The Use of Metal Nanoparticles to Produce Yellow, Red and Iridescent Colour, from Bronze Age to Present Times in Lustre Pottery and Glass: Solid State Chemistry, Spectroscopy and Nanostructure

The use of metal nanoparticles dispersed in an optically clear matrix by potters and glassmakers from the Bronze Age up to the present time is reviewed from the solid state chemistry and material science point of view. The nature of metal (gold, silver or copper), the importance of some other elements (Fe, Sn, Sb, Bi) added to control metal reduction in the glass in relation to the firing atmosphere (combined reducing oxidizing sequences, role of hydrogen and water) are considered in the light of ancient Treatises and recent analyses using advanced techniques (FIB- TEM, EXAFS,…) and classical methods (optical microscopy, UV-visible absorption). The different types of colour production, by absorption/reflection (red, yellow) or diffraction (iridescence) and the relationship between nanostructure (metal particle dispersion, layer stacking) and lustre colour are discussed. The very specific interaction between light and the metal nanoparticle makes Raman scattering a very useful "bottom up" technique to study the local glass structure around the metal particles as well as to detect incomplete metal reduction or residues tracing the preparation route, hence making it possible to differentiate between genuine artefacts and fakes.


Introduction
Potters have been using nanomaterials, namely clays, for more than 17000 years: most of the properties of traditional ceramics (notably a good shapability, low sintering temperatures and low production costs) stem from the fact that their raw materials -natural clays-are nanosized [1]: the water molecules lubricate the sliding of the clay nanoslabs against each other and the rheology is pH controlled; the nanometric size promotes ion diffusion, and hence reactivity and sintering. Sol-Gel routes mimic clay technology and extend its advantages to a variety of non-silicate compositions [1,2]. The invention of Ceramic Matrix Composites (CMCs) is also very old and derives its properties from natural nanosized mineral, asbestos, a rock yarn made of ~10nm diameter flexible fibres. The invention dates back to the Iron Age, in Corsica: women potters prepared utensils for cooking with a mixture of long asbestos yarns and clays [3]. The interest in this ancient technology is also old: for instance the first scientific study of these CMC materials was undertaken in 1838-1839 by A. Brongniart, the famous ceramist, glassmaker and solid state chemist who established the French National Ceramic Museum at Sèvres.
A new generation of engineered nanomaterials (grain size <100 nm) has been in development -if not already commercially available -for the last twenty years. Reducing the dimension of matter domains down to the nanometre scale confines the electronic and vibrational wave functions, which results in unique properties and opens a wide range of potential applications in different domains such as: i) optics and microwaves : pigments for the cosmetic industry, fluorescent markers and quantum dots, photonic crystals (multiplexing and switching in optical networks), quantum shards from Suse (Abbasid, 9 th century), c -e) Fatimid shards depicting a lion or a leopard (c) and a palmette (d, e) both excavated from Fustat (12 th century, Cairo, Egypt); for the later shard two orientations are shown; f) red Hispano-Moresque and g) Mudejar (16 th century, Andalusia) potteries (Sèvres Museum Collection); the later depicts the Latin HIS inscription, Iesus Humanitas Salvator, and after 180 d turn, an Arabic one, almulk, the Kingdom; h) and i) are modern red and gold shine lustres from Eva Hundum. (Photograph, Ph. Colomban). Dimensions: a,b,d-e ~5x5 cm 2 , c, g, h, I ~diameter ~12 cm, f diameter ~20 cm. however reported that Buontalenti, the famous glassmaker and potter of the Medici Dukes, reported that the colour of Roman alassonti vases changes with the observation angle [7]. This remark is consistent with the production of lustre glasses before Islamic Times, but, to our knowledge, no dated artefact has been found and analysed. Actually, known lustre glasses do not exhibit such a strong metallic shine as observed for lustre pottery (Figs 1 & 2), which questions the relationship between their preparation techniques and their micro/nanostructures.
Incorporation of elemental Ag at the glass surface from a precursor ("cementation") remains the usual way to obtain yellow coloured stained glass windows from the Middle Ages through the Renaissance to present times [7,11,12,28]. The production of red stained glasses by flashing/coating thin films (30 to 500 µm) of a Cu°-rich glass (Fig. 2) over a colourless glass is also dated from the Middle-Ages [29]. The very strong absorption of the light by the metal-containing film requires very thin layers of metal-containing glass (Fig. 3).  [97] The polychrome lustre decor on ceramics, an innovation of the Abbasid potters [21,22,[25][26][27][30][31][32][33][34] and thus the oldest optical device utilising nanotechnology, relies on a controlled dispersion of the metallic Ag° and/or Cu° nanoparticles (Table 3) within the upper region of the glaze [21,22,. Layers are formed from the colloidal Ag o /Cu o nanocrystals in the near-surface region (Fig. 3) of a tin (or similar multivalent metal)-containing lead silicate glaze, giving rise to an iridescent multicoloured reflective decoration. Recent studies have demonstrated that the presence of Ag 0 and Cu 0 nanoparticle "arrays" is responsible for this phenomenon. Table 1 summarizes the main places of production of studied lustre pottery. But the same studies have also exposed the complexities, with a large part of Cu and Ag atoms being oxidised and with a stacking of many metallic nanocrystal-containing layers acting as selective filters separated by a metal particle-free layer [21,22]. Such complex layering has also been observed in some butterfly wings (for instance Uraniidae) and beetles, in which a range of coloured scales/layers, with varying sizes and spacing, contributes to the variety of colours and iridescence ( [21,22,[61][62][63][64] and refs herein). Tailoring the sequence of more or less adsorbing/reflecting layers, i.e. the electromagnetic impedance of its layers is mandatory to control the light penetration from microwave to visible range [21,22,[65][66][67][68][69][70] as attempted for modern optical or microwave devices [67,68,[70][71][72][73][74][75].

