Improving the connectivity of MgB2 bulk superconductors by a novel liquid phase sintering process

This work investigates a new processing method developed to improve the connectivity of ex-situ MgB2 bulks at low sintering temperatures. Mg additions (1–10 wt.%) were mixed to pre-synthesised MgB2 to make composite powders that were sintered at 900 ∘C by the field assisted sintering technique. Addition of 10 wt.% Mg resulted in a substantial increase in density from 68% to 79% and a dramatic reduction in MgB4 from 11 to ∼0 wt.%. Pressure and dilatometry data recorded in-situ during the sintering process revealed that Mg additions led to different sintering mechanisms depending on the Mg fraction. For large Mg fractions (6 and 10 wt.%) a Mg liquid phase was formed and led to significant density improvements, and all pre-existing MgB4 was transformed into MgB2. A small amount of residual Mg remained in the bulks after the sintering process. Connectivity was improved with Mg additions, increasing four fold in the 10 wt.% Mg-MgB2 sample compared to unmodified MgB2. J c values at low field were also significantly improved by Mg additions, in particular the 6 and 10 wt.% Mg-MgB2 specimens showed J c (20 K, 0 T) values 4–5 times higher than for unmodified MgB2.


Introduction
MgB 2 is considered as a promising material in the field of applied superconductivity due to its relatively high critical temperature of 39 K (T c ) and the fact that MgB 2 can be manufactured in various forms (bulks, wires and tapes) by scalable and readily available powder metallurgy routes. Its high T c value makes it more suitable for liquid helium-free applications than conventional low temperature superconductors * Author to whom any correspondence should be addressed.
Original Content from this work may be used under the terms of the Creative Commons Attribution 4.0 licence. Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.
(LTSs) such as NbTi and Nb 3 Sn. In addition, MgB 2 does not contain any rare earth or toxic elements and is also much easier and cheaper to process than high temperature superconductors such as the cuprates [1]. This combination of properties makes MgB 2 an attractive candidate to replace 'traditional' LTS materials such as NbTi in applications including MRI machines and magnetic drug targeting devices [2,3].
MgB 2 bulks and wires are routinely produced by various processing techniques commonly grouped in two categories: in-situ and ex-situ methods. In the case of the in-situ route, Mg and B precursor powders react together to form the MgB 2 phase at moderate temperatures (600 • C-900 • C). This method has the advantage of producing sintered specimens showing high inter-grain connectivity. On the other hand, the large density difference between the precursor powders (2.10 g cm −3 for the powder mixture compared to 2.63 g cm −3 for MgB 2 ) leads to extremely porous specimens, which in turn reduces the total cross sectional area for supercurrents and also results in poor mechanical properties [4][5][6].
The ex-situ method, in which pre-synthesised MgB 2 powder is consolidated at high temperature (900 • C-1300 • C), usually with the assistance of pressure, can produce samples with significantly higher density. The field assisted sintering technique (FAST), also known as Spark Plasma Sintering or Direct Current Sintering is a popular technique that has been used successfully to produce dense MgB 2 specimens by different research groups [7][8][9][10][11]. This technique consists in driving a high current (usually at a low voltage) through an assembly composed of a graphite die in which the powder is enclosed between two graphite punches. The large current going through the punches/die and the green compact (if conductive) leads to rapid heating and densification of the powder through Joule heating. Several groups have used sintering temperatures in excess of 1100 • C to obtain dense and well-connected MgB 2 samples [5,8,[10][11][12]. However, using high sintering temperatures is costly and also promotes the formation of impurity phases such as MgB 4 that lowers the total superconducting fraction in MgB 2 bulks manufactured at high temperatures. This comes from the decomposition reaction 2MgB 2(s) ⇌ MgB 4(s) + Mg (g) that occurs around 900 • C [13,14]. Matthews et al [11] also showed that the fraction of MgB 4 in MgB 2 bulks sintered by FAST rapidly increased with higher sintering temperatures.
Therefore, the following challenge arises: how to improve the density and connectivity of ex-situ MgB 2 bulks without using high temperatures that promote the formation of impurities and increase the manufacturing cost. One possibility, explored by a few research groups is to use ultra-high pressure sintering techniques [15][16][17]. However, pressure of several GPa requires specialist industrial presses and is not desirable for large scale commercial production.
