New insights into the composition and microstructure

Archaeological Research in Asia xxx (xxxx) xxx–xxx

Contents lists available at ScienceDirect

Archaeological Research in Asia journal homepage: www.elsevier.com/locate/ara

Full length article

New insights into the composition and microstructure of ceramic artifacts associated with the production of Chinese bronzes at Yinxu, the last capital of the Shang dynasty ⁎

James B. Stoltmana, , Zhanwei Yueb, Zhichun Jingc, Jigen Tangb, James H. Burtona, Mati Raudseppd a

Department of Anthropology, University of Wisconsin-Madison, Madison, WI 53706-1393, USA Institute of Archaeology, Chinese Academy of Social Sciences, 27 Wangfujing Dajie, Beijing 100710, China Department of Anthropology, The University of British Columbia, 6303 NW Marine Drive, Vancouver, BC V6T1Z1, Canada d Department of Earth, Ocean and Atmospheric Sciences, The University of British Columbia, Vancouver, BC V6 T 1Z4, Canada b c

A R T I C L E I N F O

A B S T R A C T

Keywords: Chinese bronze manufacture Shang dynasty Yinxu Xiaomintun Miaopu North Optical petrography SEM & XRD analyses

The manufacture of Chinese bronzes has been referred to as a “co-craft” thus emphasizing the importance of pottery as a companion to metallurgy in this process. The focus of this paper is upon the ceramic products— molds, models, and cores—that are critical components of this co-craft. Optical petrography, SEM, and XRD analyses of thin sections of these ceramic products from two bronze foundries at the late Shang site of Yinxu are employed to provide new information about their composition, structure, and production. Rather than produced from unaltered loess as is generally believed, these artifacts were made from loess that had been subjected to at least two of three procedural steps: levigation, the selective addition of sand (only for outer layers of molds and for cores), and the addition of lime. The evidence for each of these steps and the reasons for them are described. These findings cast major new insights into the complexity and sophistication of the ceramic component to Chinese bronze manufacture.

1. Introduction In this paper three analytical methods, optical petrography, Scanning Electron Microscopy (SEM), and X-ray Diffraction (XRD), are applied to ceramic molds, models, and cores associated with bronze production at the last capital of the Shang Dynasty, Yinxu, in the modern city of Anyang in north China (Fig. 1).1 The goal of these analyses is to determine the mineralogical composition and microstructure of these ceramic artifacts in order to reveal the selection and processing of raw materials and technological choices made in their manufacture. These data indicate that the technology of mold manufacture used to produce Chinese bronzes was far more complex and sophisticated than previously believed. The elaborate bronzes of Late Bronze Age China have long fascinated archaeologists, historians, art historians, and the general public (Chase, 1983). A wide array of approaches and perspectives have been adopted to study different facets and issues of Chinese bronze technology, including metallurgical analysis of cast bronzes (Gettens, 1969; Su et al., 1995), replication experiments (Wan, 1972; Feng et al., 1980,

⁎

1

1982; Tan, 1999), morphological and technological analyses of casting remains (Shih, 1955; Chen, 1986; Li et al., 2007), and stylistic examination of casting designs and fabrication techniques (Bagley, 1987, 1990, 2009; Nickel, 2006). Ever since Karlbeck (1935) first concluded that Shang founders fabricated bronzes by direct casting, later commonly referred to as piece-mold or section-mold techniques, it has been generally recognized that bronze production in early China was “a co-craft based on the mastery of two different materials, metals and clay” (Li, 2007: 185). Metal melting and casting were wholly dependent upon a wide range of ceramic products (e.g., furnaces, melting crucibles, molds, models, cores). Indeed, as expressed by Nigel Wood (1989:50), “ceramic processes and materials are at the heart of early Chinese bronze casting.” In spite of such realization of the importance of ceramic technology in the production of bronzes in early China, relatively little attention has been paid to the physical composition and microstructure of foundry remains associated with bronze production, assuming that the raw materials used by Shang founders were relatively unaltered local sediments.

Corresponding author. E-mail address: [email protected] (J.B. Stoltman). Huanbei, shown in Fig. 1 and mentioned in Fig. 6, is a short-lived, early Shang site that is the apparent precursor to Yinxu.

https://doi.org/10.1016/j.ara.2017.11.002 Received 9 September 2017; Received in revised form 9 November 2017; Accepted 13 November 2017 2352-2267/ © 2017 Published by Elsevier Ltd.

Please cite this article as: Stoltman, J.B., Archaeological Research in Asia (2017), https://doi.org/10.1016/j.ara.2017.11.002


J.B. Stoltman et al.

Fig. 1. Map showing the location of the city of Anyang where the site of Yinxu is located and the location of the two bronze foundries (Xiaomintun and Miaopu North) within Yinxu.

2



Supplemental to the use of polarizing microscopes, the SEM technique in backscattered electron (BSE) mode2 was also employed in this study to examine the physical composition and microstructure of molds and models. BSE imagery not only has a much greater resolution than can be generated with an ordinary optical microscope but also reveals compositional variations. Although SEM with EDS provides information about the elemental composition of mineral phases that normally allows for unambiguous identification of most minerals, neither optical petrography nor SEM/ EDS provide quantitative results about the percentages of minerals too fine to be optically identified. For this reason we used XRD3 to determine relative weight percents of silt (quartz/feldspar/amphibole) versus phyllosilicate clay versus lime (carbonate).

