A phonon is a quantized unit of vibrational energy in a crystal lattice — the vibrational counterpart to the photon of light. Collective vibrations of atoms in a solid come in discrete packets of mechanical energy called phonons. They are the primary carriers of heat and sound in insulators and semiconductors, and they control thermal expansion, structural stability, conventional superconductivity, and much of infrared and Raman spectroscopy.
From Atomic Vibrations to Quantized Phonons
Atoms in a crystal vibrate around equilibrium, and because they are coupled by chemical bonds those vibrations propagate as waves. Lattice dynamics starts from a simple fact: the natural motions of the crystal are collective. A single atom’s displacement is tied to its neighbors, so the independent degrees of freedom are extended, wave-like normal modes, not isolated atomic wiggles.
In the harmonic approximation the potential is expanded to second order in atomic displacements. Transforming to wavevector \(\mathbf{q}\) and diagonalizing yields independent modes labeled by \(\mathbf{q}\) and branch index \(\nu\), each with frequency \(\omega_{\nu}(\mathbf{q})\).
Quantum mechanics quantizes each mode. A harmonic oscillator of angular frequency \(\omega\) has levels
$$ E_n = \bigl(n + \tfrac{1}{2}\bigr) \hbar \omega $$with \(n = 0, 1, 2, \ldots\). Each quantum \(\hbar \omega\) is one phonon. Raising \(n\) by one adds a phonon: the amplitude increases by one quantum while the spatial pattern (the eigenvector) stays the same. The zero-point term \(\tfrac{1}{2}\hbar\omega\) remains even at \(T = 0\).
This particle language is practical: instead of tracking \(10^{23}\) coupled atoms, we treat the solid as a gas of phonons. In thermal equilibrium the mean occupation follows the Bose–Einstein distribution
$$ n(\omega, T) = \frac{1}{e^{\hbar\omega / k_B T} - 1} $$so high-frequency modes stay mostly empty until \(k_B T \sim \hbar\omega\). That fact underlies the temperature dependence of heat capacity, thermal conductivity, and spectroscopic intensities.
Acoustic vs Optical Phonon Branches
A primitive cell with \(p\) atoms has \(3p\) degrees of freedom and therefore \(3p\) phonon branches in three dimensions. These split into two families with different long-wavelength behavior.
flowchart TD A[Unit cell with p atoms] --> B[3p phonon branches] B --> C[3 acoustic] B --> D[3p − 3 optical] C --> E[In-phase motion<br/>ω → 0 as q → 0] D --> F[Out-of-phase motion<br/>finite ω at q = 0] E --> G[Heat transport, sound] F --> H[IR / Raman, polar modes] C --> I[LA + 2 TA] D --> J[LO + TO branches]
- Acoustic phonons: neighboring atoms move in phase. At long wavelength this is an ordinary elastic wave — a sound wave. Frequency goes to zero as \(\mathbf{q} \to 0\): a uniform translation of the crystal costs no restoring force.
- Optical phonons: neighboring atoms (or sublattices) move out of phase. In an ionic crystal that relative motion creates an oscillating dipole that can couple to light — hence “optical.” They retain a finite frequency at the zone center.
There are always exactly 3 acoustic branches and \(3p - 3\) optical branches. Monoatomic Bravais lattices (\(p = 1\)) have only acoustic branches; silicon and zincblende (\(p = 2\)) have 3 acoustic + 3 optical. Each family further splits by polarization: longitudinal (motion parallel to \(\mathbf{q}\): LA, LO) and transverse (perpendicular: TA, TO — two each in 3D).
| Property | Acoustic phonons | Optical phonons |
|---|---|---|
| Atomic motion | In phase | Out of phase |
| Frequency at \(\mathbf{q} \to 0\) | \(\omega \to 0\) | Finite, nonzero |
| Physical analogy | Sound / elastic waves | Sub-lattice oscillation; IR/Raman |
| Dominant role | Heat transport, sound, elasticity | Light absorption, Raman, polaritons |
| Number of branches (3D) | 3 | \(3p - 3\) |
In polar materials, LO and TO modes split at \(\Gamma\) because a polar LO displacement produces a macroscopic electric field. Capturing that LO–TO splitting requires the dielectric tensor and Born effective charges — quantities DFPT supplies directly.