In this study, a novel approach that we developed to sinter pre-synthesised MgB 2 powder is reported.
Liquid phase sintering (LPS) has been used for decades to consolidate ceramic materials that are difficult to sinter by conventional methods [18][19][20]. In LPS, a liquid forms in the green body and enhances sintering by pulling the powder particles together through capillary force and by improving the atomic transport mechanisms responsible for densification.
The approach investigated here involves mixing a small fraction of elemental Mg with pre-synthesised MgB 2 in an attempt to enhance densification and improve connectivity in bulks sintered at low temperature. It is expected that Mg (which melts at 650 • C) can form a liquid phase during sintering and thus promote densification. Mg-MgB 2 composite powders were consolidated using the FAST with the aim of developing a LPS process to significantly enhance connectivity and reduce the impurity content in ex-situ MgB 2 bulks manufactured at low temperature.
Several research groups have investigated the effect of excess Mg in the case of MgB 2 bulks and wires produced by the in-situ process [21][22][23][24][25]. Zeng et al [21] and Ma et al [23] both produced MgB 2 bulk samples by using 10 wt.% of excess Mg and reported significant improvements in critical current density (approximately 2-3 fold increase at self field and 20 K) compared to their reference MgB 2 sample. Zeng et al [21] showed a reduction in MgO content with excess Mg and concluded that Mg facilitated the oxygen intake in the MgB 2 lattice, improving grain connectivity and thus enhancing J c . Ma et al [23] measured significantly higher density and connectivity values in their bulk sample containing 10 wt.% excess Mg. They suggested that the J c improvements resulted from higher density and grain connectivity. Zhang et al [24] and Guner et al [25] reported similar findings and concluded that excess Mg led to higher density and lower impurities content in the bulk improving inter-grain connectivity and J c values at low field.
Whereas the use of excess Mg is well documented in the case of the in-situ method, a process combining the ex-situ technique and Mg additions to promote LPS has not been investigated. In this work, a new method to consolidate ex-situ MgB 2 is presented and particular focus is given to the role of Mg in the sintering process and its effect on the microstructure and superconducting properties of ex-situ Mg-MgB 2 bulks.

Experimental details
Pre-synthesised MgB 2 (purity: 99%, Alfa Aesar) and elemental Mg (purity: 99.8%, Alfa Aesar) powders were mixed together manually for 5 min in a glovebox under Ar. Five compositions were investigated: the commercial MgB 2 powder and blends containing 1, 3, 6 and 10 wt.% Mg. The commercial and composite (Mg-MgB 2 ) powders were placed into a 20 mm diameter graphite die protected with graphite paper and two graphite punches were used to complete the assembly. Cold compaction of the powder was performed in the graphite die by using a uni-axial press that was operated to reach a pressure of 30 MPa for 1 min. The die assembly was then transferred into a Dr Fritsch DSP 507 FAST apparatus. Prior to sintering, the chamber was evacuated and then filled with Ar three times, to eliminate any residual oxygen in the chamber. The consolidation chamber was then continuously evacuated during the sintering process and a baseline pressure of approximately 0.4-0.6 mbar was obtained. The green bodies were heated from room temperature to 900 • C at a rate of 120 • C min −1 before being maintained at this temperature for 5 min. During this first heating step, the pressure was gradually increased to 50 MPa and then maintained during the 5 min dwell at 900 • C, before being progressively released during the final cooling step. The FAST apparatus was operated with a constant voltage and pulsed current ranging between 1.5-4 V and 0.4-1.2 kA.