This study began as an application of optical petrography to the comparative analysis of ceramic molds, models, and cores with the local sediments from Yinxu (the proper name for the Shang site located on the outskirts of the modern city of Anyang). As this work progressed, two properties of these artifacts were observed that had not previously been reported: (1) their silt content was much higher than the local loess, the generally accepted sediment used in their manufacture; and (2) lime had been added, visible as cryptocrystalline calcite, that was so fine grained, it was easily overlooked. Once these two observations had been made, SEM and XRD analyses were applied in order to confirm their veracity. 2. Methodology In this section a brief discussion of each the three methodologies is presented, starting with optical petrography, to assist the reader in relating the data to the conclusions. All thin sections were analyzed with an Olympus polarizing microscope using a basic point-counting procedure as described in several earlier publications (Stoltman, 1989, 1991, 2001, 2015). A standard counting interval of 1 mm at 10 × magnification was employed covering the total area of each thin section or until the total number of points counted (exclusive of voids) exceeded 200. The point counting was conducted “blind,” that is, without prior knowledge of the samples other than their thin section number. The results of the point counting are recorded as “bulk composition,” which is a volumetric measure comprised of the relative percentages of three variables, matrix/clay, silt, and sand (including gravel), all of which are defined here solely as size grades as follows: matrix/clay ≤0.002 mm; silt = visible grains ranging in size from 0.002 mm to 0.0624 mm; and sand = grains ranging in size from 0.0625 mm to 2.00 + mm. Voids were counted but not included in the bulk compositions because most are due either to paste preparation or firing of pottery vessels, properties not associated with raw materials. As a complement to this index, a sand-size index is also recorded for each artifact. This was devised by assigning a value within the following ordinal scale to each sand-sized (or larger) grain encountered in point counting based upon its maximum diameter: 1) 2) 3) 4) 5)

3. Local sediments and pottery at Yinxu Before discussing the ceramic artifacts associated with bronze manufacture, we present a brief review of the physical composition of the local clay-rich sediments around Yinxu and the domestic pottery containers made from them (Stoltman et al., 2009). The purpose of this review is to provide a cultural context for evaluating the compositional data for the molds, models and cores. Two types of local clay-rich sediments used for pottery manufacture at Yinxu have been identified: water-deposited alluvium and wind-deposited loess (Stoltman et al., 2009). Compositional data are available for seven examples of the former and six samples of the latter, a total of 13 local sediments in all (See Table 1). Since the compositions of the paleosols are so similar to those of the alluvium, it appears that the three paleosols had formed on top of alluvial sediments. Accordingly, we treat these seven samples as a single group for comparative purposes in contrast to the much siltier loessic sediments, i.e., the mean silt value for the loess is 19.5 ± 3.3% versus 8.6 ± 2.9% for the alluvium (Table 1). The preponderance of pottery recovered at Yinxu is gray ware (Cheng, 1960:146; Kaogusuo, 1994; Li, 1977:202). Because of its abundance, it can safely be accepted as locally made. In making gray wares, Yinxu potters used two distinct recipes. The first, which is composed of untempered, “raw” loess, was mainly for jars and bowls used for such non-cooking purposes as storage and serving (e.g., guan, pen, gui, jue, dou, etc. See Thorp, 2006:154 for images of these vessel forms). By contrast, cooking vessels, such as li-tripods and steamers

Fine = 0.0625–0.249 mm Medium = 0.25–0.499 mm Coarse = 0.50–0.99 mm Very coarse = 1.00–1.99 mm Gravel = 2.00 + mm

2 Backscattered electrons are generated from elastic collisions between energetic beam electrons and atoms within the specimen. The number of backscattered electrons generated is primarily dependent upon the average atomic number of the target, and is recorded as image brightness on micrographs (Goldstein, 2003). On BSE images, organic matter and epoxy resin impregnating the ceramic matrix appear dark (i.e. black), while mineral grains such as pyrite, carbonate, quartz, and biogenic silica appear decreasingly bright. Such marked contrast in brightness enables the mineralogical identification of clastic particles and clay, and the determination of fabric characteristics, particularly grain size distribution and porosity. BSE imaging was carried out at the University of British Columbia using a Hitachi S-3400 N SEM with an attached Oxford EDS system (Oxford Instrument INCA series). Prior to BSE imaging, the polished thin sections are coated with 150 Å of amorphous carbon in a vacuum sputter coaster to ensure sufficient conductivity, as ceramic samples are poorly conductive material. BSE observations were made at 15 or 20 KeV (beam accelerating voltage), 0 (tilt angle) and 10 mm (working distance). 3 For the XRD analyses, the samples were reduced to the optimum grain-size range for quantitative X-ray analysis (< 10 μm) by grinding under ethanol in a vibratory McCrone Micronising Mill for 7 min. Continuous-scan X-ray powder-diffraction data were collected over a range 3–80°2θ with CoKa radiation on a Bruker D8 Focus Bragg-Brentano diffractometer equipped with an Fe monochromator foil, 0.6 mm (0.3°) divergence slit, incident- and diffracted- beam Soller slits and a LynxEye detector. The long fine-focus Co Xray tube was operated at 35 kV and 40 mA, using a take-off angle of 6°. Minerals were identified using the International Centre for Diffraction Database PDF-4 + using Diffrac Plus Evaluation software (Bruker AXS). Quantitative phase analysis was done with Rietveld program Topas 4.2 (Bruker AXS). The amounts measured represent the relative amounts of crystalline phases normalized to 100%. As the crystal structure of smectite (probably montmorillonite) is disordered and not predictable, the amount was estimated by using the PONKCS method of Scarlett and Madsen (2006).

The size values for each of the grains were then summed and divided by the total number of grains counted, thus providing a mean “sand-size index,” which ranges between 1 and 5 for each thin section. This index is useful for discriminating loess from alluvium and tempered from untempered vessels. Most artifacts in this study for which the bulk compositions were recorded lack temper, so their compositions may be compared directly with raw sediments, offering the possibility that the sedimentary source for such artifacts may be identified. For tempered artifacts, however, the temper must first be identified then deleted from the computations before identification of a sedimentary source would be possible. Scanning electron microscopy (SEM) has now become a standard technique that is utilized to provide additional information about ceramic fabrics (Echin, 2009; Freestone and Middleton, 1987). Among the advantages of using SEM, commonly coupled with X-ray energy dispersive spectroscopy (EDS), are high magnification, great depth of observed field, as well as the ability to determine the chemical composition of fine-grained particles (Krinsley et al., 1998; Reed, 2005).