Phonon Dispersion Relations
The map from wavevector to frequency is the phonon dispersion \(\omega_{\nu}(\mathbf{q})\). Plotting \(\omega\) along high-symmetry paths through the Brillouin zone gives the phonon band structure — the vibrational analog of electronic bands.
Near the zone center, acoustic branches are approximately linear, \(\omega_{\mathrm{ac}}(\mathbf{q}) \approx v_s |\mathbf{q}|\), with sound velocity \(v_s\) fixed by elastic constants and mass density. LA slopes are typically steeper than TA. Optical branches are flatter near \(\Gamma\); their zone-center frequencies appear in IR and Raman spectra (subject to selection rules).
Dispersion curves encode sound velocities, optical frequencies for IR/Raman comparison, anisotropy along crystal directions, soft modes that dip toward zero (often precursors of a phase transition), and imaginary frequencies (plotted negative by convention) that signal dynamical instability.
The full spectrum is summarized by the phonon density of states
$$ g(\omega) = \sum_{\nu} \int_{\mathrm{BZ}} \frac{d^{3}q}{(2\pi)^{3}}\, \delta\bigl(\omega - \omega_{\nu}(\mathbf{q})\bigr) $$which counts modes per unit frequency. Free energy, entropy, and heat capacity are integrals over \(g(\omega)\) weighted by Bose factors.
Thermodynamics from the Phonon Spectrum
Once \(g(\omega)\) is known in the harmonic approximation, standard thermodynamic quantities follow without further electronic-structure work.
Heat capacity. Each mode contributes
$$ C_V(\omega, T) = k_B \left(\frac{\hbar\omega}{k_B T}\right)^{2} \frac{e^{\hbar\omega / k_B T}}{\bigl(e^{\hbar\omega / k_B T} - 1\bigr)^{2}} $$and the solid’s vibrational \(C_V\) is that weight integrated against \(g(\omega)\). At low \(T\) only long-wavelength acoustic modes are excited; the Debye model approximates them as linear sound waves and predicts \(C_V \propto T^{3}\) for insulators. At high \(T\), \(C_V\) approaches the classical Dulong–Petit limit \(3Nk_B\) per formula unit of \(N\) atoms (before anharmonic and electronic corrections).
Free energy and phase stability. Harmonic free energy and entropy are closed-form integrals over \(g(\omega)\). Comparing free energies of competing polymorphs versus temperature is standard when electronic energies alone are close.
Thermal expansion. A purely harmonic crystal would not expand with temperature. Real expansion comes from anharmonicity. Mode Grüneisen parameters \(\gamma_{\nu}(\mathbf{q}) = -\partial\ln\omega_{\nu}/\partial\ln V\) measure how each frequency shifts with volume; positive average \(\gamma\) yields positive expansion, while negative \(\gamma\) in some modes can produce zero or negative expansion.
Superconductivity. In conventional BCS superconductors, phonons mediate the attraction that binds Cooper pairs. Electron–phonon coupling \(\lambda\) and a characteristic phonon frequency set estimates of \(T_c\) — a natural DFPT extension beyond pure lattice dynamics.
Thermal Conductivity: Phonons as Heat Carriers
In metals, electrons usually dominate heat transport. In insulators and many semiconductors, phonons carry essentially all of the heat. In a kinetic-theory picture the lattice thermal conductivity is
$$ \kappa_L \sim \sum_{\nu,\mathbf{q}} C_{\nu\mathbf{q}}\, v_{\nu\mathbf{q}}^{2}\, \tau_{\nu\mathbf{q}} $$Three ingredients matter: heat capacity (which modes are occupied), group velocity \(v = \nabla_{\mathbf{q}}\omega\) (steep acoustic branches carry heat; flat optical branches often do not), and lifetime \(\tau\).
