Skip to content
Ungewitter

All essaysEssay

Table I, recomputed

Ungewitter’s table of vault thrusts, rebuilt row by row as three-dimensional thrust networks: where the 1890 values lie within the range of admissible thrusts, and where they do not.

On page 135 of the Lehrbuch stands a small table that carries much of the book’s second chapter. For thirty kinds of vault, five rises and six build-ups of the webs, it gives the weight V₀ and the horizontal thrust H₀ per square metre of plan. Multiply by the area a buttress carries and you have the forces it must resist;Lesson 5 used the table that way. This essay asks how good its numbers are. It rebuilds every row as a three-dimensional thrust network (Lesson 19) and lets a solver find the range of thrusts each vault can exert: first in square bays, as the table is printed, then in the slightly oblong bays the book also admits, and finally, for three rows, in finer networks, to see how much the answers depend on the mesh.

What the table says

From the Lehrbuch

„Die Tabelle ist ermittelt auf Grund konstruierter Stützlinien und mit Anwendung der einfachen Formel H·h = G·a“The table was obtained from constructed lines of thrust and by applying the simple formula H·h = G·aUngewitter 1890, p. 133
„bei mittelhohen spitzbogigen Kreuzgewölben von etwa ⅔ Pfeilhöhe ist also ein Schub zu erwarten, der ungefähr gleich ⅓ des betreffenden Gewölbgewichtes (einer Hälfte) ist“for medium-high pointed cross vaults of about ⅔ rise, then, a thrust is to be expected of about ⅓ of the weight of the vault (one half)Ungewitter 1890, p. 136

Ungewitter and Mohrmann drew lines of thrust through typical vaults and condensed them into one equation of moments: the half vault’s weight G at its lever a balances the thrust H at its lever h, the rise of the line of thrust. Besides V₀ and H₀ the table lists the levers themselves: a between 0.15 and 0.23 of the span, h between 0.7 and 0.9 of the rise, and the height z at which the thrust crosses the wall face, between 1/8 and 1/3 of the rise. The two values of H₀ are for small and large vaults. The rows hold for square or slightly oblong quadripartite vaults with little doming (p. 134).

The table does not claim that a vault pushes with exactly this force. It names one line of thrust that fits inside the vault and the force that goes with it, which is the right claim to make. Heyman’s safe theorem says that if any line of thrust in equilibrium with the loads lies within the masonry, the structure stands (Lesson 13). The actual thrust is then somewhere between the least and the greatest admissible value, depending on how far the abutments give way (Heyman 1995); a buttress that yields a little lets the vault settle towards its least thrust.

Rebuilding the vaults

Thrust network analysis carries this reasoning into three dimensions (Block & Ochsendorf 2007; Block 2009). A plan of bars, the form diagram, stands for the paths along which forces may run. Because the loads are vertical, the horizontal forces must balance in plan by themselves; their balance is drawn as a second figure, theforce diagram, in which every bar of the plan appears as a parallel line as long as its horizontal force. Once the horizontal forces are fixed, the heights of the network follow from one linear system. Doubling every force halves the heights: the scale of the force diagram is the thrust. The network is admissible if it lies between the soffit and the back of the vault everywhere.

Each row of Table I becomes a square bay of 4 m and of 8 m and, since the book admits slightly oblong bays too (p. 134), two oblong bays of the same spans whose length is 1.25 and 1.5 times the span. All are laid out by the rib generator of the Studio’s vault bay: diagonal ribs and boundary arches as circular arcs reaching a common crown, brick ribs of 12 × 25 cm (enlarged in proportion beyond a 9 m diagonal), and the web thickness and material of the row. Each compartment is cut into 8 courses parallel to its boundary arch, spanning from rib to rib, which is how the book itself reads a web: as strips that bear on the ribs (p. 51). The loads come from the geometry: webs over their true surface, ribs, backing in the pockets up to a third of the rise, plaster. For the solid-brick rows our weights come within 23 % of the book’s V₀. The thrust may enter the springers anywhere between the springing and the top of the backing.

