Fluorine loss determination in bioactive glasses by laser‐induced breakdown spectroscopy (LIBS)

Chemical analysis is crucial when developing materials since their chemical composition is inherently related to their properties. In many cases, the final composition of a specimen does not correspond with the theoretical (nominal) one, since undesired chemical reactions between raw materials or between raw materials and the environment (decomposition or evaporation during thermal treatments, irradiation, interaction with air, etc.) may occur during processing. In such cases, initial and desired stoichiometry and, potentially, reproducibility and the intended properties of the material may be compromised. Chemical analysis is particularly important in glass science and technology, owing to the high temperatures achieved during the glass melting process (usually above 1000°C), which can lead to evaporation of volatile compounds. Fluoride-containing glasses are especially affected, since fluoride species present high vapor pressures at these melting conditions1 and, thus, fluorine losses occur during glass melting, typically as either SiF4 or HF. 2 However, further volatile species present in the glass must be considered as well,3 for example, the loss as volatile SnF2 in a stannous fluoro-phosphate glass system. In the glass industry, fluoride compounds are added as opacifiers of glasses and glass-ceramics. Fluoride-containing glasses present high transparency in the UV to IR spectral Received: 12 October 2020 | Revised: 14 December 2020 | Accepted: 30 December 2020 DOI: 10.1111/ijag.15867


| INTRODUCTION
Chemical analysis is crucial when developing materials since their chemical composition is inherently related to their properties. In many cases, the final composition of a specimen does not correspond with the theoretical (nominal) one, since undesired chemical reactions between raw materials or between raw materials and the environment (decomposition or evaporation during thermal treatments, irradiation, interaction with air, etc.) may occur during processing. In such cases, initial and desired stoichiometry and, potentially, reproducibility and the intended properties of the material may be compromised. Chemical analysis is particularly important in glass science and technology, owing to the high temperatures achieved during the glass melting process (usually above 1000°C), which can lead to evaporation of volatile compounds. Fluoride-containing glasses are especially affected, since fluoride species present high vapor pressures at these melting conditions 1 and, thus, fluorine losses occur during glass melting, typically as either SiF 4 or HF. 2 However, further volatile species present in the glass must be considered as well, 3 for example, the loss as volatile SnF 2 in a stannous fluoro-phosphate glass system.
In the glass industry, fluoride compounds are added as opacifiers of glasses and glass-ceramics. Fluoride-containing glasses present high transparency in the UV to IR spectral region and low refractive index. 4 Thus, they are of interest for a broad range of optical applications, including telecommunications, optoelectronics, near-infrared or mid-infrared emissions, lenses, and filters. Fluoride-containing bioactive glasses have been studied in great detail as well, 2,5,6 as fluoride plays an important role in the development of biomaterials for dental applications. It inhibits enamel demineralization by forming fluorapatite, which is more acid resistant and has a lower solubility than carbonated hydroxyapatite. [6][7][8][9] Glass-ceramics are prepared by controlled crystallization of the parent glasses. In fluoride-containing glass-ceramics, where fluoride crystals are embedded in the glassy matrix, the retention of fluorine during melting is critical to avoid limitation of the resulting crystalline fraction. Relevant examples of the importance of minimizing fluorine loss in the achieved crystalline fraction are oxyfluoride systems for optical applications [10][11][12][13] and bioactive glass-ceramics. 