Anionic Structure in Molten CryoliteâAlumina Systems - The Journal of

Article pubs.acs.org/JPCC

Cite This: J. Phys. Chem. C 2018, 122, 21807−21816

Anionic Structure in Molten Cryolite−Alumina Systems Kelly Machado,*,† Didier Zanghi,† Mathieu Salanne,‡ Vincent Stabrowski,§ and Catherine Bessada† †

CEMHTI UPR3079 CNRS, Univ. Orléans, F-45071 Orléans, France Sorbonne Univ., UPMC Univ Paris 06, CNRS, PHENIX, F-75005 Paris, France § EMRA Aluval, F-38340 Voreppe, France J. Phys. Chem. C 2018.122:21807-21816. Downloaded from pubs.acs.org by UNIV OF LOUISIANA AT LAFAYETTE on 09/27/18. For personal use only.

‡

ABSTRACT: For aluminum production, the alumina (Al2O3) dissolution in the electrolyte is one of the important step in the industrial process. The electrolyte is a cryolitic bath containing mainly NaF and AlF3 at around 1000 °C. In this liquid, the main anionic species are fluoroaluminates ions such as [AlF6]3−, [AlF5]2−, [AlF4]−, and F−. During the Al2O3 dissolution, different kinds of oxyfluoroaluminates species are formed. However, due to the difficult experimental conditions and the large number of potential ions, no quantitative speciation could be made up to now. Here, we propose a speciation of alumina−cryolite melts combining in situ NMR experiments, classical molecular dynamics simulations, and electronic structure calculations. This allows us to establish the nature and quantities of each species, depending on the Al2O3 concentration, for two different initial cryolitic compositions. In particular, we show that the O atoms are always linked to at least two Al atoms, leading to the formation of polymeric oxyfluorides. From the dynamic point of view, the simulations also show that the lifetime of the O−Al bonds is also much longer than that of the F−Al ones.

1. INTRODUCTION In the Hall−Héroult process, alumina (Al2O3) is dissolved in cryolitic melts (NaF−AlF3) at around 1000 °C and aluminum is produced by electrolysis.1 The composition of the cryolite− alumina melt is an important factor for the efficiency of the industrial process. However, the alumina dissolution mechanism in such melts is still open to discussion.1 Although it is well accepted that the dissolution results in the formation of oxyfluoroaluminates anions [AlxOyFz]3x−2y−z, the exact speciation remains unknown. Many different structural entities in molten cryolitic−alumina systems were proposed depending on the composition of the melt and on the investigation tool. To illustrate this variety, Grjotheim2 listed 22 different complexes that might exist. The main techniques used to probe their nature were Raman spectroscopy, thermodynamic modeling, and, more recently, in situ high-temperature NMR.3−7 In summary, it is therefore generally accepted that the oxygen adopts a bridging position between two aluminum atoms. The skeleton of the resulting fluoroaluminate species are Al−O−Al at low concentration of alumina and Al O O Al at high 3−6 concentration. When the composition of the initial cryolitic melt is varied by changing the cryolitic ratio, CR = n(NaF)/ n(AlF3), similar variations occur. The dominant type of oxyfluoroaluminate complexes are based on Al−O−Al for CR < 3 and on Al O O Al for CR ≥ 3. The most cited structures are 2− [Al2OF6] , [Al2OF8] 4−, [Al2 OF10]6−, [Al2O 2F4 ]2−, and [Al2O2F6]4−. However, due to the different results obtained by different methods, there are still several aspects that are not well established about the nature, the structure, and the proportion of these complexes. © 2018 American Chemical Society

To address these issues, we propose to combine in situ NMR measurements at high temperature, molecular dynamics (MDs) simulations, and density functional theory (DFT) calculations. This approach was already successfully used to obtain the speciation in the NaF−AlF3 molten system7 for a wide range of compositions (5−50 mol % of AlF3). In this range of compositions, free fluorine (F−) and different coordination complexes were identified: AlF63−, AlF52−, and AlF4−. A majority of free fluorine (F−) was observed up to 25 mol % of AlF3, then the AlF52− species become the majority between 25−40 mol % of AlF3, with a maximum of 60% around 33 mol % of AlF3. The AlF4− entities start to form at 25 mol % and rapidly become the dominant fluoroaluminates species. The proportion of AlF63− complexes remains very minor regardless of the AlF 3 composition. Using the same methodology, we quantify the impact of adding alumina up to the saturation (8 mol % of Al2O3) on the speciation of molten NaF−AlF3 for two cryolitic ratio (CR = 3 and 2.2). These two compositions were chosen because they play a major role in the production of aluminum by electrolysis and the range of the alumina addition was chosen to cover the industrial interest. We will present as a first step the calculation of different physical quantities carried out to validate the interatomic interaction potential used in the molecular dynamic simulations. The speciation of these molten ternary systems then will be presented and discussed in details by analyzing the ion trajectories obtained from molecular dynamics simulations and the NMR chemical shift obtained experimentally. Received: July 18, 2018 Revised: September 6, 2018 Published: September 18, 2018 21807

