Hyperelastic curve fit calculator
Paste a rubber test curve and get the constants for Neo-Hookean, Mooney-Rivlin, Yeoh and Ogden models in one go, with the fit error, what each model predicts in the tests you did not run, and whether it stays stable. Or enter the constants you already have (from LS-DYNA's d3hsp, an Abaqus fit, a paper) and check them. Copy the LS-DYNA or Abaqus card at the end. Free, no sign-up, runs in your browser.
Fits, stresses and stability limits re-derived by an independent solver (stress as dW/dλ, finite-difference Hessian, scipy fits) · Updated 21 Nov 2026
Rows are "strain stress", separated by spaces, tabs or commas; lines starting with # or $ are skipped. Stresses are taken as nominal (force over original area) after conversion; true stress is divided by λ, which assumes the rubber keeps its volume. Constants are in the stress unit (MPa); LS-DYNA's g-mm-ms and Abaqus N-mm also give MPa.
| Model | μ₀ (MPa) | RMS (MPa) | Worst | Unstable at |
|---|
What is a hyperelastic material model?
Rubber does not have one Young's modulus. Stretch a rubber band and it is soft at first, then stiffens sharply near its limit, and it springs back to its original length from strains of several hundred percent. A hyperelastic model handles this by defining a strain energy W, the energy stored per unit volume, as a function of how much the material is stretched in each direction. Every stress follows by differentiating W. The models differ only in the form of W:
λ₁, λ₂, λ₃ are the principal stretches (current length over original length in three perpendicular directions) and I₁, I₂ the strain invariants built from them. The bars mean the volume change has been taken out; rubber keeps its volume almost exactly, so for fitting it is treated as incompressible:
Neo-Hookean is Mooney-Rivlin with only C₁₀. Yeoh keeps only I₁ terms. LS-DYNA's *MAT_HYPERELASTIC_RUBBER (077_H) uses the Mooney-Rivlin polynomial with N = 1 (C₁₀, C₀₁), N = 2 (adds C₁₁, C₂₀, C₀₂) or N = 3 (adds C₃₀); *MAT_OGDEN_RUBBER (077_O) uses the Ogden sum. The calculator fits all eight forms at once.
Stress in the three standard rubber tests
Rubber is tested in uniaxial tension (a dumbbell or "dogbone" strip), equibiaxial tension (a sheet inflated or pulled equally in two directions) and planar tension, also called pure shear (a wide strip pulled across its short side, so it cannot narrow). With W₁ = ∂W/∂I₁ and W₂ = ∂W/∂I₂, the nominal stress P (force over original area) is:
For Ogden the uniaxial stress is below. Its small-strain shear modulus μ₀ is the number to compare between models: every good fit of the same rubber should give about the same μ₀.
Test machines report engineering strain and either engineering or true stress. The conversions, for a material that keeps its volume:
A few reports convert with "engineering strain = 1 − extension". The extension ratio LS-DYNA prints in d3hsp is λ itself, so the engineering strain is λ − 1, and a stretch below 1 means compression.
How the fit works
For the Mooney-Rivlin family and Yeoh the stress is linear in the constants, so the best fit in the least-squares sense is a single linear solve over all the points you enter, in every mode at once. Ogden is nonlinear in the exponents α. The calculator searches the αs with a simplex method from several starting points and solves for the μs exactly at each step (variable projection). Choose "relative" error when the small-strain points matter as much as the large-strain ones; absolute error lets the high-stress end dominate.
Why one uniaxial curve is not enough
The demo points are Ogden's 1972 model of Treloar's rubber sampled at 10% to 100% strain: model output, used here because the true answer in every mode is known. Fit them and every model matches the uniaxial points closely. Now look at what the fits predict in the tests they never saw:
- Mooney-Rivlin, 2 terms: equibiaxial stress at λ = 2 of 1.988 MPa against the true 0.808 MPa, +146%. The C₀₁ term barely affects uniaxial tension, so a uniaxial fit can give it almost any value, and C₀₁ dominates the biaxial response.
- Neo-Hookean: 0.698 MPa, −14%. One constant cannot be badly wrong in a new mode, but it misses the stiffening.
- Yeoh, used beyond its data: fine at λ = 2 in equibiaxial tension (−4%), but extrapolated to 400% uniaxial strain it predicts 20.9 MPa against the true 1.71 MPa, 12.2 times too stiff. A cubic in (I₁ − 3) runs away outside the fitted range.
