2.1 Glass Preparation

A series of glasses with compositions of xCaF₂-(20-x)CaO-20ZnO-50B₂O₃-10SiO₂ (x = 0, 5, 10, 15, and 20 mol%) were synthesized to control glass dissolution behavior and to investigate the correlation between the glass structure and antibacterial properties. The chemical compositions of the prepared glasses are listed in Table 1. For convenience, the samples were designated as A 20-00, A 15-05, A 10-10, A 05-15, and A 00-20, corresponding to x = 0, 5, 10, 15, and 20 mol%, respectively. High-purity reagent-grade powders of ZnO (Sigma Aldrich, 99%), SiO₂ (Sigma Aldrich, 99%), B₂O₃ (Sigma Aldrich, 99.9%), Ca₂CO₃ (Sigma Aldrich, 99.99%), and CaF₂ (Sigma Aldrich, 99%) were used as starting materials. Each batch, weighing approximately 60 g, was mixed in stoichiometric proportions using a 3D mixer for 5 h to ensure homogeneity. The well-mixed powders were melted in 100 mL platinum crucibles at 1150 °C for 1 h in an electric furnace. The molten glass was then poured into a preheated graphite mold at 550 °C, close to the glass transition temperature (T_g). The samples were subsequently annealed at 550 °C for 1h to relieve residual thermal stress, followed by slow cooling to room temperature (23 °C). The obtained glass was crushed and dry-milled in a ball mill at 500 rpm for 2 h. The resulting powders were sieved through a 45 μm mesh to obtain particle sizes below 45 μm.

2.2 Glass Characterization

The amorphous nature of the prepared glass samples was confirmed using X-ray diffraction (XRD, D8 Advance, Bruker, Germany). Diffraction patterns were collected continuously over a 2θ range of 10-80°, using Cu Kα radiation (40 kV, 30 mA) at a scanning rate of 1.167°/min. The T_g was determined using differential scanning calorimetry (DSC, TG-DSC STS200RV, Hitachi, Japan). Approximately 15-25 mg of powdered sample was placed in an aluminum crucible and heated to 900 °C at a rate of 10 °C/min under a nitrogen atmosphere. The T_g of each sample was measured three times to ensure reproducibility. Glass density was measured using Archimedes method with ethanol (0.788 ± 2.2%) as the immersion medium. Each measurement was repeated ten times to ensure accuracy. Density was calculated using the following equation:

where ρ_sample and ρ_eth represent the densities of the glass and ethanol, respectively, and W_air and W_eth are the weights of the sample measured in air and ethanol. The measured densities and chemical compositions were used to calculate the molar volume (V_m), using the following equation:

The atomic packing density C_g was calculated using Equation (3), where X_i is the molar fraction of cation i and V_i is the ionic volume of cation i:

The ionic volume V_i was determined using Equation (4), where N_A is Avogadro’s number; m and n are the numbers of cations and anion atoms in A_mB_n, respectively; r_A is the ionic radius of the cation; and r_B is the ionic radius of anion. The ionic radii were taken from Shannon and Prewitt^{[54, 55]}:

The free volume was calculated using Equation (5) ^[56]:

