On the Structure of the Roman Pantheon

Robert Mark and Paul Hutchinson

Since the time of its construction, the bold, brilliantly simple schema of Hadrian's
Pantheon has inspired much emulation, commendation, and even fear. Modern
commentators tend to view the building as a high point in an "architectural rev-
olution" brought about mainly through the Roman development of a superior poz-
zolana concrete that lent itself to the forming of unitary, three-dimensional struc-
tures . Other factors cited for the technical success of the Pantheon include the use
of a series of massive, concentric stepped rings and the lightening of the dome by
coffering and gradated, light-weight aggregates. To investigate these theories, and
thereby to understand late Roman design rationale better, a numerical-computer
modeling study of the dome structure was undertaken. It yielded several surprises.
Analysis revealed that the stepped rings induced higher, rather than lower, crit-
ical stresses in an uncracked dome model. But by allowing the model to crack
freely, a salutary effect was caused by the rings. The cracked model closely sim-
ulated the behavior of the actual dome, which was discerned to act structurally as
an array of arches. In fact, the configuration of the dome seems to indicate that
the builders understood this-which points to the conclusion that late Roman ar-
chitectural development was not so closely tied to structural innovation as has
been generally believed.

No other single building has captured the attention of before the time of the Pantheon, but we know of none that
architects, particularly since the Renaissance, as has the even approached its scale. The largest of the earlier, extant
Pantheon. Constructed between ca. 118 and 128, it has an domes seems to have been part of a bath complex at Baiae.
interior space of awesome scale (Fig. 1), and it is the most Dating from the first half of the preceding century, the so-
completely preserved building of the Imperial Roman cap- called "Temple of Mercury" has a clear span of 21.5 meters. 3
ital. The 43.4-meter clear span of the dome was unmatched For many historians, the Pantheon represents a kind of
for well over a millennium, and not substantially surpassed culmination of the "Roman architectural revolution"
until the adoption of steel and reinforced concrete in the brought to fruition during the course of the first century
modern era. 1 through the adoption of high-quality concrete that could
The enormous influence of the Pantheon is easily traced be more readily used for the construction of curvilinear
through numerous buildings from the later Roman period, architectural forms. 4 Unlike simple lime mortar, which is
and again from the beginning of the Renaissance well into produced by adding water to a mixture of quicklime and
the twentieth century. 2 Earlier prototypes, though, are not sand, and sets when all the water has been evaporated into
so readily identified. Domed buildings were not uncommon the atmosphere or absorbed into the surrounding masonry,

The authors gratefully acknowledge the good advice and help received the Middle Ages and the Renaissance," Classical Influences on European
during various phases of this project from Tracy Cooper, Slobodan Cur- Culture AD 500-1500, ed. R.R. Bolgar, Cambridge, 1971, 259ff.). The
ci(, William Loerke, William L. MacDonald, Claudia Marchitiello Mark, lesson of Roman architecture for 20th-century design, according to Le
and Jean-Herve Prevost. Research on the structure of ancient Roman ar- Corbusier, was to be derived from "immense cupolas with their supporting
chitecture, of which this paper is a part, was supported by fellowships drums . . . all held together with Roman cement ... absence of verbosity,
and grants from the John Simon Guggenheim Foundation, the National good arrangement, a single idea, daring and unity in construction, the
Science Foundation, the National Endowment for the Humanities, and use of elementary shapes" (Towards a New Architecture [Paris, 1923],
the Andrew W. Mellon and the Alfred P. Sloan Foundations. New York, 1960, 146-47).
1 The clear span of St. Peter's in Rome (constructed 1585-90) is 42m. Metal 3 Illustrated in Giedion, 136, 137.
structures in the late 19th century reached spans as great as 113m (for the 4 This view is exemplified by John Ward-Perkins' somewhat startling re-
Galerie des Machines, Paris, 1889), and the maximum clear span of a 20th- marks about the structure of the Colosseum, completed only a half-cen-
century, thin, reinforced-concrete shell is 219m (CNIT Exhibition Hall, tury earlier, and perhaps the only other building of the period to rival the
Paris, 1958). Pantheon in influence. "Built about a framework of squared stone ma-
2 See, e.g., MacDonald, 1976. During the Middle Ages, the Pantheon was sonry which constituted the essential load-bearing skeleton .. . it was not
the subject of the "united belief that it owed its existence to sinister forces a building of any great originality . . . conservative in its structural meth-
of demons-the familiar dense row of pillars supporting the vault in Gothic ods and choice of materials as it was in design ... a strain in Roman
cathedrals was miraculously lacking- and not to the ratio or the genius architecture that was shortly to be swept away by new techniques and
of its architect" (T. Buddensieg, "Criticism and Praise of the Pantheon in new aspirations" (Ward-Perkins, 68ff.).
 ON THE STRUCTURE OF THE ROMAN PAN THEON 25

1 Pantheon, View of the interior

(photo : R. Mark)

