The team’s approach tackles the linearized gravity-wave equations, which describe small-amplitude waves superimposed on a background atmosphere. Unlike fully nonlinear time-step models, which must march forward through tiny increments of time and can model dramatic phenomena such as wave breaking and secondary wave generation, linear methods solve for the wave field directly. The trade-off is well documented in the literature: in earlier work by other researchers, a time-step model required several hours of computation while a linear method completed the same task in seconds. That speed makes linear methods far better suited for analyzing measured data, where many model runs may be needed to interpret observations.

At the heart of the new model is a multilayer method, a technique first applied to acoustic-gravity waves in 1962. The idea is elegant: the atmosphere is sliced into a stack of thin horizontal layers, and within each layer the complicated equations—with coefficients that vary continuously with altitude—are approximated by a simpler system with constant coefficients. The analytic wave solutions in neighboring layers are then stitched together by requiring continuity of the variables at each interface. The new model uses the numerical variant of this approach, in which the equation coefficients are evaluated at the middle of each layer. Crucially, this variant retains the height derivatives of background atmospheric parameters such as temperature and wind, which the older physical multilayer method discards by assuming those parameters are constant within each layer.

That distinction matters more than it might seem. In 1973, the influential physicist C. O. Hines criticized physical multilayer methods, arguing that it becomes impossible to find physically meaningful state variables when viscosity and wind are both nonzero, or when thermal conductivity and a temperature gradient coexist. As later researchers pointed out, this criticism does not apply to the numerical multilayer method, because in a purely mathematical context it is sufficient to prove that the method converges to the correct solution as the layers become infinitesimally thin. The new model sidesteps the old debate entirely, solving the equations in a way that explicitly accounts for viscosity, thermal conduction, and ion drag—the frictional force that charged ions exert on the neutral air.

The mathematical machinery draws an unexpected inspiration from a seemingly unrelated field: radiative transfer, the study of how light propagates through scattering atmospheres. Solving the linearized gravity-wave equations turns out to be mathematically analogous to solving the equations of radiative transfer, and the authors adapted techniques developed for the latter, including the discrete ordinate method with matrix exponential and the matrix operator method. The solution strategy rests on a matrix-exponential formalism and offers two classes of solvers: global matrix methods and scattering matrix methods. In the global matrix approach, all the layer equations and boundary conditions are assembled into one large banded system of linear equations, which is solved efficiently using standard LAPACK routines for complex band matrices. The scattering matrix approach instead builds up the solution layer by layer, using reflection and transmission matrices borrowed conceptually from seismology.

Both solvers must contend with a notorious numerical pitfall known as swamping. When dissipative wave modes—viscosity waves and thermal conduction waves—are included in the model, certain descending modes grow so explosively in the upward direction that the computation overflows. The new model tames this problem with scaling matrices that rescale the exponential terms depending on whether their growth rates are positive or negative, a technique standard in radiative transfer theory. The researchers also carefully distinguish ascending from descending wave modes using the real parts of the eigenvalues of the propagation matrix, and they incorporate a causality-preserving technique called the imaginary frequency shift, introduced by Harold Knight and colleagues in 2019, which ensures that the computed wave field never responds before its source acts—a fundamental physical requirement.

The model’s flexibility extends to both single-frequency waves and time-dependent wave packets. For a monochromatic source, the problem is solved at a single excitation frequency. For a wave packet—say, a pulse of air disturbance with a Gaussian envelope in time—the equations are transformed into the frequency domain, solved across a band of frequencies, and then reconstructed in time via an inverse Fourier transform. In their demonstration simulations, the team used realistic atmospheric and ionospheric data for the EISCAT Tromsø auroral location in Norway, with an altitude grid stretching from 80 to 500 kilometers. The wave packets produced temperature perturbations of roughly 31 to 33 Kelvin and vertical velocities reaching more than 20 meters per second, depending on the horizontal wavelength.

The numerical experiments delivered a clear verdict on the competing solvers: all methods achieved identical accuracy, with relative errors below one part in a million, but the global matrix method was significantly faster than the scattering matrix method, especially for time-dependent wave packets, which require numerous matrix operations in every layer. The simulations also quantified the impact of ion drag, which moderately attenuates the waves between roughly 180 and 350 kilometers altitude, where the ion number density is relatively high. Interestingly, the model showed that in the altitude range of 150 to 300 kilometers, waves computed with the full general model attain larger amplitudes than those from the simplified model, because including altitude-dependent background properties weakens the effective vertical damping.

The researchers emphasize that this model is only the first component of a larger ambition: a comprehensive tool for analyzing ionospheric gravity waves using satellite measurements. Two extension paths are envisaged. In one, the neutral-atmosphere and ionospheric equations would be solved together as a fully coupled system, with the state vector augmented by perturbed ion density and diffusion velocity. In the other, a two-step strategy would first solve the neutral atmosphere and then feed the wave-induced perturbations into established ionosphere models such as SAMI2 or SAMI3. For now, the implementation is freely available as open-source code on GitHub and archived on Zenodo, an unusual level of accessibility for a specialized numerical tool. By combining decades-old multilayer concepts with modern matrix-exponential techniques from radiative transfer, the DLR team has given atmospheric scientists a fast, rigorous, and transparent instrument for decoding the ripples that connect the weather below to the charged skies above.

Subject of Research: Numerical modeling of linearized atmospheric gravity waves using a multilayer matrix-exponential method

Article Title: A numerical model for solving the linearized gravity-wave equations by a multilayer method

Article References: Doicu, A., Efremenko, D. S., & Trautmann, T. (2026). A numerical model for solving the linearized gravity-wave equations by a multilayer method. Annales Geophysicae, 44(2), 655-688. https://doi.org/10.5194/angeo-44-655-2026

Image Credits: AI Generated

DOI: 10.5194/angeo-44-655-2026

Keywords: gravity waves, atmospheric modeling, ionosphere, multilayer method, matrix exponential, numerical simulation, ion drag, viscosity, thermal conduction, radiative transfer, open-source code, Annales Geophysicae