With the second demo, which adds equibiaxial and planar points, the same two-term Mooney-Rivlin lands at 0.815 MPa in equibiaxial tension at λ = 2 (+1%). That is why rubber suppliers test in at least two modes, and why the strain range of the test should cover the strain range of the part.
Material stability: the Drucker check
A fitted W can have regions where pushing the material further makes the stress drop. That is negative stiffness: an implicit solver stops converging and an explicit one lets elements collapse. The test, which Abaqus reports when it evaluates a fit, is Drucker's: any small strain increment must do positive work. For an incompressible material written in log strains e₁, e₂ (e₃ = −e₁ − e₂), that means the matrix of second derivatives of W must be positive definite:
The calculator checks it along uniaxial, equibiaxial and planar paths, in tension (λ up to 10) and compression (λ down to 0.1), and reports the first stretch at which each fit fails. Ogden fits whose terms all have μᵢαᵢ > 0 are stable everywhere. Mooney-Rivlin fits with negative higher-order constants are the usual offenders.
The "Hyperelastic material models" challenge: the class constants
Skill-Lync students have posted 124 reports for this LS-DYNA challenge. Everyone gets the same uniaxial dogbone data, fits Mooney-Rivlin and Ogden models with N = 1, 2 and 3 through LS-DYNA's built-in fit and copies the constants out of d3hsp. 47 reports type the same N = 1 result (C₁₀ = 0.1768, C₀₁ = 0.1474 MPa) and 27 the same N = 3 set, so the constants below are effectively the class's. Several conclude that Mooney-Rivlin N = 3 matches the data best. None of the 124 reports tests the constants outside uniaxial tension:
| Constants | μ₀ (MPa) | Uniaxial P at 100% | Equibiaxial P at λ = 2 | Unstable: uniaxial compression | Unstable: equibiaxial tension |
|---|---|---|---|---|---|
| Mooney-Rivlin N = 1 | 0.648 | 0.877 MPa | 3.02 MPa | stable | stable |
| Mooney-Rivlin N = 2 | 0.611 | 0.869 MPa | -0.49 MPa | λ 0.43 | λ 1.52 |
| Mooney-Rivlin N = 3 | 0.611 | 0.869 MPa | -3.38 MPa | λ 0.51 | λ 1.40 |
| Ogden N = 1 | 0.611 | 0.869 MPa | 1.08 MPa | stable | stable |
| Ogden N = 2 | 0.611 | 0.869 MPa | 1.08 MPa | stable | stable |
| Ogden N = 3 | 0.611 | 0.869 MPa | 1.08 MPa | stable | stable |
- In uniaxial tension, up to 100% strain, five of the six fits are the same material. Mooney-Rivlin N = 2 and N = 3 and all three Ogden fits give μ₀ = 0.611 MPa and the same stress within 0.06% from 5% to 100% strain. Only the two-term Mooney-Rivlin differs, by up to 4.7% (stiffer at small strain). Whatever makes one simulation match the data better than another, it is not the constants: "N = 3 fits best" is a difference of hundredths of a percent.
- Outside that one test they are not. Mooney-Rivlin N = 2 and N = 3 lose stability in equibiaxial tension at λ = 1.52 and 1.40, and in uniaxial compression at λ = 0.43 and 0.51: squash a seal made of the N = 3 rubber to about half its height and the model has negative stiffness. N = 3 also fails in uniaxial tension at λ = 2.64, 164% strain, not far past the 100% the data covers. All three Ogden fits stay stable from λ = 0.1 to 10.
- Ogden N = 2 is really N = 1. Its second term has α₂ = −0.0018, so μ₂α₂ = −0.004 MPa: a constant of 2.23 MPa that does almost nothing. LS-DYNA's fit put a second term where the data had no use for one.
Load the first preset to see the class constants in the calculator, then switch the model.
The "Stretch ratio ... Ogden" challenge: λ = 6, and a Poisson's ratio of 0.45
The other LS-DYNA challenge (170 reports) gives a *MAT_OGDEN_RUBBER card and asks for a 10 mm block, meshed with four solid element formulations (ELFORM 1, 2, −1, −2), to be stretched to a stretch ratio of 5. The card is Ogden's 1972 fit to Treloar's rubber: μ = 0.62, 0.00118, −0.00981 MPa (6.3, 0.012, −0.1 kg/cm² in the original) with α = 1.3, 5 and −2. Two things in the reports are worth a second look.