DOI: 10.1021/acs.jpcc.8b06905 J. Phys. Chem. C 2018, 122, 21807−21816

Article

The Journal of Physical Chemistry C

details in a previous publication.7 In the present work, only the parameters of the atomic pairs involving the oxygen atom were adjusted because the pair potential parameters for Al3+, F−, and Na+ were taken from the previous study on NaF−AlF3.7 All these values are reported in Table 1. The dipole polarizations of

2. METHODS 2.1. NMR Experiments. To carry out oxygen NMR experiments, 17O-enriched α-Al2O3 was synthesized from aluminum isopropoxide and water containing 46% of 17O isotope.8 Several compositions from the NaF−AlF3−Al2O3 (CR = 3 and 2.2) system were prepared in a glovebox under dried argon by mixing proportions of AlF3 (Alfa Aesar 99.99%), NaF (Alfa Aesar, 99.995%), and 17O-enriched α-Al2O3. All the samples were confined in boron nitride crucibles (without oxide binders) of high purity, with an outside diameter of 9 mm and a height of 14 mm. The crucible is sealed using a screw cap. For high-temperature NMR measurements, we used a Bruker AVANCE I NMR spectrometer operating at 9.4 T equipped with a CO2 laser-heated system (250 W, 10.6 μm), which irradiate the crucible directly, symmetrically, and simultaneously from above and below, making it possible to ensure a good homogeneity of heating by limiting the temperature gradients at the level of the sample. This new configuration also enables to use a greater amount of powder that in previous work9 to improve the NMR signal-to-noise ratio. The axial probe, originally intended for liquid NMR, has been specially modified in the laboratory in collaboration with Bruker for use at high temperatures. It is a multicore static probe with two radio frequency channels (channel: X and 1H−19F), which can be tuned during the temperature experiment. To preserve the radio frequency coil and the electronic components during the hightemperature experiments, a double cooling circuit is used with circulating air at ambient temperature and pressure. During the heating, the BN crucibles are kept under a continuous stream of argon to prevent their oxidation. The experimental setup has been described in details in previous publications.9−11 In the liquid state, the main concern is the isotropic part of the shielding tensor: σiso = (σxx + σyy + σzz)/3, due to the fast tumbling of molecules. The isotropic shielding tensor is usually relative to a measured reference, σref. The shielding σ and chemical shift δ, expressed in ppm, are related by δiso = 106(σiso − σref)/σref. In this study, the chemical shifts of all nuclei (17O, 27Al, 19 F, and 23Na) are referenced to H2O, 1 M aqueous solutions of Al(NO3)3, CFCl3, and NaCl, respectively. High-temperature NMR measurements give a single narrow signal for each nucleus. Its position is the average over time of the chemical shifts of all configurations sampled by the species during acquisition, weighed by their proportion.11 For each experiment, the samples were heated up to 20 °C above the corresponding melting point. After thermal homogenization, the spectra for each nucleus were recorded sequentially (19F, 27Al, 23Na, and 17O) for the same sample. The accuracy of temperatures is estimated to be within ±10 °C. To avoid any evolution of the sample at high temperature in the confined container, the duration of the experiments did not exceed 20 min. 2.2. Molecular Dynamic Simulations. We perform molecular dynamics simulations in the framework of the polarizable ion model (PIM).12,13 This model is a sum of four pairwise additive contributions: charge−charge, dispersion, overlap repulsion, and polarization. All the parameters of the interatomic potential were obtained by matching the forces and dipoles of each atom and the stress tensor of the cell calculated by the PIM with those obtained by ab initio calculations with Vienna Ab initio Simulation Package code.14−16 The analytical expressions used, as well as the ab initio adjustment procedure used to obtain the potential parameters, have been described in

Table 1. Parameters of the Polarizable Interatomic Potential for the Molten [NaF−AlF3]−Al2O3 for CR 3 and CR 2.2 System (All the Parameters are Given in Atomic Unit (au)) ion pair

