The trends resulting from the data exploration are discussed in relation to their impact on degradation risk and in relation to material/market knowledge acquired at several archives and in discussion with paint manufacturers. The use of white pigments in the 20th century will first be discussed based on general trends and specific artists. The results will be concluded with the identification of characteristic trace elements of interest.
Introduction of TiO2 by Dutch artists
It was commonly assumed, based on the occurrence of TiO2 in artworks from the United States [1, 5, 7], that TiO2 was used by artists almost as soon as it was taken into production. However, the present dataset suggests that a different trend was followed by Dutch artists. Figure 2 shows the number of paintings containing different peak areas of titanium, zinc and lead over the decades.
As discussed in the previous section, a normalized Ti peak area of 0.1, with a main peak of Zn relates to a mixture, a layered structure or the use of multiple different paints. Figure 2 shows the convincing introduction of titanium white in the 1970s, while the pigment was already in production since the twenties. Earlier, starting in the fifties, mixtures or layers of titanium and zinc white can be found. The timing is in line with findings from the Royal Talens archive in Apeldoorn , where it was found that artists’ oil paints containing titanium white were only marketed in 1952 for the Dutch market, while the first test batch was already produced in the thirties . For the French market, Royal Talens marketed titanium white oil paints already in 1950, possibly as competition for ‘Société Lefranc’ and ‘Société Bourgeois’, that sold titanium white paints since the twenties . While the relative absence of early titanium white in Dutch oil paints can be considered to be a good sign regarding degradation risk, it does not directly mean stable pigments were used from the seventies onward [1, 5]. An example is paint manufacturer Royal Talens, that only switched to rutile for oil paints in the nineties (personal communication with Bert Klein-Ovink, Royal Talens) and still uses anatase for gouache today (confirmed by XRD analysis). Furthermore, Laver notes that “Due to their greater whiteness early products based on anatase were retained for many years after rutile pigments were introduced.” . However, it is important to realize that treated anatase pigments were also brought to the market and may have been used in late 20th century paints again reducing the degradation risk . For instance, the pigment ‘Tiofine A20’ is often reported as titanium white pigment in Royal Talens recipes. The reference collection of the Cultural heritage agency of the Netherlands has a pigment labelled ‘Tiofine’, and while this may not be the same pigment, it does come from the same manufacturer. This pigment, shown to be anatase by XRD analysis, showed in STEM-EDX analysis to have a partial Al/Si coating which would reduce the pigments photocatalytic activity.
We can only speculate about the reason for the slow rise of titanium white in Dutch oil paintings. Warnings were published about the yellowing of titanium white in oil paints [1, 39,40,41] possibly creating some reluctance for the use by artists. In the meantime, the yellowing problems were partly resolved by mixing titanium white with zinc white. Manufacturers such as Webber in the United states used zinc white additions early on and marketed their product ‘Permalba’ as “Brilliant…Permanent…and easy to handle”.
Titanium white is not only absent in the white paints but also in the colored paints. Due to the high refractive index of titanium white, zinc white was preferred at Royal Talens to adjust the hue of a color paint tube (personal communication Bert Klein-Ovink). This is confirmed in the present dataset. Of all the (averaged) measurements of colored paints, 25 have lead as the main peak, 50 have zinc as the main peak and only 10 have titanium as the main peak. The latter are likely related to non-oil paintings. This suggests that, while white oil paints may be at risk for photocatalytic degradation, this risk is minor for TiO2-induced color fading.
Furthermore, most of the paintings with Ti-K as the main peak are made with non-oil binding media such as acrylics, tempera and gouache. This is in line with XRD analysis of reference gouache, acrylic and tempera paint outs containing for the most part titanium white and fillers. It is understandable since both zinc white and lead white are not stable in, e.g. acrylic emulsion paints . Furthermore, additional confirmation is found in archival recipes at Royal Talens of pure titanium white water-borne paints without the addition of any zinc white (van den Berg, unpublished research).
Figure 2 does not only shows the slow rise of titanium white and the ubiquitous use of zinc white, it also indicates the slow decrease of lead white. Until the eighties, it is found as the main element in some paints. This is in stark contrast with reports stating that lead white nearly disappears after World War 2 . Additionally, analyzing a subset of painting grounds (Fig. 3) further confirms the use of lead white for grounds far into the twentieth century. In the Netherlands, lead white was banned from interior painting as early as 1939 [42, 43]. However, protective regulations for people working with lead white were only implemented in 1988, and a full ban of lead white for outdoor paintings and artworks came as late as 2009 [42, 43]. It is clear that due to its unmistaken popularity, despite the known health hazards, lead white was hard to ban from Dutch 20th century oil paintings. Low counts of lead may originate from an added drier rather than from the lead white pigment.