Fourier transform infrared (FTIR) spectra were measured using a Perkin-Elmer FTIR 1600 spectrometer in the range of 4000-250 cm^-1 at room temperature. To minimize the influence of ambient moisture, the specimens were strictly prepared by mixing 1 mg of the synthesized glass powder with 200 mg of anhydrous KBr and pressing them into pellets. Each spectrum was obtained by averaging 64 scans with a nominal resolution of 4cm⁻¹. X-ray photoelectron spectroscopy (XPS) analysis was conducted using a surface analysis system (SPECS, Germany) equipped with a Phoibos 150 electron analyzer and monochromatic Al Kα radiation (1486.6 eV, 400 W). The C1s peak at 284.6 eV was used as a reference to calibrate the binding energies (BE) of different elements to correct for the charge effect. Spectral fitting was performed with CASA XPS software using Gaussian-Lorentzian (30% Lorentzian) line shapes, and the fits were optimized by minimizing the root-mean-square error. The local structures around silicon and boron were investigated using ²⁹Si and ¹¹B magic-angle spinning nuclear magnetic resonance (MAS-NMR) spectroscopy. ²⁹Si MAS-NMR spectra were recorded on an 11.7 T spectrometer at a resonance frequency of 99.36 MHz with a spinning rate of 10 kHz. A pulse length of 0.89 µs (π/12) and a recycle delay of 20 s were applied. 3-(Trimethylsilyl) propionic acid sodium salt-d4 was used as the chemical shift reference. ¹¹B MAS-NMR spectra were acquired at 11.7 T with a resonance frequency of 160.46 MHz and a spinning rate of 15 kHz. A pulse length of 1 µs (π/6) and a recycle delay of 3 s were applied. An aqueous solution of H₃BO₃ was used as the chemical shift reference.

Glass dissolution tests were conducted by immersing each sample in 100 mL of distilled water at 40 °C for 24-168 h. The annealed glass samples used for the tests were cut and polished on both sides to dimensions of 10 x 15 x 1.5 mm. The concentration of the released ions was quantified by inductively coupled plasma mass spectrometer (ICP-MS, Agilent 7900). The zeta potential of the glass particles was measured by particle electrophoresis using a Pen Kem System 3000 Electrokinetic Analyzer (Pen Kem Inc., Bedford Hills, New York, USA). Two bulk samples were mounted in the measurement cell, facing each other with a 100 µm gap to allow electrolyte flow. Measurements were performed in 0.001 M KCl solution at pH 7.2, adjusted with 0.05 M NaOH. The zeta potential reflects the effective surface charge of the glass particles, whereas in solution, charge reversal occurs due to cation adsorption at the surface.

Antibacterial activity was evaluated according to the ASTM E2149-10 standard. The bacterial suspensions of E. coli and S. aureus were standardized to 1.5 × 10⁶ CFU/ml using the plate count method. Each glass sample was inoculated with bacterial suspension and agitated on an orbital shaker at 350 rpm for 1 h at 37 °C to promote contact between the glass particles and bacteria. After shaking, aliquots of the bacterial solution were serially diluted with the recovery diluent, and 1 mL of each dilution was plated and incubated for 24 h. Surviving colonies were counted and compared with those of the control (0 h). Antibacterial activity was considered effective when more than 99% of CFUs were devitalized. Antibacterial activity was calculated using the following equation:

where C_control and C_survivor represent the bacterial counts in the control and test samples, respectively. The cytotoxicity tests were conducted in accordance with the Water Quality Process Testing Standard (ES 04704.1) established by the Ministry of Environment of Korea. This method evaluates toxicity using Daphina magna by observing survival or growth inhibition rates. The degree of toxicity was quantified using the toxicity unit (TU), defined as:

where c is the concentration of the test chemical (mg / L), and EC50 is the concentration that causes a 50% biological response (mg / L).

2.3 Simulation Details

In MD simulations, the accuracy of structural modeling critically depends on the selection of interatomic force fields. In this study, classical MD simulations were conducted to construct five structural models for each glass composition using a Born-Mayer-type pairwise potential, expressed as follows:

where Φ_ij is the interaction energy of the i and j ions, r is the interatomic distance, Z_i and Z_j are the effective charges, and D_ij and ρ_ij are repulsive constants and softness parameter, respectively. This Born-Mayer-type potential has been widely employed in modeling fluoride-containing glass systems incorporating silicate, calcium, zinc, and borate components owing to its proven reliability in reproducing short-range structural features and ionic interactions. The parameter sets used in this study were adopted from previous studies, which derived them from fluoride-containing crystalline systems and were successfully validated for similar multi-component glass networks^{[35, 52, 53]}. The specific numbers of ions, cell sizes, and partial charges used in the simulations are summarized in Table 2. To ensure charge neutrality within each fundamental simulation cell; the charge of oxygen atoms was adjusted within the range of -1.23 to -1.26. All simulations were performed using the Large-scale Atomic/Molecular Massively Parallel Simulator (LAMMPS) package^[57]. Initial configurations containing approximately 5,000 atoms were generated randomly within cubic periodic cells consistent with the experimentally measured glass densities. For each composition, three independent configurations were subjected to a melt-quench cycle. The long-range Coulombic interactions were calculated using the Ewald summation method with a precision of 1×10^-6, and the equations of motion were integrated using the leapfrog Verlet algorithm with a time step of 1 fs^{[58, 59]}. Each system was initially equilibrated at 300 K for 100 ps under zero pressure, followed by heating to 6000 K, and melting for 100 ps in a canonical (NVT) ensemble. The temperature was maintained for an additional 100 ps to ensure complete melting, then cooled to 300 K at 5 K/ps. Final equilibration was performed in a microcanonical (NVE) ensemble for 100 ps at 300 K. Temperature control during the NVT runs was achieved using a Nosé-Hoover thermostat^[60]. For structural analysis, atomic trajectories were recorded every 50 integration steps over the final 60,000 steps of the NVE simulation at 300 K.

To characterize the structural ordering in the simulated glass systems, both short- and medium-range orders were analyzed. The short-range order was evaluated using coordination number (CN) and bond angle distribution (BAD) analyses, whereas the medium-range order was assessed using Qⁿ distributions. The CN, which represents the number of neighboring atoms within the first coordination shell of a central atom, was determined from atomic pair correlations. BADs were obtained to describe the angular correlations between a central atom and its two nearest neighbors, providing a statistical distribution of the bond angles. The Qⁿ analysis was employed to quantify the degree of network polymerization in structural units such as [SiO₄], [BO₃], and [BO₄], where n represents the number of bridging oxygens (BOs) connected to each network former. The cutoff distances used for CN, BAD, and Qⁿ determinations were selected based on the position of the first minimum in the corresponding pair distribution functions (PDFs).

Mean square displacement (MSD) analyses were performed to investigate atomic mobility and the temperature-dependent diffusion behavior of the glass systems. For each composition, simulations were conducted at 1000, 1200, 1400, 1600, and 1800 K to ensure adequate equilibration within the glass-forming region. Each temperature run consisted of 60 ps of NVT equilibration, followed by 200 ps of NVE dynamics with a 1 fs time step. The atomic trajectories obtained from the NVE ensemble were subsequently used to evaluate the MSD. The long-time linear region of the MSD-time curve, where the MSD increased proportionally with time, was used to extract the diffusion coefficients of both the individual atomic species and the overall glass system. The MSD was calculated using Equation (6):

where N represents the total number of atoms and r_i(t) denotes the position of atom i at time t. To improve statistical reliability, the MSD values were averaged multiple times from the equilibrated NVE ensemble. The diffusion coefficients (D) for each atomic species (Zn, B, Si, and Ca) and the overall system were then determined from the slope of the linear region of the MSD curve using Einstein’s equation:

3.1 Thermal and Physical Properties

The synthesized glasses were transparent, and XRD analysis confirmed their fully amorphous nature (Fig. 1a). As summarized in Table 1, the glass density decreases, whereas the molar volume increases with increasing CaF₂ content. To further evaluate the structural compactness, the atomic packing density and free volume were calculated. As the CaF₂ substitution increases from 0 to 20 mol%, C_g progressively decreases from 0.609 to 0.574, while the free volume expands from 9.82 to 11.95. This trend is primarily attributed to the higher molecular weight of CaF₂ (78.07 g/mol) compared to that of CaO and the formation of a more open network structure. The substitution of divalent oxygen (O^2-) by monovalent fluorine (F^-) disrupts the continuous network linkage, thereby increasing the free volume^{[61, 62]}. As shown in Fig. 1b, the T_g decreases with increasing CaF₂ concentration. The observed decrease in T_g was consistent with the increase in molar volume. This topological relaxation is fundamentally driven by a decrease in 3-D network connectivity, primarily associated with the structural conversion of tetrahedral BO₄ units into planar BO₃ units (as will be detailed in Section 3.2).

