Skip to main content
Version: Latest

Pseudo-transient mechanics

With control.has_PT = yes, DES3D solves for quasi-static mechanical equilibrium within a physical timestep. The finite-element force assembly uses the spatial discretization shared with the explicit momentum path. PT adds an inner relaxation loop whose candidates all refer to the same physical increment. Configuration and diagnostic details are in Choosing a mechanical solver.

One physical increment, many trials​

Let denote the material history at the start of a physical step, its physical duration, and its reference mesh. Each candidate velocity defines a constitutive trial

is the existing material update, including applicable plastic, viscous and frictional history. Every evaluation restores first. Consequently, trials do not represent of aging or viscous relaxation. Mixed stress correction is applied before candidate force evaluation. Acceptance commits the full constitutive candidate.

With the default hydraulic content model, temperature advances once before a physical PT solve and remains fixed during its trials. The optional finite-density model instead advances its heat and water inventories after accepted mechanics. Hydraulic transport, mesh motion, stress rotation and remeshing remain outside the trial loop; see Poromechanics. Gradients, volumes, facet areas and normals remain on the reference mesh during mechanics. For moving-mesh physical steps, height-dependent foundation, water and side-support pressures include the candidate vertical displacement in their height argument. This is an incremental pressure-load correction, not a complete finite-deformation follower-load formulation.

Force balance in the admissible velocity space​

Write the undamped assembled force as

Let be the orthogonal projector onto admissible homogeneous velocity perturbations, and let be its total free rank. PT measures

For , the free residual is defined as zero. On axis-aligned roller boundaries this removes constrained force components. Oblique constraints require projection onto their free directions. Support reactions are excluded; applied tractions, body forces and foundation loads remain. Artificial damping is excluded as well. The units are N/m in 2D and N in 3D.

The affine velocity constraints enforce prescribed motion. Their homogeneous counterparts define the perturbation space. DES3D retains the ordered boundary dispatch and corner precedence of its boundary conditions; arbitrary contradictory prescriptions are not simultaneously enforced.

The default acceptance threshold uses an absolute tolerance plus a relative tolerance times the first warm candidate's residual. The alternative running-scale criterion and its limitations are given in the convergence reference.

Starting candidates​

The previous physical velocity supplies the warm start. If that candidate already meets the threshold, it is accepted. Otherwise, a physical solve can also evaluate the affine prescribed velocity with zero free motion, and retain the finite candidate with the smaller residual. Both candidates evaluate the complete physical-dt constitutive response.

This matters for a held elastic load: reusing the previous loading velocity could apply the same compression again. Zero free motion can represent its equilibrium. For a Maxwell material, however, zero velocity still permits constitutive relaxation over and need not satisfy the imposed traction. A new equilibrium velocity must then produce continued creep. The elastic and Maxwell tutorials demonstrate the distinction.

Relaxation schemes​

PT_option = 0 uses localized dual-time stress/velocity relaxation. Its factors depend on mesh heights, effective viscosity, PT_CFL and PT_Re, following the effective-viscosity/dual-time approach of Räss et al. (2022). Lagged stress is solver scratch; acceptance always checks the full constitutive candidate.

PT_option = 1 uses adaptive dynamic relaxation. A Gershgorin row-sum construction supplies nodal fictitious masses from the elastic/viscous velocity stiffness. A secant Rayleigh quotient estimates a low mode for adaptive damping; momentum is discarded when it opposes the current force. There is no lagged stress, and PT_CFL and PT_Re are unused.

The row-sum construction bounds the magnitude of eigenvalues of the mass-scaled linear reference operator. It does not prove stability of the full nonlinear constrained problem. Both options share the force criterion, material acceptance and rollback behavior.

Initial equilibrium and diagnostic time level​

Initial body-force equilibration uses homogeneous supports and the instantaneous material skeleton, without transport, viscous relaxation or physical frictional aging. Plastic history is committed only for the accepted initial candidate. Maxwell and EVP initialization use their instantaneous skeletons; a pure viscous material has no such skeleton and is rejected for this initialization. Initial equilibrium precedes frame-zero output and monitor initialization, and completion is checkpointed.

The saved PT residual force describes the accepted reference-geometry solve before hydraulic transport, mesh motion and stress rotation. For physical steps with thermal diffusion enabled, temperature has already been advanced. The report's stage=before_physical_updates refers to the later updates, not the preceding thermal stage. A force reconstructed from saved post-rotation stress can differ slightly, even on a fixed mesh. Remeshing invalidates this diagnostic until another solve is accepted. See PT force output.

Neither relaxation scheme can produce a stable quasi-static equilibrium where the model has none. Strain softening can introduce negative stiffness; changing the numerical relaxation does not supply physical regularization. The implementation notes and validation record separate demonstrated cases from unresolved convergence limits.