- Many blocks went to λ = 6. A stretch ratio of 5 means pulling the 10 mm block by 40 mm. 18 reports type a peak true strain of 1.78 to 1.79 and a peak stress of 5.02 to 5.03 MPa, which is what a 50 mm pull (λ = 6) gives. 15 of them still call it a stretch ratio of 5. If your peak true strain is 1.79, your block went to λ = e1.79 = 6.
- PR = 0.45 makes the rubber compressible. LS-DYNA turns PR into a bulk modulus. With μ₀ = 0.416 MPa and ν = 0.45, K = 4.02 MPa, only 9.7 times μ₀:
| Block | Lateral stretch | Volume ratio J | True stress |
|---|---|---|---|
| λ = 5, PR 0.45 | 0.546 | 1.491 | 3.97 MPa |
| λ = 6, PR 0.45 | 0.534 | 1.714 | 5.023 MPa |
| λ = 5, PR 0.4995 | 0.449 | 1.007 | 8.43 MPa |
| λ = 6, PR 0.4995 | 0.411 | 1.012 | 15.01 MPa |
| λ = 6, incompressible | 0.408 | 1 | 15.41 MPa |
The 5.023 MPa in the second row is the 5.02 to 5.03 MPa the reports print, recomputed here from the card alone, which also confirms how LS-DYNA builds K from PR. At λ = 6 the block's volume has grown 71%, and it carries 3.1 times less stress than a rubber that keeps its volume. At λ = 5 the gap is 3.97 against 8.54 MPa. The element formulations agree with each other in the reports because a uniformly stretched block has no gradients to tell them apart; the material card, not the element, sets the answer. Enter your run in "Check my LS-DYNA block run" above.
LS-DYNA and Abaqus card conventions
- Mooney-Rivlin and Yeoh constants are the same in both codes. Abaqus orders a polynomial N = 2 card C₁₀, C₀₁, C₂₀, C₁₁, C₀₂, D₁, D₂; LS-DYNA's 077_H card is C₁₀, C₀₁, C₁₁, C₂₀, C₀₂, C₃₀. Mixing up C₁₁ and C₂₀ when copying between them is an easy mistake.
- Ogden μ differs by a factor. LS-DYNA (and Ansys) write each term as μᵢ/αᵢ(...); Abaqus writes 2μᵢ/αᵢ²(...). The same rubber therefore has μᵢ(Abaqus) = μᵢαᵢ/2. The calculator converts.
- Compressibility. LS-DYNA takes Poisson's ratio PR; Abaqus takes D₁ = 2/K. The cards here use K from your PR and μ₀, the same relation LS-DYNA uses.
- N in the LS-DYNA card. N > 0 asks LS-DYNA to fit the constants itself from a load curve; the cards here are written with N = 0 and the constants filled in. Check the field layout against the manual for your LS-DYNA version before a production run.
What this calculator does not cover
- Viscoelasticity, rate dependence and hysteresis (LS-DYNA's NV terms, Prony series), the Mullins effect (stress softening after the first stretch) and permanent set.
- Foams and other strongly compressible materials (Blatz-Ko, Hill's foam, LS-DYNA *MAT_LOW_DENSITY_FOAM): the fits here assume a material that keeps its volume.
- Arruda-Boyce, Gent, Marlow and other forms, temperature dependence, and volumetric test data.
- Real test data has scatter: a fit that matches every point to 0.1% has usually been given too many terms. Prefer the simplest model that stays within the scatter and is stable over your part's strain range.
Skill-Lync's CAE courses go from material models and element formulations to setting up, running and checking nonlinear and crash models in LS-DYNA, HyperMesh and Ansys, the work CAE teams hire for.
Frequently asked questions
What is a hyperelastic material model?
A model for rubber-like materials that stretch elastically to large strains. Instead of a modulus it defines a strain energy W as a function of the stretches; every stress follows by differentiating W. Neo-Hookean, Mooney-Rivlin, Yeoh and Ogden are different forms of W, each with constants fitted to test data.
How do you find Mooney-Rivlin constants from test data?