Fig. 1. (a) XRD patterns and (b) glass transition temperature variation with CaF₂ content of xCaF₂-(20-x)CaO-20ZnO-50B₂O₃-10SiO₂ glasses (x = 0-20 mol %).

3.2 Glass Structure Characteristics

Fig. 2a presents the ²⁹Si MAS-NMR spectra, which exhibit a progressive shift toward lower chemical shifts with increasing CaF₂ content. Because the highly broadened ²⁹Si signal encompasses five overlapping states (Q⁰ to Q⁴), quantitative deconvolution was avoided to prevent mathematical fitting uncertainty. Instead, the structural evolution was reliably evaluated based on the overall peak shifting trend. The spectra indicated a gradual increase in the polymerized Q³ and Q⁴ species, reflecting enhanced connectivity within the silicate network. In particular, the glass containing 20 mol% CaF₂ exhibited a distinct enrichment of Q⁴ units, suggesting the formation of a more extensively cross-linked Si-O-Si framework. As shown in Fig. 3a, the Si Qⁿ distribution obtained from the MD simulation reveals that low-order units (Q⁰ and Q¹) are nearly absent across all compositions. With increasing CaF₂ content, the fraction of Q² units decreased, while Q³ and Q⁴ species increased, confirming that CaF₂ incorporation suppressed the formation of non-bridging oxygen (NBO) and promoted enhanced network connectivity, consistent with the NMR results.

Fig. 2. (a) ²⁹Si and (b) ¹¹B MAS NMR spectra of A 20-00, A 10-10, and A 00-20 glasses showing increased polymerization with CaF₂ addition.

Structural rearrangement of the borate network was examined using ¹¹B MAS-NMR and MD simulations. As shown in Fig. 2b, the spectra exhibit a clear peak shift toward higher chemical shift values with increasing CaF₂ content. This overall trend indicates an enhanced contribution from trigonal BO₃ units, as the relative intensity of the BO₃ region increased at the expense of the tetrahedral BO₄ region, reflecting a gradual structural conversion upon fluoride incorporation. Furthermore, the specific resonance positions within the BO₃ and BO₄ envelopes also exhibit noticeable shifts. The shift of the BO₄ peak can be attributed to changes in the second coordination sphere, such as an altered distribution of B-O-Si and B-O-B linkages, as the local network evolves. Similarly, the shift of the BO₃ peak is likely driven by changes in these second nearest neighbors, as well as a progressive increase in non-ring BO₃ structures at the expense of ring-type structural units. This observation is consistent with the MD results, which demonstrate a decrease in the average B-O coordination number from approximately 3.30 to 3.22, and a corresponding increase in the BO₃ fraction (Fig. 3b). Furthermore, the O-B-O bond angle distribution shown in Fig. 3c reveals an enhanced intensity near 120° (planar BO₃) and a reduced intensity near 109.5° (tetrahedral BO₄), confirming the structural transition from the BO₄ to BO₃ configuration.

Although the overall B-O coordination number decreased, detailed speciation analysis revealed that the borate network became ordered and polymerized. In the CaO-rich glass (A 20-00), the BO₃ units were distributed between Q² and Q³ configurations (Fig. 3d-f). With increasing CaF₂ content, the proportion of Q² species decreased markedly, while the fully polymerized Q³ configuration became dominant, accounting for approximately 60% of the BO₃ units in the CaF₂-rich glass (A 00-20). Similarly, within the BO₄ group, the fraction of fully connected Q⁴ species exceeded 50% at the highest CaF₂ concentration. These results indicate that despite the reduction in the boron CN, the borate network achieved a higher degree of polymerization owing to the effective elimination of NBOs.

