Introduction

As an isomer of iron oxide, ε-Fe2O3 has several excellent properties such as ferromagnetic resonance in the millimeter wave band [1] and can be used as millimeter wave absorber when doped with Ga or Al [2, 3]. The most impressive is its high coercive field, which can be exceeding 2 T for pure ε-Fe2O3 [4, 5], and even be as high as 45kOe when doped with Rh [6, 7], making it a promising magnetic recording material. However, ε-Fe2O3 is a metastable phase, which will spontaneously transform into the stable α-Fe2O3 when crystal size exceeds 50 nm or the temperature is high enough [8,9,10]. After it was first discovered in 1934 [11], researchers were not able to synthesize it artificially until 1963 [12]. Using mesoporous SiO2 as the matrix is the preferred synthesizing method [10, 13,14,15,16,17], others such as synthesizing it in solution without matrix [18] or by pulsed laser deposition [19, 20] are also used.

Surprisingly, the recent upsurge of research on ε-Fe2O3 actually originated from ancient ceramics, a research field that seems to have little relationship with modern chemical synthesis. In fact, the first dendrites α-Fe2O3 in black spots of porcelains was found by Colomban and Sagon in a Vietnamese blue-and-white stoneware in 2004 [21], but the observation of ε-Fe2O3 in ancient ceramics was mainly in the past 10 years. In 2012, Liu et al. found nano-scale ε-Fe2O3 crystals in black-glaze ceramic shard unearthed from Qingliangsi kiln site [22]. In 2014, Dejoie et al. found iron oxide crystals up to several micrometers in hare’s fur strips and oil spot patterns on black-glaze Jian wares of Song Dynasty in Fujian Province, and confirmed that they are metastable phases ε-Fe2O3 [23]. These findings ignite the research upsurge of ε-Fe2O3 in ancient ceramics.

Since then, micrometer-scale ε-Fe2O3 crystals in different ancient ceramics were found in the oil drop ware of ** was roughly 1 × 40 μm2.

The XRD spectrum was obtained by Bruker D8 DISCOVER X-ray diffractometer. The X-ray used was Co-Kα with a wavelength of 0.178897 nm, and the minimum spot size was about 400 μm.

The micromorphology observation was conducted by MIRA3 field-emission scanning electron microscope (FE-SEM, TESCAN) equipped with EDS accessory (Genesis, EDAX, USA). Before observation, carbon layer with a thickness of about 10 nm was deposited to enhance the surface conductivity. In our experiments, back-scattering electron (BSE) detector was used, the accelerating voltage was 25 kV, the beam intensity used in imaging and EDS was 12 and 18 respectively, the corresponding spot size was about 5 nm and 40 nm, and the single point testing time for EDS analysis was 120 s.

The ceramic ware body’s thermal expansion curve was measured by DIL402C thermal expansion meter of NETZSCH. In our experiments, the sample size was 25 mm × 5 mm × 5 mm, the heating rate was 5 ℃/min, the sweep gas was nitrogen gas with a flow rate of 50 mL/min.

Results and discussion

Macroscopic information

Figure 2a shows the XRD spectrum of the brown area in the sample. Figure 2b shows the thermal expansion curve measured by push-rod thermal expansion method, and its firing temperature can be calculated to be 1180 ± 20 ℃ [41, 42]. The optical image and SEM back-scattering image of this sample are shown in Fig. 2c and d.

Fig. 2
figure 2

The XRD spectrum of the sample’s brown area, the thermal expansion curve, and the schematic diagram of the experimental regions in this paper. a The experimental XRD spectrum. The peak positions of relevant phases are shown by different symbols and the Miller indexes are labeled. Here the red, green, blue, cyan and magenta stand for α-Fe2O3, quartz, mullite, ε-Fe2O3 and Al-doped ε-Fe2O3, respectively. It can be seen that the ε-Fe2O3 and α-Fe2O3 coexist here. b The sample’s thermal expansion curve. c and d are optical and SEM back-scattering images of the sample’s brown area, and the regions of the following experiments are shown by red circles

