Self-Consistent Time-Dependent Harmonic Approximation

• The paper introduces SCTDHA as a Gaussian method that replaces complex nonlinear dynamics with a self-consistently determined quadratic Hamiltonian.
• It computes time-dependent observables such as correlation functions and spectral responses across diverse quantum and classical systems.
• The technique is applied to out-of-equilibrium sine-Gordon models, anharmonic phonon dynamics, and magnetic resonance, highlighting its practical accuracy and limitations.

Searching arXiv for recent and foundational papers on the topic. Self-Consistent Time-Dependent Harmonic Approximation (SCTDHA), also encountered as Time-Dependent Self-Consistent Harmonic Approximation, denotes a family of Gaussian real-time schemes in which a nonlinear many-body problem is replaced by a quadratic Hamiltonian or Gaussian density matrix whose parameters are fixed self-consistently from instantaneous expectation values, fluctuation correlators, or averaged force derivatives. The designation is explicit for the out-of-equilibrium quantum sine-Gordon model (Nieuwkerk et al., 2018), while closely related constructions appear as dynamical extensions of the Self-Consistent Harmonic Approximation for anharmonic nuclear motion and phonons (Monacelli et al., 2020, Bianco et al., 2017, Ribeiro et al., 2017), and for coherent magnetic-resonance dynamics in ferromagnets and antiferromagnets (Moura, 2022, Villela et al., 2024).

1. General structure of the approximation

Across its different realizations, SCTDHA has three invariant components. First, the original nonlinear dynamics is projected onto a harmonic sector: a cosine interaction, an anharmonic Born–Oppenheimer potential, or a nonlinear spin Hamiltonian is replaced by a quadratic form. Second, the quadratic coefficients are not fixed a priori; they are obtained self-consistently from the state generated by the same quadratic dynamics. Third, the resulting harmonic problem is used to compute explicitly time-dependent observables such as correlation functions, resonance spectra, susceptibilities, coherent precession, or transport response.

Context	Self-consistent harmonic object	Dynamical output
Sine-Gordon model	$H_{\mathrm{SCH}}(t)$ with time-dependent $f,g,h$ (Nieuwkerk et al., 2018)	$\langle \phi\rangle_t$, mode dynamics, full distribution functions
Anharmonic nuclei and phonons	Gaussian $\hat\rho(t)$, SCHA Green function, self-energy (Monacelli et al., 2020, Bianco et al., 2017)	IR/Raman spectra, spectral functions, phonon lifetimes
Magnetic resonance	Quadratic $H_2$ or $H_0^m$ in canonical spin variables (Villela et al., 2024, Moura, 2022)	AFMR/FMR spectra, susceptibilities, coherent precession, spin pumping

A common misconception is to identify SCTDHA with an ordinary small-amplitude expansion about a fixed equilibrium point. In the formulations above, the harmonic reference is not merely an initial Taylor expansion. It is a renormalized, state-dependent quadratic theory. In the sine-Gordon case the coefficients $f(x,t)$, $g(x,t)$, and $h(x,t)$ depend on $\langle\phi(x)\rangle_t$ and on the connected variance $f,g,h$0; in nuclear dynamics the effective harmonic Hamiltonian depends on $f,g,h$1 through averages of the exact force and Hessian; in magnetic applications the effective exchange, anisotropy, and Zeeman fields are renormalized by parameters such as $f,g,h$2, $f,g,h$3, and $f,g,h$4 (Nieuwkerk et al., 2018, Monacelli et al., 2020, Villela et al., 2024, Moura, 2022).

2. Variational and Gaussian foundations

The static ancestor of SCTDHA is the Self-Consistent Harmonic Approximation. In lattice formulations, SCHA is built from the Gibbs–Bogoliubov variational principle, which minimizes the free-energy functional over harmonic trial density matrices. The trial Hamiltonian is quadratic in ionic displacements about variational centroids, and the optimal force-constant matrix $f,g,h$5 is determined self-consistently. In one explicit formulation, the renormalized matrix satisfies

$f,g,h$6

so the effective harmonic kernel is an average of the exact Born–Oppenheimer Hessian over the trial Gaussian state (Errea et al., 2013, Monacelli et al., 2020).