Cassius purple and ruby glass
Cassius Purple, a dispersion of gold particles in glass is named after Andreas Cassius, who described it in his treatise De auro (1685). It was also known before this date as it was already mentioned by Johann Glauber and Andreas Libavius a few decades before and the process is well described in a fourteenth-century Bolognese manuscript of painters' recipes [76,77]. This was not the only recipe to obtain red colours from gold; eighteenth-century versions of Antonio Neri's Arte

112
Journal of Nano Research Vol. 8 a)

Fig. 2:
Examples of lustre Fatimid glasses (a, original bottle; b, a fake [114]) and of red copper Krich glasses (b,c). The samples are from b) Aube French Country (13 th century) and c) from Amiens Cathedral (13 th to 17 th century), respectively [100]. (Photograph Ph. Colomban). Lustre vetraria describe heating a gold powder until it turned purple and using this to create red-coloured glass [77]. The first preparation of ruby glass, is assigned to Bernard Perrot an Italian glassmaker working in Orléans ( [114] and references herein). Usually, gold nanoparticles are precipitated due to tin oxide from a gold solution in regal acid and then incorporated into the glaze precursor before firing [7]. The observation of both copper and gold in red Roman glass tessarae [8] and in the Lycurgus cup [6], however questions the above proposed date for "Cassius' purple" and the gold ruby glass could be a Roman Heritage.

Ancient recipes and treatises, from Middle-Ages to 19 th century and present times
Interest in lustre techniques is however very ancient [7]. Recipes are reported in some of the oldest known technological books [76-81, and references herein]. Understanding of the ancient recipes is not straightforward and a cause of debate. With the development of a scientific approach to ceramic technology during the 18 th and 19 th centuries many research studies were conducted to understand and recreate the techniques, in particular at the end of the 19 th century by Théodore Deck (1823-1891) at the Sèvres Factory [22,82] and were even a subject of teaching [7,11,12,83]. The copies of the lustre Alhambra vase depicting a gazelle couple, made by Th. Deck, are famous (three are conserved at the Sèvres National Ceramic Museum, the Rouen Ceramic Museum and the Musée des Arts Décoratifs of Paris). In Spain the tradition has continued up to now [84]. Contemporary glassmakers and potters still employ more complex colloidal metallic dispersions of gold, silver, platinum and copper to generate desired colours in decorative glasses and enamels [22] but also try to retrieve all the parameters necessary to recreate the ancient recipes, see further. As mentioned above, the earliest lustre pottery ever found was made in Iraq during the Abbasid Dynasty (9 th century) [22,27,[30][31][32][33][34]44]. Some scholars claim that a similar lustre also exists on glass vessels from the same period or slightly before [25][26] but artefacts are very rare and poorly documented. There is only one short scientific report on lustre glass composition of pieces excavated from Fustat, the ancient Fatimid's capital that has been published, to our knowledge [85]. The first lustre glass with a reliable inscription, now at the Islamic Art Museum, Cairo, was dedicated to an official person in charge during the sole year 773 AD [25][26][27], which is 20 to 50 years before the first (dated) lustre Abbasid pottery. These are the first examples of controlled nanotechnology for optical devices. It is thought that the beautiful gold-like iridescence was used to circumvent the prohibition by Islamic rule of making profane utensils of gold.