For one distribution of forces the admissible scales form an exact interval. The distributions themselves have one factor for each level of courses and one for the thrust of the ribs at the keystone, and finding the least and the greatest thrust over all of them is not a convex problem: a search finds good networks but cannot tell how good. In a square bay, though, every course is a small hanging-chain arch between two rib nodes at the same height, and every rib a chain whose forces grow towards the springer. An oblong bay keeps that property through its two mirror axes: its four ribs are mirror images, and the courses of the long and the short compartments push in proportion to their spans, so that their resultant runs along the rib, which is the book’s own argument (p. 61). Written in the reciprocals of those forces, the heights of the network become linear, and what is left is a handful of curves 1/x. Enclosing each curve between its tangents and a chord gives a linear programme whose optimum no network can beat; cutting the ranges into smaller pieces tightens it (branch and bound), and propagating every cut through the heights narrows the other ranges at once. Optimising networks in a fixed plan is an old idea (O’Dwyer 1999); what is added here is the bound. The solver reports the best network it found together with that bound, so every number in the tables carries its own proof of how far from the true extreme it can be. The equilibrium code is checked against the parabola and the catenary, against the closed-form solution of a loaded grid, and againstcompas_tna, the open-source TNA library of the Block Research Group, which gives the same heights to nine decimal places on the same network.

Form diagram (plan)

Force diagram

Thrust surface (axonometric)

Rise
Web
Span
State

Row IIIb, 8 m: the network admits 162–185 kg/m² of plan; the book gives 140–160 kg/m², below the range. Proved to within 0.05 % (least) and 0.02 % (greatest). Thrust per springer H: 26.3 kN; Load per springer V: 56.6 kN; Thrust enters the springer at: 0.67 m; Weight of the model: 361 kg/m².

Thrust per springer H
26.3 kN
Load per springer V
56.6 kN
Thrust enters the springer at
0.67 m
Weight of the model
361 kg/m²
Figure 1Pick a row of Table I. The solver builds the vault from the gothic rib layout, cuts each compartment into courses, and finds among the force distributions the least and the greatest thrust whose network stays inside the masonry, with a bound that shows how close it came. Left: the form diagram, ribs heavy, courses light. Middle: its force diagram (vermilion), with the springer reactions dashed. Right: the thrust surface over the ribs of the vault.

Square bays

The chart sets every row against its own book value: the grey band is Ungewitter’s range of H₀ around 1, the vermilion bars the thrust-network ranges for the 4 m and the 8 m vault, converted to kg/m² with the book’s own weight so that only the ratio of thrust to weight is compared. Of the 60 comparisons of square bays, the book’s range lies inside the TNA range 29 times, partly inside 7 times, below it 24 times and above it never.