2,8,9 In 1982 Kokubo et al. 5,14 developed the commercial bioactive glass-ceramic Cerabone-AW ® (Nippon Electric Glass Co, Japan) based on a SiO 2 -P 2 O 5 -CaO-MgO-CaF 2 system. This glass-ceramic consisted of surface crystallized oxyfluorapatite and wollastonite crystals embedded in a glassy matrix. It was shown that this material formed a tight bond to bone and had improved mechanical properties compared to bioactive glasses. Moreover it could be machined into various shapes, which added an important benefit for clinical applications. 15 Fluorine loss has important consequences for the thermal behavior of glasses. 16 It affects the glass transition temperature (T g ) and influences the crystallization tendency. 2,8,9 Thus, it is necessary to know the actual fluoride content of the glass under investigation and to establish glass preparation protocols, in which the expected fluorine loss during melting is already taken into account when designing new glass compositions. For example, in a study of a fluoro-aluminosilicate glass used for ionomer cement preparation, the fluorine loss was minimized by compositional changes, which allowed for a sufficient number of aluminum ions to bind to fluoride ions, thus decreasing volatilization. 17 In another study 18 the precursor glass and the fluorinating agents (NaF, LiF, MgF 2 ) were melted separately and then added for a final melting. Moreover, additional quantities of the fluorinating agents were added in the final steps of the melting to balance the loss of fluorides. Fluorine losses in glasses are strongly dependent on the composition. While in bioactive glasses in the system SiO 2 -P 2 O 5 -CaO-CaF 2 fluorine losses are typically in the range between 5 and 23%, 2,19 in oxyfluoride glasses much higher fluorine losses between 30 and 40% have been reported 11,12,[20][21][22] ; in the system SiO 2 -TiO 2 -SrO-SrF 2 even as high as 63%. 23 The use of lids on the crucibles during melting as well as pre-calcined raw materials (decreasing the water content, thereby avoiding the fluoride loss as HF) is highly recommended to increase fluorine retention during melting. 4,12 Fluorine loss during melting depends strongly on the total fluorine content of the glasses and on the network former/ network modifier ratio, 2,4 as well as on the melting temperature and duration of the melting process. 18 Typically, an increase in melting temperature leads to an increase in fluorine evaporation and, thus, to a reduction in the fluoride content in the glasses. 13 The fluorine loss rate in oxyfluoride glasses is reported to be more important in the first minutes of melting (from 5 min at 800°C) and decreases with increasing melting time (until 30 min at 800°C). 16 The nature of the fluorinating agent is also reported to influence fluorine losses. 4 Several analytical methods are used to determine the fluoride content in materials 17,24,25 (and references therein). A potentiometric method employing a fluoride ion-selective electrode has been used for fluoride quantification in glasses. 16 Here, several dissolution processes including preparation of acidic solutions (HNO 3 ) and pH adjustments are involved, making this method time consuming. In the particular case of fluoride-containing bioactive glasses, the determination of fluoride release during glass immersion in aqueous solutions is also commonly performed by using a fluoride-selective electrode. 2 In another study, 4 the Ehrlich and Pietzka method 26 was employed. Here, powdered glass samples are mixed with SiO 2 powder and boiled in H 3 PO 4 . The evaporating SiF 6 is collected in a NaOH solution where the fluoride dissolves and subsequently precipitates as PbBrF. Addition of AgNO 3 transforms PbBrF into AgBr, which can be titrimetically quantified through re-titration of Na 2 S 2 O 3 against SCN − anions. While this method offers a high precision of ±0.01 wt%, 4 it is very complex and labor-intensive. Inductively coupled plasma -optical emission spectroscopy (ICP-OES) is a high-sensitivity atomic spectroscopy technique that is widely available and allows for simultaneous analysis of various elements. However, the fluorine ionization capability is very low in the argon plasma usually employed in this technique, 27 making its detection not feasible.
