Some Reservations about the Newport Tower C-14 Dates

J. Huston McCulloch

August, 2001

This paper was published in the Midwestern Epigraphic Journal, Vol. 15, 2001, pp. 79-92.

In a widely cited 1997 paper, Johannes Hertz raises a number of arguments against a pre-Colonial origin for the famous Newport, Rhode Island Stone Tower. Hertz insists that it was modeled after the 17th century Chesterton Mill in Warwickshire, England, and points out that a 1948-9 survey by Hugh Hencken and William S. Godfrey found indisputably colonial artifacts at the bottom of a trench that surrounds the foundations. He extensively discusses the recent carbon-14 dating of the mortar by Jan Heinemeier and Högne Jungner (HJ, 1994). According to HJ, their tests indicate that the Tower was built not earlier than 1635 AD, and most likely in the range 1651-1679.

Architect Suzanne Carlson, writing already in 1996 in response to the 1994 Danish original of Hertz's article, persuasively refutes Hertz's architectural and historical objections: Even Johannes Brønsted, whom Hertz approvingly cites, admitted that "the Romanesque lines of the tower are so striking that if the tower stood in Europe, probably no one would contradict a date in the middle ages" (in Hertz 1997, p. 75). Carlson argues that Chesterton Mill was in fact built as an observatory, and only much later converted to use as a mill. She points out that the trench discovered during the 1948-9 survey makes sense as part of a colonial repair of a pre-existing tower for use as a windmill, after an earlier mill blew down in 1675. Furthermore, this trench does not work as part of the original construction, because it lacks any evidence of the presence of the staging that would have been necessary to have supported the arches. Instead, its backfill contains thousands of mortar fragments, as would be expected if it were opened as part of a repair operation.

However, Carlson admits that she, as an architect, does not understand the highly technical carbon-14 dating of the mortar. I have had a little chemical training (as an undergraduate at Caltech), and some prior familiarity with dendrocalibration, which is an important complication in the HJ paper. Perhaps they or someone else will be able to correct me, but my reading of their paper is that although the C-14 results are certainly consistent with a 17th century colonial origin for the tower, they by no means conclusively rule out a pre-Columbian origin.

To make lime mortar as was used in the Newport Tower, limestone (mostly calcium carbonate) is roasted to make calcium oxide, or quicklime. This is first combined with some water to make calcium hydroxide, or slake lime. At the time of construction, additional water and sand are added to the slake lime, allowing it to absorb carbon dioxide from the air and to set into crystalline calcium carbonate. Some of the atmospheric carbon dioxide would contain radioactive C-14 rather than inactive C-12 or (less common) C-13, much as photosynthesis captures C-14 from carbon dioxide in the atmosphere.

As originally devised circa 1950, radiocarbon dating was based on the assumption that the proportion of C-14 in the atmosphere has been constant over time, so that the amount of C-14 left in a sample would fall exponentially with its age. However, C-14 dating of tree rings of known age subsequently demonstrated that this is not strictly true. Standard "dendrocalibration" curves have been constructed to compensate for this variation in atmospheric radiocarbon (e.g. Stuiver and Kra, 1986), and these curves are used by HJ.

HJ admit that there are two factors that potentially could make the measured age of the mortar appear younger than the true age of the construction.

The first of these is slow absorption of atmospheric carbon dioxide. The surface of the mortar sets up quickly - in as little as a few hours. The interior portions of the mortar will ordinarily set up eventually, but this requires carbon dioxide to diffuse through pores in the mortar, either in gaseous form, or in solution in the water that has been added to the slake lime. Depending on how easy it is for the gas to find such pores and work its way through them, this could take a considerable time. On p. 38, they claim that the effect should be limited to a few years, yet on p. 40, they admit that in the case of the two medieval Finnish churches discussed below, some samples still exhibited an alkaline reaction, indicating the presence of some unreacted calcium hydroxide, approximately 700 years after their construction! Because of this slow absorption, the estimated date will not reflect the actual date of construction, but some weighted average of later dates, even if at the time of testing the reaction appears to have been complete. Surprisingly, HJ do not report having even tested the Newport Tower samples for residual alkalinity.

The second problem, which they just mention and then drop, is recrystallization of the carbonate. It is well known that the calcium carbonate in bone is not very reliable for C-14 dating, because the original carbonate ions may exchange with carbonate in the groundwater that might be either too old - if it represent dissolved limestone - or too young - if it contains atmospheric carbon dioxide dissolved in rainwater. For this reason, the preferred method of dating bone is to isolate the bone collagen, which contains carbon, but in a more stable form. Similarly, lime mortar that is exposed to rain on a regular basis may contain carbonate that dates not from when the mortar first set up, but from much later rainstorms that may have drenched the structure.