Full size image

Through comparison with the database, we found that there were various phases in the analyzed region such as quartz, mullite ε-Fe2O3 and α-Fe2O3. Compared with pure ε-Fe2O3, Al-doped ε-Fe2O3 (the reference phase in the database is Fe1.71Al0.24Mg0.02Ti0.03O3, which was found in the ore) seems to be in better agreement with the experimental peak positions. It should be noted that we do not intend to claim the existence of Fe1.71Al0.24Mg0.02Ti0.03O3 in this sample, but only want to use it as evidence that the substitution of Al3+ into original ε-Fe2O3 will decrease the crystal lattice constants, and fit better with minor peaks in the experimental XRD spectrum.

It can be found in Fig. 2c and d that the background color of the brown area is dark brown, the boundary between brown area and the black glaze is bright orange, while there are many red brown regions in the brown area, whose brightness in the back-scattering image is significantly higher than the background, so it can be speculated that the iron concentrations in these regions are very high.

Typical crystal morphologies

Figure 3 shows the typical morphologies of the crystals in the sample and their Raman spectra. In Fig. 3e, four typical Raman peaks of ε-Fe2O3 crystals locate near 91 cm−1, 124 cm−1, 156 cm−1 and 238 cm−1 [29], which are indicated by green dash lines; four typical Raman peaks of α-Fe2O3 crystals locate near 225 cm−1, 295 cm−1, 410 cm−1 and 612 cm−1 [43, 44], which are indicated by red dash lines. It can be seen that the white feathery dendrites in 3(a) are ε-Fe2O3, which are located at the border between brown area and black glaze; the coarse granular crystals in 3(b) are located in the brown area, which are also ε-Fe2O3. The peak at about 295 cm−1 in spectrum (a) was lost in spectrum (b), which may be related to spinel phases such as maghemite and magnetite [44].

Fig. 3
figure 3

The crystal morphologies and Raman spectra in different sample regions. a–d come from region I in Fig. 2, and the “1” “2” and “3” in each subplot number denote optical images from Raman microscope, middle magnification SEM back-scattering images and high magnification SEM back-scattering images. e shows the Raman spectra obtained in the center of each subplot

Full size image

However, although the coarse and fine granular crystals at the center of 3(c) and 3(d) look very similar to that in 3(b), Raman spectra proved that they are α-Fe2O3, which indicates that these two phases cannot be distinguished simply according to the morphologies. It should be noted that the width of peaks at 225 cm−1and 295 cm−1 are broader than that of pure α-Fe2O3 crystals in the literature, which suggests that the Al.3+ ion may substitute into the lattice of α-Fe2O3 crystals [44].

Phase map**

In order to verify the distributions of ε-Fe2O3 and α- Fe2O3, Raman map**s were conducted by micro-Raman spectrometer in region I and region II, as shown in Fig. 4 and Fig. 5. After collecting the Raman spectra of each point, the software would automatically calculate and extract the principal components, which correspond to different phases.

Fig. 4
figure 4

Local Raman map** results in region I. a Local back-scattering SEM images in region I, where the orange square shows the Raman map** area. b Optical image corresponds to the orange square area shown (a). c Six principal Raman components obtained in the Raman map** of area (b). d Raman map** result of (b) area. In (c) and 4(d), red, green, blue and cyan represent α-Fe2O3, ε-Fe2O3, quartz and anatase respectively

Full size image
Fig. 5
figure 5

Local Raman map** results in region II. a Local back-scattering SEM images in region II, where the orange square shows the Raman map** area. b Optical image corresponds to the orange square area shown (a). c Three principal Raman components obtained in the Raman map** of area (b). d Raman map** results of (b) area. In (c) and (d), red, green and blue represent α-Fe2O3, ε-Fe2O3 and quartz, respectively

Full size image

Figure 4 shows the local Raman map** result of region I. The first four of the six principal components calculated correspond to the Raman spectra of α- Fe2O3, ε-Fe2O3, quartz and anatase respectively, as shown in Fig. 4c. The phases correspond to the other two unknown principal components are not directly given by database, but unknown 1 is probably a luminescence due to the presence of rare earths [43]. Comparing Fig. 4b and d, it can be seen that α-Fe2O3 and ε-Fe2O3 coexist in this region with irregular morphologies, and α-Fe2O3 looks brighter than ε-Fe2O3 in the bright-field image.

