I always find cases where there are multiple versions of an image a fascinating challenge. Stepping back and not necessarily accepting the conventional wisdom of what came first, or which is better than what, but looking at it through a broader historical lens demands we go back to first principles. Why should the well-known version in a museum be the prototype? Are we dealing with a smaller scale preliminary study, or a studio record? What about copies, ancient and modern? Disentangling these relationships demands a high level of critical analysis, one where there is often intriguing art history to come from the study, ranging through the intricacies of artists’ workshops to the resurfacing of styles in later Schools.

Part of the critical toolkit I’ve been building for myself recently has been how to address such challenges. I’m not going to discuss this too much here (I’ll leave that for another time) but will mention a couple of techniques I use. One is to start mapping out the inheritance tree, by which I mean the order in which each artwork was produced and what might have been the immediate parent to the following child. There can be all sorts of clues here, like compositional elements that define ‘families’ of related works, or where details become misunderstood (which I call ‘transcription errors’, like in genetic inheritance). I’ve also been recently using some AI-based similarity tools, code I wrote to help with retrieving similar artworks, but which also proves useful for identifying closely related images based on their visual appearance.

These ways of building models of artwork relationships help form what we might call a ‘coherent’ understanding of connections. That is, we have a map of linked artworks based essentially on their mutual similarity. Much of the art historical process of constructing the oeuvre of an artist is done like this – begin with a ‘keystone’ artwork, and then add others on the basis of closely shared stylistic features. Importantly, the keystone artwork should be one with a high level of believability, such as signed and dated works. I call these works ‘foundational’ because they connect directly to a time, a place, and an artist. Establishing the veracity of these foundational artworks is hugely important because they are frequently the seed-crystal for an entire oeuvre, essentially defining what we understand to be an artist’s style.

My colleague at Vasarik, Thereza Wells, is (among her many talents) a notable Leonardo scholar. Some years ago, she introduced me to a particularly intriguing example of these art networks. Working on the many versions of the Madonna of the Yarnwinder, it became apparent in credible versions by Leonardo that underdrawn features that were subsequently painted out went on to spawn a wholly distinct lineage that was picked up and repeated by other artists. It seems that Leonardo’s fame, his propensity to change his mind, and the length of time he took to complete commissions, meant that his pupils, and possibly visitors, were copying incomplete paintings in his studio and producing derivative versions that were, in turn, further reproduced and copied.

Clearly, dealing with any painting purportedly by Leonardo brings extra levels to the challenge. When the subject is the Mona Lisa, then we are definitely in territory where angels fear to tread. I’m going to take a deep breath then and talk a little bit now about a version of the Mona Lisa which has had some claim to being not only one by Leonardo, but also earlier than its world-famous sibling in the Louvre. Known as the Isleworth Mona Lisa, it has a long provenance (back to the 18^th century in fact) and some notable supporters. Held in private hands, it has also been subject to a significant amount of research over the years, including a range of scientific analysis and even some early AI comparisons. If we wanted to summarise its current status, we might though call it ‘disputed’, attached to the Leonardo canon on a stylistic rather than foundational basis, but then backed up by all sorts of claims justifying the materials found (which to my eye are significantly wanting).

The potential presence of the Isleworth Mona Lisa in Leonardo’s canon could significantly upset the status quo, so an unambiguous resolution would be good. Let’s see if we can do this.

In my last blog we looked at how we can take information on patterns of material use and combine them to give better estimates of when an artwork was produced. Fortunately, data from the scientific analyses conducted have been made available online, so we know that the painting is executed on canvas and has received a radiocarbon measurement of 300 years (± 27) ‘before present’ (which happens to be defined as 1950). Materials analysis includes data on the paint composition and the ground structure, while use of ‘brushstroke analysis’ claims it matches that of Leonardo.

From a preliminary examination of this evidence, we might immediately comment that the use of canvas rather than panel is atypical for Leonardo (and others at that time/place). In these cases I usually look at whether it might be a transfer (that is, moved from an original panel support to a new canvas one, a fate suffered by a number of Leonardo’s works); as far as I can tell from the available evidence this is not the case, so, consequently, anything we determine from the canvas and ground layers can reasonably be assumed to be true, giving us confidence in the next points. Taking the radiocarbon measurement (which was carried out by the prestigious ETH laboratory in Zurich), and correcting it using the latest calibration data gives a likely date range for the canvas production from 1498-1655:

A quick calculation shows that about 10% of the overall likelihood falls within Leonardo’s productive life. That is, the canvas could have been available at the time of Leonardo, but it is much more likely to be later. (For information, the ‘95.4% probability’ simply means that 95.4% of the overall likelihood of a particular date falls within that 1498-1655 range.)

Let’s now take another piece of evidence, though this time we’ll need a bit more explanation. This is the nature of the preparatory ground on the canvas. Various paint fragments were taken and prepared by Maurizio Seracini (who is a noted, if sometimes controversial, specialist on Leonardo’s materials and techniques) as cross-sections to show layer structure. Our interest here lies in the first two reported layers in the samples, the first of which has a red colour, the second of which is a grey. The specific pigments present are also interesting, as they confirm a suspicion: the red layer contains red earth pigments, lead white, red lead, and carbon black, while the grey layer primarily contains lead white, calcite, and carbon black. From the illustrations on the Isleworth Mona Lisa website, the carbon black in the grey layer also looks like it could be charcoal, judging by the size and shape.

I would argue that this is actually a type of ‘double’ ground, one based on a red below and a grey above, that is quite common within a restricted period of art material history. I first came across them many years ago when I was working on the analysis of a painting by Pieter Brueghel the Younger, signed and dated 1626. Since then, I have been collecting examples of them with the idea of mapping their historical and geographical use. Importantly, the composition of these grounds is all much like that described above for the Isleworth Mona Lisa.

What we need to know though is how common use of these grounds was in the past. You might be familiar with using histograms (‘bar charts’) as a way of expressing frequency of something in a particular range, but I’m going to use something slightly different, called a ‘kernel density estimate’ (KDE). I’m not going to explain here exactly what this is (there are plenty of sources on the internet for that anyway), but suffice it to say, it gives what we need – an estimate of a likelihood across a time period based on how many examples there are with particular dates. The result is like a histogram, but a continuous curve. For some data I have on red-grey grounds, this shows is that it was a popular way of preparing canvases in the mid-17^th century, then again in the earlier 18^th.

How does this help with our Mona Lisa question?

As I’ve described previously, we can combine evidence, in this case the calibrated radiocarbon curve and the occurrence pattern of red-grey grounds, to get a more refined estimate. It gives us this curve:

Again, what we are seeing is an estimate of how likely something is, in this case the combination of our canvas and its radiocarbon dating, along with the red-grey ground preparation. Very clearly, we have a combined date in the first half of the 17^th century. In fact, when we zoom in, most of the peak (90% of it in fact) covers the period 1617-1654, so a good century and more after Leonardo’s death. Consequently, we can be really quite confident that the Isleworth Mona Lisa is a later version, not an original Leonardo.

This isn’t the only Leonardo-related painting I’ve seen with this ground type (I recall a Leda and the Swan I examined some years ago), and I suspect that there may have been a workshop in the 17^th century producing these. We also know that this ground type was used in northern Europe, so the Isleworth Mona Lisa is probably not even Italian.