Zinc white and titanium white are a common mixture, and zinc white with lead white is also found possibly as leaded zinc oxide . On the other hand, the combination of lead white and titanium white is only found in grounds (Fig. 3), which may originate from multiple ground layers or mixtures. The subset of impastos does not show a single case of a titanium white/lead white mixture. This is in line with investigated recipes in the Royal Talens and Webber archives , where only one recipe called ‘mixed white’ contains all three white pigments. In this, 1956, Talens recipe titanium white is present as a minor component of the oil paint (ten times less by weight than lead white and zinc white). This finding indicates that if lead and titanium peaks are both identified in an XRF spectrum these elements are likely originating from different layers, with titanium white most likely present in the top layer.
The developments in white paint composition in the 20th century
Figure 4 shows the PCA scoreplot of the average of the white areas from all paintings, normalized to the maximum peak area. The figure can be interpreted as a reference roadmap for researchers and conservators doing XRF on modern paintings. Figure 5 presents the associated loadings for principal component 1 and 2, which describe 85% of the variation in the post-processed dataset. The PCA scoreplot shows interesting chronological features discussed below. Before doing so, we would like to underline a number of important caveats. Firstly, clusters based on paint type have been manually defined after coloring the plot for normalized Ti peak areas, normalized Zn peak areas, normalized Pb peak areas and normalized ratio: Ti/(Ti + Zn), taking into account the signal distortion between Zn and Ti peak areas (Additional file 1: Figures S20, S21). Secondly, this analysis does not provide insight into painting stratigraphy, hence combination groups are labelled: mixture or layers.
Figure 4 presents the same trend as previously depicted in Fig. 2, namely of the gradual development from lead white to titanium white via the extensive use of zinc white. It shows interesting cross-connections between artist, time and paint compositions. One important finding is the strong representation of CoBrA artist (Appel, Constant, Corneille, Jorn and others) in the ‘Zn paints’ section. Zn paints are susceptible to problematic soap formation  which may become an even larger problem than already expected.
PCA of the complete dataset shows interesting general chronological trends. However, a closer look at data subsets can be equally interesting. Investigation of the subset of artworks with titanium as the main peak shows an anti-correlation of barium and calcium (Additional file 1: Figure S22). This indicates that in most cases one type of filler or composite pigment base was used. Furthermore, investigating the subset of artworks with zinc as the main peak yield an anti-correlation of titanium and lead (Additional file 1: Figure S22). This observation supports once more that zinc white would either be combined with lead white or with titanium white but not with both.
Looking into the artist studio…
Another aim of this study was to investigate how the general trends in pigment use relate to individual artists. To this end, we selected three painters/artistic movements represented by a sufficient number of paintings in the dataset. This resulted in three examples: 11 paintings by J.C.J. van der Heyden, 7 paintings by K. van Bohemen and 30 paintings by CoBrA artists Appel, Constant, Corneille, Rooskens and Nieuwenhuis.
van der Heyden
The paints used by J.C.J. van der Heyden can be clustered in three different categories, Fig. 6a. J.C.J. van der Heyden appears to undergo an identical development in his use of white pigments as the general development in the Netherlands discussed previously. In the 1950s he used lead white and zinc white (either mixed or superimposed), in the 1960s he moves on to use a zinc and barium containing paint (either a mixture of zinc white with barium sulfate or lithopone). Finally, in the 1970s and 1980s, he uses titanium based paints, four of which are in acrylic binders. XRD analysis shows that ‘Hellende Horizon’, a painting in acrylic from the 1970s, contains rutile, the more stable form of titanium white, and calcium carbonate. This could suggest that the other acrylic paints from that time may have been produced with similar pigments and could be less vulnerable to degradation. The results from this case study coincide with the general trends observed in the full dataset and thus support this method of analysis and data processing.
In the case study of the K. van Bohemen paintings (by bivariate or PC analysis) two main types of paints can be identified: the zinc oxide based paints (post 1974) and the Zn/Ba based paints (up to 1975), Fig. 6b.
The Zn-based paints split into two groups indicating the mixtures of zinc white and titanium white (PCA scoreplot top left) and the mixture of zinc white and lead white (PCA scoreplot right). For this subset, the general trend of TiO2 appearing in the 1970s is confirmed (paintings from 1974 and 1979).
