White box / Algorithms and implementation / Nonlinear solution and state

Full Newton with line search

Algorithm / execution · assumptions, essential equations, full derivation, algorithm, code and verification references.

Browse another sheet Algorithms and implementation / Nonlinear solution and state
Algorithms and implementationNonlinear solution and stateSheet 1 / 3
Algorithm / executionROUTE-BOUNDED ROBUSTNESS TOOL · not a convergence guarantee

A consistent route tangent defines the Newton correction. A line-search factor may reduce a trial update when the residual does not decrease.

01 Model and scope

Newton linearizes the internal-force residual at the current trial displacement and solves for a correction. A line search tests a scaled correction when the full step fails to decrease the residual; convergence still requires the route's residual and increment criteria.

Route assumptions and calculation checks

  1. Stagnation and numerical residual floors are classified instead of treating every small update as convergence.
  2. Adaptive increments can cut back and retry after a rejected trial.
  3. A shared assembly interface does not mean every element/material/contact combination composes.

02 Essential equations and symbols

r(u,λ)=fext(λ)−fint(u)\mathbf r(\mathbf u,\lambda)=\mathbf f_{ext}(\lambda)-\mathbf f_{int}(\mathbf u)KT(uk)Δuk=r(uk,λ)\mathbf K_T(\mathbf u_k)\Delta\mathbf u_k=\mathbf r(\mathbf u_k,\lambda)uk+1=uk+ηΔuk,0<η≤1\mathbf u_{k+1}=\mathbf u_k+\eta\Delta\mathbf u_k,\qquad 0<\eta\le1∥rk+1∥rref≤ϵr\frac{\|\mathbf r_{k+1}\|}{r_{ref}}\le\epsilon_r∥Δuk+1∥uref≤ϵu\frac{\|\Delta\mathbf u_{k+1}\|}{u_{ref}}\le\epsilon_u

Definitions and conventions

r(u,λ)\mathbf r(\mathbf u,\lambda)
Equilibrium residual at displacement u and load factor λ.
KT\mathbf K_T
Consistent route-specific tangent.
Δuk\Delta\mathbf u_k
Newton correction at iteration k.
η\eta
Accepted line-search scale between zero and one.
r_{ref},u_{ref}
Reference scales for residual and displacement tests.
ϵr,ϵu\epsilon_r,\epsilon_u
Route-specific residual and increment tolerances.
Full Newton with line search · method mapTrial corrections are tested against the route residual; the increment converges, cuts back, or stops with an explicit status.
4 detailed steps

03 Step by step derivation

Follow the assumptions, equations and discrete operations in order. The full explanation stays visible.

1 · Define the equilibrium residual and route tangent

At a fixed load parameter, equilibrium is external force minus internal force. The tangent is the derivative of this residual for the exact selected element, material, and geometric route. A tangent from another route can converge slowly or converge to the wrong discrete problem.

r(u,λ)=fext(λ)−fint(u)\mathbf r(\mathbf u,\lambda)=\mathbf f_{ext}(\lambda)-\mathbf f_{int}(\mathbf u)KT(u)=∂fint∂u\mathbf K_T(\mathbf u)=\frac{\partial\mathbf f_{int}}{\partial\mathbf u}

2 · Compute the Newton correction

Linearize the residual at the current iterate and solve the tangent system for a displacement correction. In a full Newton step, the tangent is reevaluated as the state changes. Sparse factor reuse is valid only while the effective operator remains unchanged.

KT(uk)Δuk=r(uk,λ)\mathbf K_T(\mathbf u_k)\Delta\mathbf u_k=\mathbf r(\mathbf u_k,\lambda)uk+1=uk+ηΔuk,0<η≤1\mathbf u_{k+1}=\mathbf u_k+\eta\Delta\mathbf u_k,\qquad 0<\eta\le1

3 · Let the line search test a scaled trial

When the full correction does not reduce the declared residual merit, test a reduced step length η. The line search helps control overshoot; it cannot fix a singular tangent, an incorrect model, or a missing constitutive derivative. The exact acceptance rule is defined in the linked driver.

η∈(0,1]\eta\in(0,1]M(uk+ηΔuk)<M(uk)\mathcal M(\mathbf u_k+\eta\Delta\mathbf u_k)<\mathcal M(\mathbf u_k)

4 · Converge, cut back, or classify the stop

Require the route's equilibrium residual and increment criteria, not a small update alone. If an increment stagnates or fails, restore accepted material state and retry with a smaller load increment where supported. Numerical residual floors and iteration limits must remain distinguishable from convergence.

∥rk∥rref≤ϵr\frac{\|\mathbf r_k\|}{r_{ref}}\le\epsilon_r∥Δuk∥uref≤ϵu\frac{\|\Delta\mathbf u_k\|}{u_{ref}}\le\epsilon_u

04 Algorithm and implementation

Read the linked versioned source for the actual assembly and solver behavior. The derivation above explains the mathematics and is not a verbatim source excerpt.

05 Verification and limitations

ROUTE-BOUNDED ROBUSTNESS TOOL · not a convergence guarantee

Each reference supports its own model, observable and analysis route. Read the original contract and limitations before transferring a conclusion.

Recorded repository calculations

See this method in a calculation.

Each case keeps its mesh, measured quantities, source record and limitations attached to the result.

Browse every calculation case →

06 Sources and related sheets

Continue the reasoning

Back to sheet contents ↑