At intermediate and high \(T\) in pure crystals, the dominant resistance is phonon–phonon scattering from anharmonicity. Umklapp processes (those that do not conserve crystal momentum modulo a reciprocal-lattice vector) produce intrinsic thermal resistance and cause \(\kappa_L\) to fall with increasing \(T\). Defects, isotopes, grain boundaries, and surfaces cut lifetimes further.
That is why thermoelectric design often aims to scatter phonons while preserving electronic transport: heavy atoms, complex cells, rattler modes, nanostructuring, and alloy disorder all suppress \(\kappa_L\). High-\(\kappa\) materials (diamond, cubic BN) combine stiff bonds, light atoms, and weak anharmonicity — signatures already visible in the dispersion and DOS. Harmonic DFPT or finite-displacement calculations give frequencies and velocities; lifetimes need anharmonic force constants. Wrong acoustic slopes still mean wrong \(\kappa_L\).
How Phonons Are Computed: DFPT vs Finite Displacement
Phonons follow from interatomic force constants — second derivatives of the total energy with respect to atomic displacements. In first-principles work those derivatives come from density functional theory (DFT). Two approaches dominate.
Density functional perturbation theory (DFPT) treats a phonon displacement as a perturbation and solves a self-consistent linear-response problem for the change in density and wavefunctions. Dynamical matrices can be obtained at essentially any wavevector \(\mathbf{q}\) using only the primitive cell — a phonon of wavevector \(\mathbf{q}\) couples electronic states differing by \(\mathbf{q}\) without a real-space supercell. DFPT is natural for polar materials: dielectric constants, Born charges, and LO–TO corrections fall out of the same machinery.
Finite displacement (frozen phonon) methods displace atoms by small amounts in a supercell, compute Hellmann–Feynman forces, and assemble the force-constant matrix numerically. The approach is simple and pairs cleanly with tools such as Phonopy, but the supercell must be large enough to contain the range of the force constants and resolve the \(\mathbf{q}\)-points of interest.
Both routes build the dynamical matrix \(D(\mathbf{q})\) and solve
$$ D(\mathbf{q})\, \mathbf{e}_{\nu}(\mathbf{q}) = \omega_{\nu}^{2}(\mathbf{q})\, \mathbf{e}_{\nu}(\mathbf{q}) $$where \(\mathbf{e}_{\nu}\) is the mass-weighted pattern of atomic motion. Frequencies are square roots of the eigenvalues; imaginary \(\omega\) means a negative eigenvalue of \(D\).
| Method | Forces / response | Large supercell? | Strengths |
|---|---|---|---|
| DFPT | Analytic linear response | Often no (primitive cell) | Arbitrary \(\mathbf{q}\); polar response; efficient dispersions |
| Finite displacement | Numerical forces from displacements | Yes | Simple workflow; Phonopy-friendly; flexible for complex cells |
DFPT wins for dense dispersions and polar crystals when a mature implementation exists. Finite displacement wins with a fast force engine, zone-center-only needs, or incomplete DFPT. Hybrid workflows are common: obtain real-space force constants on a coarse grid or supercell, then Fourier-interpolate to arbitrary \(\mathbf{q}\) for bands and DOS.