3



Table 1 Bulk compositions for local sediments in and around Yinxu. Sediment type

Thin section #

% Matrix

% Silt

% Sand

Sand size index

Alluvium Alluvium Alluvium Alluvium Alluvium mean ± 1 Std Dev

HBOO-21 AYOO-34 AYOO-35 ALNO9-B7 N=4

88 94 86 94 90.5 ± 4.1

11 5 12 5 8.3 ± 3.8

1 1 2 1 1.2 ± 0.5

1.00 1.50 1.20 1.00 1.18 ± 0.24

Paleosol [“Huanbei soil”] Paleosol Paleosol Paleosol mean ± 1 Std Dev Alluv + Paleosols mean ± 1 Std Dev

AYO9-3 ALNO9-B1 ALNO9-A1 N=3 N=7

84 91 89 88.0 ± 3.6 89.4 ± 3.8

8 8 11 9.0 ± 1.7 8.6 ± 2.9

8 1 0 3.0 ± 4.4 2.0 ± 2.7

1.65 1.00 – 1.32 ± 0.46 1.23 ± 0.29

Loess Loess Loess Loess Loess Loess Loess mean ± 1 Std Dev

HBOO-20 ALNO9-A4 ALNO9-A6 ALNO9-A8 ALNO9-B4 ALNO9-B6 N=6

77 78 80 83 82 72 78.7 ± 4.0

22 21 17 16 17 24 19.5 ± 3.3

1 1 3 1 1 4 1.8 ± 1.3

1.00 1.00 1.00 1.00 1.00 1.00 1.00 ± 0.0

Because the compositional values for the paleosols and alluvium are similar, it is assumed the soils formed on top of alluvium, i.e., the two can be lumped together.

Fig. 2 is a ternary graph presenting the bulk compositional values for 46 local gray-ware vessels from Yinxu for which thin section analyses have been conducted. In Fig. 2, these values are plotted against the compositional values for the 13 local sediments, 6 loess, 4 alluvium, and 3 buried, organic-enriched soil horizons. As can be readily seen from this figure, 29 of the 32 noncooking vessels plus two cooking vessels (li) fall within the compositional range of loess. By contrast, the majority of cooking vessels (10 of the 14) is sand tempered and plot in the compositional range of alluvium. [This inference is based on acceptance of the mean sand value for alluvium of 2% as a reasonable basis for identifying all sand grains in excess of 2% as humanly added temper.] A single non-cooking jar is sandtempered, providing the exception to the “rule” that non-cooking vessels are never tempered while one of the sandy li cooking vessels

(e.g., yan and zeng), while less common but nonetheless abundant, were typically made from alluvium that was tempered either with sand or, less often, with crushed rock (i.e., “grit”) (Stoltman et al., 2009). The sand (mainly quartz) used for temper was readily available locally, but the metamorphic rocks sometimes used as temper were definitely of exotic origin. Whether such pots, or just the temper, were imported is a debatable issue. In any case the addition of temper seems clearly to have been a conscious practice to ensure that vessels subject to recurrent heating and cooling could withstand the thermal shock associated with cooking. Because of the uncertainty associated with the origins of the grittempered vessels, they are excluded from further consideration here in order to focus upon the compositions of what can safely be accepted as locally made pottery.

Fig. 2. Ternary graph comparing the bulk compositions of 46 local grayware vessels with 13 local sediments.

4



Table 2 Bulk compositions for Xiaomintun single-layer molds. Mold type

Thin section #

% Matrix

% Silt

% Sand

Sand size index

Vessel Untyped Vessel Vessel Untyped Untyped Untyped Untyped Untyped Untyped Weapon Gu Vessel Weapon Lei Gui Vessel Gu Vessel Gu Squared Zun Squared vessel Vessel Vessel Gu Mean ± 1 Std deviation

AYOO-28 AYOO-29 AYOO-30 AYOO-31 UBC2010-42 AGO8-17 AGO8-18 AGO8-19 AGO8-20 AGO8-21 AXS09-4 AXS09-5 AXS09-6 AXS09-7 10AXS-22 10AXS-23 09AG-11 09AG-12 09AG-13 09AG-14 09AG-15 09AG-17 09AG-18 09AG-19 09AG-20 n = 25

63 70 59 66 70 66 71 65 69 56 56 70 61 59 69 75 65 62 61 68 68 59 63 60 61 64.5 ± 5.1

36 28 39 32 30 33 28 33 30 42 40 28 37 36 29 25 34 36 39 30 28 40 35 36 35 33.5 ± 4.6

1 2 2 2 0 1 1 2 1 2 4 2 2 5 2 0 1 2 0 2 4 1 2 4 4 2.0 ± 1.3

1.00 1.00 1.00 1.00 – 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 – 1.00 1.00 – 1.00 1.00 1.00 1.00 1.00 1.14 1.005 ± 0.02

is a unique outlier with 29% silt content.

and a secure stratigraphic context for bronzes was established (e.g., Chang, 1980; Kaogusuo, 1994; Li, 1977; Thorp, 2006). While several bronze foundries have now been identified at Yinxu, the focus of the present study will be upon two, Xiaomintun and Miaopu North (Fig. 1). The largest bronze foundry at Yinxu, over 5 ha in area, is located in and near the Xiaomintun village (Jing et al., 2013:344). Initially discovered in 1960, the Xiaomintun foundry was extensively excavated from 2000 through 2004 (Anyang Work Station, 2007; Yue and Yue,

4. Bronze foundries at Yinxu Located on the loess-covered north China plain in Henan province, Yinxu was the last capital of the Shang dynasty, ca. 1250 to 1046 BCE. It is one of China's most celebrated sites because it was here, beginning in 1928, that the earliest evidence of Chinese writing was discovered

Table 3 Bulk compositions for Xiaomintun two-layer molds. Mold type

Thin section #

% Matrix

% Silt

% Sand

Sand size index

Coarse (outer) layer Untyped Untyped Untyped Untyped Vessel Vessel Untyped Chariot fitting Vessel Lei Mean ± 1 Std deviation

AGO8-22 AGO8-23 AGO8-24 AGO8-25 AXSO9-8 AXSO9-9 O9AG-16 AXSO9-3a AXSO9-10a 10AXS-24a n = 10

58 68 61 53 66 64 65 61 60 60 61.6 ± 4.4

23 24 30 40 23 26 25 26 21 29 26.7 ± 5.4

19 8 9 7 11 10 10 13 19 11 11.7 ± 4.2

1.39 1.17 1.62 1.08 1.32 1.29 1.26 1.21 1.22 1.20 1.28 ± 0.15

Fine (inner) layer Untyped Untyped Untyped Untyped Vessel Vessel Untyped Mean ± 1 Std deviation

AGO8-22 AGO8-23 AGO8-24 AGO8-25 AXSO9-8 AXSO9-9 O9AG-16 n=7

70 64 60 62 63 74 61 64.9 ± 5.2

27 34 38 36 36 24 35 32.8 ± 5.2

3 2 2 2 1 2 4 2.3 ± 0.95

1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 ± 0

a

Fine layers not available for 3 molds.