Figure 5 shows the local Raman map** result of region II. The three principal components calculated correspond to α-Fe2O3, ε-Fe2O3 and quartz respectively, as shown in Fig. 5c. It can be seen from Fig. 5b and d that α-Fe2O3 is still brighter than ε-Fe2O3.

Compositions of typical areas

Now, we have roughly clarified the morphologies and distributions of α-Fe2O3 and ε-Fe2O3. In order to obtain the clues of its formation mechanism, it is necessary to analyze the compositions of these phases and their surrounding materials. For this reason, we chose a series of testing positions to acquire the EDS spectra of α-Fe2O3 (PA), glass phase around α-Fe2O3 (GA), ε-Fe2O3 (PE), glass phase around ε-Fe2O3 (GE), large feathery dendrites ε-Fe2O3 (FE), glass phase around the large feathery dendrites ε-Fe2O3 (GF), common glass phase (CG), glaze (G) and body (B) on the sample surface. Several typical test areas have been marked in Figs. 3, 4 and 5, while polished cross-section samples were used to test the compositions of glaze and body. After statistical data processing, the experimental results are shown in Table 1, in which the uncertainty of the instrument is taken as 0.1%.

Table 1 Compositions of different sample areas (wt%)
Full size table

We observed that the large feathery dendrites all appeared at the boundaries between black glaze and brown area. Table 1 shows that both the SiO2/Al2O3 ratio and the CaO content of large crystals and their surrounding glass phase are between the corresponding parameters of body and glaze, which is consistent with our observation above.

In the brown area, the SiO2/Al2O3 ratio of α-Fe2O3 (2.6) and ε-Fe2O3 (2.0) are significantly lower than that of the body (3.6) and glaze (4.3), which indicates that the aluminum content is higher in the area where ε-Fe2O3 exists; the SiO2/Al2O3 ratio of glass phase near large feathery dendrites (3.8) and common glass phase (4.1) are between that of the body and glaze; while the SiO2/Al2O3 ratio of the glass phase near α-Fe2O3 (5.8) and ε-Fe2O3 (4.4) are higher than that of the body and glaze. On this basis, it can be inferred that when the iron oxide crystals were formed, the Si and Al elements in this system were redistributed, and the requirements on SiO2/Al2O3 ratio of ε-Fe2O3 (2.0) is more stringent. If we simply assume that the crystallization of Fe2O3 will affect the element distribution in the glass phase, it will be difficult to explain the selectivity to aluminum shown by α-Fe2O3 and ε-Fe2O3.

As mentioned in “Macroscopic information” Section, compared with pure ε-Fe2O3, doped ε-Fe2O3 (the chemical formula of the reference material in the database is Fe1.71Al0.24Mg0.02Ti0.03O3) seems to fit better with the experimental spectrum. According to the silicon-aluminum ratios in different regions given by EDS, it is natural to speculate that some Fe3+ may be substituted by Al3+ in ε-Fe2O3 phase. The radius of Al3+ is 50 pm, which is slightly smaller than that of Fe3+ (60 pm), and they both carry the same number of positive charges, therefore in theory Al3+ can enter the lattice of Fe2O3. In fact, in another work of ours to be published, it has been confirmed by DFT calculation that the lattice of ε- Fe2-xAlxO3 indeed exist stably, and the corresponding apparent color changes due to the change of system energy level. Previous literature also reported the substitution behavior of Al and Ti atoms in α- Fe2O3 crystals [25].