A fluorine-induced charge compensation mechanism can rationalize the polymerization of both the silicate and borate networks. Substitution of CaO with CaF₂ decreased the overall oxygen content of the glass matrix. Highly electronegative F^- ions preferentially coordinate with network-modifying cations (Ca²⁺ and Zn²⁺) to form stable [Ca-F] and [Zn-F] bonds^{[63, 64]}. In fluoride-free glass, these cations typically interact with oxygen atoms, generating NBOs and depolymerizing the glass networks. However, when bonded to F^-, the modifiers are effectively sequestered, diminishing their ability to disrupt the Si-O-Si and B-O-B frameworks.

Fig. 3. (a) Qⁿ distribution of Si species. (b) Average B-O coordination number and fractions of BO₃/BO₄ units. (c) O-B-O bond angle distributions (solid line represent BO₃ unit bond angles and dotted line represent BO₄ unit bond angles). (d-f) Qⁿ distributions of BO₃ and BO₄ units in A 20-00, A 10-10, and A 00-20 glass compositions from MD simulations.

Previous studies have reported that fluoride ions exhibit a strong preference for coordination with network modifiers rather than with network formers^{[63, 64]}. Because F^- cannot effectively penetrate the covalent glass network, it tends to form ionic [Ca-F] and [Zn-F] bonds and cluster within modifier-rich regions, rather than being incorporated into the SiO₄ or BO₄ frameworks^{[63, 64]}. This behavior was also confirmed in the present MD simulations, which revealed an increase in clustering of F^- ions around modifier-rich domains with higher CaF₂ content. This trend is presented in Fig. 4, which depicts the spatial distributions of oxygen and fluorine atoms in the A 20-00, A 10-10, and A 00-20 glasses. As the CaF₂ concentration increased, the regions enriched in modifier cations (Ca²⁺ and Zn²⁺) became progressively associated with fluorine atoms, forming distinct fluorine-rich clusters. These fluoride-enriched modifier domains expanded at the expense of the oxygen-rich network, indicating that fluorine segregation intensified with increasing CaF₂ content. Consequently, oxygen atoms that would otherwise participate in charge compensation remain available to bridge the network formers (Si and B), thereby facilitating the formation of BOs. Thus, despite the overall reduction in the oxygen content, fluorine incorporation suppressed NBO generation and stabilized the highly polymerized structural units, such as Q⁴(Si), Q³(B), and Q⁴(B).

Fig. 4. Cross-sectional images showing the spatial distribution of oxygen (blue) and fluorine (pink) in (a) A 20-00, (b) A 10-10, and (c) A 00-20 glasses, highlighting F-rich clustering near modifier-rich domains.

The interaction mechanism between fluorine and the modifier cations was further elucidated using MD simulations and XPS analyses. As shown in Fig. 5a, the Zn-O CN decreased sharply from 5.15 to less than unity with increasing CaF₂ content. In contrast, the Zn-F CN increased significantly, accounting for approximately 85.8% of the total coordination in the A 00-20 glass. A similar but less pronounced trend was observed for calcium (Fig. 5b), where the Ca-F coordination contribution reached 30.4% at 20 mol% CaF₂. The stronger affinity of fluorine for Zn²⁺ ions compared to that for Ca²⁺ ions can be attributed to the difference in their ionic field strengths, defined by the charge-to-radius ratio (Z/r²). The smaller ionic radius and higher field strength of Zn²⁺ enhance its electrostatic attraction to F^-, leading to the preferential formation of Zn-F bonds. In both cases, the total CN increases with fluorine addition, from 5.15 to 7.01 for Zn and from 8.21 to 9.24 for Ca, indicating a denser local packing environment facilitated by the smaller ionic radius of F^- relative to O^2-.