Roman pozzolana (after Pozzuoli, where it was first dis- Nonetheless, one must be cautious in characterizing the re-
covered) sets by combining chemically with water in the sistance of concrete to cracking caused by tension forces
same way as modern Portland cement. These cements do tending to pull it apart. Although modern concrete based
not need to dry out as does lime mortar; in fact they are on controlled-cured, high-quality Portland cement exhibits
"hydraulic" in that they will set even when immersed. In measurable tensile strength (the level of stress causing a
addition to the obvious advantage for underwater con- material to fail in tension) , its tensile strength is taken to
struction, large batches of pozzolana cement will cure rel- be nil in reinforced concrete design . Experience has dictated
atively rapidly, even in damp conditions. It could thus be that reinforcing steel is always required in regions of a
used for the massive, primary structural elements of large structure where calculations indicate that tension will be
buildings. Furthermore, the early compressive strength of present . Even more important, in view of our later dis-
pozzolana cement is far superior to that of lime mortar. cussion of the state of the Pantheon fabric , it is also re-
26 THE ART BULLETIN MARCH 1986 VOLUME LXVIII NUMB ER 1

Modeling the Pantheon

Analysis of the Pantheon structure was carried out using
a three-dimensional, numerical-computer (finite-element)
modeling code developed by Jean-Herve Prevost at Prince-
ton University. 8 In this approach, the configuration of a
building structure is described by a series of coordinates
taken at finite intervals . These coordinates define a mesh
that becomes the geometric model for the computer. A se-
ries of equations related to loading conditions and the prop-
erties of the building materials are then used to calculate
the displacements of all the mesh points in order to obtain
an overall deflection pattern for the model. Through equa-
tions of elasticity, the deflection pattern then gives infor-
mation about the distribution of structural forces through-
out the building.
To simplify the analysis, a typical meridional section for
the three-dimensional finite-element model of the Pantheon
structure was specified. Because of the extensive openings
in the cylindrical wall-fully one-quarter of its volume is
taken out by statue bays, passageways, and other voids-
no typical section actually exists (see Figs. 3 and 4) . But
since our interest is centered on the functioning of the dome
and the conceptual design of the basic structural configu-
ration, we need not here deal in detail with the wall. 9 In
fact, it is possible to design a solid wall that provides sup-
2 Pantheon, View from St. Peter's (photo: S.Curcic) port to the dome which is similar to that provided by the
actual, voided wall. It was determined that a 5.5-meter-
thick, solid cylindrical wall provides the same overall struc-
quired where temperature changes are likely to be tural stiffness as the six-meter-thick actual wall; and this
encountered.5 equivalent wall was used throughout the analysis (Fig. 5).
Another reason given for the success of the Pantheon A second simplification made for the modeled, typical
dome was the use of a new structural device, the series of building section concerns the structural effect of the dome
concentric stepped rings arrayed about the outer surface of coffering. The coffering, which forms a waffle pattern be-
the dome (Fig. 2) .6 According to W.L. MacDonald, "the ginning just above the springing and ending several meters
rings add to the load over the critical or haunch portion of from the oculus, is actually relatively shallow compared to
the great vault and function as buttresses, helping to bring the full dome thickness, which is taken to be uniform at
the structure into stability through compression ." 7 Indeed, 1.5 meters above the stepped rings. A volumetric analysis
the reproduction of the rings on many of the Pantheon de- indicated that less than five percent of the total dome weight
rivatives would seem to imply that later designers shared is taken out by the coffering. And since it decreases the
in this view. A final, usual observation is that the designer stiffness of the dome to only a small degree, the coffering
aimed for lightness by coffering the underside of the dome effect could be neglected in the finite-element model. An-
and using a lightweight aggregate in the upper reaches of other consideration, evident from the building plan, is that
the building, whose density is significantly less than that the porch plays no role in the supporting structure of the
of the lower, supporting structure. Our investigation of the dome. The porch is in fact hardly connected to the rotunda.
dome, based on modem structural analysis techniques, was The base of the rotunda is assumed to be fixed to perfectly
aimed at evaluating these theories, and thereby gaining rigid foundations, that is, held rigidly against all displace-
some new insight into ancient techniques of large-scale ments . The dome itself is assumed to have been erected on
building. timber centering so that, in effect, dome forces were "turned