Some rare Abbasid lustre ceramics have a complex and fine decoration, the so-called polychrome lustre in which multi-coloured iridescence can be voluntarily present in the same artefact: samples have been excavated from Suze, Samara and from other Abbasid palaces, but the most known example is the Abbasid tiles, imported from Iraq, and placed in the mihrab of the Sidi Oqba Mosque in Kairuan (Tunisia) [22,30,57,58,69]. The Islamic production technology spread to Egypt, Persia, Muslim Iberian Peninsula (al-Andalus) during medieval times (>12 th ), following the expansion of the Islamic Kingdoms, and then to Italy (>15 th ) [22]. But the polychrome lustre technology (the depicted iridescent decor shine exhibits at least two colours for the same examination angle of the artefact) was lost less than a half century after its discovery. It can be seen as the work of a unique master Potter! The French potter Théodore Deck, who produced many different types of pottery replica [82], tried to recreate a polychrome lustre in ~1885. The basic constituents were Ag 2 S, CuS (plus HgS and Bismuth for some Spanish/Italian productions) [38,47,86]. Recent studies (Table 1) mainly on Hispano-Moresque and Italian productions, more rarely on Iranian, Fatimid and Abbasid ones and on Cu° or Ag°-containing stained glasses using UV-visible absorption [42], Transmission Electron Microscopy (TEM) [21,35,39,42,44,45,46,65], EXAFS (Extended X-ray absorption fine structure), SAXS and Rutherford backscattering spectrometry [30,87,53,54] clearly evidence the heterogeneity of the lustre film: the lustre can be described as a 0.2-0.5 µm thick layered film that contains metallic (spherical or not) nanocrystals of Ag or Cu (except in rare cases where doubts are possible, no alloying is observed) dispersed in a silica-rich glass (Table 4) and a glassy (alumina-enriched, lead-poor) metal-free outermost layer. Optical microscopy with convenient high magnification and lighting is however sufficient to detect the metal particle containing (multi) layers [21,37]. EXAFS confirms that a large part of the Cu content is present in the ionic form (Cu + and Cu 2+ ), as in glass [13], and not only as Cu° metal [87]. Similar conclusions were drawn from Raman scattering both for Cu and Ag elements in lustre pottery [37] and red [13] or yellow glass [28]. The roughness of the lustre surface was evidenced by AFM [41] or by confocal optical measurements [22]. Obviously, the optical properties are dependent on the particle composition (Table 5), size and distribution. Lustre is indeed the first controlled nanostructured grating that was made by technicians. In the past few years, a great deal of interest has been shown in the preparation of similar nanoscale metal clusters embedded in a dielectric (e.g. glass) matrix, which are potential materials for filters or active optical devices or microwave absorbing materials [66][67][68][70][71][72][73][74][75].
Analyses reported above are destructive and now not allowed on rare ancient (polychrome) lustre potteries. Even optical microscopy to be efficient requires a fresh sample fracture, and in many cases appropriate polishing is mandatory. They have been performed on "recent" productions in the form of small shards or on 2 nd quality samples. We recently demonstrated the potential of Raman (micro-) spectrometry for the non-destructive classification of very different ceramics and glasses [8,[88][89][90] as well as for a better understanding of the technology used to manufacture ancient objects [91][92][93]. Portable Raman instruments make it possible to study artefacts in secure places (Museum, Auction Gallery, etc.) [92]. PIXE and associated RBS techniques also provides reliable information on the sample composition. In particular composition profiles from the surface toward the core are very useful to discriminate between Au, Cu and Ag-enriched regions [93][94][95]. If usual instruments require small samples, big instruments like AGLAE at C2RMF [93][94][95] offer the unique possibility to study big objects. Alternatively the use of portable LIBS and XRF spectrometers is a convenient way to analyse big, non-moving artefacts. In particular LIBS instrument allows the quantitative study of the glaze layers [59].
Another approach has become popular, namely the study of modern replicas, prepared at the laboratory [42,43,47,49,96,97] or better by modern potters using gas firing [98]. This point will be discussed further.
The old potter reports, such as that written by Abû al-Qâsam [78], a member of an Iranian potter family, explain that the lustre is obtained by putting a mixture of clays and ochres, silver and copper salts (sulphides, sulphates, other salts…), vinegar and dregs in specific places for decoration control of an already silicate glazed piece of pottery [22,[78][79][80][81]. Deck' recipes are given in Table 3 [82]. The whole system is then heated in a dark red (~600-700°C) kiln in a mostly reducing atmosphere (oxidizing conditions at the cycle end are sometimes noted and it is clear that temperature/atmosphere control is complex [21,22]). Usually the glaze contains some lead (Table 4) but rare lead-free glazes are also reported [27,43,44]. For a silver-based yellow stained glass the technique is very similar [7,11,12,28]. For a red copper stained glass the technique must be different [97] because of the very strong absorption of Cu° containing glasses which requires a very thin metal-containing layer (Figs 2bc & 3ef). Furthermore, strong mechanical requirements requested for stained glass necessitate the glass piece to be quite thick (2-10 mm). Usually a multilayer glass coating is obtained by flashing a glass bubble: the glass bubble is blown to a very thin thickness (tenths of microns) and splashed over a glass plate to cover it. This technique is still used in some Venetian workshops for the preparation of gilded tesserae, where a very thin gold sheet (10-20 µm) is sandwiched between a thick support and the flashed glass coating, the Cartelina [93]. This technique was also used in the 1970's to fill the gap between the two magnetic parts of the first produced video recorder heads. Another potential technique is the dipping of the glass batch extracted from the melt with the tube, alternatively in copper-containing and copper-free molten glass. The blowing makes thinner the glass layers.  [22]), c) Eva Haudum' gold-like replica [98] and d) a modern materials developed for microwave adsorption [70]; e) optical micrograph of the section of a red flashed glass on its colourless support, f) detail of the red film; note the 3 layers with Cu° and the interlayers free of metal precipitate.