The consolidated disks were typically 20 mm in diameter and 3-5 mm in thickness, and were characterised by several techniques, including x-ray diffraction (XRD), scanning electron microscopy (SEM), scanning transmission electron microscopy (STEM) and magnetometry. XRD patterns were acquired using a PANalytical Empyrean diffractometer with Cu kα radiation (λ = 0.154 nm) at 40 kV and 40 mA. The lattice parameter, crystallite size, strain and weight fraction of the different phases detected in the bulk specimens were estimated from the XRD data by using Rietveld refinement (PANalytical HighScore Plus software). The weight fraction was estimated by using the Hill and Howard method [26]. Instrumental broadening of peaks was corrected using a Si standard analysed under the same scan conditions. The SEM analysis was undertaken in a Zeiss Merlin field emission gun SEM system operating at 10 kV, equipped with an Oxford Instrument 150 mm 2 Xmax EDX detector. The STEM analysis was performed in a JEOL 3000F at an accelerating voltage of 300 kV, equipped with a high angle annular dark field (HAADF) detector and an Oxford Instrument EDX detector. STEM specimens were prepared by cutting 3 mm disks from thin slices obtained from the sintered pellets. The disks were then mechanically polished to approximately 100 µm before being dimpled and finally thinned to perforation with a Fischione 1010 Ion Mill. The bulk density of the consolidated pellets was measured in isopropanol using the Archimedes' method. Magnetisation measurements were acquired with a Quantum Design physical property measurement system vibrating sample magnetometer on cuboid samples with dimensions of approximately 2 × 2 × 3 mm. The critical current density (J c (B)) was estimated from the full magnetic hysteresis loops using Bean's model [27].
To estimate the effective cross-sectional area for macroscopic current transport in the bulk samples, four-point transport measurements and analysis of connectivity have been performed to obtain the normal state electrical resistivity as a function of temperature. Following the analysis of Collings et al [28], values of connectivity were extracted from the data using the procedure described in a previous article by Matthews et al [11]. The Materials constant K used in the connectivity analysis was estimated by fitting reference single crystal data from Eltsev et al [29]. To quantify the phase compositions, XRD patterns of all five specimens have been analysed by Rietveld refinement and the results are summarised in table 1. The fraction of MgO was identical in all four Mg-MgB 2 specimens and unmodified MgB 2 . In contrast, the amount of MgB 4 decreased drastically in samples containing Mg additions, from >10% by weight to a level that was undetectable in the 6 and 10 wt.% specimens. In fact, these two samples contained 0.4 and 4 wt.% of residual Mg respectively. The evolution of the composition with Mg additions is also shown graphically in figure 2. This can be explained by applying Le Chatelier's principle to the decomposition reaction 2MgB 2(s) ⇌ MgB 4(s) + Mg (g) .

XRD and density measurements
Le Chatelier's principle states that if a dynamic equilibrium is disturbed by a change in the reaction conditions, the position of the equilibrium moves to counteract this change. In the absence of Mg additions, the forward reaction is favoured and MgB 2 decomposes into MgB 4(s) and Mg (g) until the dynamic equilibrium for these particular conditions is reached. Indeed, table 1 shows that the sample consolidated from unmodified MgB 2 contained 11 wt.% MgB 4 . However, in the samples containing Mg additions, presumably some Mg evaporated and created a positive pressure of Mg (g) thus moving the dynamic equilibrium of the decomposition reaction to the left.
XRD characterisation of the as-received MgB 2 powder showed that it contained 7 wt.% MgB 4 prior to sintering. This means that in the Mg-MgB 2 samples containing 3, 6 and 10 wt.% Mg, Mg additions not only suppressed MgB 2 decomposition but reacted with pre-existing MgB 4 to form 'fresh' MgB 2 during sintering. In other words, Mg additions converted pre-existing MgB 4 impurities into MgB 2 in-situ.