Phonons in Quantum ESPRESSO: the ph.x Path
In Quantum ESPRESSO, the standard DFPT route centers on ph.x, with the ground-state plane-wave code and post-processing tools around it:
flowchart LR A["pw.x<br/>relax + tight SCF"] --> B["ph.x<br/>DFPT on q-grid"] B --> C["Dynamical matrices"] C --> D["q2r.x<br/>real-space IFCs + ASR"] D --> E["matdyn.x<br/>dispersion"] D --> F["matdyn.x<br/>DOS + thermo"]
- Geometry and SCF (
pw.x) — Relax first. Residual forces produce spurious soft or imaginary modes. The phonon SCF needs a tight electronic threshold; linear response inherits ground-state error. - DFPT (
ph.x) — Compute dynamical matrices on a uniform \(\mathbf{q}\)-grid (ldisp = .true.withnq1,nq2,nq3) or at selected \(\mathbf{q}\)-points. For insulators and polar compounds, compute the dielectric tensor and Born charges for non-analytic LO–TO corrections. Response thresholds (tr2_ph) are typically far tighter than ordinary SCF tolerances. - Interatomic force constants (
q2r.x) — Fourier-transform dynamical matrices into real-space force constants. Enforce the acoustic sum rule so the three acoustic branches go to zero at \(\Gamma\). - Dispersion, DOS, thermodynamics (
matdyn.x) — Interpolate force constants along a high-symmetry path for dispersion, or onto a dense mesh for \(g(\omega)\) and harmonic free energy, entropy, and heat capacity.
Practical input decks and LO–TO setup are in our tutorial on phonon calculations with DFPT in Quantum ESPRESSO. Finite-displacement workflows remain useful when you prefer supercell forces and tools such as Phonopy.
Two points often decide trustworthiness. Convergence hierarchy: phonons are second derivatives — settings fine for total energies can still shift optical frequencies by several cm\(^{-1}\) or leave acoustic branches slightly imaginary at \(\Gamma\). Method cost: DFPT with ph.x scales with \(\mathbf{q}\)-points and irreducible representations; finite displacement scales with supercell size. For large cells, that trade-off — and GPU-accelerated kernels — matters as much as the formal method.
Imaginary Frequencies and Structural Stability
One of the highest-value uses of phonon calculations is not spectroscopy but stability. Because \(\omega^{2}\) is an eigenvalue of the dynamical matrix, a negative eigenvalue yields an imaginary frequency. Codes conventionally plot imaginary modes as negative frequencies on a dispersion figure.
An imaginary mode means the energy decreases when atoms move along that eigenvector: the structure is not a local minimum — a dynamical instability, not a numerical curiosity. Typical meanings include:
- A structural phase transition to a lower-symmetry phase (for example, a soft optical mode that freezes into a ferroelectric distortion).
- A high-temperature or epitaxial structure stabilized only by entropy, strain, or kinetics.
- An incorrectly chosen magnetic order, charge order, or lattice setting.
- Incomplete relaxation or inadequate settings that create spurious imaginary modes near \(\Gamma\) — always check acoustic sum-rule enforcement and convergence before claiming an instability.
A spectrum with all real frequencies throughout the Brillouin zone is the standard proof of dynamical stability. It does not prove global thermodynamic stability against decomposition into other phases, but it is a necessary filter: a structure that fails the phonon test should not be trusted as a metastable compound without a clear stabilizing mechanism.
Soft modes that approach zero without going imaginary mark the approach to a continuous transition. Which \(\mathbf{q}\)-point goes soft (zone center vs zone boundary) distinguishes ferroelectric-type instabilities from antiferrodistortive or charge-density-wave-like patterns.
For more on the DFT engines behind these workflows, see our computational engines page and related posts on the blog.
Run it on Simatra
Simatra computes phonon dispersions, phonon densities of states, thermal properties, and electron-phonon coupling using DFPT and finite-displacement workflows on GPU-accelerated clusters built for materials modeling. Our GPU-Opt-V2 instances deliver up to 5x faster convergence and handle supercells of up to ~2,000 atoms — the scale finite-displacement phonon calculations demand. Run the open-source Quantum ESPRESSO engine or Simatra’s native KRONOS engine, a C++20, GPL-3.0 DFT code with CUDA/HIP/Metal backends, described on our computational engines page. Start a free trial with $100 in credits at app.simatra.io and map the lattice dynamics of your material today.