A third factor that might make their tests give too young a date, mentioned by Hertz but not by HJ, is the possibility of colonial repairs or "tuck-pointing" to strengthen a pre-existing structure for conversion to a windmill. In this case, the surface mortar might be colonial, while only the deeper mortar, well inside the joints, would reflect the true date. According to Hertz (p. 93), test cores were taken deep enough to avoid contamination from both rainwater and later repairs. Nevertheless, some of the samples HJ took were in fact "prized out as whole pieces of mortar, and marked as surface samples. These whole samples ... are expected to yield the most reliable results as the crushing of the mortar can be done under controlled laboratory conditions, resulting in a more effective mechanical separation of the fossil carbonate from the samples." (HJ p. 36) One of these surface samples, from the flue above the fireplace, was in fact used to construct their composite date for the Tower's construction, contrary to the impression given by Hertz. HJ themselves make no mention that the other samples were taken with a care to avoid repairs.

The primary potential source of bias in the opposite direction, according to HJ, is unburnt limestone particles that may remain in the quicklime after roasting. These will remain as calcium carbonate in the mortar, but will contain essentially no C-14, and make the sample appear older than it really is. These particles are harder than the mortar, and react more slowly with acid. In order to minimize their effect, HJ take the trouble to separate the carbon dioxide that is released from the mortar into two fractions. They argue that the first fraction should contain little if any of the fossil carbonate, and indeed the second fractions of the drilled samples give dates that are older, by as much as 265 years. For this reason they discard the second fraction dates, except for the surface samples, for which mechanical separation of limestone particles was feasible.

As a control to see how accurate their method is, HJ also date a mortar sample taken from the nearby Wanton-Lyman-Hazard (WLH) House, known to have been built between 1676 and 1698. The date they obtain is consistent with its known date of construction, and so they conclude that their date on the Tower must be accurate as well. Although testing this house was a sensible and useful control, there are three problems with its interpretation.

The first problem with the WLH House control is that due to the peculiarities of the dendrocalibration curve over the past few centuries, it tells us almost nothing about the rate of initial absorption of carbon dioxide into mortar. Because of coal burning during the Industrial Revolution, tree rings known to date from approximately 1665 to 1945 contain very similar amounts of carbon-14 today, and hence dates in this period are very difficult to differentiate. Thus, although the WLH House gives a point estimate date that can be dendrocalibrated to 1689 AD, the same raw C-14 content also dendrocalibrates, using HJ's Figure 2, to approximately 1730, 1810, 1920, or even 1945! On p. 38 they admit that there is no date between 1675 and the historical limit (1950) that can be rejected at even the 68% confidence level (1 standard error). In other words, the WLH House contains mortar whose carbonate could have set at any time between its known date of construction and the mid-20th century. If the method has a bias due to slow absorption, this control therefore tells us nothing about it.

The second problem with this control is that the sample obtained from it was apparently from its interior (in the basement), where it would have been protected from the elements, and thus not prone to carbonate substitution from rainwater. The Tower, on the other hand, would have been open to the elements for centuries before its colonial conversion to a windmill, if indeed it is medieval, and in any event has been again exposed to the weather in recent times, since at least 1837, according to Hertz.

The third problem with the WLH House as a control is that it only tells us that the mortar method correctly dates late 17th century structures. In order to demonstrate that their test results rule out a pre-Columbian Norse date for the Newport Tower, they should have also provided one or more controls known to have approximately the alleged pre-Columbian date of the Tower, and which were exposed to the same sort of weather the Tower would have received.

HJ do provide mortar dates on two medieval churches from Eckerö and Hammarland on the Åland Islands in Finland, and claim on p. 36 that this experience helps interpret the dates on the Newport Tower. However, on p. 39, they admit that there is serious uncertainty about the dates of construction of these churches, and that they are in fact using their method to establish these ages. They therefore are not true controls. And even if their true ages were known, the samples were taken from the interior, protected portions of these churches, which apparently have been continuously roofed since their construction. (See HJ, Figures 3 and 4). These tests therefore tell us nothing about the rate of substitution from rainwater that may have occurred in the case of the Newport Tower.

