Oxyfluoride glasses with composition 55SiO₂–20Al₂O₃–15Na₂O–10LaF₃ (mol%) (55Si–10La) have been prepared by melting reagent grade SiO₂ sand (Saint-Gobain, Aviles, Spain, 99.6%), Al₂O₃ (Panreac), Na₂CO₃ (Sigma-Aldrich, >99.5%), LaF₃ (Alfa Aesar, 99.99%). Pr³⁺ and Yb³⁺ were added as fluorides (Alfa Aesar, 99.99%) in 0.1–0.5 and 0.5–1 concentrations (mol%). Samples doped with only Pr³⁺ were also prepared for comparison of the optical properties. A more complete description of glass preparation was given in (de Pablos-Martín et al., 2011).

Al₂O₃ was previously annealed at 800°C for 12 h. Batch materials were weighed to obtain 100 g of glass, mixed for 1 h to ensure a good homogenization, put in a covered Pt crucible and annealed for 2 h at 1200°C. The Pt crucible was then placed in an elevator furnace for 1.5 h at 1650°C, the molten glasses were quenched in air onto a brass mold, fused again for 30 min to improve homogeneity and quenched onto a cold (−10°C) brass mold. The glasses were annealed at 600°C for 30 min for stress relaxation.

Glass-ceramics were obtained by heat treatment at 620°C for 1, 3, 5, 20, 40, and 80 h and at 660 and 680°C for 20 h. In all the cases, a heating rate of 10°C/min was used followed by quenching in air.

Heat treatments were performed on bulk specimens (size 1–1.25 mm).

Non-isothermal crystallization kinetics was studied by DTA/TG (SDT Q600—TA Instruments). Measurements have been performed on 20–30 mg of glass with particles size between 1 and 1.25 mm to reproduce bulk conditions. DTA scans were carried out with heating rates in the range 10–60°C/min.

The glass transition temperature T_g, crystallization activation energy E_a, and Avrami parameters (n, m) were calculated from DTA measurements.

The Avrami parameter n allows assessing the crystallization process and was obtained employing the Ozawa equation (Ozawa, 1970): (1)(d[ln[−ln(1−x)]]d(lnq))T=−n, where x is the partial area of the crystallization peak calculated for a fixed temperature T and q is the heating rate. By using the Kissinger equation (Kissinger, 1956) the crystallization activation energy E_a was obtained by (2)ln(qTp2)=−EaRTp+C, where T_p, R, and C are the crystallization peak temperature, the gas constant and a constant, respectively. Finally, the m parameter, representing the growth dimensionality, was obtained by the Matusita equation (Matusita and Sakka, 1980): (3)ln(qnTp2)=−mEaRTp+C′.

The heat-treated samples were milled and sieved (< 63 μm) and characterized by XRD with a Bruker D8 Advance diffractometer. Diffractograms were acquired in the rage 10 ≤ 2θ ≤ 70° with a step size of 0.02° and 1 s acquisition for each step. Crystals size, D, was estimated using the Scherrer equation (Eq. 4), where λ is the wavelength (1.54056 Å—CuKα₁), B_m the full width at half maximum of the LaF₃ peak (111) and θ its diffraction angle. The factor 0.94 corresponds to spherical crystals. Pseudo-Voigt function has been used to fit diffraction peak parameters. The instrumental broadening B_i has been also taken into account using NaF powder properly milled and sieved (<63 μm): (4)D=0.94λcosθBm2−Bi2.

Crystalline growth can be described by the following equation: (5)r=Utp, where r is the crystal radius, U the crystal growth rate, t the time, and p a growth exponent. The logarithmic form of Eq. 5 is commonly used: (6)log(r)=log(U)+plog(t).

TEM samples of glasses and GCs were prepared by cutting slices, plane parallel grinding, dimpling to a residual thickness of 10–15 µm, and ion-beam thinning using Ar⁺ ions. The angle of incidence was set to 8°, the beam energy to 5 kV, current to 5 mA, and milling time to 10–14 h. HRTEM including scanning transmission microscopy-high angle annular dark field and energy dispersive X-ray spectroscopy (EDXS) were performed with a JEOL 2100 field emission gun transmission electron microscope operating at 200 kV and providing a point resolution of 0.19 nm. The microscope was equipped with an energy dispersive X-ray spectrometer (EDXS—INCA x-sight, Oxford Instruments). EDXS analysis was performed in STEM mode, with a probe size of ca. 1 nm. In order to determine the particle distribution, we first assumed the particles to be spheres. No high contrast was obtained when working in the Scherzer focus, the shape of the particles was not well defined and difficult to measure. Thus, slightly under-focused TEM images were used to solve this problem. HAADF-STEM images were obtained where the particle shape was more distinguishable, and it is possible to measure the average diameter of the particles. By this method, only well-defined particles were measured which still resulted in a statistically well-representative data collection.