Ungewitter’s thrust against the thrust-network range

Book: Table I, H₀ min–maxTNA range, 4 m vaultTNA range, 8 m vault×k = webs thickened until a network fits0.511.52Thrust relative to the book’s mean value (book = 1)1:81:31:22:35:6Ia (1:8, a) Book: Table I, H₀ min–max: 360–400 kg/m² TNA range, 4 m vault: 288–920 kg/m² TNA range, 8 m vault: 323–544 kg/m²IaIb (1:8, b) Book: Table I, H₀ min–max: 500–550 kg/m² TNA range, 4 m vault: 395–1,234 kg/m² TNA range, 8 m vault: 442–738 kg/m²IbIc (1:8, c) Book: Table I, H₀ min–max: 700–750 kg/m² TNA range, 4 m vault: 505–1,861 kg/m² TNA range, 8 m vault: 588–1,324 kg/m²IcId (1:8, d) Book: Table I, H₀ min–max: 950–1,000 kg/m² TNA range, 4 m vault: 627–2,816 kg/m² TNA range, 8 m vault: 760–1,970 kg/m²IdIe (1:8, e) Book: Table I, H₀ min–max: 1,600–1,700 kg/m² TNA range, 4 m vault: 1,020–5,315 kg/m² TNA range, 8 m vault: 1,263–3,601 kg/m²IeIf (1:8, f) Book: Table I, H₀ min–max: 1,375–1,438 kg/m² TNA range, 4 m vault: 789–4,813 kg/m² TNA range, 8 m vault: 1,047–3,233 kg/m²IfIIa (1:3, a) Book: Table I, H₀ min–max: 160–180 kg/m² TNA range, 4 m vault: 136–228 kg/m² TNA range, 8 m vault: 148–182 kg/m²IIaIIb (1:3, b) Book: Table I, H₀ min–max: 220–240 kg/m² TNA range, 4 m vault: 186–311 kg/m² TNA range, 8 m vault: 203–249 kg/m²IIbIIc (1:3, c) Book: Table I, H₀ min–max: 300–330 kg/m² TNA range, 4 m vault: 244–499 kg/m² TNA range, 8 m vault: 271–381 kg/m²IIcIId (1:3, d) Book: Table I, H₀ min–max: 420–450 kg/m² TNA range, 4 m vault: 321–697 kg/m² TNA range, 8 m vault: 360–588 kg/m²IIdIIe (1:3, e) Book: Table I, H₀ min–max: 710–750 kg/m² TNA range, 4 m vault: 559–1,242 kg/m² TNA range, 8 m vault: 628–1,057 kg/m²IIeIIf (1:3, f) Book: Table I, H₀ min–max: 650–719 kg/m² TNA range, 4 m vault: 455–1,028 kg/m² TNA range, 8 m vault: 602–1,117 kg/m²IIfIIIa (1:2, a) Book: Table I, H₀ min–max: 110–120 kg/m² TNA range, 4 m vault: 108–163 kg/m² TNA range, 8 m vault: 118–134 kg/m²IIIaIIIb (1:2, b) Book: Table I, H₀ min–max: 140–160 kg/m² TNA range, 4 m vault: 149–223 kg/m² TNA range, 8 m vault: 162–185 kg/m²IIIbIIIc (1:2, c) Book: Table I, H₀ min–max: 190–220 kg/m² TNA range, 4 m vault: 199–321 kg/m² TNA range, 8 m vault: 222–290 kg/m²IIIcIIId (1:2, d) Book: Table I, H₀ min–max: 280–320 kg/m² TNA range, 4 m vault: 281–483 kg/m² TNA range, 8 m vault: 320–465 kg/m²IIIdIIIe (1:2, e) Book: Table I, H₀ min–max: 480–550 kg/m² TNA range, 4 m vault: 479–847 kg/m² TNA range, 8 m vault: 549–814 kg/m²IIIeIIIf (1:2, f) Book: Table I, H₀ min–max: 475–550 kg/m² TNA range, 4 m vault: 347–655 kg/m² TNA range, 8 m vault: 474–755 kg/m²IIIfIVa (2:3, a) Book: Table I, H₀ min–max: 90–100 kg/m² TNA range, 4 m vault: 110–124 kg/m² TNA range, 8 m vault: 119–120 kg/m² (×1.69)IVa×1.69IVb (2:3, b) Book: Table I, H₀ min–max: 110–130 kg/m² TNA range, 4 m vault: 149–166 kg/m² TNA range, 8 m vault: 159–161 kg/m² (×1.7)IVb×1.7IVc (2:3, c) Book: Table I, H₀ min–max: 160–180 kg/m² TNA range, 4 m vault: 202–245 kg/m² TNA range, 8 m vault: 229–229 kg/m² (×1.16)IVc×1.16IVd (2:3, d) Book: Table I, H₀ min–max: 220–250 kg/m² TNA range, 4 m vault: 273–359 kg/m² TNA range, 8 m vault: 319–354 kg/m²IVdIVe (2:3, e) Book: Table I, H₀ min–max: 400–430 kg/m² TNA range, 4 m vault: 472–646 kg/m² TNA range, 8 m vault: 555–636 kg/m²IVeIVf (2:3, f) Book: Table I, H₀ min–max: 438–544 kg/m² TNA range, 4 m vault: 394–604 kg/m² TNA range, 8 m vault: 545–761 kg/m²IVfVa (5:6 bis 1, a) Book: Table I, H₀ min–max: 80–90 kg/m² TNA range, 4 m vault: 99–101 kg/m² (×1.18) TNA range, 8 m vault: 102–104 kg/m² (×2.35)Va×1.18×2.35Vb (5:6 bis 1, b) Book: Table I, H₀ min–max: 100–110 kg/m² TNA range, 4 m vault: 135–137 kg/m² (×1.18) TNA range, 8 m vault: 138–144 kg/m² (×2.37)Vb×1.18×2.37Vc (5:6 bis 1, c) Book: Table I, H₀ min–max: 150–160 kg/m² TNA range, 4 m vault: 196–244 kg/m² TNA range, 8 m vault: 208–213 kg/m² (×1.58)Vc×1.58Vd (5:6 bis 1, d) Book: Table I, H₀ min–max: 210–230 kg/m² TNA range, 4 m vault: 266–347 kg/m² TNA range, 8 m vault: 296–299 kg/m² (×1.13)Vd×1.13Ve (5:6 bis 1, e) Book: Table I, H₀ min–max: 350–370 kg/m² TNA range, 4 m vault: 446–599 kg/m² TNA range, 8 m vault: 506–584 kg/m²VeVf (5:6 bis 1, f) Book: Table I, H₀ min–max: 375–547 kg/m² TNA range, 4 m vault: 393–583 kg/m² TNA range, 8 m vault: 589–811 kg/m²Vf
  • Flat vaults, 1 : 8. All 12 comparisons inside.Row Ib at 8 m: the book gives 500–550 kg/m², the networks admit 442–738. The book’s rule of thumb, thrust about twice the weight, sits in the middle of the range of H : V, 1.2–6.25.
  • Low vaults, 1 : 3. All 12 inside. Row IIb at 8 m: 220–240against 203–249.