When we scrutinize HJ's Table 1, we find some further problems with their dates. Pure calcium carbonate contains about 60% carbonate by weight. Several of their samples contain less than 10% carbonate, and on this criterion HJ reject them. However, they do not explain what is present, if not calcium carbonate. If the difference is primarily unreacted calcium hydroxide, there is a serious slow reaction problem that potentially affects all the dates. On the other hand, if the other material is primarily inert silica sand, there is no particular indication that the remaining calcium carbonate is in any way contaminated, and these samples should be no worse than any of the others. Since Accelerator Mass Spectrometry (AMS) methods are being used to measure the amount of C-14, valid results can often be obtained with even very small amounts of carbon. Small sample size might result in a large standard error, and therefore could be a valid criterion for forgoing the expense of a test, but is not per se a valid criterion for rejecting test results once they have been performed, so long as the standard error is not unusually large as a result. Thus, two samples contained 2.0% or less carbonate and were legitimately not even tested. However, sample 8 from the fireplace was tested despite containing only 5.8% carbonate, but then was inappropriately excluded from the final estimates of the age of the Tower, even though its standard error was only 70 years, less than on two of the five samples that were included in the final estimate (75 and 90).

It is very significant that the preferred first fraction of carbon extracted from the inappropriately excluded sample 8 gives a negative uncalibrated C-14 date of -110 BP, or in other words, 2060 AD! (BP = "Before Present", i.e. before 1950, the approximate date when radiocarbon dating was developed.) Because atmospheric testing of nuclear weapons since 1945 has made recent decades appear to be far in the future before calibration, this does not literally mean that the mortar tested in 1993 had atmospheric carbon from the 21st century. However, it does indicate, after calibration, that the carbon was from some date after 1945, long after the Tower is known to have been built. HJ make no comment on this impossible date for the construction, but instead merely drop it from consideration on the inappropriate criterion of the low carbonate concentration per se. In fact, sample 8 appears to exhibit a more severe case of substitution bias than I would have imagined possible, despite Hertz's assurances (1997, p. 93) that rainwater contamination "could be excluded."

A second problem in their dates involves Sample 2 from Pillar 7. This sample was tested twice, in two preparations of the same mortar sample. The preferred first fraction gave an age of 365 BP ± 55 the first time, and 170 BP ± 75 the second time. The difference is 195 years ± 93. We may therefore reject, at the 95% confidence level, the hypothesis that these are dates on the same sample, even though we know that they are. The first sample, 2.a.1, is based on the first 20% of the carbon dioxide captured from the sample when treated by phosphoric acid, whereas the second sample, 2.b.1, is based on the first 50%. However, the later carbon dioxide consistently gives older dates for the drilled samples, due to slower dissolution of particles of unburnt limestone that remained in the lime, whence HJ's preference for the first fraction, so that if anything this difference should make the first sample appear younger than the second. HJ make no comment on this inconsistency, but instead treat 2.a.1 and 2.b.1 as if they were two valid, independent observations. On p. 38, they add an ad hoc 60% to the reported standard errors in order to make the dates they average together consistent with a single date of construction.

A third anomaly in their data comes from the carbonate concentration they report for the WLH House sample. Although pure calcium carbonate contains only 60% carbonate, they report that this sample contained 70% carbonate! If this is true, there is something wrong with the chemical model being used that ought to be investigated. Perhaps they meant to say 70% of the theoretical maximum amount of carbonate, but that is not what they did say.

One additional factor that may affect the results, is the possibility that the lime mortar reaction may not fractionate C-14 at the same rate that photosynthesis does. Isotopic fractionation occurs because C-14 atoms are heavier than ordinary C-12 atoms, and therefore carbon dioxide molecules containing C-14 move about more slowly at any given temperature. This means that calcium hydroxide will have a higher probability of reacting with molecules containing C-12 than with those containing C-14, even if these were present in the same proportions. Photosynthesis, which the dendrocalibration curve is based on, will similarly fractionate C-14 differently than C-12, but possibly at a different rate than the mortar reaction. Furthermore, molecules containing C-14 will diffuse through the pores in the mortar more slowly, and hence will reach its interior to react in reduced proportions.