The Zn/Ba correlation, in the four black labeled paints (Fig. 6b), suggests the association of zinc and barium either as a mixture or as lithopone (co-precipitate of BaSO4 and ZnS). Until 1945, lithopone was a strong competitor for the still developing titanium white pigments in the United States . While the present data cannot differentiate between a mixture of zinc white and barium sulfate or the pigment lithopone, XRD of another sample from a different van Bohemen painting from 1966 indicates the presence of lithopone, placing the pigment on the artist’s pallet. Interestingly, investigating a subset, not necessarily requiring PCA, provides information about a paint composition (ZnO + BaSO4 or lithopone) not noticeable in the full dataset (Fig. 4). This highlights the possible benefit of doing further investigation of the data in the future.
CoBrA artists Appel, Constant, Corneille, Rooskens and Nieuwenhuis
As previously indicated, in this study it was found that CoBrA artists often used zinc white. To investigate this further, PCA was applied on a subset of 30 paintings by five CoBrA artists. The results indicated that the main difference between the paints was the zinc to lead ratio. While most paintings contain mainly zinc white, there is a selection of paintings containing zinc white and lead white. In this dataset, neither the date nor the artist coincides with the type of paint used therefore the visualization is omitted. This suggests random pigment choice in the CoBrA group. However, this did not compromise the detection of general trends in the full dataset.
Trace element indicators
The exploration of the data also resulted in the identification of two possibly indicative and characteristic trace elements: niobium and zirconium.
The determination of the crystal structure of titanium white in paintings is the first step towards a possible risk assessment. However, XRF cannot determine the difference between rutile and anatase and less available techniques such as Raman and XRD are required. Our dataset indicates an association between titanium and niobium, Fig. 7. With one exception, niobium is only present if titanium is also present. Niobium is a characteristic element associated with ore impurities and is impossible to remove via the sulfate process , the earliest production process of titanium white pigments.
STEM-EDX confirms that indeed niobium and titanium are associated,Footnote 6 Fig. 8. The chloride process, started to be developed in 1948 (pilot plant) and a full-scale production facility was opened in 1959 by E.I du Pont de Nemours and Co., Tennesee. In Europe a production facility producing chloride processed rutile opened in 1965 (British Titan Products—one of the suppliers for Winsor and Newton paints) [1, 5]. While both processes can theoretically produce both crystal structures, generally the chloride process is used to produce the more stable rutile (developed in the 1940s). Hence the absence of niobium and thus the use of the chloride process can be an indication of the presence of rutile and dates the pigment to post 1959 (or post 1965 in Europe). The presence of niobium is related to the sulfate process which is still in use today. Therefore this can indicate both types of titanium dioxide pigment and any production date. Comparing the crystal structure analyzed by XRD from micro-samples, with the scatter plot (Fig. 7) confirms this.
It should be noted that implementing the dating, colored as ‘pre’ or ‘post’ 1959 in Fig. 7, presents some challenges. Several paintings, painted before 1959, show the absence of niobium. The amount of niobium left in the pigments is dependent on the ilmenite ore. For instance Norwegian ilmenite (containing 0.007% Nb) results in pigments with 0.012% Nb, while Australian ilmenite (containing 0.10% Nb) results in pigments with 0.22% Nb . While the low niobium contents in titanium white derived from the Australian ilmenite should still be detectable by pXRF , some factors may inhibit the detection of niobium with XRF in sulfate processed titanium dioxide. This likely due to the paint composition, detection limit, set threshold (15 cnt/s), the measurement settings and subsequent signal-to-noise ratio. When the titanium counts are low, inherently the niobium counts will be low and possibly undetected. Similarly, when titanium is not the main peak but zinc or lead are strongly emitting the Nb signal may get lost in the noise (this is the case for the two red dots labelled 1956 and 1952). The latter was confirmed by investigating the reference material (Additional file 1: Figures S23, S24). In the scatter plot of the small dataset of all impasto’s the correlation is clearer and dating is consistent, which supports the idea that in some cases Nb is not detected, even though it is present. To confirm the absence or presence of niobium, and thus get an indication of crystal structure and dating, analytical improvement is required. In the case of the Bruker III-SD, this can be reached using the Cu/Ti/Al (green) filter with a voltage of 40 kV, a current of 15 μA and a measurement time of at least 120 s .
Similar to niobium, zirconium is almost never present without titanium. Zirconium can originate, among others, from the paint drier or from a pigment coating on the titanium white that decreases its photocatalytic activity . The simultaneous absence of zirconium and titanium suggest that if present, this indicates either an increased amount of drier (due to slow drying properties of titanium whites) or the detection of the inorganic coating. With the present dataset, using pXRF, the origin of the zirconium cannot be unambiguously determined.
Nevertheless, if zirconium can be related to the presence of a pigment coating, which results in reduced photocatalytical activity, this is of interest from a risk management perspective.
Further research should be performed, which notably entails optimizing pXRF settings as well as using a specifically designed set of reference material.