How well did a prince in exile know his country?

July 1711

The map marks the place where he lost his throne#

Near the town of Huși, on the bank of the river Prut, the engraver put a cross-shaped fort and one line of text: Hic Russi pugnarunt cum Turcis A° 1711, “here the Russians fought the Turks in the year 1711”.

Dimitrie Cantemir was then prince of Moldavia, a principality between the Carpathians and the river Dniester that paid tribute to the Ottoman sultan; today its land is shared by Romania, the Republic of Moldova and Ukraine. He had sided with Peter, the tsar of Russia. After their defeat at Stănilești he left with the Russian army and never came back. He drew the map in Russia in the years that followed, with the book in which he describes his country, Descriptio Moldaviae, requested by the Berlin Academy. Cantemir the man has a page of his own.

What is on the sheet

785 named places, from the capital to the last village#

A small circle is a village. A tower is a market town. A building with a cross is a monastery. Around Iași, the capital, the sheet names Galata, Cetățuia, Bârnova, Socola, Copou, in Latin letters spelled the Polish way, with cz for “ch” and sz for “sh”: Czetezuia, Husz, Suczava.

Every name was read from the scan of the Bibliothèque nationale de France, with the position of its sign measured in pixels.

What Europe had before

On a French academician’s map, Moldavia was almost empty#

Guillaume de l’Isle, geographer of the Academy of Sciences in Paris, printed a map of Hungary and its neighbours in 1703. His Moldavia has a few dozen names, Iași sits beside the Prut, and the word Deserts is written twice across the country.

De l’Isle’s sheet is drawn at about half Cantemir’s scale and covers Hungary and its neighbours, so a few dozen names is not a verdict on him. Cantemir put 785 places on the same region. The question is how accurately he placed them.

The test

The sheet laid on the ground without stretching it#

In blue are today’s waters: the rivers Siret, Prut, Dniester and Danube, and the Black Sea shore. The sheet has been turned, scaled and shifted until the 72 towns and monasteries that can be recognised by name and by the kind of sign came as close as possible to their real places. The sheet was neither stretched nor bent.

The sheet’s north is turned 8 degrees from true north. And the engraved rivers do not lie on the real ones: Cantemir’s Prut runs almost due south, the real Prut bears south-east.

How far off

Half the towns and monasteries fall within 21.5 km of their place#

Each red dot is the true place of a town or monastery. The line ties it to its engraved sign. 41 of the 72 are within 25 km; 15 are more than 50 km off.

The lines do not point at random. In the north most point west; in the south, south and south-east; in the mountains, east. An error with a direction belongs to the shape of the whole country, not to each place.

What he actually knew

Against its neighbours, a town is displaced by about 8 km#

The lines are short in the middle of the sheet and long at its edges. That says little about Cantemir: any map drawn too wide looks like this once its centre is set in place. Roman is 4.4 km out, Iași 7.9 km, and Hotin, a fortress on the Dniester in the far north, 70 km.

A better measure: each town is left out in turn and its place predicted from the towns that remain. Counted so, the displacement is 7.4 km in the middle of the country, 6.1 km in the mountains and 9.9 km in the districts between the Prut and the Dniester. In the north, where only three places are certain, it is 18.3 km; for Hotin itself, 12.5 km.

The shape of the country

His Moldavia is one and a half times as wide as it should be for its length#

Scroll and the sheet stretches like cloth until the recognised places come close to their red dots. It lengthens from north-west to south-east and hardly moves across. The stretch on screen corrects each region separately. Separately, a single stretch of the whole sheet, the same everywhere, gives the ratio ×1.58. It summarises the whole sheet and is not the same in every region: refitting with one region left out at a time gives between ×1.46 and ×1.92.

George Vâlsan, the Romanian geographer who found the map, wrote in 1926 that “the shape of the country is exaggeratedly wide”. Measured: for equal distances on the ground, the sheet shows 1.58 times as much across Moldavia as along it (between 1.42 and 1.85, depending on which towns are used).

Where the width comes from

Part of it belongs to the frame of degrees, not to the drawing#

Degrees are engraved on the sheet’s border. A degree of longitude is drawn 0.78 times the length of a degree of latitude at the top and 0.90 times at the bottom. At Moldavia’s latitude it should be 0.68. The degrees of longitude are drawn too long, then: a true grid of points put through the sheet’s frame comes out too wide by a ratio of about 1.22.