Micro and nanostructures
After the description of the main recipes in order to specify the stringent phenomena we will discuss their effects on the colour shine, the importance of the host glass/glaze matrix composition and nanostructure, the mechanisms of metal particle reduction by the combined action of furnace atmosphere and redox equilibrium, and we will try to classify the different microstructures. Finally we will consider the non-destructive Raman analysis of different glasses and glazes and the possible differentiation between genuine artefacts and fakes.
The experimental evidences are summarized in Figs 3 & 4: i) the dispersion of Ag° and Cu° nano precipitates in the glaze near its surface in a complex distribution which in many cases does not fit with the gradient expected for a steady temperature behaviour with a source of Ag or Cu at the sample surface, ii) the metal particle free surface layer is observed in most of the TEM studied lustres, iii) the higher refractivity of the uppermost surface film (less PbO) (e.g. in [35]), iv) a rather large distribution of the particle size, from 1 to 100 nm. The metal content ranges between 0.1 and 2 wt%. According to Deck's formula and corresponding Surface Plasmon Resonance peak (Table 5) the red colour (not shine) results from copper and the gold one from silver. Actually, this is more complex. We will see further that these colours are directly related to the wavelength of the plasmon resonance peak (Table 5) and it is important to say that the plasmon peak wavelength and shape depend on many parameters: particle size, inter-particle distances and electronic coupling, etc. [68,75,96] . This colour is different in its nature than the colour of the lustre iridescence.   Fig. 4. Nice TEM images are difficult to obtain because of the poor mechanics of many glazes; a new FIB device is very useful [21] for the in situ thinning of the lustre glaze. Four microstructures are observed: the metal nanoparticles close to the surface (Type 1) with or without a metal-free upper layer (Type 2), multilayer sequence (Type 3) and the sequence with inverted gradient size (Type 4). All these microstructures have been observed in the literature, and the best sheen is observed for the multilayer structures, as for opals [1,24] and beetles [21,63]. By comparison with the modern tailored nanocomposites for microwave penetration control or optical devices [70][71][72][73][74][75]99], it has been pointed out that the metal-free upper layer acts as adapted impedance layer to avoid reflection at the air/glaze interface and the successive bilayers (metal-containing, metal-free) improve the wavelength selectivity (i.e. in visible range for colour) [20,21], in other words a better diffraction of the light. It is obvious that the Type 1 microstructure does not differ from gilding, a coating with flux addition and that the so-called Meissen Gold Lustre belongs to this type ( Table 2).
A great effort has been made in the last few years to understand the different steps involved in lustre production by means of replica preparation [42,43,47,49,[96][97][98] and the comparison of the achieved microstructures with those of artefacts produced at different periods in various workshops. However, it is obvious that modern techniques of firing are different from the older ones, especially regarding the heating fuel and duration of thermal cycles. In the pastand this was true up to the first part of the 20 th century in many countriesthe duration of a firing cycle exceeded days and sometimes weeks (e.g. in Chinese Dragon kilns). Thus lower temperatures than those used by modern potters are convenient and made it possible to control the size of bubbles, precipitates, reactions in the glaze or at the glaze-body interface more precisely [19].

Colour by absorption (SPR) and by diffraction (lustre shine)
Colours. The absorption of silver, gold and copper particles in the visible range gives rise to a narrow peak, the so-called Surface Plasmon (electron) Resonance or SPR, a collective oscillation of the free electrons at the surface of metal nanoparticles, at respectively 400, 520 and 570 nm (Table the stained glass from becoming opaque, middle-aged flashed glass are constituted of stacks of thin films, alternatively clear or containing Cu° particles, with thicknesses of ~20 to 100 µm, coated on an optically clear glass plate (Fig. 2). Due to the preparation process of glass bubbling, the glass plates have a characteristic corrugated surface [100], which can be used by scholars to identify such glasses by touching with their fingers, eyes being closed for a better sensitivity! Usually the 19 th and 20 th century red glasses consist of a single thin film on the colourless support with a flat surface. In the later case, acid (HF) etching was used to mimic the colour heterogeneity of red middle-aged flashed glasses. Body: A : clay-based paste; S : highly siliceous body Metal particle size determines the position and shape of the SPR peak in a complex fashion. Furthermore the chemistry of silver is also complex. As for thallium, the lone electron pair promotes bonding between neighbouring cations and (Ag n ) m+ clusters, with m<n, can be formed [102,103], modifying the electronic absorption. The present knowledge on metal nanocomposites questions the former assignment of the red colour of the Lycurgus' cup to gold and silver particles [6]. However the gold content is very low. A large amount of copper was measured but assigned to the copper grill. It is likely that a part of the copper content was really within the glass and contributes to the dichroic effect. As described above, the colour of the lustre sheen is not directly related to the SPR but to the more or less gold diffraction of the light by the gratings formed by the metal particle-containing layers. This phenomenon can be described by Mie scattering [69] but actually consists in the well-known Bragg scattering by a poorly ordered structure [22][23][24].  [96]); lines correspond to blue, yellow, green and red laser wavelengths; b) UV-visible spectra recorded on red copper glass (19 th century Sèvres workshop) and on the corresponding colourless homologue (P3). The sample and the red film thicknesses are given.

