¹ In this article the following abbreviations will be used Callebat—Callebat, Louis, Vitruve De L' Architecture Livre VIII (Budé, ed.; Paris, 1973)Google Scholar.

GdM—de Montauzan, Germain, Les Aqueducs Antiques de Lyon (Paris, 1908)Google Scholar.

Smith—Smith, Norman A. F., ‘Attitudes to Roman Engineering and the Question of the Inverted Siphon’, History of Technology, 1976, pp. 45–71Google Scholar.

MAGR—Grenier, Albert, Manuel de l'Archéologie Gallo-Romaine (Paris, 1960), 4, 1Google Scholar.

Plommer—Plommer, W. H., Vitruvius and Later Roman Building Manuals (Cambridge, 1973)Google Scholar.

SAT—Forbes, R. J., Studies in Ancient Technology (Leiden, 1964), 1Google Scholar.

Ward-Perkins—Ward-Perkins, J. B., ‘The Aqueduct of Aspendos’, PBSR, 1955, 115–23Google Scholar.

Other abbreviations are as in L'Année Philologique.

The following will not be abbreviated but will be referred to, and may be found of value for reference

Ashby, T., The Aqueducts of Rome (Oxford, 1935)Google Scholar.

Babbitt, and Doland, , Water Supply Engineering (New York, 1955)Google Scholar.

Belgrand, M., Les Aqueducs Remains (Paris, 1875)Google Scholar.

Blanchet, A., Recherches sur les Aqueducs et les Cloaques de la Gaule Romaine (Paris, 1908)Google Scholar.

Casado, C. F., Aqueductos Romanes en España (Madrid, 1972). (see n. 50, infra)Google Scholar.

Grimal, P., Frontin: les Aqueducs (Budé, ed.; Paris, 1961)Google Scholar.

Grimal, P., ‘Vitruve et la Technique des Aqueducs’, R.Phil., 1945; 162–74Google Scholar.

Kretzschmer, F., ‘La Robinetterie Romaine’, RAE, 1960, 89–113Google Scholar.

Mortet, V., ‘Recherches Critiques sur Vitruve et son Œuvre; IV, Vitruve et l'Hydraulique romaine’, RA, 1907, 75–83Google Scholar.

Stehlin, K., ‘Über die Colliviaria oder Colliquiaria der Römischen Wasserleitungen’, Anzeiger für schweizerische Altertumskunde, 1918, 167–75Google Scholar.

Van Buren, A. W., RE VIII A, s.v. Wasserleitungen, 472–5 (1955)Google Scholar.

Van Deman, E., The Building of the Roman Aqueducts (repr. Washington, 1973)Google Scholar.

For a commentary on current studies on aqueducts and a general bibliography of the field the reader may find of interest a study at present in the press: Hodge, A. Trevor, ‘A Plain Man's Guide to Roman Plumbing’, EMC XXVII, N. S. 2, 3 (1983)Google Scholar. (This journal is also known as Classical Views, University of Calgary.)

Special mention must be made of the first three authors listed, Callebat, de Montauzan and Smith, for together they form the indispensable basis of any study of siphons. Each is marked by a wide interest and knowledge of the subject as a whole, but does have a specialised approach—Callebat is primarily interpreting a literary text, de Montauzan giving an account of the archaeological remains, while Smith is a historian of hydraulic engineering and an authority on the technical aspects. It will thus be seen that, taken together, they form a particularly powerful combination. I need scarcely mention my own debt to these three sources, for even a quick glance at my footnotes and references will make it plainer how much I owe to them than could any formal acknowledgement here. I must, however, express my gratitude and indebtedness to Prof. D. R. Townsend, of the Faculty of Engineering, University of Ottawa, a specialist in hydraulics, and to Eng. Herenguel, of the Société du Canal de Provence, for much valuable assistance on technical matters: they are in no way responsible for any mistakes in this account, which remain entirely my own contribution.

² GdM 176: ‘C'est donc, à proprement parler, comme on voit, un siphon renversé’. Callebat, 169: ‘la solution… du siphon, ou plus précisément du siphon renverse’. Smith, 51, goes one further: ‘“Inverted siphon” is, in some ways, an unfortunate term for what is in effect a pressurised pipe-line carrying water across a valley or depression. However, the term is firmly established in the technical vocabulary and this is no occasion for trying to eject it.’ He sounds regretful about it.

³ See LSJ and OLD for references. Diabetes: Col. 3, 10, 2. In the sense of a siphon this is a hapax legomenon, though the word is attested in other meanings.

⁴ Str. 5, 3, 8.

⁵ Nîmes: GdM 316, fig. 124; MAGR 98, fig. 32; for the castellum and its place in the distribution system, Callebat 149. The castellum at Pompeii is illustrated by Kretzschmer, F., La Technique Romaine (Brussels, 1966, tr. from the German, ed., Dusseldorf, 1958), 51Google Scholar, fig. 86. Towers: Kretzschmer, op. cit. 53, figs. 88–9 (elevation, section and photograph). I have marked this element ‘water tower’ in my diagram: Kretzschmer calls it a ‘Pilier de distribution secondaire’, and it is often called a ‘stand-pipe’ (a confusing term for a tank). Maiuri, A., Pompeii (Novara, 1960) 27Google Scholar, writes of these towers that they ‘by increasing the pressure, kept the supply of water under regular control’; I am not sure what he means by this. Van Buren, , RE VIII A, 474Google Scholar, states that they reduced the pressure (‘sie waren hauptsachlich notwendig zur Verminderung des Wasserdruckes’), and that this was their main function, distribution being secondary and incidental. We may also note the little-known but excellent article by Kretzschmer (n. 1, supra), whose fig. 37 bis (page 110) is a variant of my Fig. 1. His drawing is based on the actual layout at Pompeii, which differs from my hypothetical diagram principally in that there the ground is not flat, and he accordingly shows his water tower at the bottom of a hill, 20 m. below the castellum (located at the Porta Vesuvii), and with its tank 6 m. above the delivery spouts and fountains at street level. The underground pipe from castellum to water tower therefore runs downhill at a slope, and is basically the same arrangement as that illustrated by Plommer, 27, fig. 4(a), right-hand side. In this system the pressure on the delivery taps (which Kretzschmer is worried about, see n. 76 infra) is reduced, since they are subjected to the pressure of a head of only 6 m., counting from the level of the tower, and not 20 m., which is what the head would be if there were no tower and the pipe ran directly from the castellum down to the taps. In this way the existence of the tower really does reduce pressure, but only at the expense of destroying the siphon and the head, so that the water can never again rise up any higher than the level of the water tower. Kretzschmer's work is repeated in summary (including his diagram) by Eschebach, Hans, ‘Pompeii, La Distribution des Eaux dans une Grande Ville Romaine’, Doss. Arch. 38 (Oct.–Nov. 1979)Google Scholar, Aqueducs Romains, 74–80 (esp. 78). See also Kretzschmer, F., ‘Römische Wasser-Kähne’, JSGU 48 (1960–1961), 50 ff.Google Scholar, and A. Maiuri, ‘Pozzi e condotture d'acqua nell' antica città di Pompeii’, NSA 1931, 546–76.

⁶ A further complication may be noted, though we will not go into it here. This is the practice of establishing relative priorities between water supplies for drinking and other less essential uses effected by setting the take-off pipes in the castellum at differing levels, so that the lowest ones, carrying drinking water only, never ran dry. This system is described by Vitruvius (8, 6, 1–2) and attempts have been made to reconcile the castellum, as he describes it, with the castella actually found. Kretzschmer (n. 5, supra), 52, and fig. 83 (reproduced as Callebat fig. 9, page 151) seems to extend this two-level system to cover not only the offtake points from the castellum but the water towers as well, so that there exist all through the city two different sets of water towers of different heights—low ones (served from low in the castellum) providing drinking water, and high ones, more liable to run dry, providing private supplies and water for the baths. In his text he implies that all towers were of the same height (‘on érigeait done, dans la ville, des châteaux d'eau intermédiaires d'une hauteur de 6 mètres’) but in his ‘schéma’ the difference in level is very clearly shown, and is increased further by Callebat, who, in reproducing this diagram, has redrawn the high set of water towers so high that they are actually above the level of the castellum supplying them and so could never have got any water at all; but this is, of course, only a diagram, not a measured drawing. It should be noted also that in the castellum at Nimes, where there is a row of 10 circular holes in the side wall for the offtake pipes (all at the same level) and three in the floor, these three run directly into the drains and were evidently used for draining the castellum for cleaning; it has therefore nothing to do with the Vitruvian two-level system, though why three such enormous pipes were needed to drain one castellum (which also had an isolating sluice at the intake point) remains obscure. The most reasonable suggestion remains that of the excavator, Pelet, Auguste, that they supplied a naumachia (Les Archives de la Commission des Monuments Historiques, Paris, 1855–1872Google Scholar, eds. Gide et Beaudry); it was probably located in the nearby Jardins de la Fontaine, where traces of a theatre have been found. But the whole question of the hydraulic organisation of the Roman domestic water supply is one calling for further study and analysis.