Electron microscopy-based techniques, like energy-dispersive X-ray spectroscopy (EDXS) are widely used to perform chemical analysis within an SEM microscope, since the sample can be analyzed without complex preparation routes, and usually the EDXS measurement is coupled with the microstructure characterization of the material, obtaining different kinds of information in parallel. However, the analysis of light elements, that is, Be, B, C, N, O, and F, is difficult and not reliable because of their low photon energies (low fluorescence yield), leading to a low signal or low signal/ noise ratio. Moreover, low photon energy leads to a high absorption in the specimen and in the detector, resulting in an incorrect measurement of photons and consequently a shift in the peak position. Additionally, low energy peaks are positioned close to the electronic noise of the detection system (which can be seen at about 0 keV). An additional challenge in the fluorine determination by EDXS, especially in glasses, is the fluoride mobility under the electron beam irradiation, 28 which considerably affects the accuracy of the quantification. However, recent advances in energy-dispersive spectroscopy in combination with the silicon drift detector (SDD) enable an accurate quantification of low atomic number elements, including fluorine. 22,29 One of the most common methods to quantify the fluorine content in glasses is X-ray fluorescence (XRF) spectroscopy. 30 However, a large error and a high average lower limit of detection of about 1-10 μg/g is associated with XRF. 30 In some cases, a fusion process of the employed pressed powders is needed to homogenize the sample and avoid particle size effects, which would lead to a smaller error. Borate species, like Li 2 B 4 O 7 , are used as fusion material and glass-forming agent for XRF analyses. 11,21,30 More recent publications reported on wavelength-dispersive x-ray fluorescence (WD-XRF) spectrometry. 13,25 Here, an electron probe micro analyzer (EPMA) equipped with five wavelength dispersive x-ray analyzers (WDX) was used to determine the fluoride content with an accuracy of ±0.1 at%. 13,19 The measurements were performed at the surface of polished bulk samples, showing that a certain amount of sample preparation is required.
Fluorine can be also determined by quantitative solid-state nuclear magnetic resonance (NMR). 31 However, typical disadvantages are weighing imprecisions, as only about 20-50 mg of sample is used to fill the rotor, and, depending on the probe, the presence of 19 F background signals, the suppression of which requires additional measurements. Moreover, a potential systematic error may arise if the relaxation times of the sample and the employed standard differ, the correction of which requires additional measurements, as well. 31 Laser-induced breakdown spectroscopy (LIBS) is a well-established spectroscopic method for in situ elemental analysis, which has the advantage of being minimally destructive. 24,32-35 A short duration high-energy laser pulse is focused on the surface of a sample. Parts of the sample are heated, vaporized, and partly ionized at temperatures up to 10 000°C, and a plasma is generated, where different atomization, ionization, and excitation processes take place ( Figure  1A). The radiation emitted by the excited species present in the plasma when the plasma cools down provides chemical information about the sample, 36 and therefore qualitative as well as quantitative analysis is possible, with limits of detection being in the ppm range. 37 One of the major advantages of LIBS in comparison with the above-mentioned methods is the reduced analysis time, minimal to no sample pre-treatment, multi-elemental detection, high spatial resolution and the potential to carry out in-lab, in situ or stand-off (remote, also called TELELIBS 38 ) analysis. 39 It has also been shown to offer comparable results to more established analytical techniques such as XRF. 40 LIBS has been used to quantify fluorine in different materials, 33,39,41 but so far not much information is available on its application for glasses. As the above-mentioned advantages of LIBS have the potential to be very beneficial in the determination of fluorine in bioactive glasses, we decided to investigate its use for this purpose. Here, four bioactive glasses in the system SiO 2 -P 2 O 5 -CaO-CaF 2 with varying P 2 O 5 content were prepared by a melt-quench route. CaF 2 was chosen as fluorinating agent. Fluorine loss is discussed as a function of P 2 O 5 content. This work demonstrates that fluorine in melt-derived glasses can be quantified by LIBS, and that this method is useful for studying and, thus, improving the retention of fluorine during melting.