5 Modern structural concrete differs from Roman concrete in two im- partment of Civil Engineering, Princeton University (Princeton, NJ) 1983.
portant details . First, its mix consistency is fluid and homogeneous , al- 9 The actual structure of the wall, which incorporates as well a large num-
lowing it to be poured into forms rather than requiring hand-layering ber of great relieving arches (Fig. 3), is complex and has received much
together with the placement of aggregate, which , in Roman practice, often attention in the literature (see, e.g., MacDonald, 1982, 106ff.). But this
consisted of rubble. Second, integral reinforcing steel gives modern con- construction is not carried into the dome . M.E. Blake and D . Taylor-
crete assemblies great strength in tension, whereas Roman concrete could Bishop draw attention to the fact that "it cannot be emphasized too strongly
depend only upon the strength of the concrete bonding to resist tension. that the framework of the arches · · · was part of [the] wall construction
6 See, for example, Giedion, 144. and did not follow the curve of the dome . It was these relieving arches
7
that led to so many fantastic but utterly erroneous theories of [internal-
MacDonald, 1982, 110.
ribbed] dome construction" (Roman Construction in Italy from Nerva
8 J.-H . Prevost, "Dyna-Flow User's Manual," unpublished report, De- through the Antonines, Philadelphia, 1973, 46).
 ON THE STRUCTU RE O F THE ROMAN PANTHEON 27

'
i
'
I

i
I
I
I

iI
I
:

4 Pantheon, Meridional section (after MacDonald, 1982 )

density of the brick-faced concrete used in the cylindrical

wall is taken to be 1750 kg / m 3 • This value is reduced to
1600 kg / m 3 in the lower region of the dome, and to 1350
kg / m 3 for the upper region of the dome (compared with
the 2200 kg / m 3 density of the concrete used in the Pantheon
3 Pantheon, Analytical drawing of the wall (after foundations, which is also the density of standard modern
MacDonald, 1982) concrete) .

on" all at once with the removal of the centering. Most Findings of the Study of Uncracked Models
important, for the first series of model tests, tensile stresses Five integral models with different geometric and loading
throughout are assumed to be of low enough magnitude so modifications were studied in addition to the "full model"
that the structural fabric of the building remains integral; of the Pantheon already described . The modifications were
that is, cracking is nowhere permitted. 1° Finally, dimen- made with relative ease using the computer-modeling for-
sions and data on the gradation of materials used in the mulation, and they provide a basis for better understanding
construction were those reported by K. de Fine Licht. 11 The the structural behavior of the building itself. The models,

10 For further details on the rationale of these modeling assumptions , see rigidities of the constituent materials, were estimated from empirical
Mark, 19-21. equations developed for light-weight, modern concrete . (See G. Winter
and A.H . Nilson, Design of Concrete Structures, 8th ed., New York, 1972,
11 K. de Fine Licht, 89ff. Elastic moduli, which account for the relative
8, 16, 26 .)
28 TH E A RT BULLETI N MAR CH 1986 VO LUME LXVIII NUM BER 1

Model 3: Hemispherical dome of the same dimensions

as above, with concrete density distributed as in the actual
dome, but the base of the dome is now rigidly attached to
the 5 .5-meter-thick, cylindrical wall below it, as shown in
Figure 6b. In this case, the heavy wall provides consider-
able resistance to outward deformation of the dome base .
Hence, maximum hoop tension is not experienced there,
but rather in a region of the dome well above the base.
Model 4: Same configuration and concrete density dis-
tribution as Model 3, except that the attached cylindrical
wall is now carried upward along the dome, as it is in the
actual building. This configuration is the same as that of
LL.I I ......1
the full Pantheon model, but without the stepped rings (Fig.
g 6c) . Maximum hoop tension is experienced in the extension
,,
rr
of the cylindrical wall .
Model 5: Same configuration as the full Pantheon model
(Fig. 4), except that the entire structure is formed of uni-
form, "heavy" concrete with a density of 2200 kg/m3 • Thus,
overall structural behavior is similar to that of the light,
full model described below, except that stresses are higher.
Model 6: Full Pantheon model (Fig. 4) with light, gra-
dational concrete density as outlined in the previous section.
The deformations in the full-model section (Model 6) un-
der the action of gravity forces are illustrated, at much ex-
aggerated scale, in Figure 7. Although a hemisphere is not
so efficient as a parabolic dome in reducing bending forces
from this loading, bending throughout the section was
found to be relatively low. Indeed, bending forces were low
enough so that when their effect was combined with those
of the (meridional) compressive thrusts of the dome, the
combined stresses were everywhere compressive in the
plane of the section. Furthermore, indicated total levels of
stress were moderate for this type of construction. The
.. _: . highest value of compressive stress in the dome itself is 2.8
kg/ cm2 ; the maximum compressive stress at the level of the
5 Computer-drawn, finite-element section of Pantheon model footings is 8.4 kg/cm 2 • Compressive stresses of the same
order of magnitude were found in all the other models as
well.
in order of geometric and loading complexity, are desig- Tension was found to be present, but acting only cir-
nated as: cumferentially. The major cause of this tension can be dis-
Model 1: Hemispherical dome of constant, 1.5-meter cerned from Figure 7. The outward deformation at the base
thickness, nine-meter-diameter oculus, 21.65-meter inte- of the dome, and of the upper portion of the cylindrical
rior radius, the whole formed of uniform, "heavy" concrete wall, must be accompanied by increases in circumferential
having a density of 2200 kg/m 3 • The base of the hemi- length. This stretching of material in the hoop direction is
spherical dome is supported only against vertical loadings; accompanied by hoop tensile stress. For the full Pantheon
no horizontal buttressing is acting, as illustrated in Figure model, these stresses are low - the maximum indicated
6a. Because of its simplicity, this configuration is readily value appears at point (a) on Figure 7: 0.6 kg / cm2 - well
analyzed from shell theory, and therefore it was under- within the tensile strength of even low-grade concrete. They
taken to provide a check on the modeling procedure.12 But are also much lower than the magnitudes of maximum ten-
since the dome is unbuttressed, it also serves to reveal the sile hoop stress found in Models 1-5 (Table) . But a com-
limiting, maximum value of circumferential (hoop) tension parison of the result from the different models is illumi-
for a dome of Pantheon scale because expansion of the dome nating, particularly in light of the present state of the dome
base is here unconfined . (For an explanation of the phe- fabric.
nomenon of hoop stress, see the description of the full Pan- One observation is that the light aggregates used for the
theon model behavior [Model 6] below.) construction of the dome are indeed quite effective for re-
Model 2: Hemispherical dome of the same dimensions
and the same support conditions as Model 1 (Fig. 6a), ex-
cept that the density is that of the light, gradated concrete 12 See, e.g., D.P. Billington, Thin Shell Concrete Structures , New York,
1965, 41-43 .
of the actual dome .
 ON THE STRUCTURE OF THE ROMAN PANTHEO N 29