Fig. 5. MD-derived coordination numbers of (a) Zn-O and Zn-F pairs, and (b) Ca-O and Ca-F pairs in xCaF₂-(20-x)CaO-20ZnO-50B₂O₃-10SiO₂ glasses (x = 0-20 mol %).

These findings are corroborated by the XPS F 1s spectrum of the A 00-20 sample (Fig. 6a), which shows a dominant peak at 684.5 - 685 eV corresponding to ionic Ca-F and Zn-F bonds. The absence of a higher binding energy (~ 688 eV) indicates that fluorine does not form covalent Si-F or B-F bonds, confirming its selective interaction with modifier cations rather than with network formers. Furthermore, the XPS O 1s spectra (Fig. 6b) exhibit a shift toward lower binding energies with increasing CaF₂ content, reflecting a redistribution of the electron density around the oxygen atoms. As Ca²⁺ and Zn²⁺ are sequestered by fluorine, the population of NBOs associated with these cations decreases, whereas the remaining oxygen preferentially forms BOs within the silicate and borate frameworks. These results were consistent with the enhanced network polymerization inferred from NMR and MD.

Fig. 6. (a) XPS F 1s spectrum of A 00-20 glass showing ionic Ca-F and Zn-F bonds. (b) XPS O 1s spectra of A 20-00, A 10-10, and A 00-20 glasses.

Overall, these results revealed a distinct structural duality characterized by local network polymerization within a globally expanded glass matrix. At the local scale, the preferential coordination of fluorine with modifier cations limits their interaction with the glass network, thereby enhancing the polymerization of both the silicate and borate networks. The increased fractions of the Q⁴(Si), Q³(B), and Q⁴(B) species confirmed the reduction in NBOs and the formation of more interconnected local structures. In contrast, from a global topological perspective, substituting oxygens with fluoride interrupts the long-range order (IRO) of the glass matrix. In fluoride-free oxide glasses, divalent modifying cations govern the network topology by acting as ionic cross-linkers, electrostatically bridging multiple NBOs across adjacent silicate and borate segments. However, the preferential coordination of highly electronegative F^- ions to these modifiers leads to the formation of monovalent terminal complexes^[63]. Because F^- acts as a monovalent, non-bridging ligand, it occupies the primary coordination sphere of the modifiers and partially neutralizes their formal charge. This structural sequestration effectively screens the electrostatic field strength of the cations, diminishing their capacity to cross-link multiple network chains. Consequently, while the network-forming structural units may undergo local polymerized, the overall glass network experiences a substantial reduction in ionic cross-linking density and cohesive energy. This decoupling of the modifier and network-former domains leads to a topologically relaxed, more open network structure, which is macroscopically evidenced by the increase in molar volume and the reduction in T_g. As illustrated in Fig 7a-c, the borate and silicate network linkages further depict this structural evolution, revealing that the overall glass framework became progressively more open and spatially expanded with increasing CaF₂ content. To clearly visualize the spatial expansion and topological openness of the primary network-forming backbone (SiO₄, BO₃, and BO₄), modifier cations and fluorine atoms were intentionally omitted from this visualization. Because these modifiers tend to form localized clusters within the glass network, as discussed in Fig. 4, including them would visually obscure the covalent glass network.

Fig. 7. MD simulated atomic configurations of (a) A 20-00, (b) A 10-10, and (c) A 00-20 glasses. SiO₄ tetrahedra shown in blue; BO₃/BO₄ units in cyan.