⁷ ‘Open’ in the sense that it is open to atmospheric pressure, and does not have a watertight and pressure-proof cover, which would transform the whole arrangement into a pressure system. All of my diagrams showing uncovered aqueducts and tanks indicate only that they were open in this sense, not that no actual cover was provided; in fact, they usually were covered with stone slabs, as a protection both against evaporation and against dirt getting into the water.

⁸ As recognised and clearly spelled out by GdM 178: ‘Quand Frontin, quand Vitruve composèrent leurs ouvrages, la ville était toute sillonnée de conduites souterraines en plomb, qui portaient dans les différents quartiers de la ville, dans les palais comme dans les domiciles privés, l'eau sous pression issue des châteaux d'eau. Le principe du siphon, dérivé de celui de l'equilibre des liquides, était done partout en action', and Vitruvius’ inadequate treatment of it leads him to wax caustic: ‘quand on lit ensuite le texte de Vitruve sur les siphons, on est un peu étonné de ce que l'œeuvre de cet auteur passe encore pour une somme complète du savoir technique chez les Romains’. On this question Van Buren, (RE VIII A 473Google Scholar: ‘Der Gebrauch von Druckrehrleitung war im Römischen Reich weit verbreteit’) and Forbes (SAT 165: ‘In Roman times the siphon was used only in some few cases, probably because of leakages and the relatively poor materials available for high pressures’; repeated almost verbatim in Singer, C., Holmyard, E. J., Hall, A. R., Williams, T. I., History of Technology (Oxford, 1956), 2, 669Google Scholar) are here in flat contradiction, and it is Van Buren who is right. For possible Roman siphons (?) see also ILS 5767, 5773–6, 5778; but the meaning of ‘salire’, upon which the identification as siphons depends, is uncertain. The principle of water finding its own level is clearly enunciated by Pliny (NH 31, 57: ‘[aqua] subit altitudinem exertus sui’). It is surprising that anyone ever doubted that the Romans understood this, and depressing that, as late as 1968, G. E. Bean, writing on the aqueduct of Aspendos, should have felt it either necessary or fitting to say ‘It was at one time doubted whether the ancients understood the principle of piping water up under gravitational pressure; if such doubts remain, the aqueduct at Aspendos would effectively remove them’—Turkey's Southern Shore (London, 1968), 75–6Google Scholar. Six years later it was still being maintained before the Royal Society (by HRH the Duke of Edinburgh) that the Romans built the Pont du Gard because they did not know water would rise to its own level (Notes and Records of the Royal Society of London, 29, 1 (Oct. 1974), p. 18; quoted by Smith, 70).

⁹ SAT 165: ‘Siphons and bold tunnelling are typical of the water-supply systems built by the Greek tyrants and the Hellenistic kings from Sicily and Southern Italy to Asia Minor’. Smith 52: ‘Bearing in mind that the Romans learned much of their engineering from the Greeks, that they regularly (and perhaps frequently) employed Greek engineers, and that Vitruvius relied extensively on Greek sources, it is important to note that a number of Greek aqueducts featured siphons. Among examples which have been identified are those at Pergamum, Patara, Mylasa, Methymna, Catania, Selinus and Syracuse’. Apart from Pergamon and Patara, these all seem to be fairly obscure and ill published (Smith gives no references, but see Merckel, C., Die Ingenieurtechnik im Altertum (Berlin, 1899), 506 ff.)Google Scholar; Van Buren, (RE VIII A, 473)Google Scholar lists 12 others. Smith (48) generalises: ‘There is virtually no aspect of Roman structural and hydraulic engineering which had not been practised, to some extent, by earlier societies’.

¹⁰ Based on the drawing by G. Barruol and P. Martel, ‘La Voie Romaine de Cavaillon et Sisteron’, RELig. 1962, 188, fig. 19(a).

¹¹ So GdM 176 (quoted without comment by Callebat 169): ‘descendant le long d'une colline pour remonter le long de la colline en face, apres avoir franchi le thalweg’. This hypothetical valley and its shape have not been mentioned before, and the use of the definite article, ‘le thalweg’, reveals the automatic assumption that one existed.

¹² GdM 72, esp. fig. 9, page 80 ( = my Fig. 6).

¹³ A masonry-built pressure channel is attested in the siphon of Angitia, near the Fucine Lake. The conduit, 60 cm. wide, was embedded in a masonry mass 5 m. thick (Blake, Marion, Roman Building Construction in Italy from Tiberius through the Flavians (Washington, 1959), 82Google Scholar). Lead glut: GdM 206: ‘on sait du reste l'usage à profusion que les anciens faisaient de ce dernier métal’. This is relevant to Ward-Perkins' comments on the expense of lead (n. 105, infra).

¹⁴ The valley came 1½ km. after Les Tourillons. For a more detailed description of this siphon, see below.

¹⁵ MAGR 153.

¹⁶ GdM 125 ff.: Beaunant; in some older works and occasionally on signs still visible locally one finds the spelling ‘Bonnant’. It is sometimes also referred to by the name Chaponost, another village at the head of the siphon, where the header tank and ramp are. Some variation in local nomenclature is inevitable since the siphon is over 2 km. long. GdM 118 ff.: Soucieu. Both sites are readily accessible from the D 42 highway south from Lyon, and will most conveniently be found marked on either sheet 73 or 93 of the 1:200,000 Michelin map.

¹⁷ 0·56 × 1·77m. (at Soucieu, GdM 118); it should be noted that aqueduct channels were often a good deal higher than necessary for the volume of water carried, so as to provide space for a man to walk through them during inspection and cleaning, and were often only half-filled when the aqueduct was running. At Soucieu a rough calculation may be of interest, based on the cross-section of the siphon pipes. These always ran full and have a cross-section of about 530 sq. cm. each, or 4,800 sq. cm. total for the nine of them. To match this, the rectangular aqueduct channel, of known width 0·56 cm., would have to be filled to a height of 0·87 cm. We thus have a channel 1·77 m. high regularly filled to about half-way, and if it were filled much higher it would be more than the pipes could handle (the overflow in the side of the tank was at 1·30 m.). We may compare the regular water level, as marked by incrustation, in the Aqua Marcia (0·60m. deep in a channel 1·70 m. high; GdM 172); the Pont du Gard is about the same.

¹⁸ Soucieu, and probably Beaunant. The La Craponne aqueduct seems to have had fewer (GdM 218), while at the St. Genis siphon 10 holes were provided and one subsequently blocked up as unnecessary (GdM 104).

¹⁹ GdM 105, 218–9. I am forced to question de Montauzan's suggestion that taps were mounted on the pipes themselves, just after leaving the header tank. On a 30 cm. or so pipe, these would have had to be very big taps, and archaeological evidence for such taps is rare, though they do exist at Ostia. For taps and valves, see below, and n. 76.

²⁰ Based on personal observation, Soucieu. This contradicts GdM, but Burdy, Jean (‘Lyon: Lugdunum et ses 4 Aqueducs’, Doss. Arch. 38 (Oct.–Nov. 1979)Google ScholarAqueducs Romains, 66), a local authority, describes the pipes as having a ‘section légèrement ovale de diamètre moyen 27 cm’. The evidence is conflicting. At Lyon three siphon tanks are preserved, at St. Genis, Soucieu and Beaunant. All three are on the Gier aqueduct and all three header tanks, but only at St. Genis and Soucieu are preserved the holes taking the ends of the siphon pipes (Plates VI c–VII a); I have examined both, and at St. Genis the holes are round, at Soucieu they are oval. The masonry is better preserved at St. Genis, which normally ought to be decisive; I prefer Soucieu for two reasons: it fits better the known shape of Roman pipes, and, secondly, because though the 35 cm. major axis of the oval may well be due to enlargement by weathering, I do not see how the minor axis, 25 cm., can ever have been any wider than it is now. I therefore suggest an oval pipe, in spite of St. Genis; but there is, of course, no real reason why round and oval pipes should not both have been in use.