Bulk specimens were cut from the annealed glass and heat treated to obtain glass-ceramic materials. 0.1 Pr and 0.1–0.5 Pr–Yb glasses were treated at 620°C for 20 h and 40 h, and at 660°C for 20 h. 0.5 Pr and 0.5–1 Pr–Yb glasses were treated at 620°C for 40 h and 660°C for 20 h. All the samples have been polished and optically characterized by UV–VIS absorption and PL spectroscopy.

UV–VIS spectra (Lambda 950—Perkin Elmer) were acquired between 300–2200 nm.

A photomultiplier tube (PMT) R6872 for UV–VIS and a Peltier cooled PbS for NIR detection were used as detectors.

A lock-in (5210-Princeton Research Instrument) configuration with an InGaN led at 435 nm (Roithner) as source for Pr³⁺ excitation and a fiber laser at 976 nm to excite Yb³⁺ ions was used to obtain PL spectra. A 2 × 2 mm² spot was produced with a lens focusing system and the samples were excited on the side edge to reduce re-absorption processes. Emission spectra were collected by an iHR-320 (Jobin-Yvon) spectrometer equipped with two gratings: 1200 g/mm blazed at 500 nm, and 600 g/mm blazed at 1000 nm. The detection system was calibrated using an incandescence lamp with known emission spectrum. A S-20 PMT and an InGaAs PD were used for UV–VIS and IR detection, respectively. Finally, all PL spectra were properly corrected for the instrument response.

Lifetime decay curves, upon excitation at 435 nm, were acquired with a fast oscilloscope (Tektronix), and the source was modulated electronically by a controller (ITC4000-Thorlabs).

For no single exponential decay, lifetimes were calculated using the following formula: (7)τavg=∫0∞t I(t)dt∫0∞I(t)dt.

The ETE and the QE were calculated using the following equations: (8)ETE=1−τPr/YbτPr, (9)QE=ηPr(1−ETE)+2ηYbETE, where τ_Pr and τ_Pr/Yb are the Pr³⁺ lifetime, corresponding to the same excited state level, in doped and co-doped samples, while η_Pr and η_Yb are the Pr³⁺ and Yb³⁺ QEs.

Differential thermal analysis curves and the variation of glass transition temperature (T_g), crystallization starting temperature (T_x), and crystallization peak temperature (T_p), with the heating rate are given in Figures 1A,B for the samples doped with 0.1–0.5 Pr–Yb.

It was not possible to estimate T_g, T_x, T_p from DTA curves performed at heating rate of 10°C/min due to very small endothermic peak (T_g) and exothermic peak (T_x, T_p) corresponding to LaF₃ crystallization. For 0.5–1 Pr–Yb doped glass the first values were obtained from a heating rate of 30°C/min.

The stability parameter, defined as ΔT = T_p − T_g, is 114°C for both co-doped glasses for a heating rate of 10°C/min (calculated by extrapolation from the fits). The variation of T_x and T_p with the heating rate is faster than that of T_g, as confirmed by the slope parameter α in the equation T = αq, summarized in Table 1. The calculated T_g for a heating rate of 10°C/min are 570°C for 0.1–0.5 Pr–Yb and 585°C for 0.5–1 Pr–Yb. Higher T_x and T_p values indicate a delay of the crystallization onset for materials with higher concentration of dopants.