There is something in the drawing too. Read by the degrees on the border, the 72 places are still spread too wide, by a ratio of 1.30 (between 1.19 and 1.45): the north of the country is pushed east. These ratios come from separate measurements; their product need not equal the overall ratio. The direction along which the country is shortened lies 150°–164° from north. That is close to the sheet’s own vertical (164°), and also to the course of the Prut (162°) and the Siret (153°); the measurement cannot tell these apart.

The test that spoils the story

A French map of 1703 is as wide on the same towns#

It is easy to believe the width is Cantemir’s own: a country remembered wider than it was by a man who had left it. That can be tested. On de l’Isle’s map 23 towns in and around Moldavia can be recognised that are also on Cantemir’s sheet. Fitted the same way, with one even stretch and no town counted for less, the French map draws Moldavia 1.32 times as wide as it should be for its length (between 1.10 and 1.82), and Cantemir’s sheet 1.34 times (between 1.13 and 1.62). The idea that the width is Cantemir’s alone does not survive: a Moldavia drawn too wide was on paper before him.

Three limits. First, the test does not show that he copied the French map, or that the two share a cause; only that the same widening is on the earlier sheet. Second, on Cantemir’s own sheet these 23 towns give a lower ratio than the rest: the other 49 certain places give ×1.58, these 23 give ×1.34, and a random set of 23 of the 72 would give as low a ratio in only 6% of draws. The test is therefore gentler than the whole sheet. Third, a ratio taken from 23 towns is pushed up by scatter alone: in a simulation with independent random scatter of the size seen here, towns drawn with no widening at all would give a median of ×1.08 on Cantemir’s sheet and ×1.13 on the French one. The direction of the stretch is only loosely given by 23 towns: about 168° on Cantemir’s sheet (between 155° and 2°, through north) and 155° on de l’Isle’s (between 120° and 4°), against 157° for the whole sheet.

Read by each sheet’s own printed degrees, the same towns give ×1.10 on Cantemir’s sheet (between 1.02 and 1.34) and ×1.37 on de l’Isle’s (between 1.14 and 1.88). The ranges overlap; the estimate for Cantemir is the lower. On de l’Isle’s sheet the degrees are drawn close to the ratio expected at Moldavia’s latitude (0.66 against 0.68), so the widening is in the towns. On Cantemir’s it is largely in the frame.

What he brought

On the towns both maps show, Cantemir’s sit twice as near in kilometres, and about as near on paper#

The same 23 towns, the same plain fit with one even stretch, each town left out in turn and predicted from the others. The median error is 16 km on Cantemir’s sheet and 34 km on de l’Isle’s. Fitted on all 23 at once, it is 13 km against 31 km. Cantemir’s sign is nearer the true place for 19 of the 23 towns. The two medians differ by 18 km. Resampling the towns’ errors gives a 95% range of 7 to 31 km; refitting on each resample and predicting the towns left out widens it to 2 to 38 km, and resampling four regions as blocks gives 7 to 26 km. Read by each sheet’s own degrees, the conclusion is the same.

Scale changes what this means. The sheets are drawn at different scales: about 1:1,300,000 for Cantemir (BnF record; 8.4 cm to a degree of latitude on the sheet) and about 1:2,550,000 for de l’Isle, whose map covers Hungary and its neighbours. In millimetres on the printed sheets the median errors are 12.2 mm and 13.3 mm; the difference is 1.1 mm, with a 95% range from -7.2 to 6.4 mm, and Cantemir’s sign is the nearer on paper for 11 of the 23 towns. So the gain in kilometres goes with the larger scale. It shows that his towns sit nearer their true places in kilometres, not that his geographical knowledge was twice as exact; the sheets also differ in what they were made to show and in the sources open to each. De l’Isle’s scale is taken from the bar on his sheet and is only roughly known.

How far this can be trusted. Dropping the three towns where the gap is widest and fitting again on the remaining 20 gives medians of 16 km against 23 km: the gap narrows, it does not close. Holding out one region at a time and predicting it from the rest (the regions overlap), Cantemir’s towns are nearer in 4 of 5 regions with a plain turn and scale, and in 4 of 5 with one even stretch. The south, towards the Danube, depends on the fit: level with one even stretch (33 km against 33 km), nearer for Cantemir with a plain turn and scale (34 km against 47 km). And the pairing of the 23 towns rests on one reading of six pieces of the French sheet. The result is clearest for the middle and the north of the country.