²¹ The preserved sections of pipe in the Musée Lapidaire at Aries (n. 48, infra) are about 2·75 m long. For the oval shape of the pipe, see GdM 201; figs. 74–7; Callebat 162. The thickness is based on a theoretical calculation of the thickness of lead pipe needed to resist the pressure known to have been generated at the bottom of the Soucieu siphon by the volume of water contained within a pipe of this known diameter. The calculation has been done independently by de Montauzan (GdM 199), who makes the thickness 3·6 cm., and Smith (61), who favours ¾ inch ( =about 2·0 cm.). The difference seems to come largely from de Montauzan using a higher safety factor (4) and Smith (2).

²² Beaunant: GdM 127.

²³ Callebat 172; see also GdM 220. In modern pipelines the pipe is often anchored by being embedded in a (buried) concrete block every 20 m. or so, and de Montauzan suggests this may have been done with ancient siphons.

²⁴ GdM 220. But see Almuñecar and Alcanadre (infra).

²⁵ For a full discussion of these bends and how they were anchored, with particular reference to Vitruvius 8, 6, 5, see Callebat, 170.

²⁶ Soucieu: GdM 120; fig. 38. Beaunant: GdM 127; fig. 40. Plommer, 29, asks: ‘But what instances of a venter are known in western aqueducts? Almost all are laid out in one continuous and very gentle slope, often raised high above valleys and plains’. The venter bridge at Beaunant is shown in my Plates IX b, c, and needs no further comment.

²⁷ For the purpose of this see n. 68, infra.

²⁸ Smith 64. On earlier suggestions that the extra width indicated that at venter level each of the pipes split into a pair of smaller ones ‘for the purpose of developing greater pipe strength’ he comments ‘Quite what sort of an improvement this represents and whether or not it would have been worth the effort is difficult to say’. See n. 63, infra.

²⁹ ‘The difference in water-level between the tanks, h feet, is the head lost in giving the water velocity and overcoming the frictional resistance in the pipe’—Smith 55.

³⁰ SAT 168. See A. Trevor Hodge, ‘Vitruvius, Lead Pipes, and Lead Poisoning’, AJA 1981, n. 14. For a remarkable illustration of how far this process can go see the lead pipes from Djemila, Algeria, published by Birebent, Jean, Aquae Romanae (Algeria, 1962) 466Google Scholar. The photograph clearly shows how the deposit has so accumulated as to block the pipe entirely.

³¹ For the same given area enclosed, the circumference of a circle is shorter than the four sides of a rectangle, but the aqueduct channel was effectively enclosed on only three sides since the water did not normally touch the roof and on this side flowed friction-free.

³² It is the same basic principle as that of the water-tube or fire-tube boiler, which utilises it to maximise heating surface.

³³ Average aqueduct gradient: GdM 171. Figures for Beaunant and Soucieu, GdM Pl. 5. Coincidentally both siphons lose the same head (9 m.), but the gradient is steeper at Soucieu since the length is shorter.

³⁴ It will be convenient here to list the siphons so far known. Though not intended to be exhaustive the list should be helpful for most purposes. It is based on that of Smith, 54.

Termini Imerese, listed by Smith with some reservations, seems more than dubious. There are certainly remains of an aqueduct there, but I have seen no reference to a siphon.

Also of value and interest, but of Greek date, are the siphons of Patara and Pergamon. Patara was first published in Texier, , Description de l'Asie Mineure (Paris, 1849), 3, 192–3Google Scholar, and is described by GdM 193. See also Bean, G. E., Lycian Turkey (London, 1978) 90Google Scholar, and C. Merckel (n. 9, supra), 504–6: it is clearly illustrated, in reconstruction, in his Abb. 198. Its construction was remarkable. Down one side of a depression and up the other ran a parallel pair of Cyclopean walls, with the interval filled with sand and rubble. The pipes ran on top of this structure; it is uncertain whether they were of terracotta or stone. There was a series of stone blocks, shaped so as to lock together, and perforated with large holes (compare Pergamon (SAT fig. 35) and Aspendos (Ward-Perkins 119, fig. 2)), but it is uncertain whether the water ran directly through the holes (all joined together to form a stone pipe) or if they carried terracotta piping, a large number of terracotta fragments having been found. Stone pipes in an aqueduct (though not from a siphon) are neither primitive nor unusual, and have been found at Dougga (Tunisia), Poli (near Arezzo) (Lanciani/Pasqui NSA 1878; GdM 193, n. 2), Azeffoun (Algeria) (Gsell, op. cit. 1, 257), and Aspendos (supra). Aspendos is particularly significant since it is well published and dated, and was still using this technique though of Roman construction and of the first-second century A.D. Pergamon is too well known to require description. The profile of this remarkable siphon is readily accessible in, e.g. SAT fig. 35; see Callebat 169, and Graeber, F., Wasserleitungen von Pergamon (Berlin, 1888)Google Scholar, Pl. 2. It is noteworthy for its great depth, no less than 183 m. separating the header tank and the lowest point, which generated an unparalleled pressure of about 20 atmospheres. It may not have worked too well, and venter bridges were inserted to ease the pressure, but the ambition of the concept must command our respect, whether it worked or not.

³⁵ GdM 72 ff.

³⁶ The most original was that it was a bridge over the Saône. The bridge is on top of a hill and the Saône would have had to make a 6 km. detour to run under it. GdM 73.

³⁷ The reconstruction is based on a rediscovered plan of 1599 which seems to show the existence at that time of a tank on top of one of the piers and, next to it, the beginning of the downward slope of the ramp. For so unusual a monument one would naturally prefer better evidence, but in view of the parallel from Aspendos (below) this is probably conclusive enough. Since de Montauzan's publication, the piers have become so overgrown that they can barely be identified (Compare my Pl. IX. c).

³⁸ Les Tourillons is located on a side road 200 m. north of highway D489 (Lyon–St. Foy l'Argentière) at the east end of the village of La Craponne (and is marked on the enlarged Lyon section of Michelin map Sheet 93). The height above sea level of the tank at Les Tourillons was 310 m., and ground level at Fourvière, 6 km. away, is about 296 m. This gives for the second half of the aqueduct a fall of 16 m. for 6 km., or 2.6 m. per km., which with a big siphon included, is not enough (compare the hydraulic gradient of 3.4 m. per km. for Beaunant, above). But the water probably did not go all the way up to the top of Fourvière (no. 40, infra).

³⁹ GdM 79, n. 1.

⁴⁰ The only distribution tank known at Fourvière is that of the Gier aqueduct, at 292 m., and it is assumed that the other three aqueducts, including La Craponne, arrived at a lower level (so as to give an adequate gradient) even at the expense of depriving the higher parts of the town of water. The actual level for the end of the Les Tourillons siphon was thus probably lower than the 296 m. quoted in my n. 38, supra, but we have no idea how much lower, and it certainly would not be enough to affect the general picture that I have outlined since the engineers would want to keep the delivery level as high as possible. Levels quoted in this study are compiled jointly from de Montauzan, the 1:100,000 map of the Institut Geographique National de France (Sheet 43), and Matthews, Kenneth D., ‘Roman Aqueducts: Technical Aspects of their Construction’, Expedition (Journal of the University Museum, Univ. of Penna.), Fall, 1970, 2–16Google Scholar.

⁴¹ The position of Les Tourillons on top of the hill brings up a point that should perhaps be mentioned. When one is walking over the country trying to find or follow the line of a siphon it is my own experience that automatically one's eye looks for the lower points in the terrain. This is natural for an ordinary watercourse, but not for siphons, ancient or modern, which often are laid out to take advantage of the heights so as to maintain the water level. In particular, if one is standing at one end of a siphon and looking across the valley to try and locate the other end, the place to look is the highest point in the ridge facing, not the lowest.

⁴² The aqueduct and its bridge was published by Lanckoronski, K., Städte Pamphyliens und Pisidiens (Vienna, 1890), 1, 120–4Google Scholar; this is followed in 1955 by Ward-Perkins, his chief concern being to show that the whole monument dates to one single period; see also G. E. Bean (n. 8, supra). None of these accounts was apparently known to Smith, who quotes only Matthews (n. 40, supra).

⁴³ Conveniently illustrated in Boethius, A. and Ward-Perkins, J. B., Etruscan and Roman Architecture (London, 1970)Google Scholar Pl. 213. The text (409) calls it ‘the most singular building at Aspendos’.

