Several binary systems, such as MO_x-Al₂O₃ [[1], [2], [3], [4]], MO_x-Nb₂O₅ [[5], [6], [7]] and MO_x-TiO₂ [[8], [9], [10], [11], [12]] (M = rare earth, alkaline earth metal, etc.) can be vitrified without adding any traditional network-former oxides such as SiO₂, GeO₂ and B₂O₃ by containerless processing. Among these systems, BaTi₂O₅ (BT₂) glass has attracted wide attention because of its unusual dielectric, ferroelectric and optical properties, which make it a very promising candidate material in various applications. In 2006, Yu et al. first successfully vitrified BT₂ spheres with a diameter of 2 mm using an aerodynamic levitation (ADL) furnace and reported an unusual, enormous jump of dielectric constant during its crystallization process [8]. Masuno et al. reported that BT₂ glass has a refractive index > 2.1 and a large Abbe number of 21.5 [10,13], suggesting that it can be used as a high-refractive-index glass in optical components such as digital cameras, smart cellphones, endoscopes, and next-generation disk systems.

The huge prospect for industrial application has stimulated intensive research efforts on optimizing the properties of BT₂ glass by chemical doping or non-stoichiometry. For example, Er-doping onto the Ba-site was reported to enhance the thermal stability and the refractive index of BT₂ glass [14]. Ca-doping on the Ba-site was reported to improve the optical properties of BT₂ [15]. Although doping is effective to tune the properties, it also has a significant impact on the glass-forming ability of BT₂. For example, Sr- or Mg- doped BT₂ was found to have a narrow glass-forming region that only 5 % of Sr or Mg can be introduced to replace Ba [15]. Furthermore, a very small variation in the Ba or Ti nonstoichiometry can influence the glass-forming ability of BT₂ [16]. Improvement of the properties may be obstructed by the low solubility of dopants in BT₂ glass. Therefore, it is crucial to establish an in-depth understanding of the effect of doping on the glass-forming ability of BT₂.

The network skeleton of BT₂ glass is constructed by highly distorted [TiO₅] polyhedra connected with both corner-shared and edge-shared oxygen [13]. This makes BT₂ unique from other titanate-based glass, the skeleton of which is usually built up by a combination of [TiO₄] and [TiO₆] polyhedra [17,18]. Previous studies on glasses based on traditional network-former oxides (i.e., SiO₂, GeO₂, B₂O₃) suggest that doping on the network-modifier site has limited effect on the glass skeleton structure, whereas doping on the network-former site has a direct impact on the bone structure and thus may play a critical role on the glass-forming ability. Therefore, dopant is usually introduced to the network-modifier site to tune the properties while maintain the network-former bone structure untouched. However, the above doping strategy may not be applicable to novel glasses based on TiO₂ as 5 % Sr or Mg doping on the Ba-site can crystallize BT₂ and the effect of doping on the Ti-site, limited to the authors’ knowledge, has not been reported yet.

Here we study the effects of La- and Nb-doping on the glass-forming ability of BT₂. La and Nb are chosen as dopants based on the following considerations: (1) The ionic sizes of La³⁺ and Nb⁵⁺ are 1.06 and 0.64 Å, respectively. Therefore, La incorporates into BT₂ by replacing Ba (1.36 Å), and Nb by replacing Ti (0.61 Å). This enables us to compare the effect of doping site on the glass-forming ability of BT₂. (2) La₂O₃ and Nb₂O₅-based oxides are reported to vitrify by containerless processing. La and Nb in BT₂ are therefore expected to have a wider glass-forming region (as opposed to Mg and Sr, which are reported to have a narrow glass-forming region of <5 % [15]). A relatively wide glass-forming region is beneficial to revealing the structural evolution and comparing the effect of doping site, as well as minimizing experimental errors such as composition deviation from the nominal value. (3) As La₂O₃- and Nb₂O₅-based oxide glasses possess excellent optical properties, it is expected that both La- and Nb-doping are desirable to enhance the optical properties of BT₂. It is necessary to establish the glass-forming region of La- and Nb-doped BT₂ prior to investigating the optical properties. We therefore prepare La- and Nb-doped BT₂ with compositions of (Ba_1-xLa_x)Ti₂O_5+δ and Ba(Nb_yTi_1-y)₂O_5+δ to unveil the glass-forming region, and study their structural evolution with an attempt to establish a preliminary understanding between the doping-site and the structural arrangement (i.e., Ti-O bond length, [TiO_n] fraction, etc) of BT₂ glass. The physical and thermal properties of La- and Nb-doped BT₂ glasses are also investigated to facilitate understanding of the glass structure. Our results reveal that La-doping on the network-modifier site and Nb-doping on the network-former site both have significant impact on the bone structure of BT₂ glass, which may provide useful guidelines for doping strategy design to achieve desirable properties required by various applications.