Bij

αij

C6ij

C8ij

O −O O2−−F− O2−−Al3+ O2−−Na+ F−−F− F−−Al3+ F−−Na+ Al3+−Al3+ Al3+−Na+ Na+−Na+

134.38 167.58 58.60 48.20 392.57 37.97 52.4 7.57 0.23 0.37

2.01 2.14 1.79 1.84 2.48 1.88 1.97 6.50 6.37 1.38

44.00 28.20 2.00 2.00 3.11 1.89 9.43 60.0 29.98 3.79

853.00 391.70 25.00 25.00 104.93 100.43 132.55 400.00 400.00 399.88

2−

2−

the nuclei obtained from the adjustment procedure in atomic units (au) are 11.036, 8.158, and 0.887 au for O2−, F−, and Na+, respectively, whereas Al3+ ions were considered to be nonpolarizable. All the calculations were carried out at 20 °C above the melting point in cubic cells containing approximately 190 atoms for eight compositions of the NaF−AlF3−Al2O3 system at CR = 3 and 2.2. The parameters of the cubic cells (composition, number of atoms, and volume) used for simulations are provided in Table 2. Initially, the mixtures were equilibrated in the NPT ensemble following the method described by Martyna et al.,17,18 with a fixed pressure of 0 GPa and a fixed temperature. After 200 ps of equilibration with a time step of 0.1 fs, production runs of 1000 ps (with a time step of 0.1 fs) were conducted for each composition in the NVT ensemble using a Nosé−Hoover thermostat.17,18 2.3. DFT Computational Details. The first-principle calculations of NMR parameters have been performed using the density functional theory (DFT), which takes into account the periodicity of the crystalline structure through the implementation of periodic boundary conditions. The chemical shielding tensors for all atoms in the selected system were computed using the gauge including projector-augmented wave method19 implemented in NMR-CASTEP code.20,21 In the calculations, ultrasoft pseudopotentials (USPP)22 and Perdew− Burke−Ernzerhof functional within generalized gradient approximation were used. The plane-wave basis with the USPP was fixed in an energy cutoff of about 610 eV. The Brillouin zone was sampled using a Monkhorst−Pack grid23 spacing of 0.05 Å−1, corresponding to a k-point mesh of 1 × 1 × 1. The core radius of pseudopotential for NMR calculations for 19 27 F, Al, 23Na, and 17O were, respectively, 1.4, 2.0, 1.3, and 1.3 in atomic units (au). To take into account the thermal agitation of the ions due to the temperature in the NMR parameters calculation, 20 snapshots were extracted every 50 ps along the MD trajectory simulated in the NVT ensemble without further geometry optimization. This spacing ensures that the atomic configurations are uncorrelated.24 Compared to previous work,7 the number of snapshots needed to be increased to obtain a 21808


Article

The Journal of Physical Chemistry C Table 2. Molecular Dynamics Simulation Conditions CR = 2.2

CR = 3

number of atoms

number of atoms

Al2O3 mol %

O

F

Al

Na

total

V (Å3)

T (K)

O

F

Al

Na

total

V (Å3)

T (K)

1 2 3 4 5 6 8

2 4 6 8 10 12 16

107 106 105 103 101 100 100

22 23 24 25 26 27 30

45 45 45 44 43 43 42

176 178 180 180 180 182 188

3059 3064 3053 3040 3090 3107 3269

1280 1260 1249 1240 1245 1275 1342

2 4 6 8 10 12 16

100 99 101 100 98 96 97

18 19 21 22 23 24 27

50 50 50 50 49 48 48

170 172 178 180 180.0 180 188

2926 2935 3025 3032 3028 3048 3180

1290 1280 1272 1263 1254 1275 1330

converged chemical shift for the oxygen nuclei. The magnetic shielding of all the nuclei was calculated and averaged for each atomic configuration (snapshot). For each composition, the isotropic chemical shielding was obtained by averaging the values obtained from the twenty snapshots.7 Therefore, to convert the calculated isotropic chemical shielding into isotropic chemical shift, a linear regression between calculated σiso values and experimental δiso values was established on numerous compounds by keeping the same calculation parameters obtained previously. For 19F, the relationship reported by Sadoc et al.25 was used, and for 27Al, 27

Na, and 17O, the following relationships were used: δiso

23

23

−1.0764*σiso (ppm) + 590.84, δiso 19

467.9, and δiso = −0.8953*σiso

F

Na

Al

=

= −0.840*σiso (ppm) +

(ppm) + 230.95, respectively.

3. RESULTS AND DISCUSSION 3.1. Validation of the Interaction Potential for the Ternary System NaF−AlF3−Al2O3. In Figure 1, the calculated

Figure 2. NMR chemical shifts 20 °C above the corresponding melting points for 27Al (blue circle solid blue circle open), 17O (green box solid green box), 23Na (purple triangle up solid purple triangle up open), and 19 F (red diamond solid red diamond open): experimental values in open markers and results of calculation in full markers for (a) [NaF− AlF3]CR 3−Al2O3 and (b) [NaF−AlF3]CR 2.2−Al2O3. Figure 1. Calculated (■ ⬤) and experimental (□ ◯)1,26 density of the melt as a function of mol % of alumina for CR = NaF/AlF3 = 2.2 (■□) and CR = 3 (⬤◯).

To validate the potential in a more robust way, in Figure 2, we confronted the experimental chemical shifts deduced from NMR measurements with those calculated from the ionic positions derived from MD simulations according to the procedure described previously. Whatever the nuclei, the experimental chemical shifts are well reproduced by the computation. For 27Al, an increase in the chemical shift values is clearly observed upon the addition of alumina, both in the experimental and the simulation results. This trend indicates that alumina dissolution changes the local environment of the aluminum ions. The chemical shifts of 23Na and 19F nuclei remain constant (experimentally: −7.5 ± 0.3 and −198 ± 2 ppm, respectively, for

densities from molecular dynamics simulations at 20 °C above the liquidus temperature are compared to FactSage assessment26 based on the available experimental data for the mixtures [NaF−AlF3]CR+ x mol % Al2O3 (CR = 3 or 2.2 and x = 1, 8). The density is predicted with an error of less than 0.05 g cm−3 up to 8 mol % of alumina. This good agreement is however not sufficient to test the ability of the interatomic potential developed for MD simulations to describe the studied system realistically. Note also that the density is not much affected by the addition of alumina. 21809