⁴⁴ The commonest explanation is that of Van Buren, (RE VIII A 475)Google Scholar, which is repeated almost verbatim by Bean (loc. cit., n. 8, supra): ‘The purpose of this was to let the water into the open, thus allowing the air to escape from the conduit and so reducing the friction which would impede the flow’ (again repeated, in essence, in the Princeton Encyclopaedia of Classical Sites (Princeton, 1976) 103Google Scholar). This seems to be pure speculation. No evidence is offered either on whether air in the pipe does increase friction, nor on whether there was any air in it and how it got there. The question of air in siphon pipes will be discussed later, but it may be noted that in general they were supposed to be full of water. Another possible explanation, indicated by the terminology ‘pressure towers’, is that they somehow increased the pressure (compare Maiuri, n. 5 supra) enough to bring the water up to the acropolis. This is of course groundless. This was quite a shallow siphon, for the acropolis was only 55 m. above sea level and the venter bridge 30 (Ward-Perkins, 118, n. 8). The pressure needed no boosting, and the ‘pressure towers’ would not in any case have boosted it. A further explanation, favoured by Van Buren (n. loc. cit., supra), is that such ‘pressure towers’ did not increase pressure but reduced it. This is untrue also. The whole question of pressure will be discussed below; at the moment let it simply be said that the water pressure in the channel along the top of the Aspendos bridge would be exactly the same whether the ‘pressure towers’ were there or not. The term used by Van Buren, ‘hydraulische Thürm’, is to be preferred as it is strictly neutral, avoiding these implications. Ward-Perkins, 118, though using the term ‘pressure tower’, clearly understands and states that they had no effect on pressure: ‘The pressure would have remained the same in either case’. He suggests that the reason for the tanks, and hence the towers, was to provide a subdivision point to facilitate maintenance, as urged by Vitruvius (8, 6, 7). But such division points, sensible enough in the open country, where the water could easily be diverted aside (and it must be remembered that a Roman aqueduct, like a river, could not be turned off but only diverted), do not seem likely on top of a venter bridge. How could they be used, and where would the diverted water go? And Vitruvius, expressly quoted by Ward-Perkins in support, equally expressly says that nothing of the sort should be done: ‘sed ea castella neque in decursu neque in ventris planitia ncque in expressionibus neque omnino in vallibus, sed in perpetua fiant aequalitate’. We may also note that the Lyon aqueduct quoted by Ward-Perkins on this page (118) as bringing water from La Martinière and having three siphons, is the one normally called the Gier aqueduct, which actually had four. Ward-Perkins apparently used Blanchet (n. 1, supra), who is much less reliable and comprehensive than de Montauzan, though published in the same year. Stehling (n. 1, supra) 174, believes that the purpose of the tanks was to eliminate the sideways thrust of the water as it went round the bend (basically my own suggestion), but, anticipating Van Buren, links them also to the release of pressure and to the text of Vitruvius, identifying the tanks with Vitruvius' colliviaria.

⁴⁵ Bean (n. 8, supra) 75; illustrated by Ward-Perkins 119, fig. 2. For stone pipes, n. 34, supra. The single stone pipe also resulted in a venter bridge at Aspendos much narrower than the wide ones for the multiple-pipe siphons at Lyon; the narrowness of the ‘pressure tower' ramp in Ward-Perkins’ photograph (n. 43, supra) forms a striking contrast with my own Pl. IX c, X a.

⁴⁶ And so identified by Kenneth D. Matthews (n. 40, supra) in the caption to his photograph of it, page 10: ‘The buttress strengthened the wall of the reservoir on top of the tower since here the line of the aqueduct makes a bend of almost 50°’. At the same time, we must also note that the aqueduct bridge (or arcade) of Segovia negotiated, with apparent success, a bend of 55° without any buttress or other lateral support at all; at this point it is about 25 m. high.

⁴⁷ Ward-Perkins, 118, not unreasonably rejects this proposal, as ‘this hardly seems to be a sufficient reason for the adoption of so elaborate a device’, particularly since the lateral bends in the aqueduct could be largely eliminated by realigning it, and suggests that the reason for the towers ‘we can only surmise’. He may be correct in his rejection and realistic in leaving the question open. But one cannot help feeling that, whatever these towers were for, there must have been an easier way of doing it.

⁴⁸ ‘Quinzenaires’: Callebat 177; 166. Arles siphon: MAGR 85. It was located upstream of the Roman bridge and 60 m. belowwhat used to be called ‘the modern railway bridge’; this is now gone, but its site is still marked by a pair of very large and monumental stone piers. Colliviaria will be discussed later.

⁴⁹ Ashby (n. 1, supra) 35 is seriously in error in wondering why masonry was not used for siphons ‘and so to arrive at the theory of a pressure-supply; for the crushing resistance of their hydraulic cement exceeds the figure now accepted as the safe standard’. Apart from the fact that a pressure-supply was not just theory but common Roman practice (n. 8, supra), ‘crushing resistance’ has nothing to do with the matter, for a siphon-channel (or pressure-supply) is under tension, not compression—the pipe is in danger of being burst from inside, not crushed from outside. The point was clearly seen and expressed by Giovanni, G., ‘L' Aquedotto Romano di Angitia’, RPAA 9 (1935), 72Google Scholar, referring to ‘una costante pressione dal dentro in fuori, che si traduce in sollecitazioni a trazione a cui la muratura è inadatta’.

⁵⁰ Casado, Fernandez, ‘La Conduccion Romana de Aguas de Almuñecar’, AEA 77 (1949), 330–3Google Scholar (esp. his Fig. A). This is summarised in the same author's Aqueductos Romanos en España (Madrid, 1972)Google Scholar, listed in n. 1 supra. The reader is advised that in this book both the illustrations and even the pages are neither numbered nor otherwise identified, making impossible any reference more precise than I have given. The book itself, published by the Instituto Edouardo Torroja, Madrid, is a reprint of six articles by Casado, a Professor of Civil Engineering, which appeared from 1968 onwards in the journal ‘Informes de la Construcción’. Although chiefly concerned with the construction and later history of the arcades and bridges on the Spanish Roman aqueducts (which are profusely illustrated), it also has some material on siphons, and is described by Smith (69, n. 2) as being now ‘the standard work’ on water-supply in Spain. My comments which follow on the aqueducts of Gades, Alcanadre and Valencia, though without individual acknowledgement, are all drawn from this source.

⁵¹ Casado, in his AEA article (page 333), suggests that the receiving tank was on top of it and that it was 8 m. high, giving the siphon an actual head of only 1 m. for a length of over 1 km. This gives a hydraulic gradient normal for an ordinary surface aqueduct but impossibly shallow for a siphon. On the analogy of Soucieu, a 9 m. head for a siphon 1 km. long would be about right (see n. 33, supra). In his Aqueductos Romanos, however, (tenth page from the end of the book, in the section ‘Resumen’) he interprets it as ‘una chiminea de aireacíon (columnaria)’, i.e., a surge tank or ‘colonne piézometrique’ (Fig. 11, infra, and n. 86; for columnaria, see infra, page 215). This means that the siphon would end at the bottom of the tower, not the top, giving it the necessary 9 m. head, but it is not plain what need such a device would fill or how it would work, situated not in the middle of a siphon but at the end of it, and above the hydraulic gradient. There is some confusion here, and I do not myself offer an explanation.

⁵² St. Silv. 1, 3, 66–7:

‘Teque per obliquum penitus quae laberis amnem,

Marcia, et audaci transcurris flumina plumbo’.

For the Marcia and Claudia siphons in Rome see Van Deman (n. 1, supra) 139, 267, and Pl. XLIII, 2; Blake (n. 13, supra) 123; Ashby (n. 1, supra) 152, 249–51; Grimal, P., Frontin (Budé, ed., Paris, 1961), 87Google Scholar, n. 91; the idea of a siphon apparently goes back to Lanciani. P. Grimal, ‘Vitruve et la Technique des Aqueducs’, R. Phil. 1945, 165, considers the siphon a ‘procède constamment employé à Rome mème’. Ponte Lupo: Van Deman, op. cit., 61.

⁵³ Saintes: Triou, A., ‘Les Aqueducs Gallo-Romains de Saintes’, Gallia 26 (1963), 119–44CrossRefGoogle Scholar. Lincoln: Smith 71 (n. 70). He refers to Thompson, F. H., ‘The Roman Aqueduct at Lincoln’, The Archaeological Journal CXI (July, 1955) 106 ffGoogle Scholar.

⁵⁴ GdM 194–6. The statement of Forbes (SAT 167) that it was ‘a siphon of earthenware pipes embedded in concrete’ derives from the original publication of P. Secchi in Bollet. Instit., 1865, 65, who had found fragments of a large terra-cotta pipe (interior diameter 0·345 m., 0·061 m. thick); this was attacked first by di Tucci {NSA. 1879, 274–5), who demonstrated that it could not resist the pressure generated, and by Bassel (N.SA. 1882, 418), who found the lead traces, and a small fragment of lead pipe (0·105 m. in diameter, 0·012 m. thick). A second fragment has the same internal diameter but is much thicker (0·031 m.), raising the interesting possibility that there were two grades of pipes used, the thicker ones being for the bottom of the siphon where the pressure was greater; the internal diameter, of course, would have to be the same throughout. An inscription honouring the builder of the aqueduct, L. Betilienus Varus, survives (CIL I²1529); he ‘fistulas solidas fecit’, fistula being the standard Latin word for lead pipe as opposed to terracotta (tubulas) (and does ‘solidas’ refer to the casting and absence of a soldered seam?).

