After Jacques Heyman,“The stone skeleton”, International Journal of Solids and Structures (1966), doi:10.1016/0020-7683(66)90018-7.
- Against published solutions and an independent implementation
- 37 of 37 checks pass· see the checks
- Against the paper’s figures
- Heyman’s paper was not read here; the table compares with the published solutions tabulated by Rizzi et al. (2010).
Research, not a design tool. Nothing here is an engineering result.
Rigid voussoirs, no tension: find the least thickness of a circular arch and the load that makes a ring collapse, with a linear program and every hinge mechanism.
Masonry as a plastic material
Heyman’s plastic theory of masonry rests on three assumptions: the stone carries no tension, it never crushes, and the blocks never slide on each other. Under them an arch is safe if a line of thrust in equilibrium with the loads can be drawn inside the masonry. Where the line touches a face, the joint can open there and acts as a hinge; with enough hinges the arch becomes a mechanism and collapses. Strength hardly matters: stability is a question of geometry.
The arch as a linear program
Our arch is a circle cut into N rigid voussoirs by radial joints. At each joint the unknowns are the force the left part exerts on the right part and its moment about the middle of the joint. Each voussoir gives three equilibrium equations with its weight and, if there is one, the point load λP. Each joint must be in compression, n ≥ 0, and the resultant must cross it inside the stone, |M| ≤ n·t/2. Equilibrium fixes every joint force once the forces and the moment at the left springing are chosen, so each problem is a linear program in four unknowns with three inequalities per joint.
Two ways of placing the weight are compared. With true weights each voussoir weighs what its annular sector weighs and acts at its centroid, which is Milankovitch’s solution. With the weight spread uniformly along the centreline, as in Heyman’s closed form and the CCR solution, the outer part of each block is lighter than it should be.
How we reproduce it
The least thickness under self-weight is found by bisection on t/r. Each step solves the linear program for the state that breaks the joint conditions least; the thickness is accepted when that violation is not positive. The collapse load of a point load is the largest λ the linear program allows. As a kinematic check every mechanism with four hinges on alternating faces is enumerated; the smallest load multiplier among them is an upper bound, and it equals the static value. The solver is our own Rust code, compiled to WebAssembly for this page.
What the run shows
The replay is the solver’s own run on a semicircle of 101 voussoirs with true weights. The first steps bracket the answer: on the steps where the arch is too thin, the line of thrust leaves the stone and turns red. At t/r = 0.107451 the line touches five faces: the extrados at the crown and at both springings, the intrados at both haunches, 54.5° from the crown. The arch then opens as a mechanism about those five hinges.
· ·
Change the angle of embrace, the number of voussoirs or the weight model and the page recomputes live. A flat arch needs very little thickness; past a semicircle it thickens fast.
· ·
Comparison with published solutions
Rizzi, Cocchetti, Colasante and Rusconi (2010) list, in their Table 1 (p. 469), the least thickness t/r, the haunch hinge angle β from the crown and the thrust h = H/(w r), with w = γ t d, for four angles of embrace. Our runs use 2·round(800·α/90) voussoirs; β is the vertex of the parabola through the eccentricities around the haunch hinge. ✓ marks agreement within 10⁻⁴ relative for t/r and h and 0.01° for β.