The solid state chemistry
Glass and glaze host matrix composition. As a function of the raw materials used, three main compositions are observed for glasses prepared before the 19 th century and the Leblanc route of synthetic soda: i) Na-rich, (12-18 wt%) low Mg (<0.8 wt%)-content glasses prepared with natron from the Middle-East as soda source, ii) Na-Ca (5-10 wt%)-rich high MgO (3-6 wt%)-content glasses prepared from desert plant ash and iii) K-Ca-rich Middle Aged glasses prepared with forest plant ashes [89][90][91]100,[104][105][106]. Because of recycling of glass since Roman times, intermediate compositions are also observed. The composition of the analysed medieval flashed red glass of Fig.  5 belongs to the last group, i.e. a K-and Ca-rich composition [28,29,91]. On the other hand, most of the studied 19 th century red glasses are Na-rich compositions [97,100].
The series of recent analyses discussed above make it possible to give representative composition ranges of lustre glazes (Table 4). Note the large dispersion observed for lead content. In fact, as far as the glaze compositions are concerned, three main types are encountered. In some cases the composition is very close to that of Punic/Roman glass, with some addition of lead, as observed for red glass tesserae. Two other groups are recognized, the rare lead-free compositions, with Na 2 O and K 2 O ranging from 4 to 10 wt% and the mixed compositions where PbO content varies between 5 and 30 wt%, rarely 40 wt% [27,31,37,44]. Fustat Fatimid's samples have a medium PbO content [37]. The lead content is very low, some time nil in some Abbasid pottery glaze [37]. Tin oxide content close to 7-10 wt% corresponds to the threshold level for precipitation when the molten glaze is cooled down. On the contrary, huge 20 wt% content arises from the deposit of a tin-rich slip before the glaze. When analysed, the upper metal-free layer seems to have a much lower lead content [35], which is consistent with firing in a strong temperature gradient.
Reducing into metal particles: Redox short-circuit and firing atmosphere. The network of a glassy silicate is built of strong covalent Si-O bonds. Consequently the melting temperature of silica is high and the diffusion coefficients of oxygen and silicon atoms in silica are very low, even in the molten state. Fast oxygen molecule diffusion requires high temperature. Thus, the action of the kiln atmosphere on oxidation/reduction in a glass or a glaze requires some short-circuits to be efficient [21,22,66,[107][108][109][110][111].
Actually, the control of the kiln atmosphere cannot be independent of the temperature control. For instance, hydrogen is a very good fuel and its injection increases the temperature. Hydrogen is produced by the reaction of water with CO or (wood) coal but, on the other hand water excess, often associated with hydrogen production from green wood, will decrease the temperature. Ancient potters used green wood faggots to produce hydrogen in the kiln atmosphere but this also produces a large amount of water vapour, which makes it rather difficult to keep the temperature under control. The roles of hydrogen and water were extensively discussed by Malins & Tonge [111]. Fig.  3 shows a TEM image [98] obtained for a modern lustre, made in 2008 by Eva Haudum, a potter working in France [112]: Two metal-rich layers are observed which corresponds to the Group 3 multilayer lustre classification. This lustre has been obtained by 4 sequences of reduction (by CO) at 580°C followed by an oxidation, which leads to an increase of the temperature by about 100°C. The upper layer(s) free of metallic particles were obtained under oxidizing conditions. The need of reducing/oxidizing sequences as proposed before [21,22] is clearly demonstrated.
Obviously, for Type 1 microstructure where metal particles are very close to the glaze surface, redox go-between elements are not required and reduction/oxidation can be monitored by the kiln atmosphere.

Metal particles, their distribution and their preparation.
In a previous report [22] we proposed a description of the polychrome lustre formation according to the Abû al-Qâsem recipe, a mixture of dregs, vinegar, ochre, clays and silver/copper salts. This assumption was confirmed by some TEM measurements [21], which clearly demonstrate that the local surface temperature reaches temperatures much higher than expected from the kiln temperature mentioned in ancient recipes. This demonstrates that lustre cannot be obtained in constant temperature and atmosphere. Ancient al-Qâsam recipes indicate that the addition of wood to obtain smoke should be carefully controlled. The technique is called 2-fire colour. An explanation has been proposed [21,22]: combustion of organic residues from the lustre precursor paste at the sample surface gives incandescence which is well visible for samples in a dark red kiln (~600°C). This offers a control of the surface temperature in different places of an artefact, as required to control a complex colour design through that of the metal particle size and size gradient. An alternative route is a flash heating by entering the sample in the fire during a limited time in order to heat only the sample surface. This technique was used by Islamic and Roman glassmakers, as demonstrated by their use of enamelled glass with a composition very similar to that of the body ( [113,114] and references herein).