The pressure data recorded while sintering Mg-MgB 2 samples provided valuable insight on the reactions that occurred during sintering. Figure 3 shows the pressure in the sintering chamber as a function of temperature during the first heating step. Unmodified MgB 2 powder showed a broad peak in pressure at 600 • C-800 • C which corresponds to the MgB 2 decomposition reaction. This is in good agreement with the Mg-B phase diagram calculated by Liu et al [13] at 1 Torr showing that Mg starts evaporating from MgB 2 above 600 • C. In contrast, this broad peak was not observed in the samples containing Mg additions. Instead, a narrower peak was detected at 500 • C-600 • C which corresponds to Mg evaporation from the Mg additions. This is consistent with the works of Coleman and Egerton [30] and Moser et al [31] who showed that the Mg vapour pressure over solid Mg rose steeply above approximately 500 • C. These results suggest that metallic Mg evaporated at lower temperatures and formed a positive Mg (g) pressure within the green body that suppressed the evaporation of Mg from MgB 2 , thus eliminating its decomposition. This is in good agreement with the XRD results and confirms the theoretical explanation based on Le Chatelier's principle, which is also consistent with the Mg-B phase diagram [13]. Table 1 shows that the Mg-MgB 2 sample containing 1 wt.% Mg had almost the same density as unmodified MgB 2 . This suggests that most of the metallic Mg probably evaporated during sintering and did not significantly enhance densification. However, the Mg-MgB 2 samples with larger Mg additions showed a significant increase in density, reaching 79% in the 10 wt.% Mg-MgB 2 sample compared to 68% in unmodified MgB 2 . This indicates that for higher Mg concentrations, not all metallic Mg evaporated and some Mg formed a liquid phase that enhanced diffusion rates leading to faster densification. In order to confirm this theory, the piston travel data has been analysed for the Mg-MgB 2 samples and unmodified MgB 2 . Figure 4 shows the relative piston travel as a function of temperature during the first heating step of the sintering program. In unmodified MgB 2 , the curve has two distinct parts: a linear decrease between room temperature and 800 • C followed by a sudden drop above 800 • C. The first part corresponds to the compaction of the powder due to the pressure  which increased linearly from room temperature to reach 50 MPa at 900 • C. The second part corresponds to sintering of the powder with the onset of sintering at 800 • C. The curve of the 1 wt.% sample was almost identical to unmodified MgB 2 with the onset of sintering also located at 800 • C. This is in good agreement with the density measurements and confirms that all the Mg evaporated at lower temperatures and did not form a liquid phase. In contrast, the composite specimens with larger Mg additions showed very different sintering curves. The 3 wt.% sample had a very similar curve to unmodified MgB 2 up to 600 • C-650 • C, but then diverged significantly and showed a rapid decrease above 600 • C-650 • C. The onset of sintering was thus pushed to lower temperatures, around 600 • C-650 • C, which coincides with the melting temperature of Mg at 650 • C. This suggests that Mg formed a liquid phase and enhanced densification, as shown in table 1. The 6 and 10 wt.% specimens showed a travel curve similar to the 3 wt.% sample, only slightly shifted downward indicating better compaction due to the higher Mg content, and with the onset of sintering also located at lower temperatures compared to unmodified MgB 2 . This confirms that the sintering behaviour is modified with the addition of sufficient (≥3 wt.%) metallic Mg. Table 1 shows that all four Mg-MgB 2 samples and unmodified MgB 2 had very similar MgB 2 crystallite size which suggests that Mg additions did not affect the MgB 2 coarsening rate. All five samples also had similar inhomogeneous strain values and a and c-axis lattice parameters. However, the MgO crystallite size increased with higher Mg concentrations, reaching 44 nm in the 10 wt.% sample compared to 31 nm for unmodified MgB 2 . Since all the composite specimens and unmodified MgB 2 had the same MgO content, this suggests that the liquid Mg phase slightly increased MgO coarsening rate during the sintering process.     Figure 5 shows typical low magnification backscattered SEM images for unmodified MgB 2 and the 10 wt.% Mg-MgB 2 sample. Unmodified MgB 2 sintered at 900 • C did not show any sign of sintering and had a very porous microstructure. In contrast, the Mg-MgB 2 sample showed a much more connected microstructure and less porosity, indicating that densification was significantly enhanced compared to unmodified MgB 2 . This is in good agreement with the large increase in density observed for the Mg-MgB 2 sample. No MgB 4 could be seen by SEM, which is consistent with the XRD results (table 1). However, the XRD analysis indicated the presence of residual Mg (∼4 wt.%), but Mg could not be detected by SEM and EDX mapping. This could come from a lack of backscattered electron contrast between the Mg and MgB 2 phase or the fact that Mg regions were too small to be resolved by EDX.

SEM and STEM analysis
Ceramics based on the Si 3 N 4 or SiC structures are sintered using LPS by adding metal oxides and/or nitrides which react with the surface silica present in the starting powder to form a liquid phase that acts as a transport medium and thus promotes densification. Liquid phase sintered Si 3 N 4 and SiC usually contain thin films (few nm) of residual glassy material at the grain boundaries and larger pockets (few tens of nm) at triple grain junctions [32]. It was shown above that Mg melted during the sintering process and also promoted densification. Therefore, similar to the glassy phase found in LPS Si 3 N 4 and SiC, Mg in Mg-MgB 2 samples could also form thin layers at the grain boundaries and/or small pockets of residual Mg at triple grain junctions. Such small features cannot be resolved by SEM, but TEM/STEM analysis could confirm the presence of residual Mg.
