DIAGNOSIS OF SOURCES OF CURR ENT INEFFICIENCY IN INDUSTRIAL MOLTEN SALT EL ECTROLYSIS CELLS BY RAMAN SPECTROSCOPY

The purpose of this project was to employ Raman spectroscopy in the study of industrial molten salt electrolysis cells. The objective was to improve the understanding of the chemistry and electrochemistry of the relevant melt systems and, in turn, of energy loss mechanisms in the industrial processes. On this basis new ways to improve the energy efficiency of these industrial reactors might be identified. The research plan has several principal elements. First, there was the design and construction of laboratory scale representations of industrial molten salt electrolysis cells that would at the same time serve a spectrocells. Secondly, there was the mastery of the preparation of the molten salt electrolytes, what in industry is called the ''front end.'' Thirdly, there was the adaptation of commercially available Raman instrumentation in order to facilitate the proposed studies. It is the nature of the specimens that so dramatically distinguished this work from conventional Raman studies for which commercial instrumentation is designed: first, the laboratory scale electrolysis cells are large compared to typical spectrocells; and secondly, the cells operate at, what for Raman studies are, extremely high temperatures. 4 refs., 2 figs.


DISCLAIMER
Portions of this document may be illegible in electronic image products. Images are produced from the best available original document. i It is somewhat ironic that the least dense structural metals, aluminum and magnesium, which can reduce the nation's net energy consumption when used as materials of vehicular construction, are among the most energy intensive metals to produce. The extraction of light metals is achieved for the most part by molten salt electrolysis, a very energy intensive process. In 1986 the production of primary aluminum and magnesium in the United States consumed 2.3% of total generated t electricity in this country. This was not 2.3% of the electricity that went to the metals industry, but 2.3% of total generated electricity.

TABLE OF CONTENTS
The industrial molten salt electrolysis cells that produce these metals operate at power efficiencies of less than 40%. Clearly, there is much to be gained in terms of energy conservation from technological improvements in the extraction of these metals.
From the standpoint of productivity and international competitiveness, reducing the energy consumption of these extraction processes is a high priority. For example, in the United States the average value for the specific energy consumption of aluminum is 7.2 k~~/lb. When power costs were in the vicinity of $0.005/kWh this amounted to approximately 3.6 cents per pound of aluminum. However, in recent years the cost of electricity has risen to the point where some domestic producers are now paying in excess of $0.035/kWh. At the same time some foreign producers enjoy access to electricity priced below $0.010. Thus, the discovery of means to improve the energy efficiency of the smelting process is a way to offset the advantage that derives from low cost power.
This calculation requires a knowledge of metal tonnage [1], generated electricity [1), and specific energy consumption of aluminum [2] and magnesium [3]. Raman spectroscopy is a technique acknowledged to be extremely valuable in the study of ionic species in which complex formation occurs [4). Industrial molten salt electrolysis is conducted in precisely such chemistries, i.e., electrolytes in which complex ions are present.
Indeed, the electrolyte is typically designed so that the ion of the metal bein~ extracted is fully complexed. For example, in the chloride based electrolytes for electrolysis of magnesium and aluminum the ions of these metals are present as tetrachloromagnesate (MgC1 4 2 -) and tetrachloroaluminate (A1Cl 4 -), respectively. In the cryolite based electrolyte used in the Hall-Heroult electrolysis of aluminum its ion is present in a number of fluoroaluminate and oxyfluoroalumina~e eo~plexes.
Yhile molten salts containing these entities had been studied by Raman spectroscopy, the specific melt compositions were far removed from those  Other studies As part of a study of bath chemistries that would support a nonconsumable anode the Raman spectrum of Zr0 2 dissolved in cryolite was measured. Figure 1 shows the results. The bath consisted of 15 wt% Zr0 2 as weighed against cryolite before melting. The temperature of the measurement was 1020 6 -C and the excitation radiation had a wavelength of 488 nm. At first glance the spectrum appears identical to that of pure cryolite. However, the main peak is found at a value of wave number shift lower than that of pure cryolite and higher than that of cryolite saturated with alumina. Vhile these results are preliminary they take on significance when viewed in the context of the conclusions cited above about the relationship between alumina concentration and wave number shift of the major peak. While further study is necessary to explain the observation in Figure 1, this still could be a demonstration of the ability of the Raman technique to measure the oxygen potential of the bath as well as metal oxide solubility in general.
In this regard Figure  First, the scientific accomplishments. This was the first study of its kind that sought to use Raman spectroscopy as an investigative tool in the study of the chemistry and electrochemistry of industrial molten salt electrolysis. As such, there were many accomplishments at the laboratory stale that will serve as t:he basis for fuLutf:! studies.
Among these were the development of laboratory scale representations of industrial molten salt electrolysis cells that simultaneously served as spectrocells; so-called spectroelectrochemical cells. Given the severe chemistries that one encounters in such systems this was not trivial.   We.gtl I "lo Al 1 0 1 In addition to the above cited publications it is expected that there will be three publications based on the doctoral thesis of S.-Y. Yoon.

