Drude Model for Metal Electrodynamics

• The Drude model is a classical free-carrier theory that defines metal conductivity and optical response using effective mass, relaxation time, and plasma frequency.
• It serves as a baseline for extended formulations incorporating memory effects, nonlocal hydrodynamics, and atomistic approaches to capture nanoscale phenomena.
• The model underpins applications in plasmonics and electromagnetic propagation, while its limitations spark debates in Casimir physics and relativistic corrections.

The Drude model is the canonical free-carrier description of electrical and optical response in metals and doped media. In its classical form it treats itinerant electrons as carriers of density $n$ and mass $m$ (or $m^*$) with a constant relaxation time $\tau$ and plasma frequency $\omega_p$; in harmonic fields it yields the familiar conductivity and permittivity

$$\sigma(\omega)=\frac{n e^2\tau}{m}\frac{1}{1-i\omega\tau},\qquad \varepsilon(\omega)=\varepsilon_\infty-\frac{\omega_p^2}{\omega(\omega+i\gamma)}, \quad \gamma=\tau^{-1},$$

with equivalent cgs forms differing by the standard $4\pi$ factors (Riffe, 2018, Hiremath et al., 2012). In current research the model functions both as a baseline constitutive law and as the seed for extended Drude, Drude–Lorentz, hydrodynamic, multi-fluid, atomistic, fractional, time-varying, and transport-effective formulations, each introduced to capture effects absent from the local constant-$\tau$ picture (Benfatto et al., 2011, Maack et al., 2017, Ganfornina-Andrades et al., 3 Feb 2026).

1. Classical free-carrier formulation

In its minimal stationary form, the model assumes that a carrier of charge $q=-e$ and effective mass $m^*$ obeys

$m$0

with current density and polarization given by $m$1 and $m$2 (Ganfornina-Andrades et al., 3 Feb 2026). For a harmonic field, the conductivity may be written in SI form as

$m$3

or in cgs form as

$m$4

with $m$5 (Ganfornina-Andrades et al., 3 Feb 2026, Benfatto et al., 2011).

The same model can be expressed through the dielectric function. In the convention emphasized for optical propagation, the free-carrier conductivity and dielectric function are linked by

$m$6

and, at low enough frequency, $m$7, so the Drude dielectric function becomes

$m$8

This form makes explicit that the Drude weight is set by $m$9 and the linewidth by $m^*$0 (Riffe, 2018, Hiremath et al., 2012).

The model also admits a direct time-domain interpretation. The conductivity kernel is exponential,

$m^*$1

so the current is a causal convolution of the field with a single decay time. This single-timescale structure is the defining simplification of the classical Drude picture (Bonança et al., 2020).

2. Optical response, propagation, and plasma-edge physics

A Drude metal is usually interpreted through the interconnected response functions $m^*$2, $m^*$3, $m^*$4, and $m^*$5. For nonmagnetic media,

$m^*$6

with normal-incidence reflectance

$m^*$7

and skin depth

$m^*$8

These relations convert the Drude dielectric function into propagation, attenuation, and impedance data (Riffe, 2018).

The sign of $m^*$9 separates propagating from evanescent regimes. Below the plasma edge, $\tau$0, the wavevector is predominantly imaginary, and the field decays evanescently with high reflectivity. Including a bound-electron background shifts the transmission threshold to

$\tau$1

so that above $\tau$2 the metal becomes transparent in the collisionless limit (Riffe, 2018).

In the good-conductor low-frequency regime, $\tau$3, one recovers the standard skin-effect asymptotics

$\tau$4

whereas in the collisionless regime, $\tau$5, the dielectric function approaches $\tau$6 with a subleading imaginary correction of order $\tau$7 (Riffe, 2018).

The same formalism governs charge relaxation. Using the Drude conductivity kernel in the continuity equation gives

$\tau$8

For $\tau$9, charge fluctuations relax oscillatory at $\omega_p$0 with envelope $\omega_p$1; for $\omega_p$2, they decay essentially with the macroscopic relaxation time $\omega_p$3 (Riffe, 2018).

