Variance in the response of silcrete to rapid heating complicates assumptions about past heat treatment methods

Heat treatment of silcretes in the Middle Stone Age of southern Africa has been taken to indicate complex behaviour among early modern humans. This inference is based on the apparent sensitivity of silcretes to rapid changes in temperature, requiring well-regulated heating and cooling rates, and controls over maximum heating temperatures. Alternative arguments have been made that silcrete can effectively be heat treated with limited control over heating rates such that heat treatment may have been a relatively simple process. These apparently contrasting points of view elide the fact that different silcretes may respond differently to heating, and that no single approach may be appropriate in all cases. To test this proposition, we undertook a series of controlled experiments in which silcrete from two sources on the south coast of Australia were prepared into blocks of specific sizes and heated rapidly to a range of maximum temperatures in a muffle furnace. In addition to potential differences in response between sources to heat, our experiments test two factors—stone volume and maximum heating temperature—that were advanced by past explanatory models to account for the probability of sample failure (fracture) during heating. The results of our experiments suggest that the tolerance of silcretes to high heating rates is highly variable between sources within regions, and that the effect of variation between sources is stronger than the other factors examined. Additional tests on limited samples from sources in South Africa support the general relevance of our findings. From these results, we infer that optimal approaches to heating in the past were probably sensitive to the silcretes being heated.

volume and maximum heating temperaturethat were advanced by past explanatory models 23 to account for the probability of sample failure (fracture) during heating. The results of our 24 experiments suggest that the tolerance of silcretes to high heating rates is dramatically 25 variable between sources within regions, and that the effect of variation between sources is 26 stronger than any of the other factors examined. Additional tests on limited samples from 27 sources in South Africa support the relevance of our findings to other contexts of silcrete 28 heating. From these results, we infer that optimal approaches to heating in the past were 29 probably sensitive to the silcretes being heated. The use of heat to alter the physical properties of siliceous rocks is a behavioural trait so far 34 documented only among modern humans, and reasonably widespread in the global record. 35 The apparently species-specific nature of this practice, coupled with underlying assumptions 36 about the uniqueness of Homo sapiens' behavioural capabilities (Villa and Roebroeks 2014), 37 has encouraged a perception that heat treatment reveals aspects of complexity in human 38 cognition. Like many aspects of debates concerning the evolution of human behaviour, 39 however, the link between evidence and inference is not straightforward. In this particular experiments to recreate that process worked on the assumption that successful heat treatment 50 that is, having the stone remain relatively intact after heatingrequired slow heating and 51 cooling rates, and control over maximum temperatures (Brown et al. 2009). In order to 52 achieve these controls, researchers buried silcrete blocks in an insulating medium (sand) 53 before building a fire over the top. The fire was sustained for a period of hours before being 54 allowed to burn out, and the sand was allowed to cool before the silcrete was extracted. 55 Supporting research suggested that this insulated method was necessary to avoid the silcrete 56 blocks fracturing, something which occurred regularly at higher temperatures in open fires 57 (Wadley and Prinsloo 2014). This approach, which we might term 'high-cost', involves the 58 planning and execution of hierarchical actions, and warrants many of the above-mentioned 59 inferences regarding behavioural complexity. 60 An alternative 'low-cost' approach suggests that heat treatment can be carried out applies to silcrete. In support of this proposition is evidence for carbonised green wood 68 exudates (residues) on heated silcretes from archaeological sites in southern Africa, along 69 with frequent evidence for heat fracture prior to flaking that, as noted above, is difficult to ). It has also been noted that there is no ethnographic evidence that the sand-72 bath approach was ever used during heating of silcrete (Schmidt 2016). If this 'low-cost' 73 approach is a more accurate characterisation of typical silcrete heat treatment in the past, then 74 the implications for behavioural complexity are probably quite limited. While the process is 75 transformative, it is not necessarily any more conceptually complex than cooking food.

76
One potential complication to past experimental work in this debate is the tendency to treat 77 silcrete as a coherent class of rock that is likely to respond consistently to heating. Yet, 78 silcrete is heterogeneous in formation, composition and character (Roberts 2003; 79 Summerfield 1981), and responses to heating have been demonstrated to be variable at 80 regional scales (Schmidt et al. 2017c; note also Byers et al. 2014 with respect to chert).