5



Table 4 Bulk compositions for molds from Miaopu North. Mold type

Thin section #

% Matrix

% Silt

% Sand

Sand size index

Single-layer molds Untyped Untyped Untyped Untyped Untyped Weapon Vessel

09 09 09 09 09 09 10

72 57 65 60 72 63 67

27 42 33 38 26 36 31

1 1 2 2 2 1 2

1.00 1.00 1.00 1.00 1.00 1.00 1.00

Two-layer molds (fine layers) Untyped Untyped Mean ± 1 Std deviation

09 AM2 09 AM3 n=9

62 66 64.9 ± 5.1

37 31 33.4 ± 5.3

1 3 1.7 ± 0.7

1.00 1.00 1.00 ± 0

Sandy layers Untyped Untyped

09 AM2 09 AM3

66 63

30 28

4 9

1.10 1.12

Sandy molds Untyped Untyped Vessel Vessel Weapon Mean ± 1 Std deviation

09 AM1 09 AM4 09 AM13 09 AM16 09 AM17 n=7

57 60 69 64 68 63.9 ± 4.3

36 25 24 30 26 28.4 ± 4.1

7 15 7 6 6 7.7 ± 3.7

1.08 1.47 1.12 1.00 1.08 1.19 ± 0.20

AM5 AM7 AM8 AM10 AM11 AM14 AM18

quantities of ceramic mold and model fragments, broken crucibles, furnace fragments, and bronze casting-related tools mostly dating to Yinxu Phases I and II (Kaogusuo, 1994; Chen, 1986).

2009; Yue, 2006; Yue et al., 2007). Among a considerable quantity of foundry remains are a wide range of structural features, crucibles, furnaces, molds, models, cores, and tools associated with the production of cast bronzes. Structural features associated with bronze casting, most dating to Yinxu Phases III and IV, include: buildings for making and firing ceramic molds and models and for casting bronzes; storage pits for raw clay materials; pits for drying molds and models; and working surfaces related to the polishing and finishing of cast bronzes. Their distribution strongly suggests organized use of multiple crafts in the production of bronze objects during the latest phases of the Shang dynasty. Miaopu North was excavated from 1957 through 1964. Bronze foundry remains covered an area of over one hectare and yielded large

At least since the l960’s (e.g. Gettens, 1969:107–114), it has generally been accepted that bronze molds at Yinxu were made from local loess. This view received strong empirical support in a paper by Freestone, Wood, and Rawson in 1989 (Freestone et al., 1989) and is championed by Kerr and Wood (2004:102–104). Freestone et al. (1989) employed optical and scanning electron microscopy with energy dispersive x-ray analysis to determine the elemental composition and

Fig. 3. Ternary graph showing bulk compositions of 35 bronze molds from Xiaomintun plotted against local loess.

Fig. 4. Ternary graph showing bulk compositions of 14 bronze molds from Miaopu North plotted against local loess.

5. The bronze molds of Yinxu

6



loess—by immersing it in water and later pouring off some of the clay fraction that remained in suspension while the coarser, silt-rich fraction settled out and was retained (See Kerr and Wood, 2004:103). From these data, we conclude that the major ingredient of Yinxu bronze molds is what we shall henceforth refer to as “refined loess”. This finding is consistent with the chemical compositions reported by Tan (1999:219) and Freestone et al., 1989:261) that shows a higher SiO2 content for the four molds they analyzed (mean = 75.1 ± 0.6%) versus seven loess/brick samples (mean = 70.3 ± 3.6%). Since the preponderance of silt grains in the loess is quartz, this is entirely expectable. XRD analysis of four Xiaomintun molds versus a local loess sample confirms the clay reduction and the elevated quartz content of the molds (Fig. 5)4. Fig. 6 presents a visual image of the differences in clay and silt content between raw loess, a ceramic vessel, and a mold from Yinxu. Another observation from the petrographic analyses is that, with one exception, no tempers were added to the molds. The exception, outer layers of double-layered molds, was invariably tempered with the addition of sand grains (See Fig. 7). As can be seen from Tables 2 and 3, outer layers have more and coarser sands than inner layers, but inner or single (i.e., decorated, “hot” layers) were never tempered. This finding is relevant in that it stands in contradiction to three oft-cited experimental replications of bronze molds that all employed one or more tempers to prevent mold shrinkage during drying, firing, and casting (Feng et al., 1980, 1982; Tan, 1999; Wan, 1972). The refined loess of the molds offers at least four possible benefits for use in bronze casting. First, the fine texture would readily receive and retain the finely engraved and stamped decoration characteristic of Yinxu molds (See Fig. 6). Second, the elevated silt content (most of which is quartz) would provide an increase in refractoriness and resistance to shrinkage and cracking. Third, as can be seen in the SEM images (Fig. 6), the clay reduction is associated with an increase in porosity, which would facilitate the escaping of gases during firing and casting and, further, would reduce thermal conductivity thus providing slower, more uniform cooling of the molten bronze. And fourth, the reduction in clay content would reduce plasticity, which would facilitate the clean release of the molds from the models, thus ensuring that the impressions of the intricate decoration would survive undamaged to the casting stage. In this regard Hamer (1975:237), in discussing press-

Fig. 5. Graph showing the results of XRD analysis comparing the compositions of a local loess sample versus four molds from Xiaomintun.

structure of four bronze molds that were housed in the collections of the British Museum, and believed to be collected in Anyang during the 1930s. To aid in interpreting the composition of the molds, they also analyzed six loess samples from three Chinese localities, Zhengzhou (n = 3), Loyang (n = 1), and Xian (n = 1). As a result of these analyses, they concluded, “that Shang bronze moulds were produced from loess in an essentially unmodified form” (Freestone et al., 1989:270). As will be discussed below, our findings agree that loess was indeed used to fabricate Yinxu molds, but not in unmodified form. The results of the petrographic analysis of 46 Yinxu molds are presented in Tables 2 to 4 and portrayed graphically in Figs. 3 and 4. Several of the molds were composed of two layers, an inner fine layer and an outer coarser layer, and in such cases the layers were analyzed separately. By referring to Tables 2 to 4 and Figs. 3 and 4, the most prominent feature of the mold compositions can be seen to be their elevated silt values in contrast to those of the local loess. A comparison of the mean silt values for the molds, the loess, and the untempered gray-ware vessels dramatizes this difference:

4 For detailed speciation of clay minerals, generally defined by clay mineralogists as the “ < 2 μm fraction,” preferred-orientation mounts are the preferred method as it allows the measurement of the positions and behaviour of the enhanced basal reflections “as is” and after various treatments, glycolation, heating, chemical dopings of various kinds, etc. This method also facilitates the identification of mixedlayer clays. However, this method cannot be used to calculate the relative weights of the clays and other minerals in a sample as the orientation of crystallites in the powder must be randomly oriented. To do this with the Rietveld method, the crystal-structure data for each phase is used together with a model for the shapes and widths of the diffraction peaks, a model for any aberrations in the shapes and positions of the peaks and a model for the background to calculate a simulated powder-diffraction pattern for each phase. The sum of the individual calculated patterns is then fitted to the digital experimental diffraction pattern. The fitting is done by least-squares refinement of the structural parameters of each phase, together with various global experimental parameters that affect the pattern, so that the difference in fit between the whole observed and calculated diffraction patterns is minimized. The fit between the sum of the calculated patterns and the experimental pattern is adjusted by scaling the calculated contribution of each constituent phase. The relative masses of all phases contributing to the diffraction pattern can then be derived from the refinement using the scale factors derived from Rietveld refinement, the number of formula units per unit cell for each phase, the mass of the formula unit, and the volume of the unit cell. Amorphous content is not measured. In the loess sample, the diffraction peaks of illite, clinochlore and smectite are broad which suggests they have crystallite sizes much < 2 μm and are thus “clay”. Of course, there may be small amounts of mixedlayer clays not detected, e.g., illite-smectite, chlorite-smectite, etc., but these are associated with the amounts of the parent phases and in any case are not likely germane to the arguments presented in the paper. The XRD analysis does not assign anything, it is a bulk analysis. The only true mica in the loess sample might be muscovite sensu stricto but its pattern is not distinguishable from that of the illite, thus it was measured and reported as illite/muscovite. However, the peak width is so broad that this phase certainly has crystallite sizes much < 2 μm and is thus mostly illite and also “clay”.

Samples Mean %silt ± 1 Std Dev. Loess Untempered gray ware Xiaomintun, fine [inner] mold layers Xiaomintun, coarse [outer] mold layers Miaopu, fine [inner] mold layers Miaopu, coarse [outer] mold layers

6 23 32

19.5 ± 3.3 19.9 ± 3.5 33.4 ± 4.7

10

26.7 ± 5.4

9

33.4 ± 5.3

7

28.4 ± 4.1

It is unlikely that any natural sediments in the Anyang region have such elevated silt values (at least we have not encountered such sediments). Rather, we believe that this property of the Yinxu molds is due to the intentional refining of the local loess by the craftspersons as a step in the process of mold manufacture. These elevated silt values could be easily attained through levigation, i.e., water separation of the 7



Fig. 6. 3 SEM images each to the right of the sediment profile or artifact shown on the left showing the differences in clay and silt content with the mold having more silt and more voids (i.e., less clay).

moulding, observes that: “clays of low plasticity (short clays) often press well and have the merit of freeing themselves from the mould fairly quickly”. An unanticipated discovery was that the Yinxu mold makers were not satisfied simply with refining the loess. Another production step was involved! Identifying this step began with the recognition of the occasional presence of post-depositional crusts of CaCO3 on sherd surfaces and within voids. Such deposits of “caliche” were well known to the senior author, having encountered them in several different contexts around the world, but were always relegated to the “interesting but not important” category because of their unambiguously diagenetic character. It was no surprise, then, to encounter such crusts on a few sherds from Yinxu and simply to note this presence when encountered. While analyzing a group of jars from a recently excavated kiln, two

sherds were both observed to have striking occurrences of diagenetic CaCO3 (Figs. 8 and 9). The first of these jars (Fig. 8) shows a typical example of this process, displaying a surficial crust, fissure filling, and void filling of CaCO3. The second sherd (Fig. 9) had an even more prominent surficial crust, but underlying that crust (in the upper left part of the image) was clear evidence of extremely fine-grained CaCO3 penetrating about 0.40 mm into the body of the vessel. Having never seen this before, and reminded of properties observed in the bronze molds that were previously dismissed as naturally birefringent properties of the clays, a reexamination of the mold thin sections was initiated. The presence of fine-grained calcite, sometimes in small aggregates but mostly randomly dispersed, was immediately recognized as occurring scattered throughout every one of the mold bodies (See Fig. 10). The thin sections of the 23 untempered gray-ware vessels were

8



then reexamined, and not a single occurrence of what we now refer to as cryptocrystalline calcite was observed. There seems little doubt that the cryptocrystalline calcite was being intentionally added by the mold makers. But how and why? Before addressing each of these questions, we first describe the results of our SEM analyses, which confirm the presence of calcium enrichment of the molds. The initial identification of this cryptocrystalline material as calcite was dependent mainly upon its distinctive, high birefringence, but it was too fine-grained for definitive identification with an optical microscope. Thus we decided to examine uncovered, polished thin sections using BSE imaging that provides a compositional image based upon the mean atomic number of the target. Although discrete grains of calcite likewise could not be resolved with the SEM, we were able to gain compositional information because the electron beam stimulates the emission of X-rays from the sample with energies characteristic of the elements generating those X-rays. The X-ray spectrum of the fine-grained groundmass has the essential peaks of calcite (Ca, C, O) along with intermixed silica (Si) and lesser amounts of other elements (Mg, Al, Fe), possibly due to traces of clay minerals. Thus we were unambiguously able to confirm the presence of calcium carbonate in the matrix, originally identified optically (Fig. 11). Turning now to the problem of how the cryptocrystalline calcite entered the mold bodies, the most likely explanation is through what is referred to as the calcium carbonate cycle (Aitcin, 2008:26; See also St John et al., 1998:120). This cycle starts with burning limestone (calcination) to a temperature of ca. 850–900°°C (easily attainable with a wood fire), which drives off CO2 gas leaving behind CaO (variously called lime, calcia, or quicklime). When the lime is mixed with the processed loess (which may have been dried and then remoistened), it would quickly interact with water to become slaked or hydrated lime [Ca(OH)2]. The hydrated lime will eventually harden, but only when exposed to air from which it readily absorbs CO2 via the process of carbonation. This allows it to return to calcite (CaCO2) thus completing the cycle (See also St John et al., 1998:120). What is noteworthy about this process is that the calcite does not enter the mold bodies directly, but becomes calcite after having been added to the molds through the process of carbonation when the hydrated lime takes on CO2 from the atmosphere.