3.3 Ion Release Measurements

Fig. 8a-e shows the temperature-dependent diffusion coefficients of the B, Si, Ca, and Zn ions, as well as the overall glass system, obtained from the MD simulations. The diffusivities of the B and Si ions decreased with increasing CaF₂ content, consistent with the structural characteristics discussed in Section 3.2. The suppression of NBO formation enhances the local connectivity of the silicate and borate frameworks, thereby restricting the long-range motion of the network-forming cations. This reduced mobility correlates with the increased fractions of Q³ and Q⁴ species identified by both NMR and MD analyses, confirming that fluoride incorporation strengthens local network polymerization.

In contrast, the diffusivities of Ca²⁺ and Zn²⁺ increase with CaF₂ addition, reflecting changes in the local coordination environment of the modifier cations rather than alterations in the Si/B network connectivity. Structural and XPS analyses revealed that highly electronegative fluorine preferentially coordinates with these modifier cations, forming stable [Ca-F] and [Zn-F] bonds. Although these bonds reflect a strong local chemical affinity, the single-bond nature of fluorine locally perturbs the 3D connectivity of the surrounding glass matrix. As Zheng et al. recently demonstrated by atom-probe tomography (APT) and ab initio molecular dynamics (AIMD) simulations in fluorinated glasses, this fluorine-induced perturbation promotes atomic clustering and creates percolated diffusion channels within the structure^[65]. Consequently, the expanded free volume and altered coordination within these loosely packed, fluorine-rich pathways facilitate the migration of Ca²⁺ and Zn²⁺, enhancing their availability for ion exchange during dissolution.

The ICP-MS results shown in Fig. 8f corroborates these findings, indicating a substantial increase in the concentrations of released Ca²⁺ and Zn²⁺ ions with higher CaF₂ substitution. A pronounced rise in release is observed at ≥ 10 mol% CaF₂, coinciding with the simulated enhancement of the overall diffusion coefficient. Collectively, these results demonstrate that fluorine incorporation accelerates ionic transport and dissolution kinetics by creating diffusion channels around the network modifiers, despite the concurrent local polymerization of the silicate and borate units.

It is important to note that while the MD-derived diffusion coefficients were calculated at elevated temperatures representing dry bulk self-diffusion, the ICP-MS leaching experiments were conducted in an aqueous solution at 40 °C governed by surface hydration and ion exchange. Due to the inherent time-scale limitations of classical MD simulations, it is physically unfeasible to obtain a statistically meaningful mean square displacement at room temperature. Therefore, elevated temperatures were strictly employed to achieve sufficient atomic mobility. Despite these mechanistic differences, the high-temperature diffusivities serve as crucial dynamical descriptors of the intrinsic glass network. As Chakraborty et al. recently reported for similar fluorine-modified borosilicate systems, the progressive incorporation of fluorine induces the formation of loosely packed, modifier-rich domains that act as preferential pathways for ion migration^[62]. This intrinsic structural evolution, as evidenced by the enhanced atomic mobility of modifier cations in our MD simulations, logically translates to the accelerated ion exchange and leaching kinetics observed in the low-temperature aqueous environment. The outward migration of Ca²⁺ and Zn²⁺ ions through these percolated diffusion channels governs the early stages of dissolution, enhancing the overall dissolution kinetics and leading to increased Zn²⁺ availability, which is considered a primary contributing factor for improving antibacterial activity. As shown in Fig. 9, the amount of released F^- ions also increases proportionally with CaF₂ content, confirming that fluorine remains stably incorporated within the glass network after melting and actively participates in the interfacial leaching processes.

Fig. 8. Temperature-dependent diffusion coefficients of (a) B, (b) Si, (c) Ca, (d) Zn, and (e) the overall system from MD simulations. (f) 24 h Ca²⁺ and Zn²⁺ ion concentrations measured by ICP-MS.

Fig. 9. 24 h F^- ion concentrations of xCaF₂-(20-x)CaO-20ZnO-50B₂O₃-10SiO₂ glasses (x = 0-20 mol %) measured by ICP-MS.

