mathbf q\cdot\epsilon_\infty\cdot\mathbf q$ and projected Born charges. A single Gamma frequency list is insufficient without a direction. Phonopy's documentation exposes this through a specified $q$ direction. Raman and infrared comparisons must use the experimental propagation and polarization geometry. **Two-dimensional polar materials require boundary-aware long-range electrostatics.** A repeated-slab calculation with three-dimensional Coulomb interaction gives vacuum- and cell-dependent small-$q$ behavior. Coulomb truncation or two-dimensional electrostatic models restore the correct nonanalytic dispersion. Born charges, dielectric response, and effective thickness need compatible units. Large vacuum alone can converge slowly to the wrong functional form. **Electronic convergence for phonons is stricter than convergence for total energy alone.** Forces and second derivatives amplify basis, $k$-point, smearing, and SCF errors. Metals require dense Fermi-surface sampling; magnetic systems require stable spin state; pseudopotentials need appropriate cutoffs. Converge representative acoustic, optical, and soft modes rather than only total energy. A few wave numbers of error can reverse a stability claim near zero. **Phonon calculations inherit the exchange–correlation and force-model approximation.** Lattice constants, bonding curvature, dielectric response, and magnetism depend on DFT functional, pseudopotential, empirical potential, or machine-learned potential. Computing phonons at experimental versus relaxed volume can shift frequencies substantially and changes what is being tested. Numerical convergence does not remove functional error. Benchmark structure, elastic constants, and selected measured frequencies before predicting unseen behavior. **Mode animation is a diagnostic rather than a quantitative observable.** Visualization chooses arbitrary phase, amplitude, time origin, and real combination of complex eigenvectors. It helps identify rotations, translations, bond stretching, and unstable distortions. Apparent atom amplitude also depends on whether eigenvectors are mass normalized. Report the normalization when using eigenvectors in projections or coupling calculations. **Participation ratio distinguishes extended and localized vibrational character.** A mode spread uniformly across $N$ atoms has a large participation ratio, while a defect or amorphous localized mode has a small one under standard normalization. The precise definition and mass weighting must be stated. Localization in a finite supercell can change with size. A flat band is not necessarily localized, and a localized harmonic mode can hybridize through anharmonicity. **Disorder replaces exact crystal momentum with broadened or statistical mode character.** Alloys, amorphous solids, isotope mixtures, and defects break primitive translation. Supercell eigenmodes can be unfolded into primitive spectral weight, while Green-function and coherent-potential methods treat averaged response. Labeling every supercell mode by a folded $q$ is misleading. Configuration ensembles are needed for disorder-broadened spectra and heat transport. ```svg ``` **Anharmonic force constants are higher derivatives of the potential energy.** Cubic $\Phi^{(3)}$ terms couple three displacements, quartic $\Phi^{(4)}$ terms couple four, and higher orders continue. Expressed in the harmonic normal-mode basis, these become interaction vertices among phonons. Their range, symmetry, and numerical noise are more demanding than harmonic constants. A potential that reproduces harmonic dispersion can still give wrong thermal expansion or conductivity because its third- and fourth-order derivatives are inaccurate. **Three-phonon processes obey energy and crystal-momentum selection rules.** One mode can decay into two or two can combine into one when frequencies and wave vectors satisfy conservation, with reciprocal lattice vector $\mathbf G$ allowing Umklapp. Matrix elements set coupling strength, while Bose factors set temperature-dependent availability. Energy-conservation surfaces and Brillouin-zone sampling dominate numerical integration. Broadening a delta function must converge without inventing forbidden phase space. **Normal and Umklapp processes play different momentum roles.** Normal processes have $\mathbf G=0$ and conserve total phonon crystal momentum, rapidly redistributing populations without directly relaxing collective drift. Umklapp processes transfer a reciprocal lattice vector to the lattice and resist heat flow. Boundaries, isotopes, defects, and electrons also relax momentum. Treating every three-phonon event with one lifetime hides hydrodynamic regimes where normal scattering dominates. **A phonon linewidth measures decay of a mode under stated conventions.** The imaginary part of the phonon self-energy gives a half-width or full width depending on definition; lifetime may be $1/(2\Gamma)$, $1/\Gamma$, or include angular-frequency factors. Always state convention and units. Spectral linewidth includes intrinsic anharmonicity plus isotopes, disorder, electrons, boundaries, instruments, and inhomogeneity. A harmonic calculation has delta-function modes and cannot predict finite linewidth by itself. **Anharmonic frequency shifts are the real part of the self-energy.