The time-dependent extension promotes this Gaussian variational structure into real time. For pure states, the derivation in nuclear dynamics uses the Dirac least action principle; for mixed states it yields a self-consistent Liouville–von Neumann equation,

$f,g,h$7

where $f,g,h$8 is harmonic and depends on the instantaneous density through centroid positions, averaged forces, and averaged Hessians. Within this closed dynamics, energy is conserved in the absence of explicit external time dependence, and entropy is conserved for the thermal ensemble. The static SCHA is recovered as a stationary solution of the dynamical equations (Monacelli et al., 2020).

In field-theoretic language, the same structure can be derived as a Gaussian truncation of the hierarchy of correlation functions. For the quantum sine-Gordon model, SCTDHA is described as a Gaussian, time-dependent variational truncation of the BBGKY hierarchy: one retains only one- and two-point functions and sets higher connected correlators to zero. An equivalent operator derivation uses normal ordering of the cosine vertex and truncates the resulting series at quadratic order in fluctuations, leading to a self-consistent quadratic Hamiltonian with explicitly time-dependent coefficients (Nieuwkerk et al., 2018).

3. Out-of-equilibrium sine-Gordon formulation

The canonical explicit use of the name SCTDHA occurs in the out-of-equilibrium quantum sine-Gordon model for tunnel-coupled one-dimensional Bose gases. The exact evolution under

$f,g,h$9

is replaced by the time-ordered evolution under a quadratic Hamiltonian

$\langle \phi\rangle_t$0

The coefficients are functionals of the same state evolved by $\langle \phi\rangle_t$1. In particular, the effective mass term is

$\langle \phi\rangle_t$2

and analogous expressions hold for $\langle \phi\rangle_t$3 and $\langle \phi\rangle_t$4. The nonlinear cosine is therefore replaced by a time-dependent quadratic potential whose curvature and linear shift are continuously updated by the evolving mean field and variance (Nieuwkerk et al., 2018).

For translationally invariant Gaussian initial states, the dynamics reduces to a closed nonlinear system for mode amplitudes $\langle \phi\rangle_t$5, $\langle \phi\rangle_t$6, and $\langle \phi\rangle_t$7. These determine

$\langle \phi\rangle_t$8

as well as the connected two-point function through mode combinations $\langle \phi\rangle_t$9. From these quantities one reconstructs the self-consistent coefficients $\hat\rho(t)$0 and $\hat\rho(t)$1, updates the quadratic Hamiltonian, and iterates in time. Because the state remains Gaussian, not only one-point functions but also full distribution functions of subsystem phase observables can be computed. The joint distribution $\hat\rho(t)$2 of the real and imaginary parts of $\hat\rho(t)$3 acquires a multiple Gaussian integral representation that is directly sampleable (Nieuwkerk et al., 2018).

The regime of validity is controlled by the smallness of higher cumulants. The approximation performs best for large $\hat\rho(t)$4, weak nonlinearity, and moderate squeezing, and it reproduces nontrivial envelope modulation in a nonlinear-oscillator benchmark much more accurately than the simple harmonic approximation. It is not expected to work near the Luther–Emery point $\hat\rho(t)$5, where soliton and antisoliton degrees of freedom dominate, and it does not reproduce the rapid narrowing of phase distributions observed experimentally in weakly interacting tunnel-coupled gases. This suggests that either physics beyond the pure antisymmetric sine-Gordon sector or genuinely non-Gaussian effects are required in that regime (Nieuwkerk et al., 2018).