By using Eq. 2, crystallization activation energies were calculated and their values are (329 ± 16) kJ/mol and (342 ± 18) kJ/mol for 0.1–0.5 Pr–Yb and 0.5–1 Pr–Yb, respectively. These results are similar to those obtained for the un-doped (de Pablos-Martín et al., 2011) and Tm³⁺ doped glass (de Pablos-Martín et al., 2013), Avrami n parameter was calculated, using Eq. 1, from the slope of each line; a n mean value was obtained from the slope of the five lines represented in Figures 2A,B for 0.1–0.5 Pr–Yb and 0.5–1 Pr–Yb, respectively. By substituting the calculated crystallization activation energy into the Matusita equation (Eq. 3) and plotting the left side of Eq. 3 as a function of E_a/RT_p, the m parameter has been obtained from the slope of the lines represented in Figures 2C,D.

For 0.1–0.5 Pr–Yb, n = 1.23 ± 0.08 and m = 1.2 ± 0.1, while for 0.5–1 Pr–Yb n = 0.86 ± 0.08 and m = 0.84 ± 0.08. The two values of (n, m) for each composition are the same within uncertainties. The higher value obtained for the 0.1–0.5 Pr–Yb glass could be interpreted considering that the crystallization process is faster than for the 0.5–1 Pr–Yb glass. These parameters can be approximated to the nearest integer or semi-odd integer resulting in n = 1 and m = 1 for both materials. This means that the use of Kissinger equation is valid for the calculation of the crystallization activation energy and corresponds to a volumetric crystallization with crystal growth controlled by diffusion (Donald, 2004). The same (n, m) parameters were also obtained for the un-doped glass and for Tm³⁺ doped glass (de Pablos-Martín et al., 2013), confirming that dopants do not affect the crystallization mechanism but may affect the crystallization kinetics and influence LaF₃ crystals size.

X-ray diffraction measurements for GCs 0.1–0.5 Pr–Yb treated at 620°C for 1, 3, 5, 20, 40, and 80 h are given in Figure 3A, while diffractograms for heat treatment at 620, 660, and 680°C for 20 h are compared in Figure 3B. Very similar diffractograms have been obtained for GCs 0.5–1 Pr–Yb and are not represented.

In all the cases, LaF₃ was the only appearing crystalline phase confirmed by the reference (JCPDS 32-0483). All the distinguishable peaks of the diffraction pattern were labeled by Miller indexes. Crystals size was estimated using the generalized Scherrer equation (Eq. 4), to take into account the instrumental broadening, applied to LaF₃ (111) peak (2θ ≈ 27.5°).

Crystal growth exponent p has been estimated by Eq. 6. Figure 3C shows crystal size variation with treatment time, and Figure 3D shows the crystal growth exponent p for GCs 0.1–0.5 Pr–Yb.

For GCs, 0.1–0.5 Pr–Yb crystals size at 620°C, shown in Figure 3C, is almost constant ≈12 nm for different treatment times, while treating the samples at different temperatures for the same time of 20 h, crystals size shows important changes. The increase of crystals size at higher temperature is indicated by the more intense diffraction peaks and by the narrowing of the peaks. At 660°C, crystals size is ≈14 nm and at 680°C ≈ 26 nm. As a consequence, for the heat treatment at 680°C, 20 h, the material partially lost its transparency due to quite bigger crystals. In fact, even though crystals are still quite small, the phase separation droplets containing several crystals inside have quite bigger sizes (as it will be shown in next section—Section “TEM”), ranging from an average value of 37 nm at 620°C up to ≈100 nm at 680°C. The GC starts to loose transparency at temperatures higher than 660°C. Moreover, temperatures higher than the glass softening temperature, ≈670°C, are not useful for practical purpose.

Glass-ceramics 0.5–1 Pr–Yb treated at 620°C, 1 h are almost amorphous and crystal size stabilizes to a constant value, around 11 nm, for treatments longer than 3 h. The onset of crystallization is delayed compared to GCs doped with 0.1–0.5 Pr–Yb, and this is associated with the higher activation energy (342 kJ/mol vs 329 kJ/mol). This is related to the nucleating effect of fluorides which promotes the production of smaller nuclei compared to what happens with lower fluoride content, according to which smaller nuclei should be favored. In fact, it is known that fluorine content in oxyfluoride glasses acts as a nucleating agent and suppresses crystal growth by increasing nuclei quantity (Chen et al., 2007; Bhattacharyya et al., 2009).

Pr³⁺ singly doped samples showed very similar behavior and only LaF₃ crystals precipitate in the glass matrix, and the crystals size is similar to the one obtained for the co-doped samples.