** Temperature changes frequencies through explicit phonon interactions at fixed volume and implicit thermal expansion. The quasiharmonic approximation includes only the volume pathway. Cubic bubble and quartic loop terms can harden or soften modes, with principal-value integrations tied to linewidth physics. Comparing constant-volume theory to constant-pressure experiment without separating these contributions confuses mechanisms. **Four-phonon scattering matters when cubic channels are weak or temperatures are high.** Quartic interactions allow redistribution and decay processes beyond three-phonon phase space and can substantially reduce conductivity in some materials. They also renormalize strongly anharmonic modes. Computational cost rises steeply with force-constant order and $q$ sampling. A three-phonon result agreeing at one temperature does not prove four-phonon irrelevance across the range. **Self-consistent phonon methods renormalize unstable or strongly anharmonic modes.** They replace the bare harmonic reference with an effective temperature-dependent dynamical matrix derived variationally, stochastically, or from sampled forces. This can stabilize phases that have imaginary zero-K harmonic modes but exist at finite temperature. Different schemes approximate diagrams and statistics differently. Convergence requires supercell, sampling, force accuracy, and self-consistency checks, not merely disappearance of imaginary frequencies. **Molecular dynamics spectra provide a classical anharmonic route.** Velocity autocorrelation Fourier transforms yield vibrational densities; mode-projected correlations yield frequencies and lifetimes; spectral energy density resolves wave vectors in periodic cells. Finite trajectory length sets frequency resolution, thermostat choice alters dynamics, and classical occupations miss quantum statistics. Machine-learned or empirical potentials extend size and time only to the accuracy of their training forces. **Phonon Boltzmann transport converts mode properties into lattice heat conduction.** The linearized BTE balances a temperature-gradient driving term against scattering. In relaxation-time form, $\kappa_{\alpha\beta}=V^{-1}\sum_{\mathbf q\nu}C_{\mathbf q\nu}v_{\alpha}v_{\beta}\tau_{\mathbf q\nu}$. Heat capacity, velocity, and lifetime are mode resolved. Phono3py supports both relaxation-time and direct linearized-BTE solutions, reflecting that collective off-diagonal scattering can matter. **The relaxation-time approximation discards repopulation coupling between modes.** Assigning each deviation an independent lifetime is simple and often reasonable when resistive scattering dominates. The full collision matrix contains in-scattering that can preserve collective momentum and increase conductivity relative to single-mode RTA. Iterative or direct solutions recover coupled populations. Calling a linewidth-derived lifetime a transport lifetime without solving repopulation can be especially wrong in hydrodynamic materials. **Mean free path is mode dependent and directional.** A common scalar is $\Lambda_{\mathbf q\nu}=|\mathbf v|\tau$, while tensor or projected lengths matter for anisotropic transport. Cumulative conductivity versus mean free path shows which modes a device size suppresses. A single average mean free path cannot represent a spectrum spanning nanometers to millimeters. Boundary scattering then becomes geometry and direction dependent rather than a universal added rate. **Boundary-limited thermal transport is a kinetic boundary-value problem.** Diffuse surfaces randomize outgoing direction, specular surfaces preserve parallel momentum, and partial specularity depends on roughness relative to wavelength and incidence. Matthiessen-style boundary rates are approximations. Thin films, nanowires, grains, and interfaces reshape the distribution nonlocally when size approaches mean free paths. Contact thermal resistance and internal boundary suppression are distinct. **Isotope scattering arises from mass disorder without changing average force constants at leading order.** Random isotopic masses break translation and elastically scatter modes with strength tied to mass variance and eigenvector character. Isotopic purification can greatly increase conductivity in high-quality crystals. The virtual-crystal harmonic spectrum plus perturbative scattering is valid for weak disorder; large contrast or localization needs explicit disorder or Green-function methods. **Defect and alloy scattering involve both mass and force-constant disorder.** Vacancies, substitutions, interstitials, and strain perturb local bonds as well as inertia. Point-defect formulas based only on mass variance can miss dominant force changes. Dilute perturbation, $T$-matrix, supercell unfolding, and configurational averaging cover different concentrations and strengths. Adding empirical rates risks double counting if the fitted potential already includes disorder broadening. **Electron–phonon interaction links lattice modes to electronic transport and superconductivity.