[Advanced] — What happens when small strain fails, why the failure is usually about rotation rather than strain, and how rubber is modelled.
When you need it
Four situations force finite-deformation theory, and only the first is obvious:
1. Large strains. Rubber, biological tissue, foams, metal forming. Strains of 100–700%, where the distinction between strain measures is enormous. → Strain Measures
2. Large rotations with small strains. The one people miss. A fishing rod, a leaf spring, a deployable structure, a slender cantilever deflecting 30% of its length — strains under 1%, rotations of 0.3 rad, and small-strain theory reports spurious strain from the rotation alone. At 0.3 rad that’s roughly , an order of magnitude past steel’s yield strain, from a rotation that deformed nothing.
3. Buckling and post-buckling. Instability is inherently a geometric-nonlinearity problem — the structure’s stiffness changes as it deforms. → Beams and Structures
4. Contact and follower loads. Pressure that stays normal to a deforming surface, or contact areas that change with load, both depend on current geometry.
The criterion is small rotations, not small strains. That’s the correction most people need. If any part of your structure rotates more than a few degrees, switch on geometric nonlinearity — the “large displacement” toggle in every FE code — regardless of how small the strains are.
The three nonlinearities
Worth separating, because they have different causes and different fixes:
| Source | Example | |
|---|---|---|
| Geometric | equilibrium on the deformed shape | large rotation, buckling |
| Material | nonlinear constitutive law | plasticity, hyperelasticity |
| Boundary | contact, friction, follower loads | a seal, a press fit |
They’re independent. Rubber under small rotation is materially nonlinear only. A steel cantilever bending elastically through 45° is geometrically nonlinear only. Metal forming is all three at once, which is why it’s hard.
Total vs updated Lagrangian
Two ways to organise the bookkeeping, and every FE code uses one or the other.
Total Lagrangian — everything referred to the original configuration. Uses and (the second Piola–Kirchhoff / Green–Lagrange pair), which are symmetric, material, and energy-conjugate. → Stress
Natural for hyperelasticity, where a stored-energy function is naturally written in terms of the reference configuration.
Updated Lagrangian — the reference is reset to the last converged configuration at each step. Uses Cauchy stress and rate quantities.
Natural for plasticity, where the material’s state evolves and there’s no meaningful “original” configuration to refer back to.
They’re mathematically equivalent and differ in convenience and conditioning. Total Lagrangian for rubber; updated Lagrangian for metal forming is the usual split.
Hyperelasticity
The constitutive framework for large elastic strain.
Define a stored-energy function per unit reference volume, and derive stress from it:
Deriving stress from a potential is what makes the material path-independent and non-dissipative by construction — load and unload along any path and you recover exactly the energy you put in. It satisfies the second law automatically. → Constitutive Models
For an isotropic material, can depend only on the invariants of — that’s forced by symmetry, not chosen:
The standard models
Neo-Hookean — one parameter, derived from Gaussian polymer chain statistics:
Good to roughly 50% strain, and the right starting point. It’s also what most computer-graphics soft-body simulators use, because it’s cheap and stable.
Mooney–Rivlin — two parameters:
Better range, still simple. The workhorse for moderate strains.
Ogden — several parameters, written in principal stretches rather than invariants:
Excellent fit for rubber to 700% strain, and the standard for serious elastomer work. The cost is more parameters to fit and a greater risk of fitting nonsense.
Arruda–Boyce (8-chain) — physically motivated rather than phenomenological. Derived from the statistical mechanics of a network of polymer chains, so its parameters mean something (chain length, network density), and it extrapolates more honestly outside the fitted range.
Yeoh, Gent, Fung — Yeoh for filled rubbers, Gent for the limiting-chain-extensibility stiffening, Fung for biological soft tissue, which stiffens exponentially — that’s why skin and arteries feel soft then suddenly firm.
Incompressibility
Rubber is essentially incompressible — , bulk modulus thousands of times the shear modulus.
That’s not a detail; it changes the formulation. The energy function splits:
with the deviatoric (volume-preserving) invariant.