81
Developing reasonable expectations for the practice of silcrete heat treatment in the past thus 82 requires some understanding of the range of variation in tolerance for different heating rates 83 and maximum temperatures between samples from difference sources. Indeed, it seems 84 plausible that both the high-cost and low-cost approaches outlined above may be viable for 85 different silcretes, given sufficient variation in response.

86
To begin to address this problem, we undertook a set of controlled experiments on silcrete 87 samples from two nearby sources located on the east coast of New South Wales (NSW),

88
Australia. Our objective with these experiments was principally to explore the extent of inter-  2016). In those early trials we observed differences in fracture rates between sources during 93 individual heating runs. We had intended to conduct subsequent formal controlled 94 experiments on samples from the same sources but lacked sufficient material for multiple 95 replications with different factors (as described below). We thus switched to the NSW 96 sources as they were easy to accessproviding enough material for multiple replications -97 and because both are known to have been used archaeologically (Hanckel 1985 Kenna 2016). In this paper we do not 106 undertake any mineralogical analysis but concentrate on the effects of two factorsheat and 107 volumeon the probability and extent of silcrete fracture during heating. We focus on rapid 108 heating because the tolerance of silcrete to steep temperature gradients is, along with 109 maximum temperature tolerated, one of the key elements of the current debate.

110
In a classic and influential set of controlled experiments, Mercieca and Hiscock (2008) 111 explored how silcrete blocks of different volumes respond when exposed to different 112 maximum temperatures. In their experiments, blocks were cut into consistent shapes (cubes) 113 and placed in a preheated furnace, producing very steep temperature gradients. Their results differentiated three zones of responseintact, cracking and fracture (Figure 1). In the intact 115 zone, blocks had no visually-noticeable adverse effects from heating. In the cracking zone, 116 blocks had visually-noticeable cracks but remain coherent. The fractured blocks broke apart.

117
Probability of cracking and fracturing both increased in response to maximum temperature 118 and block volume, such that smaller blocks were able to withstand higher maximum 119 temperatures prior to cracking/fracturing than larger blocks. In general, no cracking or 120 fracturing was witnessed below 600°C, and in the smallest blocks, temperatures of up to 121 700°C could be tolerated without cracking. while those heated to ≤521°Cwhether on a bed of coals or buried in sand beneath the fire -129 did not. Ascribing this pattern to energy release during a phase change in quartz minerals, 130 Wadley and Prinsloo (2014: 49) concluded that: "Rapid heating or cooling through the phase 131 transformation at 573°C will cause fracture of the silcrete". Their results suggest that, size 132 effects notwithstanding, the propensity of silcrete to shatter will increase around and beyond 133 that value.

134
A limitation common to both of these sets of experiments is that they used silcrete from controlled for? 146 We should note at the outset that our experiments only tested for the probability of visually-147 noticeable fracture. It is plausible that unfractured rocks could sustain sufficient damage to 148 their internal structure as to be unusable for tool production, and equally that fractured blocks be classified as 'floating fabric' according to Summerfield (1981).

174
The silcrete samples were heated in an electrical muffle furnace preheated to three set  The benefit of this approach is to allow multiple replications, with less reliance on the 186 representativeness of single specimens.  We also tested the proposition that rates of fracture would increase significantly for samples  (Roberts 2003). Sorting in these silcretes 264 is moderate to poor, with subangular quartz inclusions up to 3 mm.

265
As with the NSW sample, nodules were cut into cubes using a brick saw and refined with a 266 trim saw, and weighed to 0.1 g. Due to the limited material available, we were only able to 267 prepare three cubes for each source in the 27 cm 3 and 64 cm 3 volumes, and nine in the 1 cm 3 268 volume. We did not prepare any cubes of 8 cm 3 . Samples were placed in the same electrical

Australian (NSW) samples
278 Table 1 outlines the percentage of surviving mass for each sample after heating; Figure 3 279 summarises the proportions of the three fracture categories by temperature and volume.  Interestingly, the BDL samples that shattered completely (i.e., 0% remaining mass) are all 287 from the 1 cm 3 volume group (Table 1). This outcome could suggest that heat fracture breaks 288 up entire samples more easily when volume is small, which explains why there is a lack of 289 'fractured' samples in this size category. However, this pattern could also relate to sample 290 recovery error. Specifically, since these samples are small to begin with, the fractured 291 fragments can scatter more easily inside the furnace. If this is the case, the post-heating 292 sample could be difficult to identify, and thus the weight of the surviving sample would be 293 assumed to be zero.