** A phonon displacement changes the electronic potential, producing matrix elements between electronic states. These govern carrier mobility, phonon linewidth from electron–hole pairs, energy relaxation, resistivity, and pairing in conventional superconductors. Screening, band occupations, interpolation, and energy conservation are central. The specialized phonon-scattering page owns device-mobility details; phonon theory supplies the lattice modes and couplings. **Remote phonons couple carriers to polar modes outside their host material.** Surface optical modes in a nearby oxide or dielectric create long-range electric fields that scatter channel carriers without atomic overlap. Frequency, dielectric response, distance, screening, and interface geometry control coupling. This is distinct from intrinsic bulk phonons and remains preserved as a specialized production page. A generic local deformation-potential model cannot reproduce it. **Coherent phonon transport requires phases or interband heat-current terms.** The particle-like BTE uses diagonal mode populations. In complex crystals with close branches, off-diagonal density-matrix or Wigner terms can contribute. Phono3py documents interband velocity-matrix formulations and notes their relevance for many close bands and glass-like systems. A mode lifetime picture becomes ambiguous when linewidths approach branch separations. **Hydrodynamic phonon flow emerges when normal scattering establishes a drifting local equilibrium.** Resistive Umklapp, impurities, and boundaries relax that drift more slowly. Poiseuille heat profiles, second sound, and nonmonotonic size effects can result within a window of temperature and geometry. Fourier's law and independent RTA modes miss collective viscosity. Demonstrating hydrodynamics requires competing rate and length-scale evidence, not a large conductivity alone. **Ballistic thermal conductance is set by modes and contacts rather than bulk conductivity.** When length is shorter than scattering lengths, reservoirs inject phonons and transmission determines heat current. Landauer expressions sum mode transmissions weighted by energy and occupation difference. Assigning $k=L G/A$ produces an apparent conductivity proportional to length, showing why a bulk material coefficient is inappropriate. Contact mode mismatch and interface transmission become part of the observable. ```svg ``` Phonon spectroscopy measures different projections of the same modes. Inelastic neutron scattering samples momentum and energy broadly and weights nuclear scattering lengths. Inelastic x-ray scattering accesses small samples and high momentum with electronic form-factor weights. Raman scattering probes near-zone-center symmetry-allowed polarizability changes. Infrared absorption probes dipole-active modes. Electron energy-loss and ultrafast optical methods add different spatial and temporal windows. A calculated frequency list must be converted through the relevant structure factor and resolution before comparison. The dynamic structure factor $S(\mathbf Q,\omega)$ connects eigenvectors to scattering intensity. One-phonon terms contain Bose creation or annihilation factors, Debye–Waller attenuation, masses, scattering lengths or form factors, polarization projection $\mathbf Q\cdot\mathbf e$, and momentum selection. Phonopy's official formulation exposes these dependencies. A mode can exist yet be invisible in one geometry because its structure factor vanishes. Raman activity follows derivatives of electronic polarizability with respect to normal coordinates. Crystal symmetry determines allowed tensor components, and polarization geometry selects them. Resonance, temperature, defects, stress, and anharmonicity alter intensities and lineshapes. Comparing only Gamma frequencies ignores whether a mode is Raman active. A finite-displacement dielectric derivative or DFPT response must use consistent eigenvector normalization. Infrared activity follows mode effective charge. Born effective charge tensors projected onto eigenvectors determine oscillator strength, while dielectric screening and LO–TO coupling shape response. Reflectivity, absorption, and loss functions are related but not identical spectra. Damping determines linewidth, and thin-film optics adds interference and substrate effects. A harmonic dielectric function with arbitrary broadening is a model, not a lifetime prediction. Neutron creation and annihilation intensities obey detailed balance. Stokes-like phonon creation scales with $n_B+1$, while annihilation scales with $n_B$. Their ratio provides a temperature check under equilibrium. Multiphoton backgrounds, incoherent scattering, and instrument resolution complicate extraction. Simulating intensities rather than overlaying dispersion curves prevents assigning a weak or forbidden calculated branch to a measured feature. Debye–Waller