Highly pure BaCO₃ (Aladdin, 99.95 wt%), TiO₂ (Aladdin, 99.99 wt%), La₂O₃ (Aladdin, 99.999 wt%) and Nb₂O₅ (Macklin, 99.99 wt%) powders were used as starting materials. Appropriate amount of each powder was weighed according to the nominal compositions of Ba_1-xLa_xTi₂O_5+δ (x = 0, 0.01, 0.02, 0.05, 0.07, 0.10, 0.20, 0.30, 0.40) and Ba(Nb_yTi_1-y)₂O_5+δ (y = 0, 0.02, 0.05, 0.08, 0.10, 0.20, 0.30, 0.40, 0.50, 0.60, 0.70, 0.80, 0.90), and mixed thoroughly with an agate mortar and pestle in ethanol. The mixture was pressed to tablet under a hydrostatic pressure of ∼200 MPa and then sintered at 950 °C for 12 h in air. After sintering, the tablet was chopped to small pieces. One piece of approximately 12 mg was used as specimen and placed on the nozzle of the ADL furnace and levitated by O₂ gas flow. The gas flow rate was monitored and controlled by a mass-flow meter. A high-resolution, charge-coupled device (CCD) video camera equipped with a viewing screen was used to monitor the levitation situation. The specimen was melted by a 100 W CO₂ laser and the temperature was measured by an infrared pyrometer. The temperature of the specimen was maintained ∼200 °C above the melting point for a few seconds to ensure the homogeneity of the melt. Once stable levitation was reached, the laser power turned off was instantly and the specimen was cooled down rapidly at a rate of hundreds of Celsius degrees per second. For each composition, dozens of glassy spheres were prepared for characterizations.

The weight of the glassy sphere was measured by an analytical balance (AUW120D, SHIMADZU) with a precision of 0.01 mg and the diameter was measured by a micrometer caliper (261-101A, Guanglu) with a precision of 0.01 mm. Density of the obtained glass sphere was calculated by:

where m and D are the weight and the diameter of the sphere, respectively. For each composition, 20 spheres were measured to obtain the mean density and the associated error. Subsequently, the molar volume V_M of each composition was calculated by:

where M₀ is the molecular weight for a 1 mol cation [19]. The glass transition temperature T_g and the first crystallization temperature T_x1 were determined using differential thermal analysis (DTA; NETZSCH STA499F3) from room temperature to 1000 °C at a heating rate of 20 °C min^-1. Raman spectroscopy measurements were carried out using a 532 nm Ar laser line in confocal Raman microprobe system (inVia Qontor).