The 10 wt.% Mg-MgB 2 specimen was further analysed by STEM, and figure 6 shows a typical HAADF image depicting a triple grain junction (indicated by the arrow) and an EDX linescan running through the small pocket located at the junction. The pocket is richer in Mg and depleted in B, which supports the theory that Mg forms a liquid phase during sintering with some residual Mg forming pockets at triple grain junctions. This is consistent to what is observed in liquid phase sintered Si 3 N 4 and SiC [32].

Microstructural evolution during the LPS process
A model for this new LPS process was developed by combining all the characterisation results presented above in order to provide an accurate description of the microstructural evolution during the sintering of MgB 2 bulks containing different fractions of Mg additions (1 up to 10 wt.%) The effects of Mg additions can be summarised as follow:  densification and also reacted with pre-existing MgB 4 to form 'fresh' MgB 2 . No residual Mg was detected and the density increased compared to unmodified MgB 2 . In addition, a reduction in the MgB 4 fraction compared to the asreceived MgB 2 powder was observed. • 6 wt.%: More liquid phase was formed and led to further density improvements and reduction in MgB 4 . No MgB 4 was detected and a very small amount (0.4 wt.%) of residual Mg remained in the bulk after the sintering process. • 10 wt.%: The large volume fraction of Mg led to better compaction at low temperature. A liquid phase was then formed at higher temperature and led to further improvements in density. No MgB 4 was detected and a larger fraction of residual Mg (4 wt.%) was present in the sample. The different steps of the LPS process are presented schematically in figure 7.

Magnetic measurements
Zero field cooled measurements in figure 8(a) show that the superconducting transition becomes narrower with increasing Mg additions. ∆T c decreased from 1.7 K in unmodified MgB 2 to 0.6 K in the sample containing 10 wt.% Mg, as shown in table 2. Matthews et al [11] reported a similar reduction in ∆T c in unmodified MgB 2 samples sintered at higher temperature. The authors attributed this to heterogeneities (composition and/or disorder) initially present in the as-received powder being annealed out at high temperature. Here, all four samples were sintered at 900 • C suggesting that the reduction in ∆T c is an effect of the Mg additions. Higher sintering temperatures and larger Mg fractions both lead to an increase in the diffusion rate, as suggested by the higher density of the 10 wt.% Mg-MgB 2 sample. In addition, it is also likely that as-received MgB 2 powder contained defects and/or local chemical heterogeneities which resulted in the relatively large ∆T c value in the sample made from unmodified MgB 2 . A possible explanation for the effect of Mg additions on ∆T c could be that even though the sintering temperature is low (900 • C) the higher diffusion rate provided by liquid Mg during sintering acts to smooth chemical heterogeneities and thus makes the sample more homogeneous, reducing ∆T c . Figure 8(b) shows that the J c − B curves for the 1 and 3 wt.% Mg samples at 4.2 K seems to be simply translated vertically compared to unmodified MgB 2 , maintaining a very similar shape. The J c − B curve of the 6 wt.% Mg sample was very similar to the 3 wt.% specimen at high field (>5 T), but J c values at low field were further improved. On the other hand, the 10 wt.% sample showed a relatively different shape; low field J c values were similar to the 6 wt.% specimen, but high field J c was substantially reduced compared to unmodified MgB 2 and all three Mg-MgB 2 samples containing lower Mg additions. The 10 wt.% sample also had an irreversibility field B irr approximatively 1.5 T lower than the other four specimens, as shown in table 2. This suggests that the MgB 2 lattice of the 10 wt.% sample was 'cleaner' (following the definition of Anderson [33]). As discussed earlier, the 10 wt.% sample was the only specimen to contain a significant fraction of residual Mg (∼4 wt.%) after the sintering process, and also showed the largest increase in density, attributed to the effect of Mg acting as a liquid phase during consolidation. It is known that the liquid phase in Si 3 N 4 and SiC acts as an impurity sink due to the much larger solubility of impurities in the liquid phase than in the solid phase [32]. It is possible that the Mg liquid phase during sintering also acts as an impurity sink and thus makes the MgB 2 lattice 'cleaner', explaining the significant reduction in B irr observed in the 10 wt.% Mg-MgB 2 bulk. Another possibility is that flux pinning in the 10 wt.