Dynamic analysis in Tellus3D: verification against exact solutions and validation from foundation tests to the Bangkok basin

This paper adds dynamic analysis to Tellus3D, an agent-driven three-dimensional geotechnical finite elementprogramwith10-nodetetrahedra,verifiesitagainstexactsolutions,andvalidatesitagainstfieldtestsandarecordedearthquake. AdynamicphaseintegratestheequationsofmotionwiththeNewmarkschemefromthestressesofanystaticphase,withaconsistentmass,Rayleighdampingpersoilandviscous,compliant,free-fieldandtiedboundaries.Theprogramreproducestheamplificationofadampedsoilcolumnwithin1.4%andthePekerisandChaosolutionsfor point loads on a half-space with root mean square errors of 0.6–1.6% and 0.2–0.5% of the static displacement.At the Garner Valley Downhole Array, a column calibrated on one earthquake predicts the surface spectra of fiveothers with a mean log residual of 0.004 ± 0.125, although no Rayleigh damping removes its over-prediction of thesiteresonances. Atthesamesitetheprogramreproducestheimpedanceofarigidslabwithin-4.5%(horizontal)and+6.9%(rocking)oncethemeshhasconverged,predictsthefrequencyofasoil-structureteststructurewithin5%ofthemeasurements,andreproducesthedispersionofasurface-wavetest. Thegroundvibrationscomputedforaconcretepile driven at Skövde have a median ratio of 1.30 to the measured ones. Finally, a model of the Bangkok basin with617,020elementsrunsthe2025 𝑀𝑤 7.7Mandalayearthquakein3.3hononeGPU,matchesexactone-dimensionalcolumnswithin6%,andshowsthattherecordedlong-periodmotionneedshorizontallyarrivingwavesthataninputfrom below cannot supply. The checks give practical rules on the mesh, the boundaries and the time step, and showthatthesoilofa three-dimensionaldynamic analysishas to becalibratedtothefoundationandthesite.