The glass forming regions of (Ba_1-xLa_x)Ti₂O_5+δ and Ba(Nb_yTi_1-y)₂O_5+δ, together with the digital camera photos of the prepared spheres are presented in Fig. 1. Based on the criteria that: (1) no recalescence occurs during cooling; (2) the sphere is colorless and transparent, the glass-forming regions for La- and Nb-doped BT₂ are identified as 0 ≤ x < 0.1 and 0 ≤ y ≤ 0.4, respectively. This suggests that Nb has a higher solubility in BT₂ glass than La. The ionic size mismatch between Nb⁵⁺ (0.64 Å) and Ti⁴⁺ (0.61 Å) is much smaller than that of La³⁺ (1.06 Å) and Ba²⁺ (1.36 Å), which is usually considered as a plausible reason for the distinction in the solubility. However, it should be pointed out that the ionic size mismatch cannot explain the dramatic difference in the solubility of Mg, Ca and Sr in BT₂. Previous study [15] shows that the glass-forming regions of Ba_1-xA_xTi₂O₅ (A = Mg²⁺, Ca²⁺ and Sr²⁺) are 0.05, 0.90 and 0.05, respectively, although the ionic size of Sr²⁺ (1.18 Å) is closer to the Ba²⁺ than Ca²⁺ (1.00 Å). Therefore, other possible factors that contribute to the distinct glass-forming region in La- and Nb-doped BT₂ should be considered.

The composition dependence of density ρ, and molar volume, V_M, of La- and Nb-doped BT₂ is shown in Fig. 2, where both ρ and V_M vary linearly with the molar fraction of LaO_3/2 (x) and NbO_5/2 (y). For La-doped BT₂, the ρ-x and the V_M - x relationships are described by:

and

respectively. The increase in density could be due to the substitution of BaO (153.326 g mol^-1) by heavy oxide LaO_3/2 (162.91 g mol^-1). The decrease of V_M with increasing x is attributed to the ionic radii difference between Ba²⁺ (1.36 Å) and La³⁺ (1.06 Å).

For Nb-doped BT₂, the ρ - y and the V_M - y relationships are described by:

and

respectively. The increasing density with increasing y is also attributed to the replacement of light oxide TiO₂ (79.89 g mol^-1) by heavy NbO_5/2 (132.91 g mol^-1). The reduced V_M with increasing y may also originate from the difference in the ionic radius between Nb⁵⁺ (0.64 Å) and Ti⁴⁺ (0.605 Å). Furthermore, the replacement of Ti⁴⁺ by Nb⁵⁺ develops more non-bridging oxygen than bridging oxygen, which may flayer-up the glass system and thus increases its molar volume [20].

DTA curves of La- and Nb-doped BT₂ glasses are exhibited in Figs. 3(a) and 4 (a), respectively, from which the glass transition temperature T_g and the first crystallization peak temperature T_x1 for each composition are obtained and plotted as a function of x and y in Figs. 3(b) and 4 (b), respectively. The difference between T_g and T_x1 (ΔT = T_x1 - T_g) is used as a measure of thermal stability of the glass. ΔT - x and ΔT - y relationships are shown in the inset figures in Figs. 3(b) and 4 (b), respectively.

The DTA heating curve for BT₂ shows characteristics of an endothermic anomaly corresponding to the glass transition, and two exothermic peaks corresponding to the first and the second crystallization temperatures, as shown in the inset figure of Fig. 3(a). For La-doped BT₂ glasses, these main features are retained. Fig. 3(b) shows that T_g, T_x1 and ΔT all present a “volcano”-type variation with increasing x, and their maximum values are reached at x = 0.05. ΔT increases from 42.8 °C for undoped BT₂ to 47 °C for x = 0.05, indicating an enhanced thermal stability of BT₂ by La doping.

For Nb-doped BT₂, the DTA curves show a dramatic change with increasing Nb-doping level, y, as shown in Fig. 4(a). With increasing Nb-doping level, the second crystallization peak is weakened and finally disappears at y > 0.20, indicating that a different structural arrangement may be induced when Nb-doping exceeds the critical level (y = 0.20). Variation of T_g, T_x1 and ΔT with increasing Nb-doping level, y, is presented in Fig. 4(b). T_g increases monotonically from 697.7 °C for undoped BT₂ to 730 °C for y = 0.20 and then decreases slightly with further increasing y. T_x1 also increases a monotonically in the composition range of 0 ≤ y ≤ 0.20, and then shows a sudden drop for y = 0.30 and 0.40. ΔT shows a complicated variation with increasing y but a maximum value (47.8 °C) is also achieved at y = 0.20. The decrease of ΔT with an further increase in y indicates that high level of Nb on the Ti-site is not beneficial to the thermal stability of BT_2, which is possibly related to the fact that Nb₂O₅ could not form skeleton network solely in the BaO-(1-y)TiO₂-yNbO_5/2 system as observed widely in Nb-doped phosphates and silicates glasses [19,[21], [22], [23], [24]]. Nevertheless, it can be concluded that an enhancement of the thermal stability can be achieved by appropriate amount of Nb-doping on the Ti-site.

