ImplicitUpdate#

Source: models/common/ImplicitUpdate.py

Update an implicit model by solving the underlying nonlinear system of equations.

Other options#

equation_system — str · required: The nonlinear system of equations to solve
solver — str · required: Solver used to solve the nonlinear system of equations
input_sensitivity_solver — str: Linear solver for the input-sensitivity (implicit-function-theorem) solve du/dg = -A^{-1} B when differentiating outputs w.r.t. inputs. Defaults to the Newton’s linear_solver; set a direct solver (e.g. DenseLU) here to keep input derivatives exact under an iterative forward solve.
param_sensitivity_solver — str: Linear solver for the parameter-sensitivity solve du/dtheta = -A^{-1} dr/dtheta when differentiating outputs w.r.t. (promoted) parameters. Defaults to the Newton’s linear_solver; set a direct solver (e.g. DenseLU) here for exact parameter derivatives under an iterative forward solve.
predictor — str: An optional predictor to provide an initial guess for the nonlinear solve.
max_substepping_level — int · default 0: Adaptive sub-incrementation depth cap, applied by the COMPILED AOTI routes only (neml2-compile). 0 (default) disables substepping. Level L lets the compiled solve recursively bisect a failing increment down to depth L (up to 2^L sub-steps), per-element, interpolating paired forces and chaining state; a sub-step still failing at depth L raises a recoverable ConvergenceError so an outer time-stepper can cut the step. The eager runtime does NOT sub-increment: evaluating an ImplicitUpdate eagerly with this set > 0 raises (substepping is an AOTI-only feature).
incremental_variables — list: Names of driving-force inputs the COMPILED substep driver treats as pure increments (scaled by the sub-step fraction rather than interpolated). Forces that already carry a ~1 counterpart are interpolated automatically and need not be listed; a lone increment with no ~1 counterpart is held constant unless listed here. Like max_substepping_level this is an AOTI-compile directive; setting it rejects eager evaluation.