factors arise from thermally averaged displacement. Harmonic eigenvectors and occupations give anisotropic mean-square displacement tensors, which attenuate diffraction and spectroscopy at large momentum transfer. Zero-point motion contributes at $T=0$. Static disorder and anharmonic motion can produce similar apparent displacement parameters. Comparing calculated and refined tensors tests eigenvectors and frequencies more strongly than heat capacity alone. Thermal diffuse scattering maps correlated phonon displacements in reciprocal space. Intensity concentrates near soft branches and can reveal instabilities away from standard high-symmetry paths. Energy-integrated x-ray or neutron diffuse maps mix modes with frequency and structure-factor weights. Simulating full reciprocal planes helps distinguish phonon diffuse scattering from static disorder and defects. Elastic constants are long-wavelength acoustic derivatives of the same energy surface. Sound velocities follow the Christoffel equation using elastic tensor and density, and must match acoustic slopes when electrostatic and internal-relaxation conventions agree. Relaxed-ion and clamped-ion elastic constants differ because internal coordinates can respond to strain. This comparison is an important low-$q$ verification of force constants. Thermal expansion links anharmonic phonon pressure to elasticity. Each mode contributes a pressure proportional to its Grüneisen parameter and energy. The elastic compliance converts that pressure into anisotropic strain. In low-symmetry materials, scalar volume Grüneisen intuition is insufficient; mode strain derivatives and tensor compliance govern directional expansion. A negative lattice-parameter expansion can coexist with positive volume expansion. Phonon drag occurs when nonequilibrium phonon momentum pulls charge carriers. It can enhance thermopower at temperatures where phonon mean free paths are long and electron–phonon momentum exchange competes with resistive loss. Standard equilibrium-band Seebeck calculations omit this coupled nonequilibrium effect. Modeling requires coupled electron and phonon BTEs or controlled approximations, with sample size and impurity sensitivity. Superconducting electron–phonon theory uses a spectral coupling function rather than phonon density of states alone. The Eliashberg function $\alpha^2F(\omega)$ weights phonons by electronic matrix elements and Fermi-surface phase space. Its integral defines coupling measures and characteristic frequencies entering approximate transition-temperature formulas. Strong coupling, anisotropy, Coulomb pseudopotential, and nonadiabaticity limit simple summaries. Kohn anomalies reveal electronic screening in phonon dispersion. A sharp feature at wave vectors connecting Fermi-surface regions arises because the electronic susceptibility changes nonanalytically. Dense electronic $k$ sampling, smearing control, and electron–phonon response are necessary. A supercell force calculation may converge slowly in real space because the screened interaction is long ranged. Temperature and doping move or weaken the anomaly. Magnetoelastic coupling makes phonons depend on spin order. Changing magnetic configuration alters bonding and force constants; displacements can modulate exchange interactions; spin fluctuations renormalize modes near magnetic transitions. A nonmagnetic phonon calculation of a paramagnet may be qualitatively wrong if local moments persist. Disordered-local-moment, spin–lattice, or ensemble approaches address different time-scale assumptions. Ferroic domain walls and interfaces host vibrational states beyond bulk branches. Broken translation, strain gradients, electrostatics, and reconstruction localize or scatter modes. Supercell modes fold and hybridize, so layer projections and unfolding aid interpretation. Interface thermal conductance depends on transmission, inelastic conversion, roughness, and nonequilibrium, not merely overlap of bulk phonon densities. Nanostructure confinement changes mode spectrum and selection. Thin membranes have Lamb and flexural modes; nanowires mix longitudinal, torsional, and bending motion; nanoparticles have discrete surface-sensitive vibrations. Bulk phonon dispersion sampled at quantized wave vectors is only an approximation when surfaces reconstruct or boundary conditions change forces. Continuum elasticity works at long wavelength, atomistics at atomic scales, and overlap provides validation. Topological phonons classify band geometry of bosonic normal modes. Symmetry-protected crossings, Berry curvature, Chern-like invariants, and boundary modes can arise in dynamical matrices with appropriate symmetries or time-reversal breaking. Eigenvector gauge and mass metric matter. A topological label requires a real frequency gap and stable structure; imaginary branches undermine a conventional harmonic band classification. ```svg ``` | Modeling choice | Physical meaning | Common failure | Decisive check | |---|---|---|---| | reference structure | point about which energy is expanded | residual force