4. Anharmonic nuclei, phonons, and dynamical response

In lattice dynamics, the static SSCHA provides the equilibrium harmonic reference by minimizing a variational free energy over Gaussian nuclear density matrices. The decisive step toward a time-dependent formulation is the identification of the free-energy curvature and its dynamical continuation. The free-energy Hessian can be written as

$\hat\rho(t)$6

which promotes the static SCHA matrix $\hat\rho(t)$7 into a fluctuation-renormalized curvature containing averaged cubic and quartic tensors. The corresponding one-phonon Green function is then written as

$\hat\rho(t)$8

with a self-energy

$\hat\rho(t)$9

Its static limit reproduces the free-energy Hessian, while finite $H_2$0 yields renormalized phonon spectra, linewidths, and spectral functions. In PbTe this framework reproduces the transverse optical phonon satellite detected in inelastic neutron scattering and the crossing between the transverse optical and the longitudinal acoustic modes along the $H_2$1 direction; in SnTe it captures the softening of the $H_2$2-point transverse optical mode and the ferroelectric transition from Fm$H_2$3m to R3m (Bianco et al., 2017, Ribeiro et al., 2017).

The fully explicit TD-SCHA for nuclear motion makes the real-time structure non-heuristic. The density matrix is constrained to the most general Gaussian form in coordinate space, and its evolution is governed by the self-consistent Liouville equation discussed above. Linearization around the equilibrium SCHA solution yields a finite-dimensional Liouvillian kernel $H_2$4 and a matrix Green function

$H_2$5

from which general response functions follow as

$H_2$6

For one-phonon observables this gives an interacting Green function satisfying a Dyson equation, and the dynamical ansatz proposed previously for the SCHA self-energy becomes exact within the Gaussian manifold. The method thereby provides IR and Raman spectra, dynamical structure factors, and other time-correlation functions without following any perturbative expansion of the nuclear potential (Monacelli et al., 2020).

This dynamical Gaussian formalism was benchmarked on a one-dimensional quartic–cubic potential, where TD-SCHA corrects the harmonic phonon energy by about $H_2$7, with about $H_2$8 error to the exact result, and reproduces the first overtone with reasonable energy and intensity. In high-pressure hydrogen phase III, using a 96-atom simulation cell, it yields vibron positions and linewidths in excellent agreement with experiment and gives access to non-Lorentzian line shapes and nonlinear driven dynamics relevant to pump-probe spectroscopy and proton-transfer problems (Monacelli et al., 2020).

5. Magnetic realizations: coherent precession, resonance, and transport

A magnetic version of SCTDHA emerges when ordered-spin dynamics is rewritten in canonical variables, typically an in-plane phase $H_2$9 and its conjugate momentum $H_0^m$0. In antiferromagnets with easy-axis anisotropy and Zeeman coupling, the nonlinear spin Hamiltonian is rotated into local sublattice frames and replaced by a quadratic form

$H_0^m$1

The renormalization factor $H_0^m$2 is obtained by matching fluctuation observables computed in the full and quadratic theories, and it defines renormalized fields $H_0^m$3, $H_0^m$4, and $H_0^m$5. The resulting equations of motion yield the full AFMR spectrum in both antiferromagnetic and spin-flop phases, while a Bogoliubov treatment provides the quantum magnon energies. In the antiferromagnetic phase the $H_0^m$6 resonance reduces to the Keffer–Kittel form with renormalized parameters; in the spin-flop phase one branch is gapless and relativistic, whereas the other is gapped and quadratic at long wavelength. Coherent-state driving then gives explicit time-dependent magnon amplitudes, sublattice precession, and susceptibilities, including the crossover from nearly circular counter-precession in the AF phase to strongly elliptical precession in the spin-flop phase (Villela et al., 2024).