I
..
J11
:(J>. '
I ~:· b '

"0
.....
.·o~.

6 Modified model sections: (a) hemispherical dome

with oculus; (b) hemispherical dome on cylindrical
I ·~ ·
~~- . ·.
:}. ::.~~
.. ~:;·::~·· .::.- ~ : - ~ -
wall; (c) Pantheon model without stepped rings
a
I bl

ducing stress. The Table indicates that if 2200 kg/m 3 heavy

aggregate had been carried into the full dome, stresses Table
would have been about eighty percent higher than in the Maximum Tension in Uncracked Models
actual building. On the other hand, the effect of the cof-
fering in reducing stress is almost insignificant because of Model Description Aggregate Hoop Stress %-increase
the small weight reduction and the fact that it does not No . (kg/cm 2) (decrease)
extend fully to the oculus. The coffering, then, can be char-
acterized as being mostly an illusionistic device.
1 hemispheri- heavy 4.9 820
Another, more surprising observation from the Table is
cal dome
that instead of increasing, hoop tensile stress is actually
reduced twenty percent by removing the stepped rings (as 2 hemispheri- light 3.3 550
was done in Model 4) .13 Furthermore, removal of the rings cal dome
does not beget bending-tension stress in the plane of the 3 hemi-dome light 1.0 67
meridional section taken through the dome; the unringed on wall
configuration turns out to be more structurally reliable than 4 full model light 0.5 (20)
the actual configuration . Although the Roman builder did w/o rings
not possess any analytical tools for making this type of
5 full model heavy 1.1 83
evaluation, other studies of pre-scientific structural devel-
opment lead us to believe that the designer of the Pantheon 6 full model light 0.6
may well have had sound technical reasons for adopting
the stepped rings. 14 Yet the indicated tensile stress levels in structural. They have only a constructional advantage:
both the unringed and the ringed dome configurations, even vertical formwork for their outer contour facilitated the
when raised by a "stress concentration factor" to account placing of concrete in regions where the natural extrados
for the local effects of wall openings (as discussed below), of the dome was highly sloped. 1s Yet, twentieth-century ob-
seem to be low enough so that no further precautions would servations of the fabric of the Pantheon itself suggest a more
have been deemed necessary. For that matter, even the lim- compelling structural explanation. The best report of these
iting, maximum value of tensile stress indicated in the heavy findings, though regrettably brief, is a half-century-old pa-
hemispherical dome (Model 1) is not greater than might be per by Alberto Terenzio, then Superintendent of the Mon-
expected to be resisted by a high grade concrete. Hence, uments of Latium.
one may well ask why any device, such as the stepped rings, Because of the spalling in 1930 of "small fragments" from
would have needed to be introduced into the Pantheon de- the dome, scaffolding was erected and used for a systematic
sign to reduce tensile stress. inspection. With the removal of plaster, the pattern of me-
One answer is that the rings were not intended to be ridional cracks illustrated in Figure 8 became evident. 16 Te-