The above results show that both La doping on the Ba-site and Nb doping at the Ti-site can be beneficial for improving T_g and T_x1 of the BT₂ glass. At the optimum doping level, an enhancement of 6 °C and 32.4 °C in T_g, as well as an enhancement of 10.2 °C and 37.4 °C in T_x1, is achieved by La and Nb doping, respectively. It’s obvious that the Nb doping has a much stronger effect on the thermal properties than the case in La doping. This is because Nb doping on the Ti- site can directly influence the Ti-O bonds and the [TiO_n] polyhedra, which are considered as the network-former of BT₂ glass. In terms of thermal stability, La doping on the Ba-site shows a simpler regulatory effect than Nb doping on the Ti-site: La doping, as well as other Ba-site dopants such as Ca and Er reported by previous studies [14,15], can improve the thermal stability of BT₂ glass and possess a optimum doping level in the middle of the glass forming region; Nb doping on the Ti-site, on the contrary, possesses no simple positive-or-negative influence on the glass thermal stability with increasing doping level. Furthermore, although Nb-doping shows a much more significant effect on T_g and T_x1 than La-doping, it shows similar effect on the maximum enhancement of ΔT. These findings indicate that the regulatory mechanism on the thermal properties of BT₂ glass is complicated and it depends on doping site and doping level, especially when the dopant is introduced to occupy the glass network. Furthermore, it should be pointed out that although the kinetic window for BT₂ is enhanced by La or Nb doping, it is still too narrow for vitrification by conventional processing methods.

Raman spectroscopy is one of the most powerful analytical tools to provide the structural information of a material and it is widely used in the analysis of titanate glasses. Previous study has shown that a major factor in the composition dependence of the glass properties is the change in the glass structure, such as the Ti-O bonding state and the [TiO_n] polyhedra [25,26].

A typical Raman spectrum of BT₂ consists of four broad peaks around 100 (Region I), 280 (Region II), 630 (Region III), and 820 cm^-1 (Region IV), as shown in Fig. 5(a). The peak at ∼100 cm^-1 corresponds to the boson peak in the collective modes of glass and amorphous materials [11,27,28]. The band at ∼280 cm^-1 is usually attributed to the vibrational mode of Ba-O bonds [15,29,30]. The other two bands at ∼630 cm^-1 and ∼820 cm^-1 correspond to the vibration/rotation of the Ti-O bonds in titanates. From the viewpoint of bond-length, BT₂ glass contains distorted [TiO₅] polyhedra that have one long Ti-O bond and four short Ti-O bonds, which are represented by the bands centered ∼630 and ∼820 cm^-1, respectively [31]. Alternatively, from the viewpoint of [TiO_n] polydedra, the bands in these two regions, from low to high wavenumber, correspond to [TiO₆], [TiO₅] and [TiO₄] polyhedra, respectively [32,33]. Consequently, the spectrum of BT₂ in the wavenumber range between 130 and 950 cm^-1 can be deconvoluted to four Lorentz peaks, as shown by the inset figure in Fig. 4(a). The fraction of [TiO_n] is calculated based on the area of each peak (S) according to:

In BT₂ the fractions of [TiO₆], [TiO₅] and [TiO₄] are 0.40, 0.45 and 0.15, respectively.