| Sample preparation
Glasses in the system SiO 2 -P 2 O 5 -CaO-CaF 2 were prepared by a melt-quench route. Four glasses were designed based on the network connectivity (NC) model. NC is the average number of bridging oxygens (BO) per network forming element (here silicon, NC Si ) in the glass structure. 6,42 In the case of bioactive glasses, Edén et al. 43 suggested an NC between 2.0 and 2.6 for optimum bioactivity, with NC = 2.4 being considered in some studies as the cut-off value for bioactivity as defined by Hill. 44 It has been reported that when adding P 2 O 5 together with stoichiometric amounts of network modifiers (and thus, keeping NC constant) the bioactivity increases with the phosphate content. 6,42,45 Therefore, in the present work, the P 2 O 5 content was varied between 0 and 5.1 mol % while keeping the network connectivity (NC)  Glasses were melted in platinum-rhodium crucibles from 150 g batches containing mixtures of the following analytical grade raw materials: SiO 2 (99.0% Carl Roth, Germany), Ca(H 2 PO 4 ) 2 •2H 2 O (Chemische Fabrik Budenheim KG, Germany), CaF 2 (Chemiewerk Nünchritz, Germany), and CaCO 3 (≥99.0% Merck, Germany). These raw materials were melted in air in an induction-heated furnace (in-house built). Crucibles were covered with lids. Melting temperatures and times were between 1400 and 1480°C and between 75 and 85 min, respectively, depending on the composition ( Table 1). The mixed raw materials were added to the crucible inside the furnace in small additions to reduce foaming. Glasses were cast into a brass mold, pressed with a carbon stamp and annealed at 670 +/-5 °C for times between 0.5 and 4 hours. Glasses were ground in a metallic mortar and sieved to obtain a particle size range <125 µm. Glass powders were stored in a desiccator until used for further analysis.

| Reference samples
Reference samples are needed to develop a fluorine calibration curve for the LIBS analysis. Seven reference samples (20 g of each) were prepared by mixing CaF 2 (99.0%, 325 Mesh Powder (45 µm), Alfa Aesar, Germany) and SiO 2 (99.0%, <125 µm, Carl Roth, Germany) according to the weight percentages shown in Table 2. These reagents were manually mixed and further homogenized in a rotatory mixer for 1 day.

| LIBS analysis
Glass samples (Table 1) as well as the reference samples ( Table 2) were pressed with a hydraulic press at a force of 50 kN (TP60, Herzog Maschinenfabrik GmbH) to form powder compacts with a diameter of ~4 cm ( Figure 1B).
Since the glass powder samples were collected from the entire volume of the bulk piece of the glasses, it is assumed that the analyzed powder samples are reasonably representative of the total volume of the glasses. Additionally, it has been reported that the particle size influences the LIBS signal; however, without any significant relationship or clear correlation. 48 In fact, the authors in ref. 48 did not observe a significant correlation between statistical particle size distribution and LIBS signal intensity. According to a recent study published in ref. 33 the particle size did not significantly affect fluorine quantification. It also confirmed that fluorine quantification using LIBS is possible without dependence on the particle size if all the samples have roughly the same particle size distribution.
The LIBS experimental set-up used in this work is based on a Nd:YAG laser (FiberLIBS, SECOPTA analytics GmbH) at 1064 nm. Excitation was performed with an energy of 1 mJ per pulse, a pulse duration tp <2 ns and a diameter of the illuminated area of <100 µm. The shot repetition rate was selected as 1000 Hz. The laser distance sensor to determine the focus height works with a semiconductor laser with a wavelength of 635 nm and a performance of 1 W.
The laser head was moved using motorized X-Y micro-translation stages, while the sample can be fixed in a sample holder system. Samples were irradiated in a pattern of 10 x 10 points, with a separation of 0.5 mm. At each point, a LIBS spectrum is obtained by accumulating 25 single shot emission spectra. Data treatment was carried out using the SEC analysis tool (SECOPTA analytics GmbH). The standard deviation (SD) is given as a measure of the data distribution in relation to the mean value.
For the development of the fluorine calibration curve 100 single spectra per sample (R1 to R7, Table 2) were averaged. The resulting reference spectra of R1 to R7 were used for evaluation. The calibration curve was obtained by the partial least squares (PLS) method included in the SEC analysis tool. Figure Table 2) are observed.