Fig. 7. Photomicrograph of a two-layered mold (AXSO9-8a) from Xiaomintun. Taken under crossed polars @ 10 × magnification. (Fine layer at top, coarse layer on bottom). Scale: quartz grain lower center = 0.30 mm.

Fig. 8. Photomicrograph of thin section AFJ08-1 from an untempered gray-ware jar from Yinxu showing post-depositional CaCO3 on the surface and as fissure and void filling. Taken under crossed polars @ 10× magnification. Scale: quartz grain near top center = 0.05 mm.

9



Fig. 9. Photomicrograph of thin section AFJO8–3 from an untempered gray-ware jar from Yinxu showing surficial CaCO3 crust + CaCO3 penetration into the vessel body (upper left). Taken under crossed polars @ 10× magnification. Scale: quartz grain near center = 0.075 mm.

Fig. 10. Photomicrograph of a Xiaomintun mold (AXSO94e) taken under crossed polars @ 25 × magnification. Note cryptocrystalline CaCO3 throughout the body but not in voids. Scale: maximum diameter of central void = 0.50 mm.

well as clay that has stiffened a bit” (Clayton, 1998:31). Moreover, if any portions of the impressions were defective, assuming the errors were relatively minor, incising and carving on the molds could easily repair the damage after the impressions had been taken.

The next question now becomes,” Why was the lime added?” The answer, we feel, resides in this salient property of lime: “Lime is a binder” (Aitcin, 2008:26). Making the same point somewhat differently, St John et al. 1998:1 note that slaked lime has “cementitious properties when exposed to air.” As the molds dried, the binding/hardening effect of the lime would stabilize them so that, with proper timing, model impressions could be effectively taken. When properly dried and hardened, the lime-carrying molds would provide a valuable benefit by avoiding a notable problem in press-molding, namely: “Very moist clay doesn't retain fine details as

6. Ceramic models and cores from Yinxu Ten models (one with two layers) and 18 cores (one with two layers) from Yinxu were analyzed using the same methodology applied to the molds. The results are recorded individually for each of these artifacts

10



Fig. 11. Energy-dispersive X-ray (EDX) spectrum of interstitial phase for single-layered mold 09AG11 for a ritual vessel from Xiaomintun documenting the presence of calcium carbonate in the matrix.

models and cores were clearly made from loessic sediments, they are unlike the molds in that most of them were made from unrefined loess. Finally, the cores differ from both the models and molds in possessing a generally coarser (i.e., sandier) texture, perhaps a reflection of functional considerations, i.e., cores are not decorated.

in Tables 5 and 6 and shown graphically in Fig. 12. Three physical properties stand out as noteworthy. The first pertains to both the models and cores: unlike the molds, they were made from unrefined loess, with but two or three exceptions in which the silt values exceed 30% (See Fig. 12). The second compositional property of note distinguishes the models from the cores; namely, the models are generally finer in texture, possessing notably less sand on average (See Fig. 12). It is presumed the lesser sandiness of the models reflects the intent of the makers to produce suitably fine-textured pastes to display reliably the elaborate decoration that is typically imparted to models but never to cores. The third property, which pertains to the models and cores alike, is that they, like the molds, have bodies that were richly endowed with cryptocrystalline calcite. The incidence of the cryptocrystalline calcite is much more variable than in the molds, but its presence is invariable. In some cores, sand-sized grains of calcite were observed. In sum, the composition of the models and cores are similar to the molds in that cryptocrystalline calcite was incorporated into their recipes, perhaps indicative of sharing common production loci. While the

7. Summary and conclusions Petrographic analyses, supplemented by confirmatory results of SEM and XRD analyses, of ceramic molds, models, and cores associated with bronze production at Yinxu have revealed two new insights into the composition of these critical artifacts for the manufacture of Chinese bronzes thus enhancing our understanding of the processes involved in bronze production. Two widely-cited views of the composition of the molds occur in the literature: (1) based upon experiments to replicate them, some form of temper must have been added (Wan, 1972; Feng et al., 1982; Tan, 1999), and (2) that they were made of unaltered loess has been suspected for decades (e.g., Gettens, 1969) and has recently received empirical support based mainly upon SEM

11



Table 5 Bulk compositions for models and cores from Xiaomintun. Artifact type

Thin section number

% Matrix

% Silt

% Sand

Sand size index

Models Model Model Model Model Model Model Model Model Model, inner layer Model, outer layer Mean ± 1 Std deviation

AGO8-11 O9AG-30 O9AG-31 O9AG-32 O9AG-33 10AXS-26 O9AG-34 UBC1O-43 AGO8-10 AGO8-10 N = 10

74 77 72 64 77 74 78 79 80 61 73.6 ± 6.4

24 21 24 32 20 24 22 19 18 38 24.2 ± 6.2

2 2 4 4 3 2 0 2 2 1 2.2 ± 1.2

1.00 1.125 1.33 1.00 1.00 1.11 1.00 1.00 1.00 1.00 1.06 ± 0.11

Cores Core Core Core Core Core Core Core Core Core Core Core Core Core Core Core Mean ± 1 Std deviation

AGO8-12 AGO8-13 AGO8-14 AXSO9-1 AXSO9-2 UBC1O-39 UBC1O-40 O9AG-21 O9AG-22 O9AG-23 O9AG-24 O9AG-26 O9AG-28 O9AG-29 1OAXS-25 N = 15

77 70 68 63 80 75 78 68 76 73 57 75 64 67 73 70.9 ± 6.4

21 29 24 26 19 25 20 26 22 24 26 20 21 21 20 22.9 ± 3.0

2 1 8 11 1 0 2 6 2 3 17 5 15 12 7 6.2 ± 5.4

1.12 1.00 1.18 1.18 1.00 – 1.20 1.24 1.00 1.00 1.40 1.00 1.23 1.00 1.32 1.13 ± 0.14

