The Collapse Load of a Non-Associated Frictional Soil
When a frictional soil dilates less than it rubs (ψ < φ), its collapse load need not be unique. The limit theorems still certify a bracket. Every single number inside it — ours included — is an estimate.
TerraBrains Research · 26 October 2026 · updated 1 November 2026
The result on one axis
The problem
Real frictional soils flow with a dilatancy angle ψ smaller than their friction angle φ. With such non-associated flow the equations can lose ellipticity at failure (Vermeer & de Borst, 1984) and the collapse load need no longer be unique: no theorem guarantees one value for a given soil, footing and load path (Radenković, 1961; Salençon, 1977), and computed limit loads depend on the initial stresses and the loading sequence (Vermeer & de Borst, 1984). A finite-element program will always print a number. The question is which statements about that number can be proved.
What is certified
Upper bound. A non-associated soil cannot be stronger than the same soil with associated flow (Radenković, 1961; Drescher & Detournay, 1993). We compute the associated upper bound by finite-element limit analysis, which yields a rigorous bound when its optimisation is solved to full tolerance: Nq ≤ 10.74 here, against the exact associated value 10.66.
Lower bound. Radenković’s theorem bounds the collapse load from below by that of a fictitious associated material whose yield surface is the soil’s plastic potential (Salençon, 1977, Appendix V.A). For ψ = 0 that material is a liquid, so the theorem gives Nq ≥ 1 and no more. For 0 < ψ < φ the same construction gives a sharper, computed lower bound: on this footing 2.4679 ≤ Nq ≤ 10.735 at ψ = 10°, and 10.6152 ≤ Nq ≤ 10.735 at ψ = φ = 25°, where the exact associated value is 10.6621.
So for ψ = 0 the certified statement is 1 ≤ Nq ≤ 10.74. It is wide because the physics is: no classical limit theorem pins the ψ = 0 collapse load closer.
What we proved next (1 November 2026)
The lower value 1 cannot be improved by the theorem. Radenković’s theorem allows any comparison material that meets its conditions, not only the plastic potential. We show that for a cohesionless soil with ψ = 0 every admissible comparison material is a liquid, so no choice gives more than Nq ≥ 1. Salençon (1977, Ch. V §8.3) called the theorem too conservative for this case; it is exactly as conservative as it looks. A sharper lower bound needs an assumption about how the soil collapses.
Two conditional lower bounds. If the soil collapses by rigid sliding blocks, Nq ≥ 8.58 — Davis’s value, which is a bound within that class only (Drescher & Detournay, 1993). If the mean effective stress stays at least the surcharge wherever the soil deforms, Nq ≥ 3.17. Our own computed collapse fields show the mean stress falling close to zero at the footing edge, so the second assumption does not hold there; neither bound is claimed for this footing.
Open. Whether any bound above Nq = 1 holds with no assumption about the collapse mechanism is, as far as we can find, an open question.
What is estimated
| Result | Nq | Status |
|---|---|---|
| Upper bound — associated flow, finite-element limit analysis | Nq ≤ 10.74 | Certified |
| Lower bound — Radenković’s theorem (ψ = 0) | Nq ≥ 1 | Certified (exact) |
| Associated-flow exact value (Prandtl 1920) | 10.66 | Reference (ψ = φ only) |
| Davis (1968) reduced strength, tan φ* = sin φ | 8.58 | Estimate; a bound only if collapse is by rigid sliding blocks |
| Potts & Zdravković (2001) Vol. 2 FE, ψ = 0, smooth (Table 6.1) | 10.03 | Estimate |
| TerraBrains regularised FE, two independent settings | 10.3 – 10.6 | Estimate |
Davis’s reduced strength is exact only for mechanisms made of sliding rigid blocks (Drescher & Detournay, 1993); outside that class it is a useful engineering estimate, not a bound. Regularised finite-element models give values that depend on the regularisation: ours, at two independent fixed settings, give 10.3 – 10.6, and weaker settings give lower values. Run on the associated case, where the exact answer is known, the same two settings come out 2 – 8 % high — so read 10.3 – 10.6 as an upper-leaning estimate. We do not claim a limit of vanishing regularisation: on affordable meshes it does not converge. We do not claim a single certified collapse load for non-associated flow, a convergence rate, a mesh-converged finite-element limit, or an error bound on these estimates.
On TerraBrains
The certified bearing-capacity analysis returns, for a rigid strip on Mohr–Coulomb soil with 0 ≤ ψ ≤ φ (any cohesion, unit weight and surcharge): the certified bracket and the problem domain it holds for; the Davis estimate; the regularised estimates with whether each one is mesh-converged; and the exact associated value where one exists. It runs on the TerraBrains compute tier, from any project's Bearing capacity analysis (Certified bracket).
References
- Radenković, D. (1961). Théorèmes limites pour un matériau de Coulomb à dilatation non standardisée. C. R. Acad. Sci. Paris 252, 4103–4104.
- Salençon, J. (1977). Applications of the Theory of Plasticity in Soil Mechanics. Wiley. Ch. V §8.3 and Appendix V.A §3.
- Davis, E. H. (1968). Theories of plasticity and the failure of soil masses. In Soil Mechanics: Selected Topics, Butterworths.
- Drescher, A. & Detournay, E. (1993). Limit load in translational failure mechanisms for associative and non-associative materials. Géotechnique 43(3), 443–456.
- Vermeer, P. A. & de Borst, R. (1984). Non-associated plasticity for soils, concrete and rock. Heron 29(3).
- Potts, D. M. & Zdravković, L. (2001). Finite Element Analysis in Geotechnical Engineering: Application. Thomas Telford. §6.5.4.4, Table 6.1 (p. 232).
- Prandtl, L. (1920). Über die Härte plastischer Körper. Nachr. Ges. Wiss. Göttingen, 74–85.
Run the bracket on your own footing.
The analysis described above runs from a project's Bearing capacity analysis. Start free and try it with your own soil.