Convert the test to stretch λ = 1 + engineering strain and nominal stress P. For uniaxial tension P = 2(λ − λ⁻²)(C10 + C01/λ), which is linear in C10 and C01, so a straight least-squares fit gives both. Fitting more terms works the same way. Ogden constants need a nonlinear fit because the exponents α are unknowns.
What is the difference between Mooney-Rivlin and Ogden?
Mooney-Rivlin writes W as a polynomial in the strain invariants I1 and I2; Ogden writes it as a sum of powers of the principal stretches, μ/α(λ1^α + λ2^α + λ3^α − 3), with the exponents fitted too. Ogden with three terms follows rubber to several hundred percent strain in every test mode; Mooney-Rivlin is reliable to about 100% and can turn unstable when it has negative constants.
Why is my hyperelastic model unstable?
Usually negative constants fitted to one kind of test. A fit to uniaxial tension can make the stress fall as the stretch rises in another mode, such as equibiaxial tension or compression. The material then has negative stiffness there (it fails the Drucker check) and the solver diverges or the elements collapse. Fit to more than one test mode, use fewer terms or a model like Ogden, and check stability over the strain range your part will see.
What Poisson's ratio should I use for rubber in LS-DYNA?
Close to 0.5: 0.495 to 0.4999 is typical, since rubber hardly changes volume. LS-DYNA turns PR into a bulk modulus K = 2μ0(1 + ν)/(3(1 − 2ν)). At 0.45, K is only 9.7 times the shear modulus, and a rubber block stretched to 6 times its length grows 71% in volume and carries about a third of the stress of real rubber.
What is the stretch ratio?
Final length divided by original length, λ = L/L0 = 1 + engineering strain. The true (logarithmic) strain is ln λ. A 10 mm block pulled 40 mm has λ = 5 and true strain 1.609; pulled 50 mm it has λ = 6 and true strain 1.792.
How many test modes do I need to fit a hyperelastic model?
At least two of uniaxial, equibiaxial and planar (pure shear) tension for a model you can trust in a real part. One uniaxial curve fixes the uniaxial response only: in our demo, a two-term Mooney-Rivlin fitted to uniaxial points up to 100% strain predicts 2.5 times the true equibiaxial stress at λ = 2.
Related on Skill-Lync
- A student report on the hyperelastic material models challenge
- A student report on the stretch ratio and ELFORM challenge
- LS-DYNA for structural mechanics course
- Consistent units calculator for FE models
- True stress-strain calculator (metals: engineering to true, MAT_024 hardening curve)
- Curve fitting calculator
- Yield strength calculator
- All free engineering tools
Sources
- R. W. Ogden, "Large deformation isotropic elasticity: on the correlation of theory and experiment for incompressible rubberlike solids", Proceedings of the Royal Society of London A 326 (1972), pp. 565-584: the three-term model and its constants for Treloar's rubber (6.3, 0.012, −0.1 kg/cm²; α 1.3, 5.0, −2.0).
- L. R. G. Treloar, "Stress-strain data for vulcanised rubber under various types of deformation", Transactions of the Faraday Society 40 (1944), pp. 59-70.
- M. Mooney, "A theory of large elastic deformation", Journal of Applied Physics 11 (1940), pp. 582-592; R. S. Rivlin, "Large elastic deformations of isotropic materials IV", Philosophical Transactions of the Royal Society A 241 (1948), pp. 379-397.
- O. H. Yeoh, "Some forms of the strain energy function for rubber", Rubber Chemistry and Technology 66 (1993), pp. 754-771.
- Abaqus Theory Guide, "Hyperelastic material behavior" (polynomial, Yeoh and Ogden forms, D₁ = 2/K₀, Drucker stability of fitted constants in the six test modes); Abaqus Keywords Guide, *HYPERELASTIC (data line order).
- LS-DYNA Keyword User's Manual, Volume II: *MAT_HYPERELASTIC_RUBBER (077_H) and *MAT_OGDEN_RUBBER (077_O): strain energy forms, PR and the N > 0 least-squares fit from a load curve.
- G. A. Holzapfel, Nonlinear Solid Mechanics, Wiley (2000): stresses in uniaxial, equibiaxial and pure shear deformation of incompressible materials.
- Skill-Lync student project reports for the "Hyperelastic material models" and "Calculate the stretch ratio by comparing the ELFORM (−2, −1, 1, 2) with Ogden material model" LS-DYNA challenges (124 and 170 reports read in full on 21 Nov 2026; counts above).