13 The more rigid comer of the stepped-ring section probably induces base of the rotunda to the summit of the dome" that he thought were
greater bending distortion of the wall under dome loading, and this results brought about by differential settlement from uneven loading of the wall ,
in somewhat larger hoop stresses within the wall . particularly near the entrance of the rotunda and in the principal niche.
14 See, for example, R. Mark and W.W. Clark, "Gothic Structural Ex- Rather than finding the vertical dislocation along wall cracks that usually
characterizes differential settlement, we have observed only traces of lat-
perimentation," Scientific American, CCLI , Nov. 1984, 176-85.
eral opening across the cracks - corresponding to the effect of hoop
is MacDonald, 1982, 110. tension . But in any event, Terenzio's inference should not obscure inter-
16 Terenzio, pl. xxv1. Terenzio also identifies fractures "reaching from the pretation of the fairl y uniform distribution of dome cracks illustrated in
Figure 8.
30 THE ART BULLETI N MARCH 1986 VOLUME LXV II I NUM BER 1

are still an order of magnitude less than the strength of a

good quality cement, would not at first appear to cause
such fracture. This question will be dealt with at length
later on; but for now it is sufficient to note from Terenzio's
findings that the uncracked models already considered in
this section may not best represent the actual Pantheon
structure. In fact, they may not even represent the actual
structure close to the time of construction. To account for
the observed behavior, the finite-element modeling code
was modified: hoop tension was no longer permitted to
develop . In effect, the model would now be cracked along
the meridians wherever any tensile hoop stress was indi-
cated, producing an array of wedge-shaped arches. 18

Findings of the Study of Cracked Models

Two cracked-section models were studied. These models
a are designated as:
Model 7: Configuration and concrete density distribu-
tion of the full Pantheon model, but without the stepped
rings - corresponding to Model 4 (Fig. 6c) .
Model 8: Full Pantheon model with light, gradational
concrete density - corresponding to Model 6 (Fig. 4) .
The simulated cracking substantially changed the be-
havior of the dome from that of the first set of models. A
three-dimensional, doubly curved dome gains structural
advantage from the interplay between the meridional and
circumferential forces . If the dome is cracked along the me-
ridians, major internal forces in the cracked region are car-
ried only in the meridional direction. Hence, the dome loses
much of its three-dimensional advantage, and behaves in-
stead as an array of arches with a common keystone. Equi-
librium demands that this loss of hoop forces is normally
compensated for by an increase in meridional forces , which
includes also increasing the meridional bending forces
7 Computer-drawn (exaggerated-scale) deformed section
within the dome.
of Pantheon model
In the cracked-section models, dome bending forces were
more than twice as great as in the corresponding uncracked
renzio inferred that the problem must have occurred very model configurations. Indeed, they were now substantial
soon after construction, because the bricks used for some enough so that when combined with the effect of com-
early crack repairs bore the same stamps as those used in pressive thrusts, tensile stresses developed in the plane of
original portions of the building. The cracks are in the lower the dome section . For Model 7, which omits the step rings,
portion of the dome, in the region analogous to that in the tensile stress due to bending was relatively high and cov-
full model which indicated hoop tension (from the spring- ered a large area of the dome extrados . The stress reached
ing, to an angle with the springing of about fifty degrees). a peak of 1.3 kg / cm 2 on the extrados where it joins the
The distribution of the cracks about the dome generally raised outer wall. For Model 8, representing the full Pan-
corresponds to openings within the upper cylindrical wall theon configuration, only highly localized tension was ob-
(as shown in Fig. 8), which serve to increase the magnitude served-at the inner corners of several of the step rings.
of local hoop stresses by a factor of two to three .17 Fractures Since these small regions of tension are in an area that is
caused by hoop tension would therefore be expected to predominantly compressive, tension cracks would not tend
originate there. Yet the predicted hoop stress levels, which to propagate and lead to any danger of structural failure.

17 This factor takes into account local openings in the wall structure as 1a Cracking of the model causes a redistribution of internal forces , so that
well as the effect of possible stress concentrati on at the boundaries of the the final equilibrium of forces must be found from the iteration of a series
openings. See e.g. , S.P. Timoshenko and J.N . Goo dier, Theo ry of Elas- of models, each carrying the cracking propagation a bit further, until con-
ticity, New York , 3rd ed. , 1970, 157ff. vergence is reached.
 ON THE STRUCTURE OF THE ROMAN PAN TH EON 31

I )
n
'1-:\'
ril
.I,n.,.'. ·~
,,, M!
·-; -,'
'
i~
\

" 'I ,'