or wrong magnetic branch | forces, stress, and alternative structures | | harmonic force constants | curvature and normal modes | short supercell truncates interactions | supercell and commensurate-$q$ convergence | | eigenvector convention | phase and mass normalization | raw vectors compared across codes | frequencies and gauge-invariant projections | | nonanalytic correction | long-range polar dipole coupling | LO–TO splitting omitted or double counted | Born charges, dielectric tensor, direction limit | | Bose statistics | quantum occupation and heat capacity | classical equipartition at low temperature | Debye limit and isotope heat capacity | | quasiharmonic theory | volume-dependent harmonic free energy | used near strongly anharmonic instability | explicit anharmonic or MD comparison | | cubic force constants | three-phonon coupling | noisy forces create false linewidths | displacement, range, mesh, and sum-rule tests | | RTA conductivity | independent mode relaxation | normal-process repopulation discarded | iterative BTE comparison | | spectroscopy simulation | mode intensity under a probe | frequency-only assignment | selection rule and resolution convolution | | imaginary mode | negative harmonic curvature | numerical artifact called phase transition | convergence and frozen-mode energy scan | Verification should begin at the force level. Translate every atom together and confirm zero net restoring force; compare symmetry-related force responses; check Newton-pair reciprocity where applicable; test displacement amplitudes; and tighten the underlying electronic or potential calculation. Force-constant symmetrization should reduce noise without hiding large violations. Archive raw and corrected constants so enforcement remains auditable. The monatomic and diatomic chains provide implementation unit tests with exact dispersions. They verify Fourier phases, cell indexing, mass weighting, branch count, eigenvector normalization, group velocity, density of states, and acoustic limits. Extending to a simple cubic central-force model tests transverse modes and elastic relations. These small tests catch errors before first-principles complexity obscures them. Supercell convergence must target real-space interaction range and the final observable. Increase size and vary shape while holding force accuracy and displacement method fixed. Compare selected frequencies throughout the Brillouin zone, acoustic slopes, free energy, imaginary modes, and thermal conductivity. Third-order constants often require a different range from second-order ones, as Phono3py permits. A converged harmonic spectrum does not establish converged scattering. Reciprocal meshes require independent convergence for thermodynamics and scattering. Heat capacity and free energy integrate smooth frequency weights and may converge on modest meshes. Three-phonon rates integrate sharp energy-conservation surfaces and need denser meshes, tetrahedra, adaptive integration, or controlled broadening. Conductivity can be dominated by a small set of long-lived low-$q$ modes, making finite meshes particularly deceptive. Energy-conservation broadening is a numerical parameter, not a physical linewidth unless derived as such. Gaussian or Lorentzian approximations to delta functions change phase space. Too narrow produces noisy mesh dependence; too broad opens forbidden processes. Extrapolate mesh and width jointly or use tetrahedron methods. Do not report the integration width as the predicted spectral width. Acoustic small-$q$ behavior needs special handling in polar, two-dimensional, and hydrodynamic calculations. Translation, rotational invariance, long-range electrostatics, and continuum slopes constrain the limit. Coarse meshes omit the longest mean-free-path carriers. Analytic Debye-like integration, adaptive sampling, or finite-size extrapolation may be needed. Simply adding the Gamma point does not represent its surrounding phase-space volume correctly. Frozen-mode energy scans verify unstable and anharmonic coordinates. Displace along a mass-consistent eigenvector over positive and negative amplitudes, relax orthogonal coordinates if physically intended, and fit quadratic, cubic, and quartic behavior. A double well supports a symmetry-breaking instability; a shallow asymmetric curve signals coupling or residual force; a purely numerical imaginary mode disappears as force accuracy improves. Temperature-dependent stabilization requires free energy, not only the zero-K curve. Molecular-dynamics validation separates potential error from harmonic approximation. At low temperature and amplitude, mode peaks should approach harmonic frequencies. Increasing temperature reveals shifts and widths. Ensure trajectory length, timestep, ensemble, cell, and potential are converged. Classical MD and quantum experimental spectra differ through occupation and nuclear quantum effects; compare peak positions cautiously and intensities through an appropriate correlation function. Spectroscopic validation should compare calculated intensities and experimental