The calculation of p exponents was carried out using Eq. 6, starting from plots of crystals size for GCs treated at 620°C for 20 h. Data are plotted in Figure 3D for GCs 0.1–0.5 Pr–Yb. The crystal growth exponent at 620°C is p = 0.040 ± 0.005, while for GCs 0.5–1 Pr–Yb the p exponent is p = 0.03 ± 0.01. The very small dependence of crystal growth on the time of heat treatment together with small values for crystal growth exponent p, indicates the presence of an inhibition phenomenon explained in detail in Bhattacharyya et al. (2009) and de Pablos-Martín et al. (2011, 2012). These previous studies showed that La and Si-enriched phase separation droplets are precipitated already during the preparation of the initial glass. Upon conversion of the glass into a nano-GCs by appropriate annealing, LaF₃ nano-crystals are formed within these droplets. Similar results have been obtained in this work as shown in Section “TEM.”

Figure 4A shows a TEM image of the 0.1–0.5 Pr–Yb glass. The starting glass presents phase separation with a narrow size distribution of the droplets between 15–40 nm and an average droplets size of 28 nm (Figure 4B). The majority of the droplets do not present any structure inside. However, in very few droplets, small crystalline domains of 5–7 nm in size have been also detected, but this incipient crystallinity in the base glass is not detectable by XRD. The chemical composition along 80 nm scanning line (Figure 4C) was measured by EDXS and represented in Figure 4D. The droplets are enriched in F, La, Pr and Yb, a clear evidence of RE incorporation inside the droplets. Furthermore, excess of Si and Al are relocated toward the periphery of the droplets, and the formation of a barrier enriched in glass formers prevent further crystal growth, during the crystallization process, due to the increase of viscosity.

Figure 5A shows an image of the GC 0.1–0.5 620°C, 20 h and bigger droplets, with average size ≈33 nm, were detected as compared to the untreated glass (Figure 5B). A feature of this glass system is the crystals formation inside the initial phase separation droplets, already enriched in crystals components in the as made glass. The size of the crystals inside each droplet is clearly observed in Figure 5C, and their size distribution is represented in Figure 5D. An average crystals size of 10 nm was obtained for this heat treatment, in agreement with the value obtained by XRD measurements. Figures 5E,F show EDXS analysis of one single droplet. Al and Si are mostly confined in the interphase and tend to be smaller in correspondence of the maximum F, La, Pr, and Yb concentration, i.e., inside the droplet. Clear presence of RE ions inside the phase separation droplets and crystals is observed for glass and GCs. The detection of Yb is quite difficult, but its presence in the crystals is observed.

UV–VIS optical density for Pr³⁺- and Pr³⁺–Yb³⁺-doped glasses and GCs are represented in Figure 6.

Glasses have lower absorbance compared to GCs that suffer Rayleigh scattering caused by density fluctuations due to the presence of nano-crystals inside the glass matrix. The strong UV absorption (Urbach tail) below 350 nm is due to electronic transitions between the ligand (oxygen mainly) and the glass network former ion (silicon). In all materials the transitions between the 4f and 4f states corresponding to Pr³⁺ and Yb³⁺ ions are clearly visible. From the left, the following Pr³⁺ transitions can be assigned: ³H₀–3P_2,1,0 at 442, 466, and 480 nm, ³H₄–¹D₂ at 590 nm, ³H₄–³F₄ at 1460 nm, ³H₄–³F₃ at 1590, and ³H₄–³F₂ at 1930 nm. The Yb³⁺ transition ²F_7/2–²F_5/2 is also observed at 980 nm. For 0.5 Pr and 0.5–1 Pr–Yb GCs (Figure 6B), a small underlying structure is observed for Pr³⁺ absorption to the ³F₃ level, reflecting the different local field felt by Pr³⁺ ions compared to glasses, a proof of Pr³⁺ incorporation inside LaF₃ crystals that causes a narrowing of the band and the Stark components can be appreciated. However, clearer evidence of RE³⁺ ions inside LaF₃ crystals appears in the PL spectra.

PL spectra for Pr³⁺ doped and Pr³⁺–Yb³⁺ co-doped glasses and GCs are given in Figure 7. Excitation has been provided by an InGaN LED centered at 435 nm and Pr³⁺ ions have been excited to the ³P₂ level. By non-radiative decay, the ³P₀ level is populated and Pr³⁺ radiative emissions from this level to the three excited states ³H_4,5,6 and ³F_2,3,4 are clearly visible in the range 450–750 nm.