Acknowledgements
Many people contributed to this research. Their efforts are gratefully acknowledged: Dr. Seok-Yeol Yoon, who conducted. most of the experiments and took almost all the reported data, designed and constructed all the high temperature spectrocells beyond the first generation including the unique "slotted cell" capable of both containment and •windowless" display of the highly corrosive fluoride melts, and recognized ways to treat Raman data so as to be able to use the Raman spectroscopy for quantitative analysis of molten salt solutions; Dr. Georges J. Kipouros, who established the facilities for purification of salts, built the first high temperature spectroelectrochemical cells, conducted the first experiments to obtain Raman spectra of "near industrial" melts, and during the early stages of this project provided much of the laboratory expertise associated with molten salt chemistry in general; Mr. John H. Flint, who assembled the system for Raman scattering  The study plans to look at conventional electrolysis, impurity effects, and unconventional bath additives. At each stage, first the Raman spectra of various melts are taken in the absence of electrolysis.
Part 1 Ibe Che~istry of Cryolite-based Melts Raman spectra will be measured in the absence of electrolysis. For many experiments a monocrystalline sapphire cell will be used. Tests will also be conducted with the melt contained in crucibles of graphite and boron nitride. In such cases the melt will be suspended from a metal wire helix held above the crucible. The results of Items 2 and 3 will attempt to determine whether Raman spectroscopy can determine bath ratio and alumina concentration, respectively.

Part 2 Conventional Hall Cell Electrolysis
In these experiments, a purified conventional bath is used. Electrolysis will be conducted in a monocrystalline sapphire cell with a graphite anode and graphite cathode block covered with a pool of molten aluminum. Bath consists of cryolite, alumina, and aluminum fluoride such that bath ratio is near 1.15. The presence ~f CaF 2 wil2 be optional. Current density will be varied from 10 mA/cm to 1 A/em .

Part 3 Effects of Impurities on Melt CbemisttY
Jla.uUlll !i!Jt!L:Lt.ll. will Ut:! mi:!.II.SUtt:!t.l ln Lhe .usem.:l!! ur elet:l.l"Olysls. Ftn many experiment& a monocry&talline sapphire cell will be used. Tests will also be conducted with the melt contained in crucibles of graphite and boron nitride. In such ~a&e& the melt will be suspended from a metal wire helix held above the crucible.
Compositions include the following:

Part 5 Effects of Unconventional Additives on Melt Cbemistxx
Repeat experiments of Part 3 substituting additive for impurity. The unconventional additive to be studied will be Li 3 AlF 6 .    As part of a study of the causes of the loss of current efficiency in industrial magnesiu~ cells. the characteristics of laboratory-scale cells are being investigated by electroche~ical and spectroscopic techniques, Specifically. to determine the factors that control the concentrations and spatial distributions of the various chemical species in the cell, Ra~an spectra are taken in situ during electrolysis.
The electrolyte consists of 11\ .MgC1 . 6~% NaCl. 18% KCl, and 6' ~aC1 2 . Cells opefate at a temperatur~ of 7~0 C and current densities up to 2 A/em . Spectral information from all identifiable species is correlated with cell operating conditions in an attempt to understand the nature of such phenomena as metal fog. streamers. and melt coloration. all of which are observed in these laboratory-scale cells.

Introduction
The extraction of light metals is achieved for the most part by molten salt elctrolysis. a very energy-intensive process. The production of primary aluminum and magnesium is estimated to have consumed 2.8% of total generated electric power in the United States during the Year 1984 [1].
It is somewhat ironic that the least dense structural metals. which can reduce net energy consumption when used as materials of vehicular construction. are among the most energy-intensive metals to produce.
Thus, research efforts are directed at reducing the energy requirements of these extraction processes.
The electrolytic production of magnesium accounts for about 70' of the total magnesium production in the Western world [2]. Magnesium production by electrolysis requires 1~-18 kWh/kg of magnesium metal. The current efficiency of the anhydrous electrolytic process (l.G. Farben-Norsk Hydro) exceeds 90%. while that of hydrous electrolytic process (Dow Chemical) is close to 80, ( As part of a study of the cauaes of loss of current efficiency, Raman spectra of laboratory-seale magnesium chloride electrolysis cells are being aeasured. Commercially available laser Raaan acattering instrumentation has been adapted to permit in situ real-time investigation of melt chemistry and to provide the basis for "fast Raman" spectroelectrochemistry in this and other melt systems. The results of the Raman work are combined with those of other techniques in order to reveal the mechani•ms and kinetic pathways that decrease current efficiency in magnesium cells.
This paper reports some preliminary Raman data for the electrolysis of anhydrous magnesium chloride.