With La-doping, the main features on the Raman spectra remain unchanged (Fig. 5(a)). The wavenumber for each peak obtained from spectrum deconvolution is plotted as a function of the La-doping level, x, as shown in Fig. 4(b) The peak at ∼280 cm^-1 shows a linear shift towards higher wavenumber with increasing x, which can be attributed to the incorporation of La onto the Ba-site in BT₂. Previous studies [[34], [35], [36]] suggested that rare-earth oxides such as La₂O₃ possess Raman bands in the frequency region below 400 cm^-1 and a band observed at 283 cm^-1 should correspond to La-O and/or [LaO_n] polyhedral vibration in La₂O₃-containing glasses. Therefore, it is reasonable to attribute the band at ∼280 cm^-1 to both Ba-O and La-O bonds in La-doped BT₂. Based on the harmonic oscillator approximation, the Raman wavenumber, ω, is determined by:

where k is the force constant and μ is the reduced mass. The ionic radius of La³⁺ (1.06 Å) is smaller than that of Ba²⁺ (1.36 Å), resulting in a shorter bond length and thus a higher force constant k [37]. The relative atomic mass for La (138.91) is close to that of Ba (137.23). As a result, the Ba-O band shifts towards higher wavenumber with increasing x. The band at ∼630 cm^-1 doesn’t shift, indicating that the long Ti-O bond length remains stable with La doping. In contrast, the band at ∼780 cm^-1 shifts to low wavenumber with increasing x. Sakka et al. suggested that there is a relation between the Ti-O bond length and the peak wavenumber of the crystals: the peaks generally shift toward higher wavenumbers as the shortest Ti-O interatomic distance decreases, and this relation can also be applied to alkali-TiO₂ glass system [31]. Here a reduction in the wavenumber indicates that the shortest Ti-O bond in the [TiO₅] polyhedra is elongated with increasing x, which relaxes the distorted [TiO₅] network. Furthermore, as shown in Fig. 4(c), the fraction of each type of [TiO_n] remains almost unchanged with La-doping.

For Nb-doped BT₂ glasses, the Raman spectra retain the main features of BT₂ (Fig. 6(a)). The spectra are also deconvoluted to four peaks to obtain the corresponding wavenumber of each peak, which is plotted as a function of the Nb-doping level, y, as shown in Fig. 6(b). The composition dependence of the wavenumber of all the four peaks shows a break at y = 0.20, suggesting a dramatic change in the structural arrangement of the glass at the vicinity of this critical composition. In the composition range 0 ≤ y ≤ 0.20, the wavenumbers present systematic and linear shift with increasing y. Specifically, the band centered at ∼280 cm^-1 corresponding to the Ba-O bond, shifts toward low wavenumber with increasing y, indicating that the incorporation of Nb on the Ti-site has a significant influence on the local structure of Ti-site in BT₂. The bands centered at ∼630 cm^-1 shift to high wavenumber with increasing y. As discussed previously, this band in BT₂ glass corresponds to the vibration/rotation of Ti-O bonds in [TiO₆] polyhedra. If we consider the partial substitution of Ti⁴⁺ by Nb⁵⁺ in the [TiO₆], according to Eq. (5), the peak should shift towards low wavenumber as Nb is larger in the ionic size (0.64 Å) and heavier in the relative atomic mass (92.91) than Ti, which disagrees with the observed upward shift in wavenumber. This indicates that the incorporation of Nb into BT₂ is more than a simple substitution effect. Furthermore, the band centered at ∼780 cm^-1 remains stable, whereas the one centered at ∼850 cm^-1 shows an increase in its wavenumber with increasing y, suggesting that some of the short Ti-O bonds are further shortened. As the fraction of [TiO₄] polyhedra is much lower than that of [TiO₅] and [TiO₆], as shown in Fig. 6(c), its contribution to the overall structural change is small. Furthermore, as the fraction of [TiO₆] increases with increasing y and exceeds that of [TiO₅], the structural change induced by Nb incorporation is dominated by the shortening of the Ti-O bond in [TiO₆]. When the Nb-doping exceeds the critical level (y = 0.20), the wavenumber of each peak shows an abrupt jump/drop. Jehng and Wachs [38] investigated the Raman spectra of a wide variety of niobates and they attributed the bands at 500-700 cm^-1 and 850-1000 cm^-1 to slightly and heavily distorted [NbO₆] polyhedra, respectively. It should be noted that niobates are usually constructed by [NbO₆] polyhedral with a different degree of distortion whereas [NbO₄] tetrahedra are extremely rare and only observed in several compounds since the Nb⁵⁺ ion is too large to force a way in the oxygen-anion tetrehedra [39]. Therefore, it is reasonable to attribute the discontinuity in the Raman peak shift at the vicinity of y = 0.20 to the assumption that the Raman bands in Region III and IV may be dominated by the vibrations of [NbO₆] polyhedra [39,40] rather than [TiO_n] polyhedra. The Raman peak shift suggests that Nb-doping introduces [NbO₆] polyhedra into BT₂, which relaxes the [TiO_n] polyhedra by decreasing the Ti-O bond length and increasing the fraction of [TiO₆] polyhedra. It starts to participate in the construction of a new network structure in BT₂ glass with increasing doping level.