"1 . -•, ~-
~

µ.!: n~ :~ 1!1 ~:
~ CJjWf . Mr :!ll1
~~~1d1Rl~ 1·_ ~~

...
I

"
8 Crack patterns observed about the interior
of the dome (after Terenzio)

The step rings thus make an obvious and, quite possibly, tunda wall would be expected to produce vertical cracks .
crucial difference in the performance of the cracked-section And although Terenzio does not illustrate cracking in the
dome. wall as he does for the dome, the cracking of the supporting
The extent of the meridional cracking in the actual dome cylinder of Model 8 extends all the way down to 7.6 meters
as shown by A. Terenzio (Fig. 8) agrees remarkably well above the rotunda floor (see note 16) . Hence, the upper
with the simulated cracking in Model 8. The cracks that portion of the modeled wall is no longer acting structurally
he documented continue up the dome to an average of about as a cylinder, but rather as a circular array of independent
fifty-seven degrees above the springing. The cracked region piers, which are more than able to support the dome ad-
simulated in the Pantheon model, by repeated cancelling equately, according to the analysis .
of the hoop tension as it developed (see note 18), reaches The coincidence of the behavior of the cracked model
fifty-four degrees above the springing. Above the cracked with that of the actual Pantheon structure indicates that to
region, the crown of the dome, including the boundary of all intents and purposes, Roman pozzolana concrete could
the oculus, is entirely in compression with but moderate not be counted upon to exhibit any tensile strength. This
levels of stress . Below the dome, hoop tension in the ro- inference seems to be at odds with the position taken on
32 THE ART BULLETIN MARCH 1986 VOLUME LXVIII NUMBER 1

shower" wetting the surface during an otherwise hot, sunny

day - could easily induce tensile, thermal stresses as high
as 15 kg/ cm 2 at the surface of the dome. 21 A possible ad-
ditional cause of large amounts of widespread tension
across the dome is the shrinking of the concrete during cur-
ing, particularly on rigid form work that has not yet been
decentered.
From our analysis and observations, it appears that the
Roman builders were well aware of the problem of tensile
cracking, and they took steps to cope with it. The perfor-
mance of the stepped rings in the cracked Pantheon models,
unlike their negative role in the uncracked models, is ac-
tually close to that described by MacDonald. Because of
the meridional cracking, and the consequent behavior of a
9 Hadrian's Villa, Tivoli (design, 118-US), 'Teatro Marittimo ." large portion of the dome as an array of arches, the struc-
Segment of concrete barrel vault with surcharge (photo: R. tural action of the stepped rings can be likened to that of
Mark) surcharge over the haunch of an arch (or of a barrel vault
where every section acts as a simple arch). Such a sur-
Roman construction by most historians as well as by con- charge, integral with the vaulting, is illustrated in Figure
temporary technical commentators. 19 It is even at odds, we 9, a section through the remains of a barrel vault at Had-
suspect, with inferences from the results that one might rian's Villa in Tivoli that was constructed contempora-
gain from tensile tests of the concrete used in the Pantheon. neously with the Pantheon. 22 Evidently the stepped rings
As already noted, the reported maximum stress levels due are the translation of extensive Roman building experience
to dead loading are all low - too low by themselves to with planar arches to the construction of a dome of un-
cause such extensive cracking. But the presence of even precedented scale. The rings may even have been placed
these relatively low tensile stresses over extensive regions as a "fix" intended to stabilize the already-cracked Pan-
of the structure creates an environment that is particularly theon structure - a hypothesis supported by Terenzio's
sensitive to cracking from other transient loadings that pro- identification of the dome cracking near the time of initial
duce tension. 20 One of these is differential temperature. Ra- construction. The original bronze tile roofing of the Pan-
pid cooling - as an extreme example, caused by a "sun theon, too, might have been put in place as an afterthought

19
The idea of a monolithic concrete dome seems to have been first pro- circumferential or 'hoop' tensions in the lower part [which Mainstone
posed near the close of the 19th century by the then Slade Professor of estimated to be about 5kg / cm 2 ] could be resisted without cracking" (De-
Fine Arts and Director of the Fitzwilliam Museum at Cambridge, J. Henry velopments in Structural Form , Cambridge, MA, 1975, 116, 118). Cowan
Middleton: "It would have been impossible for the Romans to build and reported also on the results of modern tests of concrete from some Roman
vault their enormous spans if they had used vaulting of brick and ma- ruins in Libya. These indicated a compressive strength of about 200 kg /
sonry. . . Roman concrete was quite devoid of any lateral thrust, and cm 2 , not unlike that of modern concrete (Masterbuilders, 56). Based on
covered its space with the rigidity of a metal lid .... The construction an empirical relationship developed for modern concrete, the tensile
of the enormous cupola is .. . as free from thrusts as if it were cut out strength of the Libyan specimen could be expected to be almost 15 kg/
of one block of stone." (The Remains of Ancient Rome, London, 1892, cm 2 (Winter and Nilson , as inn. 11, 22).
r, 66; II, 131.) In this account, the massive walls retain little structural 20
Cracks caused by transient loadings in regions of a structure already
function except to provide vertical support for the dome. Nonetheless,
subject to constant (dead-weight) tension will tend to remain open after