Article


Table 3. RDF Analysis at Different Compositions Obtained by MD Simulation

CR 3

CR 2.2

Figure 3. Total ionic conductivity obtained from molecular dynamic simulations (⬤■) and experimental results (◯ □).1,26 The melting point as a function of the alumina content is represented by the line.

position of the first maximum (Å)

position of the first minimum (Å)

Al2O3 mol %

Al−O

Al−F

Al−O

Al−F

1 2 3 4 5 6 8 1 2 3 4 5 6 8

1.752 1.754 1.747 1.749 1.748 1.728 1.704 1.730 1.755 1.753 1.751 1.760 1.739 1.744

1.776 1.778 1.771 1.773 1.772 1.777 1.752 1.778 1.779 1.777 1.775 1.760 1.763 1.744

2.066 2.061 2.060 2.069 2.062 2.059 2.065 2.083 2.098 2.103 2.108 2.116 2.108 2.092

2.210 2.212 2.205 2.199 2.199 2.190 2.183 2.177 2.179 2.176 2.180 2.170 2.166 2.159

Figure 4. Viscosity obtained from molecular dynamic simulations of 5 ns (⬤ ■) and experimental results (◯ □).1,26 Figure 6. Activation energy of Al−O, Al−F, and Na−F pairs as a function of mol % of alumina for CR = NaF/AlF3 = 2.2 (●) and CR = 3 (blue box solid).

addition of oxide. The number and the nature of the anions or cations around these nuclei probably remain identical whatever the alumina content in the bath. Finally, for 17O, it is more difficult to establish clear trends. We observe a decrease in the experimental chemical shift from 24 ppm at 1 mol % Al2O3 to 9 ppm at 8 mol % of Al2O3 for CR 3 and from 21 ppm at 0.7 mol % Al2O3 to 9 ppm at 8 mol % of Al2O3 for CR 2.2; however, these values are within the error range of the calculated values. The large error bars in the calculations are caused by the small number of oxygen atoms in the simulation cells, whereas experimentally they were evaluated according to the temperature incertitude, the linewidth and the repeatability of the experiments. In the calculations, the uncertainties were estimated according to the dispersion of the linear regression between chemical shift and chemical shielding (±5 ppm for 27Al and 23Na, ±7 ppm for 19F, and ±12 ppm for 17O). To test the classical interaction potential used in the MD simulations, the total ionic conductivity from the simulated ionic trajectories was also calculated. This quantity was computed according to27

Figure 5. Radial distribution functions obtained by molecular dynamics simulations of molten system [NaF−AlF3]CR 3−4 mol % Al2O3.

CR 3 and −7.5 ± 0.3 and −197 ± 2 ppm, respectively) as alumina is added in the range of the studied compositions. From a qualitative point of view, this indicates that these nuclei present a first solvation sphere that does not evolve markedly with the 21810


Article


Figure 9. Cage relaxation time (τ) for Al−F (up) and Al−O (down) pairs as a function of alumina concentration for cryolitic ratio of 3 (■) and 2.2 (○).

i e2 1 jj σ= lim t →∞ jjjj kBTV 6t jj k

2y

∑ qiδri(t ) i

zz zz zz zz {

(1)

where e is the elementary charge, qi is the formal charge of the atom i, and δri(t) is the displacement of ion i in time t. This physical value, which describes the ability of the electrolytes to transport all the charges without distinction, is of particular interest to the aluminum production industry to improve the electrical efficiency of the electrolysis cells. In many cases, these experimental data are difficult to obtain due to the corrosive environment and the high temperatures. For this reason, in Figure 3, the calculated values are compared with the experimental conductivity obtained from the thermodynamic software FactSage.1,26 The calculated values reproduce quite closely the experimentally observed trends. Over the range 0−5 mol % of Al2O3, the addition of alumina is associated with a decrease in the melting point (1009−961 °C for CR 3 and 988− 940 °C for CR 2.2). Then, to dissolve more alumina, up to 8 mol % of Al2O3, the melting point is increased significantly from almost 127 °C (CR 3) or 185 °C (CR 2.2). Despite these quite large variations in the melting temperature, the corresponding electrical conductivity does not vary much. The viscosity of the melts was calculated using the formalism of Green−Kubo for long trajectory (5 ns) and compared with the experimental data1,26 (Figure 4). A good agreement was obtained. Alumina addition slightly increases the viscosity, given the appearance of oxyfluoroaluminate complexes larger than the fluoroaluminate species are poorly connected to each other. According to these results, the interatomic potential model developed is able to predict the dynamic and transport properties in the molten phase of the [NaF−AlF3]CR−Al2O3 mixture over a broad range of alumina compositions (0−8 mol %); note that no experimental information was used in the parameterization procedure. Using the data from the MD simulations, it is also possible to describe the local structure of the simulated liquid state. 3.2. Analysis of the Radial Distribution Functions (RDFs). In the structural analysis of the molten liquid, the radial distribution function (RDF) gives the first information about the structure. The RDFs for the Al−F and Al−O pairs present a first

Figure 7. Radial distribution functions (RDF) of Al−Al ionic pairs of the [NaF−AlF3]CR−Al2O3 molten system as a function of alumina concentration (CR = 3 top and CR = 2.2 down).