  • Round vaults, 1 : 2. 4 inside, 6 partly inside, 2 below. Row IIIb at 8 m:140–160 against 162–185. The book’s values sit near the least thrust the networks reach, sometimes just above it, sometimes just under.
  • Pointed vaults, 2 : 3 and 5 : 6. 22 of 24 below. The book’s ⅓ of the weight for a 2 : 3 vault meets a least thrust of the networks of 0.36 of the weight. For the thinnest webs no network of this pattern fits at all, and the solver proves it: webs and ribs had to be thickened by up to ×2.4 before one did (9 comparisons). Row IVb at 8 m: 110–130against 159–161, with webs and ribs ×1.7 as thick; at the least thickness that works, only one network fits, so the range closes to a single value.
Table I and the thrust-network ranges, per m² of plan (kg/m²)
RowRise f : sSpanV₀ bookV₀ modelH₀ bookH₀ TNAH : V TNAGap least · greatest, %ThickenedBook
Ia1:84 m200260360–400288–9201.44–4.60 · 0.1as builtinside
Ia1:88 m200243360–400323–5441.62–2.720.09 · 0.1as builtinside
Ib1:84 m270303500–550395–1,2341.46–4.570 · 0.1as builtinside
Ib1:88 m270286500–550442–7381.64–2.730.1 · 0.1as builtinside
Ic1:84 m370401700–750505–1,8611.36–5.030 · 0.01as builtinside
Ic1:88 m370383700–750588–1,3241.59–3.580.1 · 0.03as builtinside
Id1:84 m500515950–1,000627–2,8161.25–5.630 · 0as builtinside
Id1:88 m500497950–1,000760–1,9701.52–3.940.08 · 0.1as builtinside
Ie1:84 m8508411,600–1,7001,020–5,3151.2–6.250 · 0as builtinside
Ie1:88 m8508241,600–1,7001,263–3,6011.49–4.240 · 0.1as builtinside
If1:84 m7257491,375–1,438789–4,8131.09–6.640 · 0as builtinside
If1:88 m8138621,375–1,4381,047–3,2331.29–3.980 · 0.09as builtinside
IIa1:34 m230293160–180136–2280.59–0.990.09 · 0.03as builtinside
IIa1:38 m230274160–180148–1820.64–0.790.08 · 0as builtinside
IIb1:34 m310340220–240186–3110.6–10.09 · 0.01as builtinside
IIb1:38 m310321220–240203–2490.66–0.80.07 · 0.01as builtinside
IIc1:34 m420447300–330244–4990.58–1.190.1 · 0.03as builtinside
IIc1:38 m420428300–330271–3810.64–0.910.09 · 0.04as builtinside
IId1:34 m570573420–450321–6970.56–1.220.09 · 0.05as builtinside
IId1:38 m570552420–450360–5880.63–1.030.09 · 0.01as builtinside
IIe1:34 m1,000932710–750559–1,2420.56–1.240.1 · 0.05as builtinside
IIe1:38 m1,000911710–750628–1,0570.63–1.060.09 · 0.1as builtinside
IIf1:34 m913937650–719455–1,0280.5–1.130.1 · 0.04as builtinside
IIf1:38 m1,1721,217650–719602–1,1170.51–0.950.1 · 0.01as builtinside
IIIa1:24 m260333110–120108–1630.42–0.630.08 · 0.02as builtinside
IIIa1:28 m260309110–120118–1340.45–0.520.06 · 0.06as builtoverlaps
IIIb1:24 m350385140–160149–2230.42–0.640.09 · 0as builtoverlaps
IIIb1:28 m350361140–160162–1850.46–0.530.05 · 0.02as builtbelow
IIIc1:24 m480504190–220199–3210.41–0.670.1 · 0.02as builtoverlaps
IIIc1:28 m480480190–220222–2900.46–0.60.07 · 0.1as builtbelow
IIId1:24 m700644280–320281–4830.4–0.690.1 · 0.05as builtoverlaps
IIId1:28 m700619280–320320–4650.46–0.660.1 · 0.01as builtoverlaps
IIIe1:24 m1,2001,042480–550479–8470.4–0.710.09 · 0.01as builtinside
IIIe1:28 m1,2001,017480–550549–8140.46–0.680.09 · 0.03as builtoverlaps
IIIf1:24 m1,0001,047475–550347–6550.35–0.660.1 · 0.1as builtinside
IIIf1:28 m1,3001,408475–550474–7550.36–0.580.09 · 0.1as builtinside
IVa2:34 m29034890–100110–1240.38–0.430.08 · 0.01as builtbelow
IVa2:38 m29032090–100119–1200.41–0.410 · 0×1.69below
IVb2:34 m380400110–130149–1660.39–0.440.08 · 0.02as builtbelow
IVb2:38 m380372110–130159–1610.42–0.420 · 0×1.7below
IVc2:34 m530518160–180202–2450.38–0.460.08 · 0as builtbelow
IVc2:38 m530490160–180229–2290.43–0.430 · 0.01×1.16below
IVd2:34 m750656220–250273–3590.36–0.480.06 · 0.01as builtbelow
IVd2:38 m750627220–250319–3540.43–0.470.06 · 0as builtbelow
IVe2:34 m1,3001,050400–430472–6460.36–0.50.09 · 0.03as builtbelow
IVe2:38 m1,3001,021400–430555–6360.43–0.490.09 · 0.09as builtbelow
IVf2:34 m1,3131,210438–544394–6040.3–0.460.1 · 0.03as builtinside
IVf2:38 m1,8091,715438–544545–7610.3–0.420.1 · 0.01as builtbelow
Va5:64 m34038580–9099–1010.29–0.30 · 0.05×1.18below
Va5:68 m34035480–90102–1040.3–0.310 · 0.04×2.35below
Vb5:64 m450442100–110135–1370.3–0.30 · 0.04×1.18below
Vb5:68 m450410100–110138–1440.31–0.320 · 0.01×2.37below
Vc5:64 m650571150–160196–2440.3–0.380.03 · 0.1as builtbelow
Vc5:68 m650539150–160208–2130.32–0.330.01 · 0×1.58below
Vd5:64 m900721210–230266–3470.3–0.390.09 · 0as builtbelow
Vd5:68 m900689210–230296–2990.33–0.330 · 0×1.13below
Ve5:64 m1,5001,151350–370446–5990.3–0.40.08 · 0.07as builtbelow
Ve5:68 m1,5001,118350–370506–5840.34–0.390.06 · 0.08as builtbelow
Vf5:64 m1,6251,452375–547393–5830.24–0.360.1 · 0.06as builtoverlaps
Vf5:68 m2,4312,174375–547589–8110.24–0.330.1 · 0.01as builtbelow