Lead oxide is easily dissolved at 80-100°C by the acetic acid from the vinegar and dregs mixture and therefore the roles of these reagents added to the clay/ochre are clear. Namely, they increase the porosity of the top surface of the glaze to maximize the reaction with the molten Ag-or Cu-rich melt. The clays-salts mixture hinders the fast evaporation of acetic acid, which improves the dissolution of the lead oxide-based glaze. The near surface region of the glaze becomes porous. Further heating causes the melting of silver and copper salts entering the meso-porosity formed by the action of the acetic acid, and their ions diffuse to some extent into the glass network. The low temperature (200-400°C) ion-exchange between a solid open framework, like that of a glassy silicate, and a molten nitrate is well documented [115][116][117]. Metal particles form by coalescence or liquid separation between the molten glaze and the metal. This is consistent with the spherical metal particle shape [22,35] and size distribution, and is close to that observed in sol-gel prepared nanocomposites by H 2 reduction [70][71][72]. This causes a decrease of particle concentration (gradient) from the sample surface (the source) toward the pottery body centre. Effectively such a microstructure is observed in Mesopotamian (Fig. 3), Italian and Moresque monochrome lustre as well as in some modern reconstructions. The multilayer structure of some Fustat (Fatimids, 11 th -12 th century) and Termez ("Iran", 13 th -14 th century) samples [22,35], differs from the other lustres (except for some instances, Mesopotamian /Abbasid samples from [45,118]) and may have required a more complex synthesis. Furthermore, the worm-shape of the Ag precipitates in some samples [22] proves that the melting temperature of silver (~960°C) was reached in some spots close to the sample surface. The observation of facetted crystals also proves that a temperature very close to 1000°C was achieved at the sample surface. Diffusion controlled phenomena cannot lead to such worm-shapes or facetting. This temperature causes the complete volatilisation of PbO from the upper glaze as the melting temperature of PbO is ~880°C. This temperature corresponds to that required for modern (i.e. acetic acid free) lustre techniques [22,119].
The lead diffusion in the glaze makes that residual acetate still remains at the glaze surface. Acetates are thermally rather stable and temperatures higher than 300°C are required to decompose them in carbon-rich residues under a reducing atmosphere. However, if a sufficient amount of oxygen is added after formation of the carbonaceous residues i.e. above 500°C, a glowing phenomenon can be observed similar to what is usually observed in some sol-gel synthesis routes [2,120]: acetate residues burn and the surface temperature of the ceramic can reach a temperature of 1000°C or more, as required to melt Ag and Cu. The high crystalline quality of metal nanoprecipitates is also consistent with intense local heating [21]. Such a phenomenon could explain the special composition of the lustre surface: less PbO because its evaporation is very active above 900°C, more Al 2 O 3 because a reaction with clays, from the precursor mixture, is also possible and no metal precipitates because of surface oxidation. This also explains the decrease of the particle size from the surface to the pottery core, i.e. with the temperature gradient within a layer. The composition of the mixture (more or less vinegar and dregs, Ag/Cu content…) will control the surface temperature as well as the embedded metal dispersion and allows one to set the final lustre colour in different places of the same item. The strong temperature gradient arising from the surface residue combustion explains why the precipitates located just below the glaze surface have a larger size and also the formation of elongated particles. The large size gradient for metal particles fit well with a flash temperature which prevents homogenisation. It is clear that wetability and metal/molten silicate surface energy also may control the particle shapes: separation of two liquid phases between the molten metal and the silicate melt is another possibility. However, this does not explain the multilayer pattern. This special pattern involves one, two or more successive deposits of the metal source mixture and/or associated repeated flash reduction/oxidation conditions. However, only the surface exothermic combustion can explain the achievement of two or more coloured types of lustre on the same artefact, as achieved by Abbasid potters [21,22,58]. This multi-step route allows the preparation of polychrome lustre glazes (i.e. different lustre colours), on the same artefact, by control of the temperature generated on different places on the pottery by the acetate residues combustion.
To summarize, the elongated shape of the bigger particles can be explained by an anisotropic flash heat. In contrast, a homogeneous temperature will give coalescence of the globular particles by Ostwald's ripening process. The production of the flash temperature can be explained by i) the exothermic combustion of the surface organic residues; this technique allows to control, empirically, the flash temperature in different spots on the sample; ii) by in-and out-taking of the pottery/glass in/from the kiln, as usually done by glass-makers for enamelling and also today by modern potters producing monochrome lustre pottery; iii) by multiple modifications of both temperature and atmosphere (H 2 , CO, O 2 ), as experimented by Eva Hundum replicas, as long as the source of metal element at the sample surface is not consumed. This technique was actually already reported in Piccolpasso's treatise [80]. He says that big broom fagots must be added ones after the other during firing. As recently explained by Allan Caiger-Smith, a modern lustre potter [121], the big size of the fagots requires the end of their burning before to add a new one, and thus an oxidizing sequence takes place between each reduced one; iv) or by a combination of these actions. Obviously lustre pottery and glass are a craftsman's masterpiece!