Figure 8. Cage autocorrelation functions out (black line) and in & out (red line) for the mixture CR 2.2 + 4 mol % Al2O3. The larger figure corresponds to the Al−F function and the smaller one to the Al−O function.

21811


Article


Figure 10. Visualization of one snapshot from the molecular dynamic trajectory for CR 3 (left) and CR 2.2 (right) with 3 mol % of alumina (sky blue circle solid, aluminum; gray circle solid, fluorine; red circle solid, oxygen). Sodium atoms were purposely omitted for the best observation. The averaged length of Al−F and Al−O bonds are 1.83 and 1.79 Å, respectively.

exchange of sodium ions takes place easily between the first and second coordination spheres of aluminum and fluorine ions. The peak positions of the radial distribution function are presented in Table 3. To quantify the “strength” of the bond, we plot the activation energy of the Al−O, Al−F, and Na−F bonds as a function of the molar percentage of alumina (Figure 6). This activation energy was deduced from the potential mean force (PMF)29 PMF (kJ mol−1) = −kBNAT ln[g (r )]

(2)

where kB is the Boltzmann constant (1.38064852 × 10−23 J K−1), NA is the Avogadro number (6.022140857 × 1023 mol−1), and T is the temperature in kelvin. This energy barrier is notably larger for the Al−O pair than for the Al−F pair. There is therefore a competition between the ions O2− and F− to surround the Al3+ ions. These energies, however, are much greater than the energy required to link alkaline ions to fluorine ions (see Figure 6). The RDF of the Al−Al pair also indicates the formation of an intermediate order in the structure of the liquid around the aluminum ions upon addition of alumina. In Figure 7, we observe a localized pre-peak at a distance much shorter (∼3.2 Å) than the main peak (∼6 Å). The amplitude of this peak increases as a function of the alumina concentration. This shorter distance between aluminum ions could be explained by the formation of dimers of the Al−X−Al type (with X = F or O) due to the existence of bridging fluorine or oxygen ions, which connect the anionic complexes (AlFx)3−x observed in the binary systems.6,7,9 Nevertheless, the fact that the height of the peak changes with the quantity of oxygen suggests that the Al−O−Al type bonds become preponderant in the organization of the liquid structure with the addition of alumina. 3.3. Cage Autocorrelation Functions. The dynamics associated with the relaxation of the first coordination sphere of Al3+ ions was studied by means of cage correlation functions that allow to investigate the dynamics of the exit or the entrance of F− or O2− ions inside this sphere.30 Because these functions have an exponential decay as a function of time (Figure 8), the corresponding relaxation times τ can be defined. For example,

Figure 11. Visualization of a oxyfluoroaluminate chain in [NaF− AlF3]CR 3−[Al2O3]6 mol % system (sky blue circle solid, aluminum; gray circle solid, fluorine; red circle solid, oxygen). Sodium atoms were purposely omitted for the best observation.

narrow and very intense peak, as shown in Figure 5, characteristic of a strong interaction leading to the formation of well-defined coordination complexes. Moreover, a marked minimum is observed with a very low value of g(r). This indicates that fluoride ions must pass through a large barrier to exit the first sphere of solvation of aluminum ions.28 On the other hand, the first peak of the RDF of the Na−F pair is much less intense and wider, revealing that the first sphere of coordination is less structured around Na+ cations (Figure 5). The higher value of g(r) on the first minimum suggests that the 21812


Article


Figure 12. Calculated atomic fractions of fluorine as a function of alumina addition: CR = 3 top and CR = 2.2 down.

the exit time is taken when the cage “out” function takes a value of 1/e, which means that 1/e of the aluminum ions in the system have lost a fluorine or oxygen in their coordination sphere. Another autocorrelation function measures the entry and exit of the cage (cage “in & out”), allowing to determine how fast the first solvation sphere of a given ion is modified. As we can see in Figure 8, the two curves cannot be superimposed, contrary to other systems such as fluoroberyllates in which there is a privileged coordination for the multivalent cationic species. Here, the loss or the gain of an F− does not necessarily lead to the arrival or departure of another one because aluminum can adopt different coordination numbers. On the contrary, for the Al−O pairs, the out and the in & out functions have a similar decay, indicating that an oxygen ion leaves the aluminum shell only when it is replaced by a new one. The extracted relaxation times for the Al−F and Al−O pairs are represented in Figure 9. Whatever the system, the Al−F relaxation times are very short, in the order of 6−9 ps. The Al−O relaxation time is higher, in the order of 100−400 ps. This indicates that the Al−O exchanges are several orders of magnitude slower than the Al−F ones. The presence of oxygen ions therefore tends to stabilize the species present in the liquid. Finally, the Al−F and Al−O relaxation times are always lower for the cryolitic ratio (CR) of 2.2 system than for CR 3 system. The exchanges between the fluorine (or oxygen) and aluminum ions are “slower” with the higher CR. This is in agreement with the activation energy results presented previously.