For samples singly doped with Pr³⁺, the emission at 1.05 µm corresponding to the transition from ¹D₂–³F_3,4 levels is observed and the emission at 600 nm corresponding to the ¹D₂–³H₄ transition, in both doped and co-doped samples, overlaps with the ³P₀–³H₆ emission band. The population of ¹D₂ level is due to multi-phonon relaxation from the ³P₀ level, and this contribution is stronger in glass than in GCs, meaning the incorporation of Pr³⁺ ions inside LaF₃ crystals, where phonons are much smaller.

However, in co-doped samples, the ³P₀ level is also quenched by the presence of Yb³⁺ ions, producing a DC signal in the range 950–1150 nm, as a consequence of an ET process. The Pr³⁺ transition ¹G₄–³H₅ at 1.3 µm is not observed in any samples, glass or GCs. Moreover, the ³P₀–¹G₄ transition at 950 is not osberved either.

For all samples, a more evident distinction in the Pr³⁺ emission between glass and GCs is observed. Again, co-doped glasses present the lowest Yb³⁺ DC signal at 976 nm while the Pr³⁺ transition ³P₀–³H₆ gets smaller passing from glass to GCs. Glass does not show sharp Stark splitting while a clear splitting of the ³H_5,6 and ³F₄ is visible in the GCs, a convincing proof that Pr³⁺ ions are incorporated into LaF₃ crystals. However, Yb³⁺ ions should be incorporated in LaF₃ crystals as well. In fact, a difference in the local environment between Pr³⁺ and Yb³⁺ does not seem favorable for ET processes. Additionally, DC emission gets stronger in GCs.

Xu et al. (2011) studied a 0.5–0.5 Pr–Yb doped oxyfluoride GCs containing LaF₃ crystals and found that the visible emission increases more than NIR emission passing from glass to GCs. Moreover, NIR emission of Yb³⁺ ions did not increase monotonously with the heat treatment temperature or time. These results are in contradiction with ours. They concluded that Yb³⁺ ions are not favored to be incorporated inside LaF₃ crystals, while we observed incorporation. TEM images (Figure 5), and particularly the elemental analysis with 1 nm resolution, showed an enrichment of Yb³⁺ inside the droplets, while no significant Yb³⁺ concentration was detected in the glass matrix.

While for 0.1–0.5 Pr–Yb doped samples the effect of the increase of temperature seems comparable to the increase of heat treatment time, for 0.5–1 Pr–Yb doped materials the increase of temperature produces the most evident improvement of Yb³⁺ DC emission at 976 nm. This may be explained considering that a doping with 0.1–0.5 Pr–Yb produces a lower nuclei density, but bigger crystals are still possible by rising the annealing temperature (Figure 3C). In particular, bigger crystals can host more RE³⁺ ions and a heat treatment at 620°C, 40 h thus produces an improvement of DC signal.

For 0.5–1 Pr–Yb, due to the quite higher fluoride content into the initial melt, the as made glass has a higher nuclei density thanks to the nucleating action of fluorine. The smaller initial nuclei make it more difficult to produce bigger crystals, due to the reduction of the effective cross-section to capture other crystal forming ions and to the presence of a diffusion barrier of higher viscosity around LaF₃ crystals that is formed earlier and which is expected to be thicker. In addition, in a glass doped with more RE³⁺ ions, viscosity increases compared to un-doped glass (or to a less doped glass), at the temperatures of nano-glass ceramic formation (T_g + 20–80°C) (de Pablos-Martín et al., 2013). As a consequence, RE³⁺ ions diffusion inside crystals can require longer times, and therefore, the best improvement is obtained by rising the temperature until a decrease of viscosity starts to allow more RE³⁺ ions diffusion inside the LaF₃ crystals but avoiding the growth of nanocrystals above 20 nm.

As suggested by van Wijngaarden et al. (2010), the cross relaxation scheme, Figure 8A, is the most common scheme for Pr–Yb DC. The two step ET process firstly allows Yb³⁺ ions excitation to the ²F_5/2, and then Pr³⁺ ions from the ¹G₄ level transfer energy to Yb³⁺ ions.