A closely related construction was developed for ferromagnetic resonance and spin pumping in a ferromagnetic-insulator/normal-metal junction. There the magnetic Hamiltonian is replaced by a quadratic trial Hamiltonian $H_0^m$7 in $H_0^m$8 and $H_0^m$9, with two self-consistent renormalization parameters $f(x,t)$0 and $f(x,t)$1 for exchange and Zeeman terms. The magnon dispersion becomes

$f(x,t)$2

and coherent microwave driving produces a stationary displaced thermal state. The transverse magnetization executes elliptical precession, the susceptibility acquires the conventional FMR form with renormalized $f(x,t)$3 and $f(x,t)$4, and the pumped spin-current density is

$f(x,t)$5

For the bcc case considered, the self-consistent parameters vanish at $f(x,t)$6; the spin-mixing conductance is $f(x,t)$7, and the additional Gilbert damping is $f(x,t)$8, both in agreement with the reported literature values (Moura, 2022).

These magnetic examples clarify an important point about terminology. In some implementations, “time-dependent” means that the harmonic coefficients themselves are explicit functions of time, as in the sine-Gordon model. In magnetic resonance, by contrast, the renormalized quadratic Hamiltonian may be time independent once the self-consistent parameters are fixed, while the time dependence resides in the equations of motion, Heisenberg evolution, or coherent-state response under a drive. The approximation remains “time-dependent” because the full dynamical spectrum and driven motion are extracted from the self-consistent harmonic theory (Villela et al., 2024, Moura, 2022).

6. Regime of validity, limitations, and methodological position

SCTDHA is a Gaussian approximation, and that fact determines both its strength and its limitations. Its strength is nonperturbative renormalization within the Gaussian sector: harmonic frequencies, masses, fields, and susceptibilities are not fixed by a bare Taylor expansion but by self-consistent averages over the fluctuating state itself. This allows it to describe regimes where naive harmonic theory is unstable or qualitatively wrong, as in strongly anharmonic phonons, temperature-dependent AFMR, or out-of-equilibrium phase dynamics in the sine-Gordon model. Its limitation is equally clear: all connected cumulants beyond second order are discarded or projected out. Consequently, solitons, antisolitons, breathers, strong multi-particle scattering, strongly non-Gaussian tunneling between well-separated minima, and genuine thermalization mechanisms beyond quadratic mode mixing are not captured (Nieuwkerk et al., 2018, Monacelli et al., 2020).

Methodologically, SCTDHA sits between static SCHA and fully nonlinear treatments. Relative to the simple harmonic approximation, it introduces state-dependent masses or stiffnesses and, in the sine-Gordon formulation, a nonzero linear term $f(x,t)$9 that tracks finite $g(x,t)$0. Relative to static SCHA, it adds explicit real-time evolution, dynamical response functions, and coherent-state dynamics. It is related to time-dependent Gaussian variational approximations, BBGKY truncations at second cumulant, and self-consistent phonon theory; it differs from truncated Wigner or semiclassical sampling approaches by remaining fully quantum within the Gaussian sector (Nieuwkerk et al., 2018, Monacelli et al., 2020).

Several limitations recur across applications. In the sine-Gordon problem, the approximation is reliable for large $g(x,t)$1, weak nonlinearity, and moderate squeezing, but not near the Luther–Emery point $g(x,t)$2. In nuclear dynamics, the Gaussian ansatz is not designed for strongly multiwell, strongly non-Gaussian distributions, even though it treats anharmonicity nonperturbatively within a single Gaussian manifold. In magnetic SCHA, the renormalization parameter can vanish abruptly near the ordering temperature, a well-known pathology of harmonic approximations, even when the estimated transition temperature remains close to experiment. The most accurate interpretation is therefore not that SCTDHA solves nonlinear dynamics exactly, but that it provides a controlled, computationally tractable, self-consistent quadratic dynamics whose reliability is highest when the evolving state remains close to Gaussian and the dominant nonlinear effect is a renormalization of harmonic parameters rather than the generation of intrinsically non-Gaussian excitations (Villela et al., 2024, Ribeiro et al., 2017, Moura, 2022).