Table 6 Bulk compositions for model and cores from Miaopu North. Artifact type

Thin section number

% Matrix

% Silt

% Sand

Sand size index

Models Model

09 AM-21

70

24

6

1.29

Cores Core Core, red zone Core, brown zone Core Mean ± 1 Std deviation

09 AM-12 09 AM-15 09 AM-15 09 AM-19 N=4

68 74 63 58 65.8 ± 6.8

20 21 30 25 24.0 ± 4.6

12 5 7 17 10.2 ± 5.4

1.20 1.40 1.47 1.11 1.30 ± 0.17

analysis of molds and sediments (Freestone et al., 1989). The research presented in this paper fails to support either of these alternatives. Instead, we found that, while the local loess was the primary material used in their manufacture, it was altered in at least two of three ways, depending upon the artifact type being made. One approach does involve the addition of sand temper, but only for limited purposes. Sand temper (or a sandy sediment) was used in three cases: (1) the external layers of multilayer molds, (2) the internal layers of multilayer models, and (3) for cores. What these three have in common is that none is ever decorated, so their coarser texture would have no deleterious effect on that important feature. Moreover, the coarser particles of these particular layers would provide a “sandwich” of thermal resistance to shrinkage surrounding the decorated, “hot” surfaces (i.e., the inner mold layers and the outer model layers) during episodes of intense heat (i.e., firing and pouring of molten metal). The second form of loess alteration is levigation. Our research confirms the suggestion of Kerr and Wood (2004:103) that water separation was used to reduce the clay fraction/increase the silt fraction of molds “by stirring up the loess with an excess of water and pouring off the top layer that stayed in suspension.” This was surely the process used to produce the elevated silt content that we have documented for the molds, for the one outer model layer, and a few of the remaining models and cores.

Fig. 12. Ternary plot showing the bulk compositions of cores and models from Xiaomintun and Miaopu North in comparison to the local loess.

12



The third form of loess alteration involved the addition of lime to the three artifact types, molds, models, and cores. This could only have been accomplished realistically by burning limestone (CaCO3) to produce lime (CaO) that, when mixed with water and the loess, (whether refined or not), yielded slaked lime [Ca(OH)2], which, when encountering air would provide the “cementitious properties” (St John et al., 1998:1) that would bind these ceramic artifacts together despite the reduced clay content of those made of refined loess. It was the recarbonated slaked lime that was seen as fine-grained calcite (CaCO3) during the petrographic analyses (cf. Karkanas, 2007:775). It is interesting to note that observation of thin sections of two bronze molds and a model from the famous Houma site in Shanxi province (Bagley, 1996; Xu, 1996; So, 1980) have confirmed that the addition of lime to molds was a practice that continued into Eastern Zhou times (771-221 BCE). In sum, the steps involved in the manufacture of molds, models, and cores reflects a sophisticated knowledge of loessic sediments and how to control them so as to successfully produce the wide range of complex, elegant Chinese bronzes that are truly one of the marvels of the ancient world.

London. Jing, Z., Tang, J., Rapp, G., Stoltman, J., 2013. Recent discoveries and some thoughts on early urbanization at Anyang. In: Underhill, A.P. (Ed.), A Companion to Chinese Archaeology. Wiley-Blackwell, West Sussex, U.K., pp. 343–366. Kaogusuo, 1994. (Institute of Archaeology, Chinese Academy of Social Sciences). Yinxu de faxian yu yanjiu (Archaeological excavation and research at the Ruins of Yin). Kexue chubanshe, Beijing. Karkanas, P., 2007. Identification of lime plaster in prehistory using petrographic methods: a review and reconsideration of the data on the basis of experimental and case studies. Geoarchaeology 22 (7), 775–796. Karlbeck, O., 1935. Anyang moulds. In: Bulletin of the Museum of Far Eastern Antiquities, Stockholm. 7. pp. 39–60. Kerr, R., Wood, Nigel, 2004. Science and civilization in China. In: Kerr, R. (Ed.), Chemistry and Chemical Technology, Part XII: Ceramic Technology, vol. 5 Cambridge University Press, Cambridge. Krinsley, D.H., Pye, K., Boggs, S.J., Tovey, N.K., 1998. Backscattered Scanning Electron Microscopy and Image Analysis of Sediments and Sedimentary Rocks. Cambridge University Press, Cambridge. Li, C., 1977. Anyang. University of Washington Press, Seattle. Li, Y.-t., 2007. Co-craft and multicraft: section-mold casting and the organization of craft production at the Shang capital of Anyang. In: Shimada, Izumi (Ed.), Craft Production in Complex Societies: Multicraft and Producer Perspectives. University of Utah Press, Salt Lake City, pp. 184–223. Li, Y.-T., Yue, Z., Liu, Yu, 2007. Cong xiaomintun dongnandi chutu taofan tan dui yinxu qingtongqi de jidian renshi (A discussion on Yinxu bronzes through the examination of ceramic molds excavated from Xiaomintun Southeast). Kaogu 3, 52–63. Nickel, L., 2006. Imperfect symmetry: re-thinking bronze casting technology in ancient China. Artibus Asiae 66 (1), 5–39. Reed, S.J.B., 2005. Electron Microprobe Analysis and Scanning Electron Microscopy in Geology, 2nd ed. Cambridge University Press, Cambridge. Scarlett, N.V.Y., Madsen, Ian C., 2006. Quantification of phases with partial or no known crystal structures. Powder Diffract. 21 (4), 278–284. Shih, C.-j., 1955. Yin dai de zhutong gongi (Bronze casting tvechniques of the Shang Dynasty). Bull. Inst. Hist. Philol. 26, 95–130. So, J.F., 1980. New departures in eastern Zhou bronze designs: the spring and autumn period. In: Fong, Wen (Ed.), The Great Bronze Age of China. Alfred A. Knopf, Inc., New York, pp. 249–263. St John, D.A., Poole, Alan W., Sims, Ian, 1998. Concrete Petrography. Arnold, London. Stoltman, J.B., 1989. A quantitative approach to the petrographic analysis of ceramic thin sections. Am. Antiquity 54 (1), 147–160. Stoltman, J.B., 1991. Ceramic petrography as a technique for documenting cultural interaction: an example from the upper Mississippi valley. Am. Antiquity 56 (1), 103–120. Stoltman, J.B., 2001. The role of petrography in the study of archaeological ceramics. In: Goldberg, P., Holliday, V., Ferring, R. (Eds.), Earth Sciences and Archaeology. Kluwer Academic/Plenum Publishers, New York, pp. 297–326. Stoltman, J.B., 2015. Ceramic Petrography and Hopewell Interaction. The University of Alabama Press, Tuscaloosa. Stoltman, J.B., Jing, Zhichun, Tang, Jigen, (Rip) Rapp, George, 2009. Ceramic production in Shang societies of Anyang. Asian Perspect. 48 (1), 182–203. Su, R., Jueming, Hua, Kemin, Li, Lu, Benshan, 1995. Zhongguo shanggu jinshu jishu (Metallugical technology of ancient China). Shandong kexue jishu chubanshe, Jinan. Tan, D., 1999. Zhongguo qingtong shidai taofan zhuzao jishu yanjiu (Casting technology using ceramic section-molds during the bronze age in China). Kaogu xuebao 1999 (2), 211–250. Thorp, R.L., 2006. China in the Bronze Age. University of Pennsylvania Press, Philadelphia. Wan, C.-p., 1972. Shang Plan: Anyang qingtong rongqi de zhuzao ji qi jishu de fazhan (Part I: The Casting and Technical Development of Anyang Bronze Vessels). In: Chi, Li, Chia-Pao, Wan (Eds.), Gu qiwu yanjiu zhuankan di wu ben: Yinxu chutu wu shi san jian qingtong rongqi zhi yanjiu (Archaaeologia Sinica, New Series, no. 5: Studies of Fifty-Three Ritual Bronzes). Institute of History and Philology, Academia Sinica, Taipei, pp. 1–54. Wood, N., 1989. Ceramic puzzles from China's bronze age. New Scientist 121 (1652), 50–53. Xu, J., 1996. Summary of chapter 1 (in English). In: Xiating, Li, Zhiming, Liang (Eds.), Art of the Houma Foundry. Princeton University Press, Princeton, pp. 75–85. Yue, Z., 2006. 2000–2001 nian Anyang Xiaomintun dongnan di yindai zhutong yizhi fajue baogao [Excavation report of the bronze foundry at Xiaomintun Southeast during 2000–2001]. Kaogu xuebao 2006 (3), 351–384. Yue, H., Yue, Z., 2009. Shilun Yinxu Xiaomintun daxing zhufan de zhuzao gongyi he qixing (Casting techniques and vessel form of the large-sized in-situ mold found at Xiaomintun, Yinxu). Kaogu 6, 72–76. Yue, Z., Wang, Xuerong, He, Yuling, Jinqiong, Tang, Shishan, Niu, Fei, Gu, 2007. Henan Anyang shi Xiaomintun shangdai zhutong yizhi 2003–2004 nian de fajue (2003–2004 Excavation at the Xiaomintun Bronze Foundry site in Anyang, Henan). Kaogu 1, 14–25.

