[Advanced] — The six missing equations. Where the physics stops being universal, where the modelling error lives, and the rules that separate a model from a curve fit.

The closure problem

The balance laws give four equations for ten unknowns. The gap is closed by a constitutive equation — a relation between stress and deformation that describes this particular material.

This is the only part of the theory that is chosen rather than derived. Conservation of momentum is as close to certain as engineering gets. Whether steel obeys is an empirical claim with a limited domain of validity.

When a simulation is wrong by a factor of three, look here first. Not at the mesh, not at the solver — at whether the material model was ever valid for what you asked it to do.

The rules a model must obey

Not anything goes. Four restrictions, and they eliminate most proposals.

1. Material frame indifference (objectivity)

The response cannot depend on the observer. Rotate the whole experiment — or watch it from a rotating frame — and the material must behave identically.

Formally, under a superposed rigid motion , the stress must transform as .

The practical consequences:

  • A law cannot depend on directly — it must go through or , where the rotation cancels. → Strain Measures
  • A law cannot depend on velocity , only on the rate of deformation — never the spin . A fluid in rigid rotation must feel no viscous stress
  • Ordinary time derivatives of stress are not objective. picks up spurious terms under rotation, which is why finite-strain plasticity needs objective rates — Jaumann, Green–Naghdi, Truesdell. Using the wrong one produces oscillating stress under simple shear, a famous and genuinely confusing failure

2. Thermodynamic admissibility

The Clausius–Duhem inequality must hold for every conceivable deformation path. → note 06

This forces viscosity positive, conductivity positive, moduli positive definite, and hyperelastic stress to derive from a stored-energy potential. The Coleman–Noll procedure extracts these systematically.

3. Material symmetry

If the material has symmetries, the model must respect them. Isotropy — no preferred direction — is the strongest and most useful:

An isotropic material’s response can depend only on the invariants of the deformation.

That’s not a convenience, it’s forced. And it’s remarkably restrictive: it takes the 81 components of a general fourth-order stiffness tensor down to two independent constants. → Linear Elasticity

SymmetryIndependent elastic constants
Fully anisotropic (triclinic)21
Orthotropic (3 orthogonal planes)9 — wood, rolled sheet, unidirectional composite
Transversely isotropic5 — fibre composites, muscle, bone
Isotropic2 — most metals, glass, unfilled polymers

4. Determinism and local action

Stress at a point depends on the history of deformation at that point (and its immediate neighbourhood), not on what’s happening across the body. Non-local models exist — they’re needed for strain-softening and damage, where local models produce mesh-dependent nonsense — but they’re a deliberate departure.

The classification

The useful way to organise the zoo: what does stress depend on?

ClassDepends onExamples
Elasticcurrent strain onlysteel below yield, rubber
Viscousstrain rate onlywater, air, oil
Viscoelasticstrain and rate, with memorypolymers, tissue, asphalt
Plasticstrain history — irreversiblemetals past yield
Viscoplastichistory and ratecreep, high-rate forming

Elastic ⟹ reversible and path-independent. Load and unload and you return along the same curve, with no energy lost. That’s the defining property, and it’s what fails for everything else.

Elasticity

Linear (Hooke):

Valid for small strain, and enormously useful precisely because it’s linear — superposition works, and you can solve one load case and add. → Linear Elasticity

Hyperelastic — for large elastic strain, defined by a stored-energy function :

Deriving stress from a potential is what makes the material path-independent and non-dissipative by construction — it satisfies the second law automatically. → Finite Deformation

The common forms, in increasing sophistication: neo-Hookean (one parameter, good to ~50% strain), Mooney–Rivlin (two, better range), Ogden (several, excellent fit for rubber to 700%), Arruda–Boyce (physically motivated, based on polymer chain statistics rather than curve-fitting).