Non-destructive analyses
Host matrix and the nano-particles. Among the non-destructive available methods, Raman spectroscopy is one of the most versatile techniques, especially for the analysis of glassy and nanosized materials [4,122]. Raman spectroscopy probes molecular and crystal lattice vibrations and therefore is sensitive to the composition, bonding, chemical environment, (nano)phase, and (nano)crystalline structure of the sample material [4]. First, the micronic resolution of common optic instruments is accessible because a laser is used as the excitation source. The signal intensity depends on the laser wavelength (high with violet excitation, very poor with IR) and of the coupling between the laser light and electronic levels of the analysed matter, in other words of the colour of both: thus for a lustre material strong differences are expected for laser excitation below the SPR (absorbent material) and above (transparent material, Fig. 5). Note that in the former case the Raman signal will be very weak because of the huge absorption, both of illuminating and scattered light (Fig. 7).
All applications in the science, art and technology of glass, glazes and enamel consist of a controlled modification of the three-dimensional Si-O network by the replacement of Si 4+ covalently bonded atoms by ionic bonded ones. This decreases the number of Si-O bridges and the connectivity of the network, i.e. modifies the glass nanostructure and associated properties such as the melting temperature, viscosity, thermal expansion…. The SiO 4 tetrahedron is a strong chemical (silicate melts remain polymerised) and vibrational entity and the Raman signature reveals the different SiO 4 conformations and their relationship with the signature of parent crystalline homologues: nesosilicates built of isolated SiO 4 tetrahedra, soro-silicates of tetrahedras sharing a bridging oxygen … up to tectosilicates where oxygen atoms are shared between adjacent tetrahedra [88,89 and references herein]. Consequently, the Raman spectra of glasses were decomposed following a model which considers the two main components, the Si-O bending and stretching bands at ~500 and 1000 cm -1 as multiplets enveloping curves of the series of Gaussian-shaped components associated at each type of SiO 4 tetrahedron (Q n components), linked by 0 (isolated SiO 4 tetrahedron), 1 (two tetrahedral sharing one oxygen to form Si 2 O 7 entity), 2, 3 or 4 common oxygen atoms. The transition from a highly connected tetrahedral structure (such as quartz and amorphous silica) to a weakly or un-connected tetrahedral units (such as CaSiO 3 wollastonites or lead-based glaze) as caused by the addition of fluxing agents is reflected in the Raman spectrum by modifications of the intensity of the Si-O stretching multiplet (centre of gravity ~950 to 1150 cm -1 ) vs. the bending one (centre of gravity ~500 to 600 cm -1 ) [88]. The relationship between the Raman index of polymerisation (Ip = A 500 /A 1000 with A being the area under the Raman band), the glass composition and the processing temperature is now well documented [28,37,88,89,92,96,[123][124][125][126][127]. For instance this has been used to classify silver-containing yellow stained glass replicas [28] and ancient red or yellow stained glasses [96,100]: for instance a clear relationship was established between the polymerization index, the Area of Q n components and the b* colour coordinate as a function of the silver content and its degree of reduction [28].  Fig. 6: a) Raman spectra recorded on the Cu-containing model Eu-doped glass with same amount of copper as Cu° nanoprecipitate or Cu 2+ ions (after [96]). b) Details presenting the low wavenumber region (cluster & ion translational modes) of the model Eu-doped Cu° glass and of two ancient lustre glazes, the Fustat Fatimid red palmetto shard (Fig. 1d-e) and the Suse Abbasid polychrome shard (see [37]).
We will take an example to show the potential of the technique. Fig. 6a compares the Raman signatures obtained for the same model glass (a Na-rich composition [96] where the copper reduction is controlled by Eu ions doping) with 0.25 Cu wt% in the following different forms, Cu°, Cu + (not shown) and Cu 2+ . Differences are obvious, both on UV-visible absorption spectra (Fig. 5a) Journal of Nano Research Vol. 8 123 Wavenumber / cm -1 Fig. 7: Raman spectra recorded on the red flashed glasses (13 th to 19 th century) using different laser wavelengths in the red and colourless places. A baseline has been subtracted for spectra recorded on coloured glass because the huge absorption of the light and the very poor Raman intensity.