Oblong bays

The book applies the table to oblong bays as well, as long as they are not too far from the square, and gives the rule: the rise ratio is to be taken over the span in the direction of the thrust sought (p. 137). Across the vessel, a bay of span s and length λs is read as a vault of span s. The 120 oblong comparisons test that rule with the thrust across the vessel. At 1 : 1.25 the book’s range lies inside the TNA range 25 times, partly inside 9 times and below it 26 times; at 1 : 1.5 the counts are 24,7 and 29. Over all180 comparisons, square and oblong: 78 inside, 23 partly, 79below.

Where both the square and the 1 : 1.5 bay fit as built, the ratio of thrust to weight across the vessel changes by a median of +3 % at the least and −2 % at the greatest end (28 pairs). What changes more is whether a network fits at all. With the crowns level, the courses of the long compartments have to climb from the same ribs to the same crown over a longer span, and the pattern needs thicker webs:42 of the 120 oblong comparisons had to be thickened, up to ×8.2, against9 of the 60 square ones. Row IIIb at 8 m and 1 : 1.5: 140–160against 170–177 kg/m², with webs and ribs ×1.38 as thick.

The thickest cases are the 4 steep bays IVf, Va, Vb, Vf at 8 m and 1 : 1.5, which need ×5.5 to ×8.2. At the least thickness that works, their networks touch the back of the webs along the crown lines and the soffit at the haunches of the last course before each wall arch, as the thinnest arch touches its ring at alternate faces (Lesson 13): that course, the longest in the bay, sets the thickness. The condition is one of geometry. The course carries its own, nearly uniform load, and the shape of its funicular does not depend on how heavy that load is, so a bay twice as large with webs of the same thickness needs twice the factor. In all 15 pairs of bays without fill that had to be thickened at both spans, the 8 m bay needs twice the factor of the 4 m bay, to the precision of the bisection. Row Va at 1 : 1.5 needs ×2.75 at 4 m and so ×5.51 at 8 m. The filled rows do not follow that rule: their fill, a fixed height above the crown, does not grow with the bay.

For the thrust across the vessel the book’s rule holds about as well as it does for the square; for the thin pointed webs the course pattern says that an oblong bay asks more of the masonry.