3.4. Speciation of the Melt. The MD simulations allow to obtain the positions in the three-dimensional space of all ions present in the simulation box along the trajectory. By defining a cutoff radius as the first minimum of the RDF for each atomic pair, it is possible to count the number of a specific atom around a selected ion, thus giving an instantaneous coordination number. The coordination of each atom is thus characterized by a large number of atomic configurations (20 000 configurations on average) extracted from a trajectory of 5 ns, which provides the speciation of the system with a good reliability. From our calculation, the statistical analysis of the Al3+, F−, and O2− ions trajectories indicates that there are no isolated [AlOFx]x− species. All the aluminum ions in these monomers are connected by at least one bridging oxygen atom. And, all oxygen ions are connected to two or three aluminum ions. This result is in agreement with several previous studies.5,6,9,31 The species founded are F−, [AlFx]3−x, [Al2Fx]6−x, [AlyOFx]3y−2−x, and [AlyO2Fx]3y−4−x (Figure 10). In all the studied compositions, no [Al2O2Fx]2−x species were found. However, at high alumina concentration, long chains are formed such as AlyO3Fx3y−6−x (y = 4 and x = 10, 11, 12, or y = 5 and x = 14, 15, 16, or y = 6 and x = 19) (Figure 11). The corresponding proportions can be expressed using various definitions, which give complementary information. First, we quantify them by using their fluorine atomic fractions in Figure 12, where the label “others” are the different types of long chains. 21813


Article


Figure 13. Calculated anionic fractions as a function of alumina addition: F−, ■; AlFx3−x, red circle solid; Al2Fx6−x, green diamond solid; AlyOFx3y−2−x blue triangle up solid; and AlyO2Fx3y−4−x, pink triangle down solid (CR = 3 top and CR = 2.2 down).

create Al−O bonds and consequently fluoride ions are released in the bath, which can have strong implications on the industrial process. The anions [AlyOFx]3y−2−x (y = 2 and x = 6, 7, 8 or y = 3 and x = 11, 12) are the first complexes with oxygen to be formed when alumina is added, whereas the anions [AlyO2Fx]3y−4−x (y = 3 and x = 8, 9 or y = 4 and x = 13) appear for a higher alumina concentration only (Figure 13). By representing separately the fluoroaluminates (Figure 14), we notice that dimers [Al2Fx]6−x disappear above 4 mol % (for CR 3) and 6 mol % (for CR 2.2) of alumina. The 5-fold coordinated aluminum (AlF52−) concentration is affected by the addition of alumina, its anionic fraction decreasing from 30 to 20% for CR 3 and from 50 to 30% for CR 2.2 due to the strong participation of this species in the formation of oxyfluoroaluminates. This has already been observed by Raman spectroscopy.6 The fluoroaluminate AlF63− also decreases from 22 to 12% and from 11 to 7% for CR 3 and CR 2.2, respectively. Considering all these results, five forms of oxyfluoroaluminates appear with addition of alumina; fluoride ions are released when oxygen atoms are introduced in the melt and the

The fluorine atoms are mainly engaged on fluoroaluminates complexes (AlFx3−x). Five types of oxyfluoroaluminates were found in our simulation: [Al2OFx]4−x, [Al3OFx]7−x, and [Al4OFx]10−x anions with one oxygen atom, [Al3O2Fx]5−x and [Al4O2Fx]8−x with two oxygen atoms, and even longer chains for high alumina concentration were observed. The first oxyfluoroaluminates to be formed are composed of one oxygen atom. For CR 3, [Al2OFx]4−x are the major oxyfluoroaluminates from 0−8 mol % of Al2O3, whereas for CR 2.2, the predominant species are [Al2OFx]4−x and [Al3OFx]7−x until 3 mol % of Al2O3 and [Al2OFx]4−x for a higher alumina concentration. The speciation was also quantified by using the anionic fractions, as shown in Figure 13, where anionic fractions lower than 1% are ignored. An analysis of Figure 13 shows that when alumina is added to the system, the amount of free fluorine increases around 20%, whereas the amount of fluoroaluminates [AlFx]3−x (x = 4, 5, or 6) decreases from ∼54 to ∼32% for CR 3 and from ∼70 to ∼43% for CR 2.2 (0−8 mol % of alumina). The quantity of dimers [Al2Fx]6−x (x = 9, 10) also decreases with the addition of alumina. These results show that Al−F bonds are broken to 21814


Article


investigations confirm the presence of free F− and different ionic complexes [AlF 6]3−, [AlF5]2−, [AlF4]−, [Al2F9]3−, [Al2F10]4−, [Al2OF6]2−, [Al2OF7]3−, [Al2OF8]4−, [Al3OF11]4−, [Al3OF12]5−, [Al3O2F8]3−, [Al3O2F9]4−, and [Al4O2F13]5− in which anionic fractions evolve as a function of alumina content. The anionic structure is strongly dependent on the addition of alumina and cryolitic ratio.