the theory was recounted in Anderson and Spiers' text (The Architecture
the termination of the transient. This contrasts with the effect in regions
of Greece and Rome, London, 1903), which, in a later edition, was cited
of constant compression where the cracks tend to close. This study is the
by D.S. Robertson. Echoing Middleton, Robertson wrote: "Concrete, as
first instance of our finding large fields of tension extending around an
the Romans knew it, was in effect a new and revolutionary material. Laid
entire perimeter of a building. (In Gothic buildings, tension was found to
in the shape of arches, vaults and domes, it quickly hardened into a rigid
be present only at discrete, highly localized regions such as at the ends
mass, free from many of the internal thrusts and strains which trouble
of flying buttresses .)
the builders of similar structures in stone or brick." (Greek and Roman
21
Architecture, Cambridge, 1969, 233; originally published as A Handbook For the calculation of thermal stress, we have assumed physical prop-
of Greek and Roman Architecture in 1929.) By now, acceptance of Mid- erties that would generally be associated with the concrete described in
dleton's theory is almost universal. For Mario Salvadori, "the Pantheon, n. 19 and that the surface of the dome is 10°C cooler than the mass of
a triumph of concrete architecture, could only be conceived and built after the concrete below the surface. Theory is given by Timoshenko and Good-
the discovery of pozzolana concrete by the Romans ... the [thickness of ier, as inn. 17, 433-37. Differential settlement (seen. 16) and earthquake
the base of the dome] is so large that the tensile hoop stresses in it are loadings (which could produce differential settlement) can also induce
well below the resistance of the concrete . ." (Why Buildings Stand Up , additional tensile stresses. But without full records, their effect is difficult
New York, 1980, 230-33). A similar observation had also been made by to predict, and it is highly unlikely that a uniform pattern of cracks would
Henry Cowan: "In the Pantheon the thickness of the concrete is so great have resulted in the vault.
in the lower portion of the dome that the tensile stresses are low" (The 22The section illustrated in Figure 9 is actually from a large-diameter,
Masterbuilders , New York , 1977, 74). Rowland Mainstone, on the other annular barrel vault framing Hadrian's 'Teatro Marittimo." Even though
hand, is more circumspect: "Structurally the new forms ... still fell short the axis of the barrel is curved slightly, the overall structure behaves es-
of a full exploitation of the increased possibilities of action ... provided sentially as a linear barrel vault.
that the conditions of support at the base were appropriate and that the
 ON THE STRUCTUR E OF THE ROMAN PANTHEO N 33

- to protect the cracked, outer regions of the dome from

the elements. Our modeling, then, has led to a new view
of the influence of the actual structural behavior on the
final design of the Pantheon, and also, for that matter, to
a reinterpretation of the Roman architectural revolution.

Roman Structural Design

There is no question that during the zenith of Imperial
Rome's power and wealth, Roman architecture acquired
new aspirations and techniques of construction. The pe-
riods of exceptional commercial and political activity of
any civilization are usually symbolized by large-scale
building. And the architecture of the Hadrianic era, for
which the Pantheon is probably the prime example, was
no exception. Yet our study of the structure of the Pantheon 10 Hadrian's Villa, Tivoli , Entrance Pavilion to Piazza d'Oro
leads us to question the generally held belief that the success (photo: R. Mark)
of this new architecture was dependent upon the devel-
opment of a unique Roman building technology.
Rome's contribution to monumental architecture derives to ensure continuity with earlier design and construction
mainly from the widespread use of the circular arch, which techniques .24
allowed large wall openings for light and access, and the In this light, there is special significance in the fact that
spatial development of the arch form in concrete: in barrel the circular arch and pozzolana cement both found major
vaulting (generated by a lateral translation of the arch), use in Roman substructure and other utilitarian construc-
groined vaulting (formed by intersecting barrel vaults), and tion long before being adapted for "high" architecture. Sig-
hemispherical domes (the circular arch, rotated) . The three- nificant masonry-arch bridges had been built for crossing
dimensional form of many of the concrete structures of, the Tiber as early as 109 B.C. (the Pons Mulvius with spans
for example, Hadrian's Villa is striking indeed (Fig. 10), as as great as eighteen meters) and 62-21 B.C. (the Pons Fa-
is their apparent plasticity (because one sees only the final bricius with twin, twenty-four-meter spans). Furthermore,
rendering, and not the intricate timber form work needed Vitruvius, writing a century before the Hadrianic era, dis-
to create the complex shapes). Yet, the basic constructional cusses in some detail the composition and application of
idea generating the domes and vaulting of this architecture concrete made from pozzolana.25
remains that of the planar arch with its characteristically The decision to begin to employ concrete for large vauits
deep voussoirs, surcharge over the springing, and substan- and domes, such as the Temple of Mercury, of which Vi-
tial buttressing to resist outward thrusts. truvius made no mention, was constructional, we believe,
Roman structural development was nowhere near so rad- rather than structural. It was based primarily on extensive
ical as that of the late nineteenth century, when the intro- building experience. We do not doubt that vault construc-