% Mg sample is not as good as in bulks containing less Mg additions.  At 20 K, the evolution of the J c − B curves with increasing Mg additions is very similar to that observed at 4.2 K. Self field J c (20 K) increased more than four fold by adding 6 and 10 wt.% Mg compared to unmodified MgB 2 . This is consistent with the SEM analysis ( figure 5) showing that the microstructure of the 10 wt.% Mg sample sintered to a much greater extent than unmodified MgB 2 . Figure 9 shows that connectivity increased significantly with Mg additions compared to unmodified MgB 2 . In particular, the connectivity increased from only 6% in unmodified MgB 2 up to 25% in the sample containing 10 wt.% Mg. This increase in connectivity almost exactly matches the increase in self field J c (20 K) discussed above. A strong correlation between connectivity and self field J c values was also observed by Matthews et al [11] in unmodified MgB 2 sintered at temperatures ranging from 900 • C to 1200 • C. This confirms that connectivity has a significant effect on J c values at low field, and also suggests that connectivity can be improved by using Mg additions instead of higher sintering temperatures. The 10 wt.% Mg specimen showed the best connectivity, suggesting that if there is a layer of residual Mg at grain boundaries it must be sufficiently thin not to adversely affect connectivity, which is consistent with the STEM analysis. The residual resistivity also decreased with increasing Mg fractions, indicating that Mg additions made the MgB 2 phase cleaner.

Conclusion
This work introduced a novel LPS process that we developed to improve the connectivity of ex-situ MgB 2 bulks. Mg additions (1-10 wt.%) were mixed to pre-synthesised MgB 2 to make composite powders that were sintered at 900 • C by FAST. Mg had a very substantial effect on the density of the bulks and their composition. Density increased from 68% to 79% and the fraction of MgB 4 was reduced from 11 to ∼0 wt.% by adding 10 wt.% Mg to unmodified MgB 2 . Pressure and dilatometry data recorded in-situ during the sintering process revealed that Mg additions led to different mechanisms depending on the Mg fraction. For small Mg fractions (1 wt.%) all the metallic Mg evaporated during sintering and generated a positive Mg (g) pressure which modified the dynamic equilibrium of the MgB 2 decomposition reaction causing less MgB 4 to be formed. No liquid phase was formed and thus no increase in density was observed. For intermediate Mg fractions (3 wt.%) some Mg evaporated and MgB 2 decomposition was suppressed. A liquid phase was formed, enhancing densification and reacting with pre-existing MgB 4 to form 'fresh' MgB 2 . No residual Mg was detected and the density increased compared to unmodified MgB 2 . For large Mg fractions (6 and 10 wt.%) more liquid phase was formed and led to further density improvements, and all pre-existing MgB 4 was transformed into MgB 2 . A small amount of residual Mg remained in the bulks after the sintering process. Connectivity was also improved with Mg additions, increasing four fold in the 10 wt.% Mg-MgB 2 sample compared to unmodified MgB 2 . J c values at low field were also significantly improved by Mg additions, in particular the 6 and 10 wt.% Mg-MgB 2 specimens showed J c (20 K, 0 T) values 4-5 times higher than for unmodified MgB 2 . Matthews et al [11] obtained similar J c values by sintering unmodified MgB 2 at 1200 • C (instead of only 900 • C in this study), suggesting that Mg additions enabled the production of well connected bulks at much lower temperatures. The optimal amount of Mg additions is the stoichiometric quantity needed to fully react with all the preexisting MgB 4 in the MgB 2 powder, which was found to be around 6 wt.% for the MgB 2 powder used in this study.
The LPS method developed in this work provides a practical alternative to the conventional ex-situ process by allowing well connected and MgB 4 -free bulks to be manufactured at lower temperatures, minimising the detrimental effect of coarsening, improving the production rate and significantly reducing the production costs.

Acknowledgments
The Authors would like to thank the UK Engineering and Physical Science Research Council and the Department of Materials, University of Oxford for financial support (EP/P026427/1, EP/ M508111/1, Manufacture using Advanced Powder Processes and Process). The Authors acknowledge the use of characterisation facilities within the David Cockayne Centre for Electron Microscopy, Department of Materials, University of Oxford. The data are available at https://doi.org/10.5287/bodleian:xr11rwggY.