⁵⁵ Smith, passim. Statements of the established viewpoint are numerous: see Forbes (SAT 165; n. 8, supra); Ashby (n. 1, supra), 35 (‘a lengthy and confused discussion’—Smith, 53; ‘sorgfältige Darlegung’—Van Buren, RE VIII A, 474Google Scholar): ‘Thus every attempt was made to avoid the natural creation of a pressure supply, with which these siphons have been confused’. Lanciani, R., Ancient Rome (London, 1894), 60Google Scholar, was one of the first to set the pattern: ‘it would have been impossible to substitute metal pipes for channels of masonry, because the Romans did not know cast iron, and no pipe except cast iron could have supported such enormous pressure’. It is not clear what ‘enormous pressure’ he means, since he has already mentioned Lyon and Alatri, and even the fact that the latter withstood 10 atmospheres. Presumably he knew these were made of metal pipes. See also Van Buren, loc. cit. Ward-Perkins (117) is an exception, seeing the explanation as economic: n. 56, infra. The Pelican paperback on the mechanics of structures, the standard handbook in the field, repeats orthodox doctrine: ‘for lack of pipes capable of conveying liquids under pressure the Romans incurred enormous expenses in building masonry aqueducts upon tall arches’—Gordon, J. E., Structures (Harmondsworth, 1978) 117Google ScholarPubMed.

⁵⁶ The point is clearly made by Ward-Perkins (117): ‘It is evident that the Republican engineers were perfectly capable of conveying water in bulk under the considerable pressure of 100 m., or approximately 10 atmospheres. That their successors did not normally choose this method of crossing an obstacle can only mean that for all ordinary purposes high level bridging was found to be cheaper and easier to maintain’. See also GdM 208: ‘La question économique était done seule en cause’. Van Buren, (RE VIII A, 474)Google Scholar believes that at Aspendos the siphon was a device to avoid building a high level bridge all the way across, on the principle that siphons were cheaper than bridges. On the question of aesthetics, one should not be too quick to reject the idea that the Romans may sometimes have put ‘un magnifique étalage de force et de grandeur’ before strict economics. Moreover, the highest aqueduct bridge in the world is the viaduct at Roquefavour, 10 km. west of Aix-en-Provence, carrying the water supply for Marseille. Opened in 1847, it is 82·65 m. high and 375 m. long, and was specifically designed both to imitate the Pont du Gard (it is a three-tier bridge) and to surpass it (being half as big again); which it does so effectively that tourists have been known to mistake it for a Roman antiquity. It seems certain that, in the 1840's, there must have been a cheaper way of doing it, and the money was judged well spent for the imposing effect. The same reasoning may have applied in antiquity to the Pont du Gard. Certainly a bridge is much more visibly spectacular than a siphon, as is demonstrated to-day by the general fame of Roman aqueduct bridges and the general obscurity of siphons twice or three times their size. A further point influencing the choice of a bridge rather than a siphon for the Gard crossing may have been the local availability of good stone: the quarries are only 600 m. from the bridge, on the left bank of the river. This would mean very cheap transport costs, which, in the days before brick and concrete, were often the single most expensive item in a building.

⁵⁷ The limitations of this approach are discussed by Smith, 59–61.

⁵⁸ ‘[the siphon] has rarely been considered (and not at all recently) and even when it has, a number of misconceptions and inaccuracies have been expressed so convincingly that they have gained general currency’—Smith, 51; he had apparently not seen Callebat, to whom this does not apply.

⁵⁹ At the risk of apparent triviality, we may offer the fictional illustration of the little Dutch boy who stuck his finger in a leaking dyke. What he had to hold back was a column of water as thick as his finger and 10 feet or so high, not the weight of the whole North Sea. That the story is apocryphal does not affect the reality of the principle involved.

⁶⁰ The reader will note the resemblance and the relevance of diagram e to the long arcades or arches crossing the Roman Campagna. If these had been replaced with pressure pipes laid on the ground, the pressure in them would nowhere have been very high, since the vertical column of water supported would be the height only of the existing arcades. Roman pipes could have coped with this without a moment's difficulty, and frequently did, as a matter of course. The use of arcades and an open channel had no doubt many other important advantages to recommend it, but avoidance of high pressure was not one of them.

⁶¹ Van Buren, (RE VIII A, 473)Google Scholar falls into this fallacy: ‘es war wesentlich, den Wasserdruck in gewissen Abständen zu vermindern mit Hilfe einer Erfindung, die dem Wasser die Rückkehr zu seinem Niveau gestattete, bevor es in einem weiteren Siphon eintrat’.

⁶² ‘During normal operation’ is a vital phrase. When the siphon was being emptied, or filled up from dry, conditions were quite different.

⁶³ ‘Dividing up a pipe’: for suggestions that the multiple pipes of the Lyon siphons were subdivided yet further in passing over the venter, see Blanchet (n. 1, supra) 34, n. 4; also my n. 28, supra. The relevant formula expressing the interrelation of pressure, volume of water and pipe resistance, is given and explained in GdM 182. It is

where R is the tensile breaking point of lead ( = 1· 35 kg. per sq. mm.); this remains the same for all siphons provided they are made of the same kind of lead; ϵ = the pipe thickness; D the diameter; and H the pressure, expressed in vertical metres of water supported (pressure is often also expressed in terms of atmospheres. Roughly speaking, 10 m. = 1 atmosphere, so that if a siphon goes down to 50 m. below the header tank, the pressure in the pipes at the bottom is 5 atmospheres). Where only one of these four terms is not known but the other three are, the equation can be solved, giving the unknown quantity. That is, given pipes of a certain thickness and known diameter (and we know the tensile strength of lead), we can then solve for H, telling us how far down (in metres) such pipes can go before they burst; or,

de Montauzan uses the example of some very thin pipes, only 0·6 cms. thick, and of diameter 57·9 cm. (which is very large indeed). Filling in these values in the equation, we get

that is, H = 28·15 metres, or under 3 atmospheres.

Coincidentally, this is the exact height of the Ponte Lupo. If these very improbably weak pipes could substitute for something as big as the Ponte Lupo (‘of unusual size’—Van Deman, 126) before bursting, we may guess at the capabilities of the smaller pipes actually in general use.

Likewise, if we know the depth of the valley (and hence the pressure) and have a tiny fragment of pipe, giving us the thickness, then the equation can be rearranged to tell us the diameter, D:

this is the size of pipe that can be used of this thickness to cross this valley without bursting; of course, there is no safety factor in this equation.

For a general outline of the stresses operating on pipes under pressure, see Gordon (n. 55 supra) 119–23. It may be noted that longitudinal stress on the walls of the pipe, for any given pressure, is only half circumferential: in other words, if a pipe does crack, the crack is certain to run lengthways along it, not around it.

⁶⁴ SAT 165: ‘It is not known how the Greeks got rid of possible air pockets in these siphons’. We should note the difference between air locks and air pockets. Properly speaking, an air pocket is simply a stationary bubble somewhere in a pipe. An air lock is such a bubble big enough to block the pipe entirely and, because of pressure or a bend in the pipe, stuck immovably in position. Unlike the air pocket, it therefore will block the water flow. The air pocket, on the other hand, causes trouble only in the event of water hammer (n. 74, infra). Of course, the two terms are often used loosely without this distinction being observed.

⁶⁵ This is one reason why Van Buren's suggestion (repeated by Bean) that the ‘pressure towers’, on the Aspendos siphon were meant to release air (n. 44, supra) will not work. If air was released as the water entered the tank it would be replaced with more entrained on the other side as it left. If no more was entrained, then why the second tank at the far end of the bridge to get rid of it, and, again, to let more in ? Air, it should be emphasised, cannot get into the circuit while it is in the pipes; one therefore does not need a series of tanks, each intended to eliminate air accumulated since the last one. The thesis advanced by Van Buren needs much more explanation and supporting evidence if it is to be properly discussed.

⁶⁶ Plommer, reviewing Callebat, (CR XXV, 1975, 222)Google Scholar remarks: ‘After a copious downhill run and after any sort of sudden upward angle, air locks are surely likely to form in a pipe’. There are three things to be queried in this statement. First, the implication that water in a filled pipe runs faster downhill. Water is incompressible so if the pipe is full the water in it forms a single unbroken column moving at uniform speed, uphill and down alike. Second, the ‘upward angles’ where air locks form are at the peaks (if any) of the siphon profile, where the pressure is low, not where a downward angle simply flattens out as it runs on to a horizontal venter. Third, the phrase ‘are surely likely to form’ is disturbing, implying supposition on a matter where firm factual evidence was available. See also n. 67, infra.