3. Extended Drude, memory functions, and interband separation

Once many-body effects are retained, the constant-$\omega_p$4 model is replaced by an extended Drude description with frequency-dependent scattering and mass. In cgs notation, the inversion formulas are

$\omega_p$5

equivalently

$\omega_p$6

with $\omega_p$7 and $\omega_p$8 (Benfatto et al., 2011).

For multiband pnictides this inversion is only meaningful after explicit interband subtraction. In LaFePO, density-functional calculations identify a narrow interband peak at $\omega_p$9–$$\sigma(\omega)=\frac{n e^2\tau}{m}\frac{1}{1-i\omega\tau},\qquad \varepsilon(\omega)=\varepsilon_\infty-\frac{\omega_p^2}{\omega(\omega+i\gamma)}, \quad \gamma=\tau^{-1},$$0 and a broad interband continuum from $$\sigma(\omega)=\frac{n e^2\tau}{m}\frac{1}{1-i\omega\tau},\qquad \varepsilon(\omega)=\varepsilon_\infty-\frac{\omega_p^2}{\omega(\omega+i\gamma)}, \quad \gamma=\tau^{-1},$$1 to $$\sigma(\omega)=\frac{n e^2\tau}{m}\frac{1}{1-i\omega\tau},\qquad \varepsilon(\omega)=\varepsilon_\infty-\frac{\omega_p^2}{\omega(\omega+i\gamma)}, \quad \gamma=\tau^{-1},$$2. Using the experimental plasma frequency $$\sigma(\omega)=\frac{n e^2\tau}{m}\frac{1}{1-i\omega\tau},\qquad \varepsilon(\omega)=\varepsilon_\infty-\frac{\omega_p^2}{\omega(\omega+i\gamma)}, \quad \gamma=\tau^{-1},$$3 reported by Qazilbash et al., the intraband-only analysis gives $$\sigma(\omega)=\frac{n e^2\tau}{m}\frac{1}{1-i\omega\tau},\qquad \varepsilon(\omega)=\varepsilon_\infty-\frac{\omega_p^2}{\omega(\omega+i\gamma)}, \quad \gamma=\tau^{-1},$$4, whereas the total conductivity without interband subtraction yields $$\sigma(\omega)=\frac{n e^2\tau}{m}\frac{1}{1-i\omega\tau},\qquad \varepsilon(\omega)=\varepsilon_\infty-\frac{\omega_p^2}{\omega(\omega+i\gamma)}, \quad \gamma=\tau^{-1},$$5; the paper characterizes this as a “substantial failure” of the extended-Drude-model analysis on raw optical data (Benfatto et al., 2011).

A related development is the extended Drude–Lorentz model, in which the measured conductivity is decomposed into one correlated intraband extended Drude channel plus Lorentz oscillators for interband structure. Applied to K-doped $$\sigma(\omega)=\frac{n e^2\tau}{m}\frac{1}{1-i\omega\tau},\qquad \varepsilon(\omega)=\varepsilon_\infty-\frac{\omega_p^2}{\omega(\omega+i\gamma)}, \quad \gamma=\tau^{-1},$$6 at $$\sigma(\omega)=\frac{n e^2\tau}{m}\frac{1}{1-i\omega\tau},\qquad \varepsilon(\omega)=\varepsilon_\infty-\frac{\omega_p^2}{\omega(\omega+i\gamma)}, \quad \gamma=\tau^{-1},$$7, $$\sigma(\omega)=\frac{n e^2\tau}{m}\frac{1}{1-i\omega\tau},\qquad \varepsilon(\omega)=\varepsilon_\infty-\frac{\omega_p^2}{\omega(\omega+i\gamma)}, \quad \gamma=\tau^{-1},$$8, $$\sigma(\omega)=\frac{n e^2\tau}{m}\frac{1}{1-i\omega\tau},\qquad \varepsilon(\omega)=\varepsilon_\infty-\frac{\omega_p^2}{\omega(\omega+i\gamma)}, \quad \gamma=\tau^{-1},$$9, and $4\pi$0, this framework yields a dome-like mass renormalization without pseudogap corrections and, at optimal doping $4\pi$1, $4\pi$2, corresponding to coherent $4\pi$3 and incoherent $4\pi$4 intraband spectral weight. When pseudogap effects are included for the underdoped samples, the reported values rise to $4\pi$5 and $4\pi$6 (Lee et al., 2021).