| Results for reference samples and calibration
The fluorine emission lines in the spectral range between 680 and 700 nm were used for analysis. Figure 2B shows the fluorine signals of the references (Table 2) centered at 685.6 and 690.2 nm. Considering the LIBS spectra in the range between 680 and 700 nm ( Figure 2B), the fluorine calibration curve was calculated (shown in the supplementary material, Figure S1). The spectra were smoothed each by five-point averaging. Normalization was performed relative to the background at 681 nm. With these settings, a coefficient of determination of R 2 = 0.9896 and a root mean square error (RMSE) of 0.1887 was obtained. Cross validation was used to verify the method. The coefficient of determination for the cross validation was determined as R 2 = 0.9608 and RMSE = 0.3856. The calibrated fluorine concentrations were in the range between 0.5 and 5.8 wt% to cover the concentration range of the analyzed glasses (Table 1).

| Results for glass samples
Spectra of four different glass samples were recorded under the same conditions as the calibration samples. Figure 3A shows the LIBS spectra over the full range. There are some truncated peaks, especially for Ca, owing to an intensity overload of the detector for these wavelengths. Nevertheless, quantification of fluorine ( Figure 3B) could be performed if 25 spectra were accumulated, in order to get enough signal intensity of the peaks for quantification.  ones. The fluorine loss associated with the glass melting process was calculated and is shown in Table 3, as well. Fluorine losses are between 13 and 55%, with the phosphate-free glass P0 being the one with the lowest fluorine loss (around 13%) and P5 (highest phosphate content) the one with the highest fluorine loss (around 55%; Table 3). The intermediate glasses P2 and P3 present intermediate fluorine losses. Focusing on glasses of a similar system (SiO 2 -P 2 O 5 -CaO-CaF 2 ), Brauer et al. 2 reported losses between 5 and 23% in glasses in the system SiO 2 -P 2 O 5 -CaO-Na 2 O-CaF 2 . However, those glasses were melted for 1 h at 1430°C, which is lower than the melting temperatures in the present work (Table 1). In ref. 19 fluorine losses in SiO 2 -free glasses in the system P 2 O 5 -CaO-Na 2 O-CaF 2 were determined by EPMA and with a fluoride ion probe, giving 18% and ~21% fluorine loss, respectively. However, in that work, melting conditions were softer, with 950°C for a duration as short as 5 min.
Möncke et al. 31 reported losses below 15% in NaPO 3 -AlF glasses. In ref. 13 fluorine losses from 15% to 21% were reported when the melting temperature of Er 3+ -doped 75NaPO 3 -25CaF 2 glasses increased from 900°C to 1000°C. Xu et al. 49 reported fluorine losses above 50% based on XPS analysis of stannous fluorophosphate glasses doped with PbO and B 2 O 3 [93], exposed to a maximum temperature of 500 °C; and around 20% in fluorophosphate glasses based on NMR analysis. 50,51 The nominal CaF 2 contents of the glasses in the present study (Table 1) were very similar (F content slightly varying between 5.35 and 5.70 wt%), as the glasses differ mostly in their SiO 2 and, most importantly, in their P 2 O 5 content. Moreover, the melting conditions were similar as well. Thus, the large differences in fluorine losses between glasses P0 and P5 are most likely to be connected to the P 2 O 5 content.