⁶⁷ This situation is potentially very damaging to the pipe, especially in the case of water hammer {infra). In modern siphons the danger is avoided by placing air pressure release valves (ventouses) at such points (infra). Plommer (loc. cit. n. 66, supra) is here also to be questioned: ‘if, as Callebat says, in modern hydraulics valves are placed at the high, not the low, bends in the pipe [there is no “if” about it; they are (n. 81, infra; this is easily confirmed by conversation with any engineer)] the great mass of rising air [what mass ? how great, and where did it come from?] would have been served in the ancient aqueduct by the castella above the venter on either side of the valley’—i.e., by bubbling up the two sides of the U and out the top. Apart from the fact that this envisages the air on one side successfully bubbling up backwards against the downward flow of the current, it does not correspond with what Callebat describes and clearly illustrates: he is not talking about the two high points in a U but the high peak in the middle of something like a W, where any air that collected would be trapped with a lower point on either side of it, and could not escape because it would mean the bubbles going downhill to get round the bend. Of course, they might be carried down and round the bend by the current, once the siphon was flowing. It would depend on various things—turbulence of the flow, how far down the bend was, how sharp, and so on.

⁶⁸ de Montauzan suggests (GdM 218) that the very slight slope observable in venter bridges is to avoid a strict horizontal, where any small accidental bump in the pipe might create a short section above the horizontal, where entrained air could lodge. Smith (58) accepts this view: the slope ‘could be construed as an air release device’. It is also possible that the Les Tourillons tank is, at least partly, an air release device of this sort, since it is on a stretch of pipe that is, in effect, a ‘flat siphon’, where any inequality of the ground would create a bump that might harbour entrained air (not air coming out of solution, for such a bump would be below the hydraulic gradient and under pressure), but I do not think this likely. De Montauzan (75) seems to be somewhat confused on the issue, and suggests that the tank is in effect a ventouse to release air somehow collected in the upstream (le Tupinier) siphon which might otherwise burst the geniculus formed as the pipe climbed up out of the valley over the edge on to the la Craponne plateau (Stehlin (n. 1, supra) 172 takes this interpretation a step further, suggesting that this is one of Vitruvius’ colliviaria; compare n. 44, supra). This is unconvincing for two reasons. It does not account for the presence of this air in the siphon. Second, this being a fairly high point in the siphon, there is no question of any great static pressure, whether from air or water; and it is most unlikely that a geniculus could be burst by the inertial thrust (for inertia, see infra) of air, as opposed to water, rushing round it (GdM 75—the ‘afflux violent de l'air en ce point haut’). n. 98, infra.

⁶⁹ Stehlin (n. 1, supra) 169; Callebat 173, fig. 14. We may note here that whatever the turbulence of water rushing round a geniculus, it will not create an air pocket unless there is air present in the water already; any void created purely by the turbulence will be a vacuum, not an air pocket (compare n. 66, supra).

⁷⁰ As urged by Vitruvius in his famous passage about ‘red rock’ (8, 6, 8; Callebat 179 ad loc).

⁷¹ This is what Vitruvius (8, 6, 6) means when he says that in the absence of a venter bridge (with its two obtuse geniculi, one at either end) the single acute geniculus that it replaces, at the bottom of the V, will burst. Plommer seems not to understand this, and thinks that ‘Vitruvius is careless in his use of geniculus, for after saying that there shall be none, he speaks of two, one at each end of the horizontal stretch (venter)’—Plommer, 28, n. 1. Callebat (179) tries to differentiate between the terminology for these two different types of bend, geniculus being reserved for the obtuse ones at the ends of the venter.

⁷² Ward-Perkins (118) points out that the sideways geniculus at Aspendos could easily have been avoided by straightening out the aqueduct bridge. Norman Smith suggests to me (private communication) that the bends may have been caused by a change in plan.

⁷³ This siphon carries an irrigation channel of the Société du Canal de Provence over the River Arc and also highway RN7 (from which it may easily be inspected) 3 km. east of Aix-en-Provence. I am indebted to Eng. Herenguel, of the Société's head office at Le Tholonet, for much valuable information on the operation of this and other siphons.

⁷⁴ Private letter from Prof. D. R. Townsend (n. 1, supra).

⁷⁵ Recognised by Vitruvius (8, 6, 9).

⁷⁶ Hodge (n. 30, supra). The best account of Roman taps, a much neglected subject, is that of Kretzschmer (n. 1, supra), who makes it plain that ‘la robinetterie romaine consiste uniquement en robinets du type à boisseau [ = rotary-plug], en bronze coulé’ (89). Used for both taps and stopcocks (Fr. robinets de débit, robinets de communication (92)) the rotary-plug tap was usually of small size, though two are known at Ostia, one in the museum and one in situ on a water main on the decumanus, to which Kretzschmer (93) gives diameters of 20 and 30 cm. (‘C’est un véritable monstre!’—90); Haberey, Waldemar, Die Wasserleitungen nach Koln (Bonn, 1972), 116Google Scholar and fig. 83, gives for the same tap (the one in the museum) a diameter of 16–20 cm.

The chief characteristic of the Roman rotary-plug tap, as relevant to siphons, is that while it is quite resistant to inertia, it is very vulnerable to static pressure. It can therefore be used on a pipe carrying quite a large volume of water running fast, where, on shutting off the current the stress to be faced would be that of inertia, but must not be employed on the lower parts of a siphon, where the pressure is high even if the water is standing still. This is because inertial thrust constitutes an impetus directed in one direction only, which the side walls of the central cylinder (Fr. noix), turned across the supply pipe to cut it off, are perfectly capable of withstanding, while static pressure is exerted, inside the tap as in the pipe and elsewhere, in all directions equally—up, down and sideways; and while the sides and bottom of the cylinder and the housing inside which it turned could resist the pressure, there was nothing to hold the cylinder down (in a modern tap there is); thus, under the upwards pressure it would simply pop up out of the surrounding housing, wrecking the tap (Fr. sauter) and releasing the water. Given that the Romans evidently had no other kind of tap, then on this basis alone taps on the lower sections of siphons, for whatever purpose, seem to be completely excluded (see colliviaria, infra). Attempts were made to include in the design some device that would hold the cylinder, the core of the tap, down and in place (93; Nemi), but they do not look effective. Another approach to the same problem involved the shape of the plug or central cylinder. This could be made either tapered or cylindrical, and turning in a housing either conical or parallel bore to match the plug. A tapered plug in a conical housing was easier to turn but more liable to leaks, as the plug was more liable to lift up under pressure. The cylinder was therefore the form normally used on pipes where the water pressure was relatively high (though, as we have seen, any really high pressure would burst the tap), such as water mains. If there really were taps on the siphon pipes, as de Montauzan suggests (n. 19, supra), this is the type they must have been. In either form, the tap was evidently very stiff to turn, and required an iron bar or spanner to turn it off or on, like a modern street hydrant. Nearly all preserved taps have at the top, in place of a handle, a small, square loop of metal, to take the end of this iron bar or key; they were not meant to be turned with the fingers like a modern tap. See also Kretzschmer (n. 5, supra) 57, fig. 97.

⁷⁷ Its chief uses seem to have been in presses for olives and grapes, which sometimes used a large wooden screw, and, made of bronze, in some surgical instruments (notably the vaginal speculum; Scarborough, J., Roman Medicine (London, 1969)Google Scholar, Pl. 43.).

⁷⁸ e.g., at the mouth of the entry channel into the castellum at Nimes. Sluices could be quite sophisticated devices, and de Montauzan reproduces a remarkably complex one from a distribution castellum at Tebourba (Tunisia), giving simultaneous but independent control of two offtake channels, the sluicegate itself having four different settings (GdM 317, fig. 125). We may also note a small limestone water distribution tank preserved in the museum at Vaison-la-Romaine, from the House of the Messii. It has two, possibly more, compartments, regulated by two stone sliding sluices, themselves pierced by a pair of round holes. I do not understand how it worked. See also Hodge, A. Trevor, ‘How did Frontinus measure the Quinaria?’, AJA 88 (1984 forthcoming)CrossRefGoogle Scholar, for the use of sluices to measure the discharge of an aqueduct.