■

AUTHOR INFORMATION

Corresponding Author

*E-mail: [email protected]. ORCID

Kelly Machado: 0000-0001-6045-5734 Mathieu Salanne: 0000-0002-1753-491X Notes

The authors declare no competing financial interest.

■

ACKNOWLEDGMENTS This study was financially supported by the ANR MIMINELA project of the French National Research Agency. All the computations presented in this work were performed at the “Centre de Calcul Scientifique en Région Centre” facility (CCRS, Orléans, France) under the CASCIMODOT program. We are grateful for the technical support by François Vivet in this work. The authors would like to thank Drs Aydar Rakhmatullin and Mallory Gobet for the synthesis of 17Oenriched α-Al2O3.

■

REFERENCES

(1) Thonstad, J.; Fellner, P.; Haarberg, G. M.; Híveš, J.; Kvande, H.; Sterten, Å. Aluminium Electrolysis, Fundamentals of the Hall-Heroult Process, 3rd ed.; Aluminium-Verlag: Düsseldorf, 2001. (2) Grjotheim, K. Aluminium Electrolysis: Fundamentals of the HallHéroult Process; Aluminium-Verlag: Germany, 1982. (3) Sterten, Å. Structural Entities in NaF-AlF3 Melts Containing Alumina. Electrochim. Acta 1980, 25, 1673−1677. (4) Kvande, H. Structural Species in CryoliteAlumina Melts as Determined From Vapor Phase Studies. Electrochim. Acta 1980, 25, 237−240. (5) Daněk, V. Physico-Chemical Analysis of Molten Electrolytes; Elsevier Science: Amsterdam, 2006. (6) Robert, E.; Olsen, J. E.; Daněk, V.; Tixhon, E.; Ostvold, T.; Gilbert, B. Structure and Thermodynamics of Alkali Fluoride-Aluminum Fluoride-Alumina Melts. Vapor Pressure, Solubility, and Raman Spectroscopic Studies. J. Phys. Chem. B 1997, 101, 9447−9457. (7) Machado, K.; Zanghi, D.; Sarou-Kanian, V.; Cadars, S.; Burbano, M.; Salanne, M.; Bessada, C. Study of NaF−AlF3 Melts by Coupling Molecular Dynamics, Density Functional Theory, and NMR Measurements. J. Phys. Chem. C 2017, 121, 10289−10297. (8) Kim, H.-E.; Zinkle, S. J.; Allen, W. R. Preparation and Characterization of 17O-Enriched Alumina for Nuclear Fusion Damage Experiments. J. Am. Ceram. Soc. 1990, 73, 425−429. (9) Lacassagne, V.; Bessada, C.; Florian, P.; Bouvet, S.; Ollivier, B.; Coutures, J. P.; Massiot, D. Structure of High-Temperature NaF-AlF3Al2O3 Melts: A Multinuclear NMR Study. J. Phys. Chem. B 2002, 106, 1862−1868. (10) Sarou-Kanian, V.; Rollet, A.-L.; Salanne, M.; Simon, C.; Bessada, C.; Madden, P. A. Diffusion Coefficients and Local Structure in Basic Molten Fluorides: In Situ NMR Measurements and Molecular Dynamics Simulations. Phys. Chem. Chem. Phys. 2009, 11, 11501− 11506. (11) Nuta, I.; Veron, E.; Matzen, G.; Bessada, C. High Temperature NMR Study of Aluminum Metal Influence on Speciation in Molten NaF-AlF3 Fluorides. Inorg. Chem. 2011, 50, 3304−3312.

Figure 14. Description in terms of different aluminum-bearing species without oxygen in NaF−AlF3−Al2O3 system: F−, ▲; AlF63−, ⊗; AlF52−, ◇; AlF4−, ●; Al2F93−, ★; Al2F104−, ☆ (CR = 3 top and CR = 2.2 down).

fluoroaluminate fractions decrease. The mono-oxyfluoroaluminates [AlyOFx]3y−2−x grow until 3 mol % of Al2O3 and then slowly decrease. The anionic complexes [AlyO2Fx]3y−4−x are formed after the addition of 6 mol % of alumina. Finally, for both systems, the dominant oxy-anions are [AlyOFx]3y−2−x in the range of the studied compositions.