| Quantity | Published | Stabileo | Difference |
|---|---|---|---|
| α = 60°, true weights against Milankovitch | |||
| Least thickness t/r | 0.0228489 | 0.0228488 | −0.00047 % |
| Haunch hinge β | 39.4606 | 39.4606 | +0.00000° |
| Thrust H/(w r) | 0.825052 | 0.8250517 | −0.00004 % |
| α = 60°, weight on the centreline against CCR | |||
| Least thickness t/r | 0.0228482 | 0.0228481 | −0.00050 % |
| Haunch hinge β | 39.4608 | 39.4608 | +0.00000° |
| Thrust H/(w r) | 0.825094 | 0.8250944 | +0.00005 % |
| α = 90°, true weights against Milankovitch | |||
| Least thickness t/r | 0.107478 | 0.1074779 | −0.00005 % |
| Haunch hinge β | 54.4840 | 54.4840 | −0.00004° |
| Thrust H/(w r) | 0.620881 | 0.6208812 | +0.00003 % |
| α = 90°, weight on the centreline against CCR | |||
| Least thickness t/r | 0.107426 | 0.1074263 | +0.00027 % |
| Haunch hinge β | 54.4963 | 54.4963 | +0.00003° |
| Thrust H/(w r) | 0.621772 | 0.6217714 | −0.00009 % |
| α = 120°, true weights against Milankovitch | |||
| Least thickness t/r | 0.327607 | 0.3276064 | −0.00017 % |
| Haunch hinge β | 63.7402 | 63.7402 | −0.00000° |
| Thrust H/(w r) | 0.342263 | 0.3422628 | −0.00006 % |
| α = 120°, weight on the centreline against CCR | |||
| Least thickness t/r | 0.326547 | 0.3265466 | −0.00012 % |
| Haunch hinge β | 64.0072 | 64.0072 | +0.00004° |
| Thrust H/(w r) | 0.349556 | 0.3495564 | +0.00011 % |
| α = 145°, true weights against Milankovitch | |||
| Least thickness t/r | 0.763995 | 0.7639949 | −0.00001 % |
| Haunch hinge β | 49.6698 | 49.6698 | +0.00002° |
| Thrust H/(w r) | 0.0391491 | 0.0391490 | −0.00019 % |
| α = 145°, weight on the centreline against CCR | |||
| Least thickness t/r | 0.740638 | 0.7406376 | −0.00005 % |
| Haunch hinge β | 56.2271 | 56.2272 | +0.00006° |
| Thrust H/(w r) | 0.0681781 | 0.0681781 | −0.00000 % |
The same table gives Heyman’s closed form, which is thinner: t/r = 0.105965 at α = 90°, 1.4 % below Milankovitch, and 15.5 % at α = 145°. As Rizzi et al. explain, it makes the resultant tangent to the intrados instead of the line of thrust. We show it for comparison and do not check against it.
With few voussoirs the hinges can only sit on joints, so t/r approaches the published value from below: 0.107374 with 51 voussoirs, 0.107451 with 101 and 0.107464 with 201. An independent implementation of the same model in Python with the HiGHS solver gives the same values to 4 × 10⁻⁶.
A ring under a point load
The second tab loads a semicircular ring, inner radius 2.35 m, thickness 0.30 m, depth 1 m and γ = 20 kN/m³, without backfill, with a vertical load on the extrados above a quarter of the clear span. No published collapse load is compared for this ring: it is checked against the kinematic bound and the independent implementation. The load grows until the thrust line touches four faces; then the arch turns into three rigid bodies.
| Voussoirs | Static (kN) | Kinematic (kN) |
|---|---|---|
| 10 | 4.2833 | 4.2833 |
| 20 | 3.7497 | 3.7497 |
| 31 | 3.7441 | 3.7441 |
| 40 | 3.6444 | 3.6444 |
| 80 | 3.5888 | not enumerated |
| 160 | 3.5881 | not enumerated |
With 31 voussoirs the hinges form at 29.0° intrados, 81.3° extrados, 139.4° intrados, 180.0° extrados, measured from the left springing. Up to 40 voussoirs, where every mechanism is enumerated, the two bounds agree to 10⁻¹³.
Limitations
Rigid voussoirs, no sliding and unlimited compressive strength; the conditions are checked at the joints only; no backfill, no settlement. The point load is vertical and applied on the extrados. Agreement with published analytical solutions and with another implementation is a numerical comparison, not a laboratory test and never an assessment of a real arch.
Heyman (1966) · Rizzi et al. (2010)
Cite this
Stabileo Research, “How thin can a stone arch be?”, reproduction of Jacques Heyman (1966), doi:10.1016/0020-7683(66)90018-7. Implementation: research/mechanics/heyman-1966 at commit 5319f7214747. Engine: the Stabileo engine at the same commit.