and Raman signature: a narrow SPR is observed for Cu° containing, red glass although absorption is low for light-blue Cu 2+ -containing glass sample and nil for Cu + -containing glass [96]. Fig. 7 compares the spectra recorded on copper red stained glasses from different origins: ancient middleage glass pieces from the Amiens Cathedral (Picardie), assigned to be made between the 13 th and the 17 th century, from a ancient church of Aube Country, close to Reims city, the place where Kings of France ware sacred and 19 th century glass produced at the Sèvres Brongniart Factory, using various laser excitations. The 406 nm laser spectra were recorded using macro configuration on the glass support. Similar spectra are obtained with micro configuration on the copper free layer sandwiched between copper-rich ones. The differentiation between K-rich (with ~950-1050 to 1090cm -1 doublet [89]) and Na-rich glass (strong 1090 cm -1 peak [89][90][91]) is obvious. As noted for the model Eu-containing copper glass, the Raman signature of Cu° containing glass differs strongly from that of the host matrix : the relative intensity of the Si-O bending multiplet vs. the stretching is modified, i.e., the polymerisation index is displaced, indicating that the environment of the metal particle is different from the host matrix. Similar conclusions have been made by S. Pérez et al. for silver-containing yellow glass [28]. Additional bands are observed when analysing the red layers : in many cases the signatures of haematite (-Fe 2 O 3 ) with main peaks at 225, 290, 410, 610 and 1310 cm -1 [128] and that of carbon (1350-1600 cm -1 doublet) [4] are observed. The formation of haematite as a result of the Fe 2+ ion oxidation in Fe 3+ promotes the reduction of copper ions in metal (see equation 8). Obviously the modification of the glass Raman signature is stronger with green excitation than with red one because the red wavelength (632 nm) is higher than the SPR peak (565 nm, see Fig. 5 and Table 5), which avoids the resonance phenomenon and hence the sensitivity to the metal particle environment. With 532 nm laser excitation the coupling with the electronic SPR is quasi maximal and in this case the Raman scattering probes only the Si-O bonds interacting with the Cu° nanoparticles (Fig. 6). Note that the very strong absorption of the light by Cu° containing glass makes the recording difficult (both the illumination and scattered light are strongly absorbed) and gives rise to green fluorescence which overlaps with the Raman spectrum [100]. Baseline subtraction is required to compare the spectra and this hinders a good measurement of the band intensity, as possible for model glass where the Cu content is limited and the particle distribution homogeneous [96].  Fig. 8 : a) Raman spectra recorded on the brown "lustre" decor of the bottle depicting a bird (Fig. 2) and on the Arabic inscription [114] of the glass artefacts assigned as Fatimid production (Fig. 2), b) Raman spectrum recorded in some place of a Fatimid lustre pottery, after [37]).
Analysis of the very low wavenumber region (< 80 cm -1 ) allows to characterise the mobile cations having a large polarisability, Ag + and for some extent Cu + /Cu 2+ ions: an example is given in Fig. 6 in model glass and in ancient Fatimids (Fustat) and Abbasid (Suse) lustres [37,96]. Similar spectra have been reported for silver yellow glasses [28]. The assignment of these modes remains subject of some debate: however silver ions can form (Ag n ) m+ clusters with mixed ionic and covalent character and the mass effect can displace the wavenumber of the Translational mode toward lower values [96,102,103]. Part of Cu and Ag element is however oxidized [29,37]. Some papers however assign the observed low wavenumber features to size effect of Ag° metal "nanoparticles" [129][130][131][132] and ways to discriminate between overlapping contributions remain a source of debate.
Differentiating genuine artifacts and forgery. Fig. 2 shows two glass artefacts, a bottle and a cup, assigned to Fatimids production. Both present some iridescence due to the microcracking of the glass as observed for ancient glasses, especially those recovered from excavations (this iridescence is not sheen and arises from multiple reflections at the glass-air interface). Fig. 8 compares the Raman spectra obtained on the brown lustre: for the glass bottle depicting a bird, a typical Fatimid's topic, a peak at 223 cm -1 is observed although a 300 cm -1 peak (with its overtone at 600 cm -1 ) is recorded on the Arabic inscription of the cup [114]. The latter peak is characteristic of a CdS phase, a modern chemical compound [133] and the observation of the overtone indicates a resonance spectrum. Actually the brownish colour fits better with that of CdSe and it is not possible to exclude the presence of CdSe mixed with CdS (the very small peak at ~250 cm -1 corresponds to the signature expected for CdSe [134]). We can thus conclude that at least the Arabic inscription is not original. As noted by Carboni, Pilosi and Wypyski in the Kingery's Series book on Prehistory and History of Glassmaking [135], "the high cost of medieval Islamic gilded and enamelled glass may also have encouraged nineteenth century collectors and glassmakers to 'restore' or 'embellish' authentic pieces". The Raman technique offers some tools to recognize these embellishments and restorations. On the contrary the 223 cm -1 peak observed on the bottle corresponds to As 4 S 4 [136], a compound used for a long time as a pigment, compatible with the raw materials of the lustre precursor paste. It can be a residue of the sulphur mixture used to produce the lustre.
Obviously, the lustre on glass appears different from the "true" lustre of Abbasid and Fatimids potteries and thus the relation between both techniques claimed by some authors [25,27] may be questioned. The lack of TEM examination does not permit us to go deeper in the discussion. In some places on the lustre glaze (Fig 8b gives the example of the Fatimids shard with palmetto decor) a single 390 cm -1 narrow peak is observed [37]. This could correspond to Pb 3 O 4 [137] but other characteristic components of this compound are not observed. Such single peak may better correspond to a resonance Raman spectrum, of a coloured materials (a sulphur ?). Studies on lustre glass are very rare [85] and this deserves further analyses.

Conclusion
The review of the present knowledge on ancient materials in which the size and distribution as well as the reduction of copper and silver metal nanoprecipitates were controlled shows the high level of empirical control achieved by ancient potters and glass makers. It is obvious that different routes have been used. The number of studied samples remains limited: for instance no real data are available by TEM for Cassius purple glaze and for red flambé potteries.