In the superconducting state, a further caution appears. A reverse-process analysis of superconducting optical spectra shows that constructing a “residual” optical self-energy by subtracting the condensate from $4\pi$7 produces a quantity that is not Kramers–Kronig paired and yields an unphysical negative optical effective mass: in the clean example it is negative below $4\pi$8, and in the dirty example below $4\pi$9. The conclusion is that the residual optical self-energy is difficult to interpret as a physically meaningful quasiparticle self-energy (Hwang, 2019).

4. Nonlocal hydrodynamic and multi-fluid Drude theories

Spatial dispersion is incorporated by the hydrodynamic Drude model, where the electric field couples to a polarization current $\tau$0 inside the metal:

$\tau$1

$\tau$2

Here $\tau$3 is the hydrodynamic parameter. The grad–div term is essential: replacing it by a curl-free approximation produces spurious resonances below $\tau$4 (Hiremath et al., 2012).

The rigorous weak formulation places $\tau$5 in $\tau$6 and $\tau$7 in $\tau$8, leading directly to a Nédélec finite-element discretization. Benchmarks against Ruppin’s nonlocal Mie theory for a cylindrical nanowire of radius $\tau$9 give a local-drude localized surface plasmon at $q=-e$0 and a nonlocal main resonance at $q=-e$1, followed by higher-order hydrodynamic resonances above $q=-e$2 at $q=-e$3, $q=-e$4, $q=-e$5, $q=-e$6, and higher. In a V-groove channel-plasmon-polariton geometry, the main resonances shift from $q=-e$7 and $q=-e$8 in the local model to $q=-e$9 and $m^*$0 in the hydrodynamic model (Hiremath et al., 2012).

For semiconductors, a single carrier fluid is often insufficient. The two-fluid hydrodynamic model assigns each carrier species its own $m^*$1, $m^*$2, and $m^*$3, so that

$m^*$4

This produces two longitudinal branches: an optical mode with finite $m^*$5 at $m^*$6 and an acoustic mode with $m^*$7. In nanospheres, the acoustic branch gives extinction peaks below the localized surface plasmon that are absent in the single-fluid hydrodynamic Drude model (Maack et al., 2017).

5. Nanoscale, atomistic, and radiative reformulations

At subnanometer scales, one route is to port Drude conduction onto an explicit atomic lattice. The $m^*$8FQ method represents each atom by a fluctuating charge $m^*$9 and couples neighboring atoms through a Drude-like exchange matrix and classical electrostatics. In stretched $m$00 nanorods and $m$01 dimers it reproduces charge-transfer plasmons, boundary dipolar plasmons, monoatomic-junction behavior, and jump-to-contact discontinuities with quantitative fidelity to TDDFT/RPA benchmarks, while remaining computationally scalable (Giovannini et al., 2019).

Another nanoscale issue concerns radiation damping of point dipoles. For Lorentz–Drude dipoles in the radiative limit, the damping term in the material equation should contain only non-radiative loss if the dipole is driven by the total field at its location. In that case the effective polarizability is

$m$02

and for $m$03 the optical theorem enforces $m$04 and $m$05. Self-consistent FDTD implementations confirm that purely radiative dipoles should be propagated with $m$06 in the oscillator equation when the drive is the total field (Wang et al., 10 Apr 2025).

For small gold nanoparticles, a different modification introduces a confinement restoring force. The model writes

$m$07

with a fitted confinement coefficient

$m$08

where $m$09, $m$10, $m$11, and $m$12. In this fitted range, decreasing diameter increases the restoring force and makes the classical confinement effect more dominant; increasing nanoparticle diameter increases the real part of the permittivity and decreases the imaginary part (Kheirandish et al., 2019).

6. Time-varying, fractional, and generalized constitutive frameworks

A recent generalization allows the carrier density $m$13, effective mass $m$14, and collision rate $m$15 to vary explicitly in time. The single-particle equation becomes

$m$16

which leads to closed response kernels in mixed time–frequency, two-times, and two-frequency form. In this framework, non-adiabatic modulation and time-dependent losses induce “temporal blurring,” “selective gating and suppression,” and “low-frequency spectral reshaping.” In the static limit it reduces exactly to the classical Drude permittivity (Ganfornina-Andrades et al., 3 Feb 2026).