resolution, not only hand-selected frequencies. Apply isotope composition, temperature, pressure, polarization, momentum path, and domain orientation. Convolve intrinsic lineshapes with instrument response. Multiple modes closer than resolution can appear as one peak, and disorder can relax momentum selection. Reserve independent spectra or conditions after calibrating the force model. Transport validation needs geometry and boundary conditions matching the measurement. Bulk conductivity, thin-film in-plane conductivity, cross-plane conductivity, transient grating, time-domain thermoreflectance, and ballistic devices sample different mean-free-path and frequency windows. Interface resistance, electrons, radiation, porosity, grain size, and contacts may contribute. Fitting one boundary specularity to one thickness is not a bulk phonon validation. ```svg ``` ```flowchart Define structure, dimensionality, charge, magnetism, boundary conditions, temperature, and target observable -> Relax forces and stress on the intended electronic or interatomic potential surface -> Choose finite displacement, DFPT, analytic model, or fitted force-constant method -> Converge basis, k mesh, smearing, supercell, displacement, SCF, and force accuracy -> Enforce and audit translational, rotational, space-group, and electrostatic constraints -> Build D(q), diagonalize frequencies and eigenvectors, and track branches or subspaces -> Check acoustic limits, LO–TO terms, imaginary modes, frozen-mode energies, and elastic slopes -> Integrate Bose thermodynamics or add volume-dependent quasiharmonic free energy -> Fit or compute anharmonic force constants and converge linewidth, self-energy, and BTE meshes -> Map modes through Raman, IR, neutron, x-ray, heat-flow, or electron-coupling forward models -> Validate across temperature, isotope, pressure, size, polarization, and momentum -> Archive structures, force constants, conventions, meshes, corrections, hashes, and uncertainty ``` Phonon diagnostics are most effective when divided into structure, forces, harmonic algebra, electrostatics, anharmonic interactions, statistics, transport, and measurement. A Gamma acoustic gap points to force invariance or pinning. A wrong LO–TO split points to Born charges, dielectric response, or boundary model. Negative conductivity contributions point to collision or tensor conventions. A frequency match with wrong Raman activity points to eigenvectors or response derivatives. Retuning one spring constant should follow these checks, not replace them. | Symptom | Likely layer | Targeted investigation | |---|---|---| | acoustic modes do not approach zero | translational sum rule or force noise | raw force-constant row sums and tighter forces | | imaginary mode disappears with supercell | force-range truncation | cell-shape and size sequence | | LO frequency depends on vacuum | inappropriate three-dimensional electrostatics | 2D Coulomb treatment and directional limit | | branches swap erratically | frequency-only sorting near crossings | eigenvector/subspace overlap and symmetry labels | | free energy changes with path mesh | path used instead of volume integration | full weighted Brillouin-zone mesh | | linewidth scales with chosen broadening | unresolved conservation surface | joint mesh–width convergence | | RTA and iterative conductivity diverge | strong normal-process repopulation | collision-matrix eigenmodes and hydrodynamic scales | | calculated mode is absent experimentally | selection rule or weak structure factor | probe-specific intensity and geometry | The acoustic sum rule should be tested before and after correction. Report maximum raw translation violation, correction norm, and changes in target modes. A large correction can make a dispersion look plausible while masking insufficient SCF, small supercell, inconsistent long-range subtraction, or a force bug. Rotational constraints are similarly essential for flexural modes whose incorrect linearization can dominate two-dimensional heat capacity and transport. Force-constant units and mass conventions deserve an explicit ledger. Codes may store energy per squared length, force per displacement, or already mass-weighted matrices. Frequencies may be angular frequency, cycles per second, wavenumber, or energy. Eigenvectors may include $1/\sqrt M$ or not. Convert through one dimensional audit and validate against a chain model before combining data from different tools. Primitive, conventional, and supercell mappings determine branch count and phases. A conventional cell produces folded modes compared with a primitive cell. Force calculators, phonon tools, and visualization must share atom mapping and lattice conventions. Apparent extra optical branches may be folded acoustic branches. Unfolding recovers spectral weight but does not change the underlying supercell eigenproblem. Nonstoichiometric, charged, or metallic structures complicate long-range response. Charged supercells carry compensating backgrounds; polar metals screen macroscopic fields differently from insulators; defects break