Oblong bays, s × λs: Table I for the rise over the span s against the thrust-network ranges across the vessel, per m² of plan (kg/m²)
RowRise f : sSpan sH₀ bookBay 1 : 1.25Bay 1 : 1.5
H₀ TNAGap, %BookH₀ TNAGap, %Book
Ia1:84 m360–400294–8780 · 0.09inside298–8580 · 0.1inside
Ia1:88 m360–400324–5410.09 · 0.03inside324–5420.09 · 0.05inside
Ib1:84 m500–550402–1,1830 · 0.09inside407–1,1600 · 0.1inside
Ib1:88 m500–550443–7350.09 · 0.09inside444–7370.09 · 0.1inside
Ic1:84 m700–750512–1,8380 · 0.02inside517–1,8210 · 0inside
Ic1:88 m700–750589–1,3110.09 · 0.1inside589–1,3310.09 · 0.04inside
Id1:84 m950–1,000635–2,7970 · 0inside639–2,7840 · 0inside
Id1:88 m950–1,000761–1,9280.1 · 0.01inside761–1,9420.1 · 0.01inside
Ie1:84 m1,600–1,7001,028–5,3080 · 0inside1,033–5,3030 · 0inside
Ie1:88 m1,600–1,7001,265–3,6460 · 0.05inside1,265–3,7040 · 0.03inside
If1:84 m1,375–1,438794–4,8180 · 0inside798–4,8240 · 0inside
If1:88 m1,375–1,4381,049–3,3700 · 0.09inside1,050–3,4320 · 0inside
IIa1:34 m160–180139–2200.1 · 0.06inside140–2060.09 · 0.03inside
IIa1:38 m160–180147–1760.08 · 0.01overlaps156–166 (×1.05)0 · 0.02overlaps
IIb1:34 m220–240190–3000.09 · 0.07inside192–2850.1 · 0.02inside
IIb1:38 m220–240202–2420.07 · 0.01inside209–2280.04 · 0.01overlaps
IIc1:34 m300–330249–4640.1 · 0.1inside253–4680.1 · 0.03inside
IIc1:38 m300–330270–3680.09 · 0.04inside269–3480.08 · 0.1inside
IId1:34 m420–450328–7030.1 · 0.03inside332–6610.1 · 0inside
IId1:38 m420–450359–5720.09 · 0.03inside358–5340.09 · 0.09inside
IIe1:34 m710–750568–1,2530.1 · 0.01inside574–1,2590.06 · 0.01inside
IIe1:38 m710–750626–1,0770.09 · 0.01inside628–1,0520.09 · 0.02inside
IIf1:34 m650–719458–1,0250.1 · 0.02inside460–1,0220.09 · 0.01inside
IIf1:38 m650–719604–1,1290.1 · 0.02inside605–1,1490.1 · 0.01inside
IIIa1:24 m110–120113–1540.09 · 0.03overlaps124–1450.05 · 0.09below
IIIa1:28 m110–120128–129 (×1.06)0 · 0below125–130 (×1.44)0.01 · 0.07below
IIIb1:24 m140–160153–2120.09 · 0.01overlaps165–2020.06 · 0.09below
IIIb1:28 m140–160175–176 (×1.03)0 · 0below170–177 (×1.38)0 · 0.03below
IIIc1:24 m190–220202–3340.09 · 0.03overlaps209–3110.1 · 0.02overlaps
IIIc1:28 m190–220231–2740.05 · 0.04below243–256 (×1.08)0 · 0below
IIId1:24 m280–320283–4970.1 · 0overlaps295–4860.09 · 0.04overlaps
IIId1:28 m280–320327–4380.07 · 0.03below338–4170.08 · 0.09below
IIIe1:24 m480–550480–8710.07 · 0.01overlaps494–8810.09 · 0.09overlaps
IIIe1:28 m480–550544–7830.09 · 0.1overlaps555–7760.09 · 0.03below
IIIf1:24 m475–550357–6560.08 · 0.02inside367–6550.05 · 0.09inside
IIIf1:28 m475–550480–7570.09 · 0.06overlaps488–7610.1 · 0.1overlaps
IVa2:34 m90–100101–108 (×1.45)0 · 0.01below105–135 (×1.99)0 · 0.01below
IVa2:38 m90–100101–110 (×2.89)0 · 0.01below102–138 (×3.98)0 · 0.1below
IVb2:34 m110–130136–145 (×1.45)0 · 0below140–181 (×1.99)0 · 0.09below
IVb2:38 m110–130135–147 (×2.89)0 · 0below136–185 (×3.98)0 · 0.01below
IVc2:34 m160–180194–2410.08 · 0.02below201–253 (×1.33)0 · 0.02below
IVc2:38 m160–180195–207 (×1.93)0 · 0.01below197–253 (×2.65)0 · 0.02below
IVd2:34 m220–250260–3690.09 · 0.02below275–3770.09 · 0.08below
IVd2:38 m220–250283–303 (×1.39)0 · 0.02below287–354 (×1.9)0 · 0.03below
IVe2:34 m400–430447–6580.09 · 0.01below449–6700.09 · 0.08below
IVe2:38 m400–430507–534 (×1.16)0 · 0below507–653 (×1.59)0 · 0.01below
IVf2:34 m438–544381–6000.1 · 0.02inside366–5990.08 · 0.06inside
IVf2:38 m438–544667–708 (×3.1)0.06 · 0below614–926 (×7.18)0 · 0below
Va5:64 m80–9099–107 (×2)0 · 0.01below95–125 (×2.75)0 · 0below
Va5:68 m80–9099–109 (×4)0 · 0.02below92–131 (×5.51)0 · 0below
Vb5:64 m100–110135–145 (×2)0 · 0.01below129–168 (×2.75)0 · 0below
Vb5:68 m100–110135–148 (×4)0 · 0.01below125–177 (×5.51)0 · 0below
Vc5:64 m150–160204–213 (×1.33)0 · 0.01below192–245 (×1.83)0 · 0.01below
Vc5:68 m150–160204–215 (×2.66)0 · 0below188–247 (×3.67)0 · 0below
Vd5:64 m210–230288–3340.06 · 0below273–344 (×1.33)0 · 0below
Vd5:68 m210–230290–311 (×1.93)0 · 0.01below267–343 (×2.65)0 · 0below
Ve5:64 m350–370479–5760.08 · 0.01below471–587 (×1.1)0 · 0.01below
Ve5:68 m350–370504–524 (×1.6)0 · 0.01below465–587 (×2.2)0 · 0below
Vf5:64 m375–547417–5780.1 · 0.06overlaps411–5680.08 · 0.01overlaps
Vf5:68 m375–547622–7930.09 · 0.1below662–889 (×8.22)0 · 0.01below