Conflict of interest There are no known conflicts of interest for any of the authors participating in the publication of this paper. References Aitcin, P.-C., 2008. Binders for Durable and Sustainable Concrete. Taylor & Francis, London. Anyang Work Station, Institute of Archaeology, Chinese Academy of Social Sciences, 2007. 2000–01 Excavation of the Shang Bronze Foundry Site at Xiaomintun Southeast in Anyang. Chin. Archaeol. 7, 32–39. Bagley, R.W., 1987. Shang Ritual Bronzes in the Arthur M. Sackler collections. A. M. Sackler Foundation, Washington, D.C. Bagley, R.W., 1990. Shang ritual bronzes: casting techniques and vessel design. Arch. Asian Art 43, 6–20. Bagley, R.W., 1996. Introduction. In: Li, X., Liang, Z. (Eds.), Art of the Houma Foundry. Princeton University Press, Princeton, pp. 69–74. Bagley, R.W., 2009. Anyang Mold-Making and the Decorated Model. Artibus Asiae 69 (1), 39–90. Chang, K.-C., 1980. Shang Civilization. Yale University Press, New Haven. Chase, W.T., 1983. Bronze casting in China: a short technical history. In: Kuwayama, G. (Ed.), The Great Bronze Age of China, A Symposium. University of Washington Press, Seattle, pp. 100–123. Chen, Z., 1986. Yinxu daofan jiqi xiangguan wenti (Ceramic molds from Yinxu and related issues). Kaogu 3, 269–277. Cheng, T.-K., 1960. Archaeology in China, Volume II, Shang China. W. Heffer & Sons, Ltd., Cambridge. Clayton, P., 1998. The Clay Lover's Guide to Making Molds. Lark Books, Asheville, North. Echin, P., 2009. Handbook of sample preparation for scanning electron microscopy and Xray. Springer, New York. Feng, F., Wang, Z., Bai, R., Hua, J., 1980. Shangdai qingtongqi shizhu jianbao [Preliminary report of the replication of casting Shang bronzes]. Kaogu 1980 (1), 91–94. Feng, F., Wang, Z., Hua, J., Bai, R., 1982. Yinxu chutu Shangdai qingtong gu zhuzao gongyi de fuyuan yanjiu [A replication study of the casting techniques of bronze guvessels unearthed from Yinxu]. Kaogu 1982 (5), 532–539. Freestone, I.C., Middleton, A.P., 1987. Mineralogical applications of the analytical SEM in archaeology. Mineral. Mag. 51, 21–31. Freestone, I. C.,Wood, N., and Rawson, J. 1989 Shang dynasty casting moulds from north China. In Cross-craft and Cross cultural Interactions in Ceramics. Ceramics and Civilization, vol. IV, edited by McGovern, P. E., Notis, M. D., and Kingery, D., pp. 253–274. The American Ceramic Society, Inc., Westerville, Ohio. Gettens, R.J., 1969. The Freer Chinese bronzes. In: Volume II Technical Studies. Smithsonian Institution Freer Gallery of Art, Oriental Studies, No. 7. Washington, D.C. Goldstein, J., 2003. Scanning electron microscopy and X-ray microanalysis, 3rd ed. Kluwer Academic/Plenum Publishers, New York. Hamer, F., 1975. The Potter's Dictionary of Materials and Techniques. Pitman Publishing,

13