A different generalization replaces integer derivatives by fractional ones:

$m$17

so that

$m$18

The model retains the classical Drude form for $m$19, while enabling power-law frequency dependence of the absorption coefficient and wavevector that is common in complex media (Karpiński et al., 2021).

At a more abstract level, the dielectric function can be treated as a meromorphic transfer function and expanded by the Singularity Expansion Method. The resulting generalized Drude–Lorentz model contains a pole at the origin, a sum of Drude terms, and generalized Lorentz terms with numerator

$m$20

rather than the classical Lorentz numerator alone. This form complies with causality, Hermitian symmetry, and the complex-pole structure of passive linear media. Reported fits span metals, dielectrics, and graphene; for Au the fitted model used $m$21, $m$22 and achieved $m$23 and $m$24 (Soltane et al., 2024).

7. Effective scope, transport bounds, and controversies

The Drude form is not a universal microscopic law. In the Drude–Kadanoff–Martin transport model, the low-energy density correlator implies

$m$25

with collective mean free path $m$26. For lattice models with local and bounded interactions, sharp bounds on the retarded density Green’s function show that the model cannot pertain at microscopic energy scales because lattice spectral weight is exponentially suppressed at high frequency. The same analysis yields a lower bound on $m$27, implying that systems with $m$28 much shorter than the lattice length scale cannot have conventional Drude peaks (Chowdhury et al., 22 Sep 2025). This suggests that the Drude peak should be regarded as an emergent low-energy feature whose validity is itself constrained.

The same effective viewpoint appears in holography. In a large-$m$29 axion model of momentum relaxation, the leading AC thermal conductivity is exactly Drude,

$m$30

and the theory exhibits a coherent-to-incoherent crossover near $m$31 (Andrade et al., 2015). The Drude response there is not fundamental electron kinetics; it is the manifestation of a single long-lived quasinormal mode.

Thermodynamic interpretation also has limits. For the classical Drude, Drude–Sommerfeld, and extended Drude–Sommerfeld models, one analysis shows that under time-dependent driving the instantaneous entropy production rate $m$32 can become negative, while the time-averaged rate and total entropy remain non-negative. The result does not contradict passivity because $m$33 and the integrated entropy is positive whenever $m$34 (Bonança et al., 2020).

The most persistent controversy concerns Casimir physics. In the Lifshitz formulation, the Drude prescription suppresses the transverse-electric zero Matsubara mode, and one 2011 analysis argues that the standard Matsubara insertion of the Drude permittivity contains a contribution non-perturbative in the relaxation parameter $m$35, producing a residual entropy and violating the Nernst theorem (Bordag, 2011). Klimchitskaya and Mostepanenko argue more specifically that the disagreement between Drude-based Casimir predictions and precision measurements originates in the low-frequency transverse-electric evanescent sector, not in dissipation as such; for Au at $m$36 they report a plasma-to-Drude pressure ratio growing from $m$37 at $m$38 to $m$39 at $m$40, approaching the classical factor of $m$41 (Klimchitskaya et al., 2023).

A separate, explicitly critical line of argument concerns relativity. One 2016 paper argues that the standard neutral-wire picture of a current-carrying conductor is incompatible with special relativity because Lorentz contraction changes the electron linear charge density while leaving the ionic lattice density unchanged. In that analysis the wire acquires a second-order radial electric field

$m$42

directed toward the conductor, with magnitude of order $m$43 (R et al., 2016). This claim is presented not as a rejection of Maxwell–Lorentz electrodynamics, but as a criticism of the nonrelativistic assumptions built into the classical Drude background model.

Taken together, these developments define the modern status of the Drude model. It remains the standard free-carrier constitutive law, but in contemporary use it is rarely taken as complete: multiband optics require interband separation, nanoscale electrodynamics requires nonlocality or atomistic discretization, ultrafast photonics requires time-varying kernels, and transport theory places explicit limits on where a conventional Drude peak can exist (Benfatto et al., 2011, Hiremath et al., 2012, Chowdhury et al., 22 Sep 2025).