How sure are these numbers?

Two questions decide whether a range in the tables can be trusted: has the solver really found the extremes, and would a finer network give other ones? The first is answered by the bound that comes with every result. Of the180 comparisons with a network, all have the least thrust proved to within 0.1 %. The greatest thrust is harder (the network has to flatten against the soffit, and many distributions come close), and the search for it must be helped: the solver tightens the ranges of all forces against each other before it branches, and refines every network it finds in a small neighbourhood. Its gap is at most 0.1 %, with a median of 0.02 %.That the thickening factor is the least that works is proved the same way for 48 of the 51 thickened comparisons (for the others the bisection met a factor it could neither prove nor refute):below it, the linear programme has no solution at all.

The second question is answered by solving rows Ib, IIIb, IVd at 8 m again with sixteen and thirty-two courses per compartment, each finer run starting from the coarser optimum. The least thrust hardly moves, by at most2.9 % from eight to thirty-two courses. The greatest thrust fallsin every row, by 0.3 % to 5.6 %: a coarse network spans each course and each rib with a few long bars, which may cut across the curved soffit between their nodes, and so admits shallower states than the vault really allows. How far would it fall with infinitely many courses? Richardson extrapolation answers that from three meshes if the changes shrink by a steady factor, and Roache’s grid convergence index turns the answer into a band (Roache 1994).Here no sequence shows the order a smooth discretisation would have (the changes shrink too slowly, too fast or change sign), so every band in the mesh table uses Roache’s conservative factor of 3.On that estimate the greatest H : V of an infinitely fine pattern lies within 0.468–0.537 for IIIb and within0.465–0.477 for IVd. For the flat vault Ib it lies within 2.525–2.636.The least thrust moves less: 0.458–0.497 for IIIb, 0.423–0.428 for IVd; for Ib the finer proofs again leave more open than the mesh changes. Read the upper ends of the ranges in the tables as generous by up to about 6 %, and the bands as estimates, not proofs.