In the end HJ apparently use five dates - two dates on the questionable surface sample from the flue, the contradictory dates from the two preparations of sample 2 from pillar 7 (first fraction only), and the first fraction of sample 12, from pillar 6 - to date the Tower. These average to 222 BP ± 30, but they adjust this standard error upward by the above-mentioned 60% to obtain 48 years instead. The point estimate dendrocalibrates to 1665 AD, but also to approximately 1790, as well as 1940, by their Figure 2. The first branch of a 68% confidence interval (1 standard error) dendrocalibrates to 1651 - 1679 AD, while the first branch of a 95% confidence interval (2 standard errors) dendrocalibrates to 1635 AD - 1698 AD. Only the periods 1698 - 1720 and 1810 - 1920 lie outside the later branches of the 95% confidence interval.

Although HJ conclude from their results that the Newport Tower could not have been built before 1635, I regard this as inconclusive evidence against an earlier date for construction, for several reasons:

1. Two of the dates used were from a surface sample that may have represented a colonial or even later repair to an earlier structure.

2. The other samples used may have been biased by slow reaction and/or substitution from rainwater. The inappropriately excluded post-1945 date on one of the samples tested demonstrates that rainwater substitution is an important factor.

3. The Wanton-Lyman-Hazard House does nothing to demonstrate that the slow reaction and substitution biases are not a problem, because of the flatness of the dendrocalibration curve since 1665 AD on the one hand, and the fact that its sample was not exposed to the weather on the other hand.

4. The two Finnish churches do little to verify the Tower date, since the true dates of these churches are unknown, and since the samples were taken from the interior of the churches, where they were not exposed to the weather and potential carbonate substitution. Indeed, the fact that a few of the samples taken from them were still alkaline indicates that slow reaction may be a serious problem of mortar dating in general.

5. There are several inconsistencies in the results and unanswered questions that remain to be addressed.

To be sure, none of these considerations proves that the Newport Tower is any older than 1635. I am merely returning a provisional "Scotch verdict" of "not proven colonial."

References

Carlson, Suzanne. "Tilting at Windmills: Newport Tower," NEARA Journal vol. 30, no. 3 & 4, Winter/Spring 1996. Published by New England Antiquities Research Association, c/o NEARA Publications, 94 Cross Point Rd., Edgecomb, ME 04556, (207) 882-8155.

Heinemeier, Jan, and Högne Jungner. "C-14 Dating of Lime Mortar," Arkaeologiske Udgravninger i Danmark 1994 (Archaeological Excavations in Denmark 1994), Copenhagen 1994, pp. 35-40. Cf. also illustrations (in immediately preceding Danish version), pp. 28, 31, 32.

Hertz, Johannes. "Round Church or Windmill? New Light on the Newport Tower," Newport History vol. 68, part 2, 1997. Published by Newport Historical Society, 82 Touro St., Newport, RI 02840, (401) 846-0813. Originally published in Danish in the Annual Report of the Danish National Museum, Copenhagen, 1995.

Redwood Library and Athenaeum. "The Old Stone Mill," website online at http://www.redwood1747.org/tower/millmenu.htm, undated.


Postscript

Since the above was written, the following very pertinent papers and websites have come to my attention:

Berry, Mark. "Chesterton Mill, Warwickshire," website at http://www.windmillworld.com/uk/chesterton.htm

Carlson, Suzanne. "Loose Threads in a Tapestry of Stone: The Architecture of the Newport Tower," NEARA Journal, vol. 35, no. 1, Summer 2001, pp. 11-36. Online at http://www.neara.org/CARLSON/newporttower.htm

de Bethune, Andre J., "On the Carbon-14 Analyses of Mortar from the Newport Tower: Theoretical Considerations," Newport History vol. 69, part 1, 1998, pp. 19-25.

White, Robert. "Newport Stone Tower," website at http://www.easternct.edu/personal/faculty/whiter/Stones/Newport-Stone-Tower.html with excellent photos of Tower and similar Lavatorium at St. Baaf's Abby, Ghent.

Www.360portugal.com. "Tomar -- Templar's Castle", website with QuickTime 360-degree panoramas of the Newport-like round Charola chapel. Check out Exterior of Charola, Interior of Charola, and View of Charola through doorway of Igreja Manuelina. Drag your cursor left and right, up and down, to move viewpoint. Requires free download of QuickTime.


Written and maintained by J. Huston McCulloch
Send comments to: mcculloch.2@osu.edu
Last revised 8/29/01


Up to Archaeological Outliers page


Technical note -- I find that this and other webpages often do not print out completely using Netscape 4.61 even though the complete text is visible onscreen, and even though they do print out correctly using MS Internet Explorer 5. This is apparently either a bug in Netscape, or a bug in Windows' compatability with Netscape.