To summarize the above, La-doping on the Ba-site and Nb-doping on the Ti-site both influence the structure arrangement of BT₂ but in a different way. For La-doping, incorporation of La into BT₂ increases the length of the short Ti-O bonds and therefore relaxes the distorted [TiO₅] polyhedra. For Nb doping, below the critical doping level (y = 0.20), the introduction of [NbO₆] polyhedra changes not only the Ti-O bond length in [TiO₆] polyhedra, but also the fraction of [TiO₆] and [TiO₅] in the glass. These results trigger the speculation that the network-modifiers in BT₂ glass possess a very important role in the glass-forming process than those in traditional network-former oxides such as SiO₂, GeO₂ and B₂O₃, which provides a useful direction for modifying the properties of these novel glasses by doping.

(Ba_1-xLa_x)Ti₂O_5+δ (La-BT₂) and Ba(Nb_yTi_1-y)₂O_5+δ (Nb-BT₂) glass spheres with a diameter of ∼1.6 mm were prepared by a containerless aerodynamic levitation method and their glass-forming region, density, thermal stability and structural arrangement were investigated. La-BT₂ shows a narrow glass-forming region of 0 ≤ x < 0.10, while Nb-BT₂ possesses a wider glass-forming region of 0 ≤ y ≤ 0.40. Both La and Nb doping can enhance the thermal stability of BT₂ glass and both exist an optimum doping level. Raman spectroscopy analysis indicates that La doping on the Ba site elongates the four short Ti-O bonds and thus relaxes the distorted [TiO₅] polyhedra in BT₂ glass. Nb doping on the Ti site introduces [NbO₆] polyhedra into the glass-former network and thus has a complicated effect on the structural arrangement of the glass. It shows a critical doping level, below which the incorporation of Nb compresses the long Ti-O bond in the [TiO₅] polyhedra. When the doping level exceeds a critical value, [NbO₆] polyhedra participate in the construction of the network skeleton. Therefore, both the La doping and the Nb doping can strongly affect the network structure of BT₂ glass.

As La₂O₃ and Nb₂O₅ are desired components to enhance refractive index, the optical properties of La-doped and Nb-doped BT₂ glass will be explored in later study. It is worth mentioning that the enhancement of optical properties may be limited by the narrow glass-forming region of La-doped BT₂. A simple calculation of the refractive index, n, according to the empirical Gladstone-Dale equation shows that La-doping only increases n from 2.20 for BT₂ to 2.23 for 10 % La-doped BT₂. Therefore, a wide glass forming region is a prerequisite to achieve desirable property by doping. The preliminary results from this work shed light on the relationship between doping-site and glass forming region in BT₂ glass and may provide strategy for improving properties of this type of novel glasses by chemical doping.

This work was supported financially by the National Natural Science Foundation of China (Nos. 51971138, 51727802 and 51821001), the National Natural Science Foundation of China-Excellent Young Scholars (No. 51922068), the National Key Research and Development Program (No. 2017YFA0403800), and the Shanghai Pujiang Program (No. 19PJ1404400). The support of synchrotron radiation phase-contrast imaging by the BL13W1 beam line of Shanghai Synchrotron Radiation Facility (SSRF), China, is gratefully acknowledged.