Viscosity

Newtonian:

Stress proportional to rate of deformation. Water, air, thin oils, and most gases. → Navier–Stokes

Non-Newtonian — apparent viscosity varies with shear rate, and the everyday examples are worth knowing because they make the concept concrete:

  • Shear-thinning (pseudoplastic) — viscosity drops with shear rate. Paint (brushes easily, doesn’t run), blood, ketchup, polymer melts. The most common non-Newtonian behaviour by far
  • Shear-thickening (dilatant) — viscosity rises. Cornstarch in water, which you can run across
  • Bingham plastic — behaves as a solid below a yield stress, flows above. Toothpaste, drilling mud, mayonnaise

The power law covers thinning () and thickening () with two parameters, and it’s the workhorse model.

Plasticity

The genuinely hard one, because stress depends on history, not on the current state.

Three ingredients:

1. A yield criterion — when does plastic flow start?

von Mises is the standard for metals, built on the second deviatoric invariant. → Failure and Yield

2. A flow rule — in what direction does plastic strain develop?

Associated flow takes , so plastic strain is normal to the yield surface. Correct for metals; wrong for soils and concrete, where dilatancy requires a separate plastic potential.

3. A hardening law — how does the yield surface evolve?

  • Isotropic hardening — the surface expands uniformly. Fine for monotonic loading
  • Kinematic hardening — the surface translates. Necessary for cyclic loading, because it captures the Bauschinger effect: a metal yielded in tension yields earlier in subsequent compression. Isotropic hardening cannot represent that, and using it for fatigue gives wrong answers

The strain decomposes additively (in small strain):

with only the elastic part generating stress. Numerically, plasticity is solved by return mapping: take an elastic trial step, and if it lands outside the yield surface, project back onto it.

Path dependence is the practical consequence. You cannot jump to the answer — you must integrate the loading history in increments, which is why plastic FE analysis is so much more expensive than elastic.

Viscoelasticity

Both elastic and viscous, with memory. Polymers, biological tissue, asphalt, and the reason silly putty both bounces and flows.

Built from springs and dashpots:

  • Maxwell (spring + dashpot in series) — models stress relaxation: hold strain constant, stress decays
  • Kelvin–Voigt (parallel) — models creep: hold stress constant, strain grows
  • Standard linear solid (three elements) — captures both, and is the minimum useful model
  • Prony series — many Maxwell elements in parallel, fitted to data. What FE codes actually use

The general form is a hereditary integral — stress depends on the entire strain history, weighted by a decaying relaxation function:

Time–temperature superposition is the practically important trick: for many polymers, raising temperature is equivalent to slowing the test. The WLF equation quantifies it, and it’s how a short high-temperature test predicts decades of room-temperature creep.

Getting the parameters

Every model has constants, and they come from experiments:

TestGives
Uniaxial tension, yield stress, hardening curve, UTS
Compressionasymmetry (concrete, soils, and some polymers differ in tension)
Torsion directly, and large shear strains without necking
Biaxial / bulgeessential for rubber — uniaxial data alone badly under-determines hyperelastic parameters
Creep and relaxationviscoelastic constants
Split-Hopkinson barhigh strain-rate response
Nanoindentationlocal properties, thin films

Two honest warnings:

Fitting uniaxial data does not determine a multiaxial model. This is the classic rubber mistake — an Ogden model fitted only to a tension test can be wildly wrong in biaxial or shear states, and it will still produce smooth, plausible-looking results.

Parameters have a validity range. Fitted to 0–5% strain, at 20 °C, at 0.001/s. Used at 40% strain, 200 °C, and 100/s, they are extrapolation dressed as data.

The practical advice

  1. Start with the simplest model that could work. Linear elastic isotropic answers more questions than people expect
  2. Know which assumption you’re leaning on, and check whether the result violates it. If a linear-elastic run predicts stress above yield, the answer is not “the part yields a bit” — it’s “this analysis is invalid, rerun it plastic”
  3. Match the model to the loading. Cyclic loading needs kinematic hardening. Long duration needs creep. High rate needs rate dependence
  4. Rubber needs multiaxial data. Non-negotiable
  5. Sanity-check against a hand calculation. A beam formula, an equilibrium check, an energy balance
  6. Report the model and its parameters alongside any result. A stress number without its constitutive assumptions isn’t a result, it’s a number

The balance laws are physics. The constitutive model is a hypothesis. Keeping those categories separate is the single most useful habit in this field.


↑ Back to Continuum Mechanics