And then it needs to be pointed out that the crystallization of ε-Fe2O3 may require a relatively low CaO content in the system. In previous literature, the content of CaO in typical black glaze with ε-Fe2O3 crystallization is less than 8%, and the content of iron oxide is mostly 5–8% [24, 26, 28, 30, 35]. However, in this sample, the CaO content is less than 5% in the regions where ε-Fe2O3 crystals exist. On the contrary, the CaO content of the black glaze is close to 20%. Although the Fe2O3 content of the black glaze is also about 6%, there is still no ε-Fe2O3 crystallization.

In addition, we found that in the brown area, the mass fraction of Fe2O3 is more than 35% where α-Fe2O3 and ε-Fe2O3 exist, and is only about 17% where the large feathery dendrites ε-Fe2O3 are located, which may be due to the fact that the large feathery dendrites are very thin, while the Fe2O3 crystallized in the brown area are thicker. As previously reported, the Fe2O3 content of large ε-Fe2O3 feathery dendrites on the surface of black glaze given by SEM–EDS was also between 13 and 24% [35]. One possible explanation is that the iron crystallized in these brown areas may at least partly come from the inside of the body.

According to Robert Tichane’s monograph Ash Glazes, potassium carbonate will be absorbed by bisqued body, and then react in the body to make it more vitrified and to make it give up its iron oxide to the surface. We think that this also explain the reason why the K2O content was much higher in the region of “Huoshihong”[40], and why should K2CO3 be used to reproduce the brownish color on Bizen stoneware [33]. When chlorine exists in the body, the iron in the body will diffuse outward in the form of highly volatile FeCl3 and gather on the surface when heated [45]. From Table 1, we can easily find that only in the areas where ε-Fe2O3 and α-Fe2O3 exist can chlorine be detected, which partly proves the rationality of the outward diffusion and aggregation mechanism of FeCl3. Moreover, the content of K2O in these two crystals and their surrounding glass phases are also higher than that in the body and glaze. We believe that it is related to the firing process of Deqing kiln.

It is generally believed that the Deqing kiln in the Tang Dynasty did not use sagger but used lamination naked firing technique [46], which inevitably made wood ash and other substances fall on the unglazed surface with the flow of high temperature air. Yin et al. had systematically studied the proto-porcelain kilns in Deqing area, and found that the glassy surface (or kiln sweat) on the inner wall has a high potassium content—the mass fraction of K2O is generally above 5%, while the mass fraction of K2O in the kiln wall fragments is only about 2% [47]. This shows that the extra K2O must have come from the wood and other fuels used. The distribution of brown areas in our sample is irregular, but most of these areas appear on the unglazed side rather than on the bottom, which is consistent with the speculation that wood ash falling increases the local potassium content.

However, we cannot completely rule out the possibility that these brown areas come from a thin layer of body-protective glaze [48, 49], or are formed spontaneously during the firing process. As for the growing mechanism of ε-Fe2O3 crystals, scholars also have not reached a consensus. Kusano et al. regarded ε-Fe2O3 as epitaxial product on spinel substrates such as (Mg, Fe)(Al, Fe)2O4 and γ-Fe2O3, while Hoo et al. thought that ε-Fe2O3 was grown on radial mullite whiskers. These problems deserve to be further studied.

Conclusion

The ε-Fe2O3 was discovered in the ceramic ware of Deqing kiln in Tang Dynasty, which is the earliest known artificially synthesized ε-Fe2O3. Different from previously found large feathery dendrites ε-Fe2O3 on the surface of black glaze, the ε-Fe2O3 crystals in this sample are only located in the brown area between adjacent black glazes and coexist with α-Fe2O3. By analyzing the compositions of α-Fe2O3, ε-Fe2O3 and glass phases around them, we speculate that some Fe3+ ions may have been replaced by Al3+ ions during the formation of ε-Fe2O3 phase, and the crystallization of ε-Fe2O3 may require that the CaO content in the system should not be too high. In addition, the iron of α-Fe2O3 and ε-Fe2O3 in the brown area may partly come from the interior of the body. Both the potassium carbonate brought by wood ash in the kiln and FeCl3 diffused outward from the body contributed to the aggregation of inner iron on the surface.