Numerically, incompressibility causes volumetric locking — standard displacement elements become artificially stiff and give badly wrong answers. The fixes are mixed u–p formulations (pressure as a separate unknown, i.e. a Lagrange multiplier), selective reduced integration, or F-bar methods.
If a rubber FE model comes out far too stiff, locking is the first thing to check. → FEM
Fitting hyperelastic parameters
The practical warning, and it’s a serious one.
Uniaxial tension data alone does not determine a hyperelastic model. Multiple parameter sets fit a tension curve nearly identically and then diverge wildly in biaxial tension, planar shear, or compression.
The consequence: an Ogden model fitted only to a tension test can be badly wrong in a real multiaxial state — and it will produce smooth, plausible-looking contour plots while being wrong.
You need at least:
- Uniaxial tension
- Equibiaxial tension (a bulge or inflation test)
- Planar shear (a wide, short specimen — “pure shear”)
Then fit all three simultaneously. Volumetric data too, if compressibility matters.
And check stability. The Drucker stability criterion requires the incremental work to be positive; a fitted model can be unstable at strains you’ll actually reach, producing negative stiffness and a solver that diverges for no visible reason. Most FE codes will report this if asked — ask.
Objective rates
The subtlety that catches people implementing finite-strain plasticity.
For a rate-form constitutive law you need a stress rate. But is not objective — under superposed rigid rotation it picks up spurious terms, so a law built on it would produce stress from pure rotation.
Several objective rates exist:
- Jaumann rate — corrects using the spin tensor . The most common
- Green–Naghdi rate — uses the rotation from the polar decomposition
- Truesdell / Oldroyd rates — used in fluid mechanics and viscoelasticity
They give different answers at large strain. The classic demonstration: simple shear to large strain with the Jaumann rate produces oscillating stress — physically absurd, and a well-known artefact. Green–Naghdi doesn’t.
The practical advice: for large-strain plasticity, know which rate your code uses, and be suspicious of results at shear strains above ~1. For most engineering problems the difference is negligible; where it isn’t, it’s dramatic.
Numerical solution
Finite-deformation problems are nonlinear, so there’s no single solve. You iterate.
Newton–Raphson, incrementally:
for each load increment:
predict displacement
repeat:
compute internal forces from current geometry and stress
residual = external - internal
if |residual| small: converged, next increment
form the TANGENT stiffness at the current state
solve for a correction
The tangent stiffness changes every iteration because it depends on the current geometry and stress state. It has two parts:
The geometric stiffness is what makes buckling emerge naturally — compressive stress reduces effective stiffness, and when it drives an eigenvalue to zero, the structure buckles. It falls out of the formulation rather than being added.
Practical convergence advice:
- Smaller load increments — the single most effective fix for a diverging nonlinear solve
- Arc-length (Riks) methods for post-buckling, where load decreases as displacement increases and a load-controlled solver cannot follow the path
- Check the tangent is consistent with the stress update; an inconsistent tangent still converges, just slowly and unreliably
- Watch for negative Jacobians — an inverted element, meaning the mesh or the increment is too aggressive → Kinematics
The honest summary
Finite deformation is where the clean structure of linear elasticity — superposition, uniqueness, closed-form solutions — is lost:
| Linear | Finite | |
|---|---|---|
| Superposition | ✅ | ❌ |
| Unique solution | ✅ | ❌ (buckling has several) |
| Closed-form results | many | very few |
| Solve | one linear system | incremental iteration |
| Geometry for equilibrium | undeformed | current (unknown) |
Use small strain whenever you legitimately can. Not laziness — the linear theory is better understood, faster, more reliable, and gives you superposition. Reach for finite deformation when rotations exceed a few degrees, strains exceed a few percent, or instability is the thing you’re studying.
Related
- Strain Measures — why small strain fails under rotation
- Stress — the three stress tensors this needs
- Constitutive Models — hyperelasticity in context
- FEM — solving these
- Continuum mechanics map
Software engineering
Infrastructure & security
Computer science
Hardware & engineering