Three things are therefore kept apart. Measured: the sheet is ×1.58 too wide, as drawn. What belongs to the sheet: fitted separately, its frame of degrees alone gives ×1.22 and the 72 places read through those degrees give ×1.30; these are not an exact split of ×1.58. On the 23 shared towns the estimate by printed degrees is higher for de l’Isle’s sheet (×1.37) than for Cantemir’s (×1.10), with overlapping ranges. Distinctive of Cantemir: the 269 places set around the towns, at twice the scale of the French sheet. His towns sit nearer in kilometres than de l’Isle’s, but about as near on paper; the width is not his alone.

The degrees on the border

The latitudes are good. The longitudes are those of his time#

The dotted lines are the true parallels. Cantemir’s latitudes are, at the median, 0.29 of a degree too far south. Once that shift is removed, an error of about 20 km remains, varying from place to place.

The longitudes are counted from Ferro in the Canary Islands, the usual zero for longitude at the time. They put Moldavia 4.6 degrees farther east than it is; Vâlsan had written “more than four degrees”. Longitude was the weak side of maps made before precise clocks.

The villages

A village is harder to recognise than a town#

Many village names repeat: Moldavia has several villages called Popești and several called Bălțați. A name on the map was accepted in two cases only: when a single place in all of historical Moldavia bears that name today (63 villages), or when a place of the same name lies within 12 km of the point where the places already identified around it say the sign should be (134 villages).

The second rule was put to a test: with the expected point of every sign moved 80 km in a random direction it still finds 9 matches instead of 134, about 7% as many. Some of these identifications may therefore be mere coincidences of name. None of them enters the numbers on the shape of the country. And of the 86 villages with a unique name, 43 fall within 15 km of the spot predicted from the towns alone; 28 are more than 50 km away. Of the 86, 23 were set aside as probable namesakes.

Where the map is weakest

The mouths of the Danube and of the Dniester: displaced even against their neighbours#

Chilia, a port on the northern arm of the Danube delta, is 99 km from its place on the unstretched sheet and still 46 km out when measured against its neighbours. Cetatea Albă, the fortress at the mouth of the Dniester: 75 km and 32 km. Among the certain places inside Moldavia, no other is displaced by more than 30 km against its neighbours.

The south between the Prut and the Dniester had not been Moldavian for a long time: the forts were Turkish and the plain, the Bugeac, was held by Tatars. The sheet has few villages here and much empty paper.

The engraver

The hand-drawn copy has the same shape#

Cantemir’s manuscript has not been found. There is, however, a hand-drawn copy, larger than the printed sheet, made by the geographer Jean-Baptiste d’Anville, who wrote in 1759 that Cantemir’s son Antioh, while Russian ambassador in Paris (1738–1744), had allowed him “to copy in full” the map his father had drawn.

On 100 towns and monasteries read separately on the two sheets, a plain enlargement lays one over the other: for half the places the two signs come within 1.2 km of each other, and the ratio of length to width differs by 1.8%. The too-wide shape is in both.

What that settles is limited. If d’Anville copied the manuscript, the width was already there before the engraver. But he is the only witness to what he copied, and Alexandrescu (2011) keeps the view that he worked from the manuscript while calling the proof thin; if he copied the print, the agreement is expected and says nothing about the engraver. The copy shows that the two sheets share a model, not which model.

A second control concerns the paper and the scan, not the engraving: the BnF holds two impressions of the same plate. Laid directly over each other by image matching (75 points, spread over about two thirds of the sheet’s width and three quarters of its height; no model read anything), they differ by 0.6 km at the median after a turn and a scale, and the ratio between their axes is 1.00. Matches were accepted within 1.8 km of the best turn and scale; allowing up to 14 km adds only five more points, which gives no sign of a larger deformation among the points added. A deformation shared by both impressions would not show. Paper or scan differences between these two impressions cannot explain errors of 20–30 km, or a ratio of ×1.5.

Where the picture can mislead

What this measurement cannot say#

A village of today does not have to stand where it stood in 1716: many moved, and the coordinate used is the centre of the present settlement. Reading the names has an error too: in a second, independent reading of part of the sheet, 82% of the names came out the same letter for letter, and for 3.1% the sign was placed more than 4 km from its position in the first reading.

The red line of any one village therefore says “about this far, about that way”. The shape of the whole country is computed on 72 towns and monasteries recognised by name and by the kind of sign.