The same rows with finer form diagrams: H : V per abutment, 8 m vaults; ∞: estimated band for an infinitely fine diagram
RowCoursesNodesH : V leastH : V greatestGap least · greatest, %
Ib, 8 m82251.6382.7320.1 · 0.1
Ib, 8 m164251.642.5970.1 · 0.1
Ib, 8 m328171.6412.5790.1 · 0.11
Ib, 8 m∞estimatenot bounded2.525–2.636proof still open · monotone, p ≈ 2.9, Fs 3
IIIb, 8 m82250.4640.5270.05 · 0.02
IIIb, 8 m164250.4710.5140.04 · 0.01
IIIb, 8 m328170.4770.5020.07 · 0.01
IIIb, 8 m∞estimate0.458–0.4970.468–0.537monotone, p ≈ 0.1, Fs 3 · monotone, p ≈ 0.2, Fs 3
IVd, 8 m82250.4260.4720.06 · 0
IVd, 8 m164250.4260.4730.09 · 0
IVd, 8 m328170.4260.4710.1 · 0
IVd, 8 m∞estimate0.423–0.4280.465–0.477oscillating, Fs 3 · oscillating, Fs 3

Reading the difference

For flat and low vaults the two methods agree: Ungewitter’s thrust is one admissible state, and not an extreme one. For round vaults the book’s thrust lies at the bottom of the range, for pointed vaults below anything our networks reach. That does not make the table wrong, and it does not make it unsafe. It shows two different pictures of the same vault.

Ungewitter’s line of thrust runs from the top of the crown to the wall face at a fifth to a quarter of the rise, with the weight of the half vault acting close to the wall. It treats the backing and the springer courses, which were built with the wall before the vault (p. 85), as part of the abutment; the thrust passes through solid masonry in the haunch. Our networks spring at the corner of the bay and must stay within webs of 12 to 30 cm all the way down; the steep lower parts of a pointed vault are hard to follow with a funicular, so the network has to stay shallow and its thrust rises. A higher backing would move the result towards the book.

The second reason lies in the form diagram. Forces here run only along courses and ribs, Ungewitter’s own idealisation. A real web is a shell and can carry load across the courses too; denser diagrams and a true optimisation over all force distributions find lower thrusts and thinner admissible vaults (Maia Avelino et al. 2021). The least thrust reported here is proved for the pattern, but over all form diagrams it is only an upper estimate of the true least thrust, and the need to thicken the thinnest webs says more about the pattern than about the vaults, which stand with half-brick webs in hundreds of churches.

So the comparison tells the designer something useful. Where the book lies inside the range, its thrust is a fair middle value. Where it lies below the range of a slicing analysis, the book has counted on the favourable behaviour of the haunch masonry and of the webs as shells. A buttress sized for Table I then relies on that behaviour, and the margin of the kern rule (Lesson 5) is what covers it.

What the computation does not show

  • The proofs hold for the course pattern only; a denser form diagram, with forces running across the courses too, can reach lower thrusts.
  • 8 courses per compartment; the least thrust is converged, the greatest is generous (see the mesh table and its estimated bands).
  • Oblong bays are compared across the vessel only, by the book’s rule of the transverse span; the thrust along the vessel, against the transverse arch, is balanced by the next bay.
  • The web surface between the ribs is a model, not a survey: each course follows the middle part of its boundary arch, set on the ribs.
  • Row f, the filled vault, has a fill up to the floor; the network may use all of it, so its range is wide.
  • Strength, sliding and the deformation of the abutments are not checked; TNA, like Ungewitter’s table, is about equilibrium and geometry.

Sources

  1. Ungewitter, G., and K. Mohrmann. 1890. Lehrbuch der gotischen Konstruktionen, 3rd ed., part I. Leipzig: Weigel. Table I, p. 135; pp. 51, 54, 61, 85, 93, 133–137.
  2. Block, Philippe, and John Ochsendorf. 2007. “Thrust Network Analysis: A New Methodology for Three-Dimensional Equilibrium.” Journal of the IASS 48 (3): 167–73.
  3. Block, Philippe. 2009. “Thrust Network Analysis: Exploring Three-Dimensional Equilibrium.” PhD diss., MIT.
  4. Block, Philippe, and Lorenz Lachauer. 2014. “Three-Dimensional (3D) Equilibrium Analysis of Gothic Masonry Vaults.” International Journal of Architectural Heritage 8 (3): 312–35.
  5. Heyman, Jacques. 1995. The Stone Skeleton. Cambridge: Cambridge University Press.
  6. Maia Avelino, Ricardo, Antonino Iannuzzo, Tom Van Mele, and Philippe Block. 2021. “Assessing the Safety of Vaulted Masonry Structures Using Thrust Network Analysis.” Computers & Structures 257: 106647.
  7. O’Dwyer, Dermot. 1999. “Funicular Analysis of Masonry Vaults.” Computers & Structures 73: 187–97.
  8. Roache, Patrick J. 1994. “Perspective: A Method for Uniform Reporting of Grid Refinement Studies.” Journal of Fluids Engineering 116 (3): 405–13.
  9. Block Research Group, ETH Zürich. compas_tna (MIT licence), the reference for the algorithm; this site’s solver is an independent re-implementation in TypeScript.