4. CONCLUSIONS The local structure in molten fluoride salt NaF−AlF3 (CR 3: 75−25 mol % and CR 2.2: 69−31 mol %) with different alumina additions (0−8 mol %) has been described. The framework of this work was to build a polarizable interaction potential that reproduces the structure of the molten NaF−AlF3−Al2O3 mixtures. The comparison of experimental NMR results with the data obtained from the combination of MD simulations and DFT calculations provides an essential and particularly solid approach for the evaluation of the ionic potential of atomic interactions used in our molecular dynamic simulations. This comparison enables us to validate this potential with confidence and determine the speciation in the molten phase. Our 21815


Article

The Journal of Physical Chemistry C (12) Madden, P. A.; Wilson, M. “Covalent” Effects in “Ionic” Systems. Chem. Soc. Rev. 1996, 25, 339−350. (13) Hutchinson, F.; Wilson, M.; Madden, P. A. A Unified Description of MCI3 Systems With a Polarizable Ion Simulation Model. Mol. Phys. 2001, 99, 811−824. (14) Aguado, A.; Bernasconi, L.; Jahn, S.; Madden, P. A. Multipoles and Interaction Potentials in Ionic Materials From Planewave-DFT Calculations. Faraday Discuss. 2003, 124, 171−184. (15) Salanne, M.; Siqueira, L. J. A.; Seitsonen, A. P.; Madden, P. A.; Kirchner, B. From Molten Salts to Room Temperature Ionic Liquids: Simulation Studies on Chloroaluminate Systems. Faraday Discuss. 2012, 154, 171−188. (16) Ercolessi, F.; Adams, J. B. Interatomic Potentials From FirstPrinciples Calculations: the Force-Matching Method. Europhys. Lett. 1994, 26, 583−588. (17) Martyna, G. Constant Pressure Molecular Dynamics Algorithms. J. Chem Phys. 1994, 101, 4177−4189. (18) Martyna, G.; Klein, M. L.; Tuckerman, M. E. Nosé-Hoover chains: The Canonical Ensemble Via Continuous Dynamics. J. Chem. Phys. 1992, 97, 2635−2643. (19) Pickard, C. J.; Mauri, F. All-electron Magnetic Response With Pseudopotentials: NMR Chemical Shifts. Phys. Rev. B 2001, 63, No. 245101. (20) Clark, S. J.; Segall, M. D.; Pickard, C. J.; Hasnip, P. J.; Probert, M. I. J.; Refson, K.; Payne, M. C. First Principles Methods Using CASTEP. Z. Kristallogr. - Cryst. Mater. 2009, 220, 567−570. (21) Segall, M. D.; Lindan, P. J. D.; Probert, M. J.; Pickard, C. J.; Hasnip, P. J.; Clark, S. J.; Payne, M. C. First-Principles Simulation: Ideas, Illustrations and the CASTEP Code. J. Phys.: Condens. Matter. 2002, 14, 2717−2744. (22) Yates, J. R.; Pickard, C. J.; Mauri, F. Calculation of NMR Chemical Shifts for Extended Systems Using Ultrasoft Pseudopotentials. Phys. Rev. B 2007, 76, No. 024401. (23) Monkhorst, H. J.; Pack, J. D. Special Points for Brillouin-Zone Integrations. Phys. Rev. B 1976, 13, 5188−5192. (24) Robinson, M.; Haynes, P. D. Dynamical Effects in ab initio NMR Calculations: Classical Force Fields Fitted to Quantum Forces. J. Chem. Phys. 2010, 133, No. 084109. (25) Sadoc, A.; Body, M.; Legein, C.; Biswal, M.; Fayon, F.; Rocquefelte, X.; Bouchera, F. NMR Parameters in Alkali, Alkaline Earth and Rare Earth Fluorides From First Principle Calculations. Phys. Chem. Chem. Phys. 2011, 13, 18539−18550. (26) Bale, C. W.; Chartrand, P.; Degterov, S. A.; Eriksson, G.; Hack, K.; Ben Mahfoud, R.; Melançon, J.; Pelton, A. D.; Petersen, S. FactSage Thermochemical Software and Databases. Calphad 2002, 26, 189−228. (27) Sindzingre, P.; Gillan, M. J. A Computer Simulation Study of Transport Coefficients in Alkali Halides. J. Phys.: Condens. Matter 1990, 2, 7033−7042. (28) Corradini, D.; Madden, P. A.; Salanne, M. Coordination Numbers and Physical Properties in Molten Salts and Their Mixtures. Faraday Discuss. 2016, 190, 471−486. (29) Hansen, J.-P.; McDonald, I. R. Theory of Simple Liquids, 3rd ed.; Academic Press: New York, 2006. (30) Rabani, E.; Gezelter, J. D.; Berne, B. J. Calculating the Hopping Rate for Self-Diffusion on Rough Potential Energy Surfaces: Cage Correlations. J. Chem. Phys. 1997, 107, 6867−6876. (31) Gilbert, B.; Mamantov, G.; Begun, G. M. Raman Spectra of Al2O3 Solutions in Molten Cryolite and Other Aluminum Fluoride Containing Melts. Inorg. Nucl. Chem. Lett. 1976, 12, 415−424.

21816


Anionic Structure in Molten CryoliteâAlumina Systems - The Journal of

Recommend Documents

Anionic Structure in Molten CryoliteâAlumina Systems - The Journal of