translational symmetry; free carriers screen LO modes and can create coupled phonon–plasmon excitations. Applying an insulating Born-charge correction blindly is inappropriate. The electronic boundary and carrier state are part of the lattice model. Temperature-dependent effective potentials offer a route between harmonic theory and molecular dynamics. Fit force constants to finite-temperature force ensembles, stochastic configurations, or MD trajectories, then diagonalize an effective dynamical matrix. The resulting modes are temperature-dependent quasiparticles if their peaks remain identifiable. Fit order, training distribution, regularization, and residual force correlations control meaning. An excellent force fit does not guarantee correct free energy unless sampling and entropy are consistent. Free-energy integration handles anharmonicity beyond quasiharmonic theory. Thermodynamic integration connects a reference harmonic or machine-learned potential to the target potential by averaging energy differences along a coupling path. Reversible scaling and temperature integration provide alternatives. Phase transitions and poor overlap require staged paths. Statistical and integration error must be below the phase free-energy difference, often only a few meV per atom. Isotope effects provide a clean mass-sensitive validation. Harmonic frequencies scale approximately as $M^{-1/2}$ for modes localized on the substituted species, while force constants remain nearly unchanged in the Born–Oppenheimer limit. Zero-point volume and anharmonic renormalization add smaller deviations. Isotope thermal conductivity tests mass-disorder scattering. Failure of these trends can identify normalization or disorder errors independently of electronic bonding. Pressure-dependent phonons test volume derivatives and phase stability. Hydrostatic compression typically stiffens positive-Grüneisen modes, while selected modes soften toward pressure-induced transitions. Compare at relaxed cells under the same stress and include nonhydrostatic experimental conditions where relevant. Raman pressure coefficients, elastic constants, and equation of state jointly constrain the potential more strongly than ambient dispersion alone. Finite temperature can broaden the concept of a phonon beyond a sharp quasiparticle. When linewidth is much smaller than frequency and neighboring separation, a Lorentzian quasiparticle is meaningful. Strong damping, central peaks, relaxational dynamics, and overlapping branches require full spectral functions or correlation matrices. Reporting a lifetime for an overdamped mode creates false precision. Wigner or Green-function descriptions can bridge particle and coherence regimes. Amorphous solids support vibrations but not exact crystal-momentum phonons. Normal modes of a finite disordered structure include propagons, diffusons, and locons under useful classifications. Allen–Feldman-like diffusivity describes harmonic mode coupling where group velocity is ill-defined; anharmonicity adds temperature dependence. Calling every amorphous vibration a phonon is common shorthand, but crystal BTE formulas need justification. Liquids lack a stable reference lattice over long times, yet vibrational spectra and collective density modes remain measurable. Instantaneous normal modes, velocity correlations, and dynamic structure factors describe short-time motion. Imaginary instantaneous modes do not mean a crystalline instability in the same sense. Solid phonon thermodynamics should not be transplanted directly across melting. Machine-learned interatomic potentials can make anharmonic phonon studies tractable at large scale. They must reproduce energies, forces, stresses, harmonic and anharmonic derivatives across strained, displaced, thermal, and defect configurations. A low random-test force RMSE may miss rare configurations controlling scattering. Compare force constants, dispersion, Grüneisen parameters, linewidths, and phase free energies against the electronic reference. Reproducibility requires the complete lattice-dynamics ledger: reference cell and atom mapping; masses and isotopes; force method and electronic settings; displacement patterns and amplitudes; supercells; raw and symmetrized force constants; phase and eigenvector conventions; nonanalytic correction; $q$ meshes and paths; thermodynamic normalization; anharmonic cutoffs; broadening; BTE solver; boundary scattering; and probe-specific postprocessing. Plots alone cannot reconstruct a calculation. The final interpretation should separate displacement pattern, frequency, occupation, group velocity, lifetime, mean free path, and probe intensity. A high-frequency optical mode can carry little heat; a low-frequency acoustic mode can be invisible in Raman; a flat mode can have high density of states but small velocity; a large linewidth can destroy the quasiparticle picture. No single phonon dispersion plot contains all of phonon physics. Read phonon theory through a force-constant-mode-statistics-and-interaction lens rather than an atoms-as-bouncing-balls lens.