Xiang et al. (2014) found that, increasing Yb³⁺ concentration to 10 mol%, cooperative energy transfer (CET) process, Figure 8B, becomes increasingly important and for very high concentration as 20 mol% CET process is the main ET process.

In this study, the absence of the emission at 1.3 µm (¹G₄–³H₅) from Pr³⁺ ions indicates that ¹G₄ level is not populated or that this level is highly quenched in glass as well as in GCs. Considering that even for glass samples doped with only 0.1 mol% of Pr³⁺ this emission is not observed, we are tempted to affirm that this transition hardly occurs in our samples. Furthermore, the ³P₀–¹G₄ transition at 950 nm is not observed in any sample and this is in agreement with the absence of population of the ¹G₄ level and finally, it could be a proof of the fact that the CET from the Pr^{3+ 3}P₀ can be relevant for co-doped samples. Hence, it is possible to conclude that the ET between Pr³⁺ to Yb³⁺ (²F_7/2–²F_5/2) is improved in GCs respect to glass and CET could be quite relevant.

As suggested by Gao and Wondraczek (2013), the ET from the ¹G₄ level of Pr³⁺ to the ²F_5/2 level of Yb³⁺ is rather unlikely because the ¹G₄ level is almost 200 cm^–1 lower than the Yb^{3+ 2}F_5/2 level and an opposite back ET, from the Yb^{3+ 2}F_5/2 to the Pr^{3+ 1}G₄ level, should be favored. Considering that the ET from ¹G₄ of Pr³⁺ to the ²F_5/2 of Yb³⁺ is not observed in our measurements, the back ET from Yb³⁺ to Pr³⁺ ions was also studied by direct excitation of Yb³⁺ ions with a laser fiber at 976 nm. The corresponding PL measurements for GCs 0.1–0.5 Pr–Yb GC are given in Figure 9. The same results have been obtained for the GCs 0.5–1 Pr–Yb.

As clearly observed, no Pr³⁺ emission is present for direct excitation of Yb³⁺ at 976 nm, meaning the absence of a back ET mechanism. Therefore, the first order ET from ¹G₄ of Pr³⁺ to ²F_5/2 of Yb³⁺ is quite unlikely.

Figure 10 shows Pr³⁺ lifetime at 610 nm (³P₀–³H₆) for 0.1–0.5 Pr–Yb and 0.5–1 Pr–Yb co-doped samples, respectively. Lifetimes have been calculated by best fit and in all cases a bi-exponential fit has been necessary. Fast decays correspond to ET between neighbor ions while longer lifetimes give indication about radiative emission lifetime, although there can be also not negligible contributions from ET over long distances (Katayama and Tanabe, 2013). Lifetime uncertainty is ≈ 5%. Pr³⁺ emission in GCs has a more evident non-exponential profile. Pr³⁺ decays in co-doped samples is faster than glass, and this is a further proof of a more efficient ET mechanism between Pr³⁺ and Yb³⁺ ions. For Pr³⁺ singly doped samples, lifetime increases passing from glass to GCs and their values are summarized in Table 2.

ETE and QE of all co-doped samples have been calculated using Eqs 8 and 9. An estimation of the highest theoretical QE was obtained setting η_Pr and η_Yb in Eq. 9 equal to 1. The values are summarized in Table 2. ETE is quite smaller for glass than for GCs and for 0.1–0.5 Pr–Yb composition it is 11% for glass and almost 60% for GC 620°C, 40 h and the highest QE is 159%. For 0.5–1 Pr–Yb, the highest ETE value, obtained for GC 660°C, 20 h, is 46% and the highest QE is 146%. Therefore, the best results in terms of ETE and QE are obtained for the 0.1–0.5 Pr–Yb GC 620°C, 40 h, and this could be explained considering the higher ratio between Pr³⁺ and Yb³⁺ ions that should favor a more uniform Yb³⁺ distribution around Pr³⁺ increasing the probability of DC emission.

Figure 11 shows Yb³⁺ emission at 976 nm for both co-doped compositions. Near single exponential decay are observed for GCs samples, while non-radiative relaxation channels are more important for glasses. GCs lifetimes increase as compared to glasses and this is a further proof of Yb³⁺ ions inside LaF₃ crystals. All Yb³⁺ lifetimes are summarized in Table 3.