duction of the new industrial materials brought forth a true tion using cut stone or brick would have been more costly,
revolution in building design .23 A far better analogy is pro- given the availability of sources of Roman building ma-
vided by the late twelfth-century development of the High terials, the organization of construction labor, and the ev-
Gothic cathedral, for which the stage was first set with the ident speed of the work. 26 But one must bear in mind that
significant improvement of medieval building technique the skilled labor, and hence the great expense of erecting
during the century preceding. A second major factor lead- the timber centering needed for both types of construction,
ing to the technical success of the Gothic was that new probably would have been much the same.
buildings, even though they often took what appeared to Concrete does afford one important structural advantage
be unexplored paths, retained many elements from earlier over common stone vault construction, however. It allows
designs . In effect, an earlier building acted as an approx- the type of gradation, without special effort , in the weight
imate model confirming the stability of new, larger con- of materials that was found to be so advantageous in the
struction . And perhaps most important of all, the Gothic Pantheon analysis . Nevertheless, this did not make con-
building organization allowed apprentices to rise through crete the only method of choice. Stone and brick continued
the ranks, even to become master builders, which served to be used extensively for monumental construction in the

23 The tensile strength of metals, for example, allowed tall structures, such 24 Mark, 10-11.
as the Eiffel Tower, to be anchored into foundations instead of requiring, 2sVitruvius, The Ten Books of Architecture , trans. M. H . Morgan, New
as was necessary for all past monumental construction, great masses of York, 1960, 46ff.
masonry to ensure stability. Another consequence was the relatively sud- 26See W.L. MacDonald, The Architect, ed. S. Kostof, New York, 1977,
den achievement of a new reach in building spans (seen. 1) . 40ff.
 34 THE ART BULLETIN MARCH 1986 VOLUME LXVIII NUMBER 1

provinces . And although no other domed building of the fact that it was not caused by it gives a new vantage point
Imperial era approached the span of the Pantheon, a num- from which to reexamine the whole issue of structure and
ber of masonry-domed buildings of very large span were style in later Roman large-scale building design.
constructed much later on. Among these were the similarly
scaled domes of the Cathedral of Florence and St. Peter's Earlier studies by Robert Mark often have focused on
in Rome, which were erected over tall crossing piers that Gothic architecture (Experiments in Gothic Structure, 1982,
made them subject to still another crucial set of structural and a related article in Scientific American, November,
problems. 27 Indeed, the brick, 0.46-meter-thick hemispher- 1984), though his research has ranged from Christopher
ical dome used by Christopher Wren to enclose the 30.8- Wren (Scientific American, July, 1981) to this recent in-
meter-span interior crossing of St. Paul's Cathedral is a vestigation of Roman architecture. [School of Architec-
structure that is valid to compare with the Pantheon dome. ture, Princeton University, Princeton, NJ 08544]
The ratio of thickness to span of Wren's dome, 1:67, if
applied to the 43 .3-meter-span of the Pantheon, gives an Paul Hutchinson, an engineer with historical interests,
equivalent thickness of 0.65 meters instead of the actual earned his degree from Princeton University in 1984. [N.G.
1.5 meters . The outward thrust of the thinner brick dome Jacobson and Associates, 414 Skinner Bldg., Seattle, WA
would thus be similar to that of the actual lightweight con- 98101]
crete dome, and although compressive stresses in the brick
dome itself would be somewhat greater, they would still be
well within an acceptable range.
None of this depreciates the achievement of the Pantheon
Bibliography
in a pre-scientific age . But the technical underpinning of
that achievement has been largely misunderstood. Rather
de Fine Licht, K., The Rotunda in Rome: A Study of Hadrian's Pantheon,
than representing any break with earlier tradition, the basic Copenhagen, 1968.
form and support of the dome came directly from long Ro-
Giedion, S., Architecture and the Phenomena in Transition, Cambridge,
man experience with masonry circular-arch construction MA, 1971.
and from using concrete as a substitute for masonry in a
MacDonald, W.L., The Architecture of the Roman Empire, 1, rev. ed.,
wide variety of structures, including even rather large
New Haven, 1982
domes. The coincidence of the observed cracking within
the Pantheon dome with the model predictions for dome _ _ _ _ _ , The Pantheon: Design, Meaning and Progeny, Cambridge,
MA, 1976.
behavior based on concrete having nil tensile strength
should put to rest the idea that Roman pozzolana somehow Mark, R., Experiments in Gothic Structure, Cambridge, MA, 1982.
accomplishes feats that no modern designer would expect Terenzio, A., "La restauration du Pantheon de Rome," La conseroation
of unreinforced concrete. Roman structural development des monuments d'art & d'historie, Paris, 1934, 280-85, pls . xxv, xxv1.
was facilitated by the use of pozzolana concrete. But the Ward-Perkins, J., Roman Imperial Architecture, New York, 1981.

27
See H. Dorn and R. Mark, "The Architecture of Christopher Wren, "
Scien tific American , CCXLV, July 1981, 168ff.
Copyright of Art Bulletin is the property of Routledge and its content may not be copied or
emailed to multiple sites or posted to a listserv without the copyright holder's express written
permission. However, users may print, download, or email articles for individual use.