Ehrt 52 reported that phosphate-rich glasses show very high fluorine loss, in contrast to fluoride-rich glasses with low fluorine loss; and that fluorine loss decreased with decreasing phosphate content. Different mechanisms for fluorine losses exist for melt-derived glasses as extensively discussed by Semrau. 53 Traditionally, it was assumed that fluoride ions replace oxygen ions in silicate glasses, thereby forming Si-F bonds and leading to SiF 4 volatilization. However, this mechanism has been proven not to be the main responsible for fluorine loss during glass melting, 53 as silicon ions have a higher affinity for oxygen ions than for fluoride ions, SiF 4 formation would be only plausible in highly cross-linked glasses or high fluoride concentrations. Nevertheless, it has been widely shown that in bioactive glasses, fluoride complexes modifier cations (Ca 2+ , Na + etc.) exclusively, and does not form Si-F bonds, as confirmed by 19 F-and 29 Si MAS NMR spectroscopy and molecular dynamics simulations. 6,19,46,47,54,55 Thus, fluorine loss in the present glasses is more likely to occur via formation of HF from the reaction of water vapor with metal fluorides (CaF 2 in our case) at high temperatures, as illustrated by the following reactions 53 :   This mechanism depends strongly on the water content (both as H 2 O and OH − ) of the raw materials used. 4 Brauer et al. 2 reported on glasses in the very similar system SiO 2 -P 2 O 5 -Na 2 O-CaO-CaF 2 with varying CaF 2 concentration, showing a water loss on ignition at 500 °C between 0.8 and 2.1%, decreasing with increasing CaF 2 content. The decrease of water loss on ignition with increasing CaF 2 content suggests that CaF 2 in the glass impedes absorption of atmospheric water. In that reference, relative fluorine loss did not show any correlation with CaF 2 content, and it was thus assumed to be dependent on atmospheric humidity levels during melting, indicating fluorine evaporation as HF rather than as SiF 4 . From all the above-mentioned, fluorine loss in the P 2 O 5 -free glass P0 is likely to occur through HF volatilization following Eqs. 2 and 3. 53 Phosphate glasses (P2-P5) can undergo an additional fluoride loss mechanism through POF 3 evaporation 4,53,56 : Hydrolysis of POF 3 leads to the formation of HF as well: Möncke et al. 4 reported that P-F bonds could only be detected in (Na 2 O-Al 2 O 3 -P 2 O 5 -F) (NAPF) glasses prepared at melting temperatures up to 850°C. P-F bonds were no longer detectable when the glasses where melted near 1000°C. P-F bonds were not detected in ref. 50 on fluorophosphate glasses melted at 1150°C for 5 min either, in good agreement with Eqs. 4 and 5. According to Brauer et al. 46 orthophosphate groups are charge-balanced by cations (Ca 2+ in our glasses) and by Ca-F complexes, which agrees with the chemical reaction in Eq. 4. Thus, P 2 O 5 -containing glasses undergo fluorine loss via POF 3 in addition to HF evaporation mechanisms (Eqs. 2 and 3), explaining a higher fluorine loss in the P-containing glass P5 in comparison with the P-free glass P0 (Table 3). A combination of these reactions is likely to occur in the intermediate P2 and P3 glasses. However, a change of the trend in the P3 glass is observed (Table 3), which may be explained by a slightly lower melting temperature and shorter time in the P3 glass (Table 1) in comparison with those of P2 and P5 glasses (fluorine loss has been reported to be critical in the first minutes of the melting and decreases at longer periods 16,19 ), or even by a heterogeneous distribution of the powdered sample.
Another parameter that may influence fluorine loss is the viscosity of the glass melts. In general, viscosity increases with increasing phosphate content and decreases with fluorine concentration. 57 In ref. 58 viscosity-temperature curves of the most common bioactive glasses were reported.

| CONCLUSIONS
Fluorine loss during the melting of bioactive glasses in the system SiO 2 -P 2 O 5 -CaO-CaF 2 system with varying P 2 O 5 concentrations has been quantified using the LIBS method. LIBS analysis offers a reduced analysis time and minimal sample preparation. Results suggest that with increasing phosphate concentrations in the glass more fluorine was lost during melting. While silicate glasses typically lose fluorine as HF, in phosphate-containing silicate glasses additional mechanisms of fluoride loss exist, in which the volatilization as POF 3 takes place. Thus, although the network connectivity was the same in all the investigated glasses, P 2 O 5 strongly influenced fluoride retention in the glass melt. While both fluorine and phosphate have been shown to have various beneficial effects on bioactive glass properties, our results suggest that fluorine loss measurements need to be performed to ensure a proper evaluation of the composition-properties relationship of the glasses.
Improving the melting conditions to minimize fluorine losses is key. The preparation of glasses in the studied system at lower melting temperatures and shorter melting times (to reduce the fluoride loss, while ensuring a homogeneous melt) is part of a future work. Fluorine losses of those glasses will be compared to the ones reported in the present study, to investigate the influence of the melting conditions on fluoride retention. The structural characterization of the investigated glasses is part of a forthcoming publication, which will provide the relationship between atomic structure and fluorine loss.