3.4 FTIR Results and Zeta Potential

As shown in Fig. 10a, the intensity of the OH^- stretching band (3400-3500 cm^-1) decreases systematically with increasing CaF₂ content. This trend is consistent with previous studies reporting reduced molecular water and hydrogen-bonded linkages in the presence of fluoride^[66]. Although KBr pellet-based FTIR detects both bulk and surface hydroxyl groups, as well as potential trace moisture, the reduction in the OH^- band intensity under identical sample preparation conditions strongly indicates a compositional trend. The substitution of hydroxyl groups by F^- ions, facilitated by their comparable ionic radii, reduces the overall concentration of hydroxyl species.

The corresponding change in the surface electrostatic behavior is illustrated by the zeta potential data in Fig. 10b. The magnitude of the zeta potential increases with CaF₂ addition, correlating with the reduction in the OH^- intensity. This behavior indicates that F^- incorporation decreases the density of negatively charged surface sites, thereby shifting the surface potential toward less negative values. This result aligns with the F^-/OH^- exchange mechanism proposed in Section. 1, suggesting that modifier cation (Ca²⁺, Zn²⁺) accumulation may further enhance the surface charge development. The increase in surface charge density has important implications for antibacterial performance. Negatively charged bacterial cell walls are electrostatically repelled by negatively charged glass surfaces, whereas positively charged surfaces disrupt bacterial membranes via electrostatic interactions, leading to cell lysis. Therefore, the observed enhancement in the zeta potential with higher CaF₂ content can be directly associated with improved antibacterial efficiency. Additionally, increased zeta potential may improve the dispersion stability of glass particles by minimizing agglomeration and maintaining a uniform charge distribution, facilitating consistent interfacial ion exchange and sustained antibacterial activity^{[67- 69]}.

Fig. 10. (a) FTIR spectra showing decreased OH^- stretching intensity with increasing CaF₂ content. (b) Zeta potential variation of xCaF₂-(20-x)CaO-20ZnO-50B₂O₃-10SiO₂ glasses (x = 0-20 mol %).

3.5 Antibacterial Properties and Cytotoxicity Evaluation

Fig. 11 shows the antibacterial activities of the prepared glasses against E. coli and S. aureus. The antibacterial efficiency was evaluated using the R-value, where R = 2.0, 3.0, and 4.0, corresponding to antibacterial activities of 99.0%, 99.9%, and 99.99%, respectively. After 24 h of exposure, all glass samples exhibited antibacterial activity exceeding 99.9 %, demonstrating strong inhibition of both bacterial species. The enhanced antibacterial effect with increasing fluorine substitution correlated with the increased release of Zn²⁺ ions from the glass network. To further distinguish the compositional effect on antibacterial behavior, an additional 1 h antibacterial assay was conducted. For E. coli, the A 20-00 glass exhibited an antibacterial efficiency of 91.6%, whereas the A 00-20 glass achieved 97.0%. For S. aureus, all glass compositions showed near-complete bacterial inhibition (~ 99.9%), confirming the broad-spectrum antibacterial activity of the fluorine-containing glass series. Overall, these results confirm that all compositions achieved ≥ 99.9% antibacterial efficiency against both E. coli and S. aureus, underscoring their strong potential as antibacterial glass materials.

The cytotoxicity of the glass powders was evaluated using the toxicity unit (TU) approach, in which a TU value ≥ 1.0 indicates high toxicity. In this study, 5 g of each glass sample was dissolved in 1 L of purified water, and the resulting solutions were analyzed. All samples exhibited TU values of approximately 0.9, indicating nontoxicity according to standard criteria. These results indicate that the prepared glasses exhibit excellent antibacterial performance without cytotoxic effects, suggesting their potential applicability as biomedical and antimicrobial coatings.

Fig. 11. Antibacterial activity of xCaF₂-(20-x)CaO-20ZnO-50B₂O₃-10SiO₂ glasses (x = 0-20 mol %) against E. coli and S. aureus after 24 h incubation.