294
In comparison to BDL, the rate of heat fracture among BNS samples is much higher. Over     Based on the second model that excludes the 1 cm 3 samples, Figure 4 summarises the 335 modelled effects of volume and source on heat fracture. In essence, greater stone volume 336 leads to heightened probability for heat fracture to occur (and hence lowers the chance for the 337 samples to remain intact or "survived"). While this effect is present in both sources, the 338 actual probability for heat fracture is offset by inter-source variation. Namely, the degree of 339 heat fracture is overall quite low among the BDL silcrete. Even for the 64 cm 3 group where 340 the effect of volume on heat fracture is the greatest, the BDL silcrete has around 70% chance 341 of remaining relatively intact (i.e., losing only up to 10% of original weight to heat fracture).

342
On the other hand, the degree of heat fracture is notably higher for BNS silcrete. As sample 343 volume increases, the probability for substantial fracture (losing up to 50-100% of original 344 weight) to occur rises sharply. Looking at the 64 cm 3 group again, the BNS samples have 345 only 18% chance of staying intact but 64% chance of becoming "exploded". In summary, 346 while the influence of stone volume on the extent of heat fracture holds for both BDL and 347 BSN silcrete, the two sources exhibit different tolerance to rapid heating. Overall, BSN 348 silcrete has a greater probability for heat-induced fracture than BDL silcrete. Interestingly, 349 the chance for BNS samples to become "fractured" is relatively stable across different 350 volumes. This could be explained by the fact that, because BNS silcrete is less resistant to 351 heat fracture, when samples experience heat fracture, particularly those in the larger size 352 groups, they are more likely to become "exploded" rather than "fractured"-i.e., if heat 353 fracture occurs, the larger stones are more likely to suffer greater fragmentation.

390
Heat treatment of siliceous rocks prior to knapping has a reasonably long history of research, 391 albeit generally of low intensity. That intensity of research has recently increased in response 392 to the argument that the appearance of heat treatment may carry implications for the 393 evolution of human behaviour. As noted at the outset, the validity of those implications 394 largely depends on how heat treatment was conducted in the deeper past, and whether the 395 underlying process was elaborate, with high set up costs, delayed returns, and significant 396 sensitivities to variation in heating parameters, or whether the process was expedient, and 397 with relatively low sensitivities.
Our purpose with this experimental program was not to resolve those debates, but to explore 399 whether there may be flaws in one of its assumptionsnamely that responses of silcrete to 400 heating are consistent, such that valid general statements could be made about the likely way 401 in which heat treatment was conducted in the past. Some past work suggests that this 402 assumption is problematic (Schmidt et al. 2017c), and our results here appear to confirm this.

403
In our principal set of experiments, using silcrete from two sources located not far from one 404 another, we found quite dramatic differences in response to rapid heating. One source

412
Our results support some elements of past research but not others. In our tested samples, 413 increasing the volume of the pieces increases the probability that they will fracture when 414 rapidly heated. However, given that we included samples from the same source as Mercieca 415 and Hiscock (2008), this finding does not constitute fully independent support for that 416 proposition. Perhaps more surprisingly, we found no significant effect of maximum 417 temperature on probability of fracture in our NSW samples. This is at odds with expectations 418 from both the volume-temperature interaction model, but also with the suggestion that 419 heating to or beyond 573°C increases fracture probability. generating prior predictions about the ways any given silcrete might respond to heating. To 427 the extent that this is possible, it would be a more sound basis for generalised statements.

428
Perhaps a more important limitation is one that is not particular to our research, but which is 429 fairly pervasive, and that is the assumed association between fracture and failure. As we 430 noted early, the fact that a given block survived heating to 700°C has few necessary 431 behavioural implications. The surviving block, though coherent, could be unworkable, and      heat fracture among the tested NSW silcrete (exclude 1cm 3 samples).