⁷⁹ The round holes in the stonework are of 0·40 m. diameter, but there is no lip, flange or socket cut into the stone to receive the end of the pipe. It must therefore have been held in place by a ring, or collar, of cement. There is no telling how thick this was, and the real size of the pipe thus remains unknown. But it must have been big, the biggest known (so far as I am aware). The three drain holes in the floor do have a recessed seating for a circular lid, with sockets for three projecting lugs on the lid to hold it firmly in place; the pipe may somehow have been fitted into this seating too. These three lids would effectively shut off the three drain holes, but there is no trace of any device (such as sluices) to shut off the 10 offtake pipes above. There is, of course, as noted above, a sluice at the point where the aqueduct enters the castellum, but even this is not devoid of problems. What would happen if it were closed, to cut off the castellum ? Would not the water in the aqueduct simply back up and overflow ? All in all, the operation of the Nimes castellum raises a number of questions as yet unanswered.

⁸⁰ Plommer 27, fig. 4b, shows a drawn reconstruction of a section of piping from the bottom of a siphon. Included in it is a device marked ‘Air valve’ (Plommer, 26–8, following Vitruvius, believes that there was air in the bottom of siphons), which looks like a short escape pipe or vent controlled by a rotary plug tap mounted on it. The tap would, of course, be under high pressure and I should think a rotary plug would be quite unsuitable (Kretzschmer, n. 5, supra, 52, worries about ordinary house taps of this type bursting under very much lower pressures, about 0·6 atmospheres, a twentieth of what they would face on a big siphon). Plommer is not concerned with these technicalities over kinds of taps and it would be both pointless and unfair to try to read too much into the details of his drawing. But it remains the only time I know of that anyone has even tried to draw a valve of any kind connected with a siphon.

⁸¹ ‘An air valve should be located at each high point to allow the escape of air and gases and to admit air sufficiently rapidly to prevent the creation of a partial vacuum in the pipe line’—Babbitt, and Doland, , Water Supply Engineering (New York, 1955), 154Google Scholar.

⁸² This is essentially how modern pipes and drains are cleaned. It will be noted that if the drainage point is to be used for cleaning as well, then the simple hole-and-bung which I have suggested above is much superior to a tap or valve, because you can get your hand into the pipe, which cannot be done through a tap. The mechanism of a tap could also easily become jammed with the sediment.

⁸³ n. 30, supra.

⁸⁴ It is difficult to be consistent and accurate in one's use of technical terms in siphon hydraulics. A term accepted in engineering is sometimes different from that employed by archaeologists for the same thing, and either group will sometimes differ internally as well. It is also sometimes more precise to use the French term, so much of the basic literature on ancient siphons being in French. Though I have tried to avoid it, a residue of imprecision and inconsistency in terminology must remain. I can only ask the reader to be on guard against it.

⁸⁵ Callebat 181, fig. 16 (P).

⁸⁶ The evidence for the use of such a device in antiquity seems to rest with a certain M. Flachat, who, in examining what was thought to be a Roman siphon at Constantinople, found what seemed to be one: ‘Dans le bas est une ventouse appuyee contre une tour qui s'élève plus haut que l'aqueduc; ce qui fait voir l'ancienneté et la nécessité des ventouses’. GdM, 190, quoting this account, observes that “ventouse” is the wrong term, ventouses being only in the top sections of siphons, and this is a ‘cheminée’ or ‘tube piézométrique’. The problem of getting the tube up to the hydraulic gradient (‘on n'imagine pas en effet un tube vertical de plus de 100 mètres de haut’), here solved by running it up the side of a handy tower, could also be met by running the tube not vertically but sideways up the natural slope of the hill. As for function, such devices are ‘simples régulateurs de pression pour parer aux arrets brusques’. A similar ‘cheminée d'équilibre’ is conjectured for Saintes by A. Triou (n. 53, supra).

⁸⁷ n. 51, supra.

⁸⁸ Front. 1, 25, mentions him as one of two possible inventors (the other is Agrippa) of the quinaria-sized adjutage. It is all we know of his ancient reputation or achievements in waterworks. Does this or does it not amount to much ? Plommer, who has a great respect for Vitruvius and his ‘precise Hellenism’, thinks that it does, and that he ‘may have been one of the most influential watermen of his age’ (Plommer, 30). Smith, 58, thinks it does not, and remarks ‘He claims very little personal involvement in engineering work, and later on the only mention of Vitruvius by Frontinus is in connection with plumbing and pipe sizes’—which can hardly be meant as a compliment since Smith sandwiches it in between statements that Vitruvius was not ‘a particularly well-educated man’ and probably ‘did not know what he was talking about’ (see also Smith, 50).

⁸⁹ ‘It is an interesting thought that the status and importance of Vitruvius was very likely far greater during the Renaissance than ever it had been under the Romans. This same exaggerated importance attaching to a sole survival has worked its deceptive influence in modern times’—Smith, 50.

⁹⁰ That I am not exaggerating here is plain from the remark of de Montauzan: ‘On est un peu étonné de ce l'œuvre de cet auteur passe encore pour une somme complète du savoir technique chez les Romains’ (GdM 178). The answer is given by Pierre Grimal (R.Phil., 1945, 174): ‘Vitruve ne prétend donner un tableau complet de la technique en usage à Rome à son temps’, and if modern writers insist on taking it as a ‘tableau complet’ or ‘somme complète’ that is not the fault of Vitruvius.

⁹¹ ‘Sa réflexion s'appuie sur la pratique contemporaine; elle n'en est pas esclave’—Grimal, loc. cit. For our purposes, of course, a ‘réflexion esclave’ would be much more informative and hence valuable.

⁹² His shortcomings are listed by Callebat, xlix, and include mistakes, omissions, bad reasoning, superficiality, uncritical judgement, overweening self-confidence, and what we may perhaps compendiously call ‘coffee-table culture’. To this imposing list Smith, 58, adds the frequent obscurity of his exposition, possible misreading of Greek sources, and, to cap it all, the probability that he did not know what he was talking about anyway: ‘and if Vitruvius did not know what he was talking about then it is hardly surprising that we cannot find out’.

⁹³ ‘It is implicit in nineteenth century and later writers that Vitruvius is ultimately a reliable source. Any failure to comprehend him is adjudged to be a failure of modern translation and intelligence’—Smith, 58. The matter is clouded by the reference to translation. Smith always refers to, and apparently uses, Vitruvius in translation and not in the Latin text, a potential source of trouble. But his main point, the (unverified and untested) assumption of Vitruvius' reliability, seems a good one. One suspects that, in approaching the subject (so far as possible) in a spirit of intellectual and comprehensive humanism, untrammelled by banausic details of technical reality, Vitruvius and his nineteenth century commentators may have had much in common.

⁹⁴ Evidence for the absence of air pressure is threefold: (1) There is none in the bottom of modern siphons, as more than one engineer has stressed to me personally. (2) It is contrary to the principles of siphon hydraulics as set out in any textbook of water supply engineering. (3) Smith, the reader must be reminded, writes on Roman siphons as a recognised authority on the history of engineering technology and a specialist in the history of water supply; and his verdict is plain: ‘One point is clear. Vitruvius is proposing some sort of device to relieve air pressure in the siphon. In fact he is advising on a non-existent problem. In a siphon carrying water under pressure the problem of air-locking cannot arise simply because the pressure is high. There is no air pressure to relieve when the siphon is in flow’—Smith, 57.

⁹⁵ It is not realistic to seek a way out by suggesting that Vitruvius may be referring to the operation of filling the siphon, when air pressure may be present and cause problems. Interruption of the supply for repairs and maintenance is a specialised and uncommon practice, to which Vitruvius rarely refers and not in the present context. Re-establishment of the supply and filling the siphons, a natural corollary of interruption, he never speaks of at all. If we insist on the ‘filling the siphon’ argument to explain the presence of ‘vis spiritus’, where on any normal reading of the text he is quite patently talking about something else much more obvious and straightforward, i.e., normal operation, then again we are faced with two distasteful alternatives. Either there is a large undetected lacuna in the text, where he made clear his shift of topic from normal operation to temporary interruption, or his exposition is hopelessly tangled and incomprehensible.

⁹⁶ n. 5, supra. For other contradictions, see Smith 52–4.

⁹⁷ Very full exposition in Callebat, 171–2.

⁹⁸ Water can generate static pressure, friction, or inertial thrust. Air can in theory also generate all three and sometimes does in practice (windmills are moved by the inertia of air, and spacecraft burned up by friction), but in siphons the second two are negligible.

⁹⁹ Compare Vitruvius (1, 1, 7), ‘spiritus naturales’. Does this refer to inertial thrust or ‘airpockets’ (so the Loeb translator) ?

¹⁰⁰ SAT 165.

¹⁰¹ Belgrand, M., Les Aqueducs Romains (Paris, 1875), 71Google Scholar; Smith, 61. Belgrand, in spite of this experiment, used masonry channels in his own aqueducts, apparently not always with success: ‘Belgrand eût pu tirer un enseignement de ce fait; car il est notoire que le pont-aqueduc, construit par lui sur la vallée de l'Yonne, a des fuites importantes’.—Blanchet, (n. 1, supra) 27, n. 5. In the history of siphons Belgrand is to-day known chiefly for his advocacy of a Vitruvian siphon dipping across the valley in a single sweeping unbroken curve, like a roller coaster, running where necessary in cutting or embankment to maintain the even profile (GdM 184, fig. 71).

¹⁰² The formulae are πr² for the area (cross-section) of a circle, πd for the circumference (r being the radius, d the diameter), π is common, so the difference proportionately between area and circumference is the difference between r² and d. d = 2r, so the difference is essentially that between doubling the radius and squaring it (each then being multiplied by π). An example will make the point clear, if we consider two pipes, of 20 cm. and 40 cm. diameter:

¹⁰³ I equate capacity to cross-sectional area of pipe, omitting such other factors as velocity or increased friction from increased surface area, which would only complicate matters.

¹⁰⁴ Smith 62, GdM 205. This is seriously misquoted by Forbes, who refers (SAT 168) to the Beaunant siphon ‘which carries the water 17 m. over a river with 7 lead pipelines (diameter 270 mm., 35 mm. lead) for which some 10,000 tons of lead were used’. The 10,000 tons figure refers to the total for the 9 siphons in the Lyon area, not just Beaunant; and Beaunant had 9, possibly 8, pipes, not 7.

¹⁰⁵ Smith 62; Ward-Perkins 117: ‘What limited [the siphon's] application on a larger scale was the fact that the best available medium for conveying water under heavy pressure was lead piping; and not only was lead an expensive commodity, but the amount required to carry a flow equivalent to a normal specus across a valley was disproportionately large’. I would query the view that lead was expensive, at least to produce; transporting it is a different matter: n. 13, supra.

¹⁰⁶ Callebat 177–9. Of particular interest is the section of terracotta pipe from Limoges, which is not round but square.

¹⁰⁷ Examples of both are illustrated by de Montauzan (GdM 204, figs. 80, 81), reproduced also by Callebat 173, fig. 14. The solder (based on an analysis of that used in some of the pipe joints preserved in the Museum at Aries) was a mixture of 84 parts lead to 60 tin (GdM 204, n. 1).

¹⁰⁸ Vitr. 8, 6, 8. M. Flachat, (GdM 190) who examined the still operational aqueduct of Constantinople, identified as Roman, observed on its course a siphon, evidently multiple-pipe. The pipes were normally buried under a metre of earth and broken tiles, but had been exposed for repairs. They were ‘cordés avec des bandelettes de chanvre, comme les bâtons de tabac (=cigares]’. He does not say what the pipes were made of. De Montauzan, not unreasonably, thinks that the rope lashing was round the joints, though Flachat does not say so (compare GdM 204, 190).

¹⁰⁹ I may remind the reader of the strength of this objection with an example that happens to be familiar to me. Railwaymen on a steam engine often talk of ‘blowing up’; it means getting up steam pressure by turning on a blower that fans the fire. Not unnaturally, the term has often alarmed the unwary, who think it means the boiler is about to burst.

¹¹⁰ A quick survey of all the Greek compounds in syn- produced nothing hopeful, but I cannot claim to have been exhaustive.

¹¹¹ See Stehlin, Mortet (n. 1, supra), Van Buren, (RE VIII A, 472–5)Google Scholar; GdM 187–90 reads ‘colliviaria’, interpreted as drain-cocks, but his exposition (188) of their function in releasing the ‘vis spiritus’ is not clear. Callebat, 175, follows him in retaining the MSS reading, ‘transmis unaniment par nos mss et porteur, sans distorsions étymologiques, d'une signification satisfaisante’. He sees the colliviaria as drain cocks (‘les organes de vidange (tuyaux munis de robinets)’), and expressly warns against the translation ‘ventouse’, ‘traduction souvent proposée, quelle que soit d'ailleurs la leçon retenue’, ‘ventouses’ being air valves. Plommer, 28, n. 1, says ‘Gundermann's “colliquiaria” seems obviously right, against the “colliviaria” of the manuscripts’, but does not say why. His interpretation is clear from his text: ‘along the venter itself we must contrive small valves to relieve air pressure in the pipes. One is amazed at the technical refinement of all this’.

¹¹² The [water] pressure in the venter is determined solely by the head above it and it cannot be relieved by valves, stand-pipes, water-cushions—whatever they are—or anything else’—Smith 57.

¹¹³ n. 86, supra.

¹¹⁴ Smith, Norman A. F., Man and Water (London, 1976) 217Google Scholar n.: ‘Not even de Montauzan is free from defects, while Ashby, pp. 34–7, is curious and confused, and A. Choisy … is idiotic’. Smith, 57: ‘The prize for incompetence must go to the eminent French architectural historian, Auguste Choisy. His extraordinary layout is shown in fig. 2 [=fig. 12, present study]. Quite apart from failing to mention what the Romans used for sky hooks, the arrangement achieves nothing. Choisy is merely substituting two siphons for one, he is adding to the flow length, and achieving not one iota of pressure reduction or anything else. And in any case these idiotic elaborations do not even match the Vitruvian text’.

¹¹⁵ Smith, n. 38, makes it clear that his knowledge of Aspendos is limited to Matthews' publication (n. 40, supra), a popular work that treats of Aspendos only marginally; he is also fully conscious of the lacuna.

¹¹⁶ de Montauzan's objection about the difficulties of a tube 100 m. high (which is what would be needed for Soucieu or Bcaunant) does not apply because we are here dealing with shallow siphons. Les Tourrillons is built on top of a hill about 50 m. high, which therefore replaces part of the vertical rise needed, and Aspendos (though believed by Smith (loc. cit.) to be ‘evidently a big siphon’) is quite shallow (n. 44, supra).

¹¹⁷ Asked to suggest any device that, hypothetically, a siphon such as the Roman ones ought to have had, other than air valves at the high points and drain cocks at the bottom, none of my colleagues in engineering could think of one. Both de Montauzan and Callebat favour drain cocks. (GdM 187; Callebat 175).

¹¹⁸ As noted by, e.g., GdM, loc. cit. His description however becomes confusing when he tries to relate them to the ‘vis spiritus’.

¹¹⁹ ‘Aquarum ductus per medias possessiones diriguntur, quae a possessoribus ipsis vice temporum repurgantur; propter quod et levia tributa persolvunt’—Gromatici Veteres, edd. Lachmann, C. et Rudorff, A., (Berlin, 1848–1852), 1, 349Google Scholar (ex libris Magonis et Vegoviae auctorum).

¹²⁰ GdM 287–90.

¹²¹ There are other well-known problems in Vitruvius' description of siphons, but the problem is usually one of deciding what he meant or recommended rather than elucidating how the existing siphons actually worked. The reader is referred to Callebat, 167–81.

¹²² SAT 161.

¹²³ Smith, 64, is very worried (‘the issue of overriding importance’) about ‘how the Roman engineers matched the flow of an open channel to the flow capacity of a multi-piped siphon’, the hydraulic calculations being very involved; and he leaves it as an open question. It does not seem to me very important. The Romans must have often worked on a trial and error basis, with siphon tubes stopped up (as at St. Genis) or water allowed to overflow as necessary (GdM 218). The very fact that the water in a Roman aqueduct channel did not always run at the same depth, but had high and low seasons, like a river, would, I should think, by itself largely invalidate the calculations Smith is thinking of.

¹²⁴ Significantly, the Roman aqueduct at Cherchel has been replaced by a modern one following roughly the same course. In the modern conduit, siphons are used to cross the valleys where the Romans used bridges, for the low cost of cast iron pipes and the high cost of quarrying stone have now made siphons cheap and bridges dear. In Roman times it was evidently the opposite (Leveau, P. and Paillet, J-L., L'Alimentation en Eau de Caesarea de Mauritanie (Paris, 1976), 78Google Scholar; GdM 208).

¹²⁵ GdM Pl. 1 and 2, passim, but especially on the southern section of the Gier aqueduct; its course, especially as shown in the earlier Pl. 1 map (after Delorme), from St. Chamond to Mornant (whether with or without the St. Genis-Chagnon cut-off) is so serpentine as almost to defy credulity.

¹²⁶ GdM 205.

¹²⁷ It will be appreciated that these figures are all fairly rough. The Craponne aqueduct, for example, may have had fewer than nine pipes (GdM 218). But they do give the order of magnitude involved.