Literature
Reasons for the loss of current efficiency are discussed in a recent review of the chemistry and electrochemistry of magnesium production [4]. More information i• given in the monograph by Strelets (5]. Of particular concern to the Raman work is information on melt structure: aagnesium does not exist as a discrete cation in chloride melts but instead i~_the form of the chloroeomplex, MgCl (6)(7)(8). Vibrational spectra confirm thi~ (9)(10)(11)(12)(13). However, the purpose of these studies was to deteraine the structure of molten salts, not to.understand the electrolytic production of aagnesium.
As a consequence, melt spectra were not taken during electrolysis. nor were they taken of melts resembling industrial compositions.

Jxperillental
A detailed description of the instruaentation is given in previou• reports (14,15]. Very briefly, a monochromatic linfarly polarized laaer beam from eit~er an Ar laser (Coherent Innova eo-C) or Kr laser (Coherent Innova eO-X) irradiates the electrolysis cell which is held inside a specially designed furnace. The scattered light is focused onto the entrance slit of the spectrometer (Spex Industries Triplemate 1403). An intensified silicon photodiode array tEG&G PARC Model 1420-3) serves as detector. The amplified signal is digitized in the detector controller (EG&G PARC Hodel 1218) and transmitted as data to the optical multichannel analyzer (EG&G PARC, OMA, Model 1215).
The polarization state of the exciting radiation is set by a polarization rotaor ( ..L I or 'I) . The beam then passes horizontally through the electrolysis ce11 6 The scattered radiation is collected at 90 and is imaged onto the vertical polarization analyzer (I~ always).
The spectrome~yr slit width is 100 pa. equivalent to -6 em .
Typically, the spectra were recorded for approximately 1 minute, corresponding to 200 scans on the OMA, which was calibrated using the emission lines of a neon lamp in the green. All spectra reported in this article were obtained using the 514.5 nm line of argon as exciting radiation.
The electrolysis cell is constructed of optical-grade square fused quartz tubing. 1" on inside edge, joined to round tubing 41 mm 0.0.
The cap is a compression fitting made of 304 stainless steel and has ports for the cathode, anode, inert gas inlet. and thermocouple.
The cap also has a sidearm for gas outlet.
The preparation of anhydrous salts for electrolyte formulation has been described previously[ l4).
In a typical experiment, the electrolysis cell is charged with salt and aaooable~ with the cap ~nd electrodes in the glove box. The charged cell is placed in tne electrical resistance furnace with windows[l4]. and the salt is melted under high purity argon.
Results and Discussion Figure 1 shows Raman spectra of pure aolten MgCl 2 at 740°C. !~ere is a strong polati&ed peak at 20~ c~ !nd a weak depolarized peak at 385 em-. These results are essentially identical to those of Huang and Brooker [10]. The difference 1e that it took the authors of this report only 4 minutes U$Jng the OMA to measure these spectra which Huang and Brooker recorded tor eeveral hours.
This shows the 5p~~d. accuracy, and sensitivity of the adapted instrumentation to be satisfactory.   Somewhat surprisingly. the addition of CaC1 2 to the alkali chloride melt did not give rise to any distinct peaks . lt was hoped that one or more such peaks could be used as a standard for calibrating composition during electrolysis experiments. The cac1 2 content is not expected to change while magnesium is produced. Figure 4 shows Raman spectra of a aelt representative of industrial composition Ill' MgC1 2 • 6% CaC1 2 • 65% 0 NaCl, 18ll J<Cl) taken at a temperature of 750 c in the absence of cu~lent.
The prominent polarized peak a! 205 em in pure MgCl has shifted to 249 em 1 .
During electrolysfs there was a decrease in peak height withou! 1 change in wavenumber for the peak at 249 em . Figure 5(a) shows a photograph of the cell taken after l minute of e~ectrolysis at a current de n sity of 100 mA / cm . The composition of the electrolyte was ll% MgC1 2 . 65' NaCl . 6!1 ca:: J . and 18% KCl . Temperature was 75o 0 c. On th~ left is a l / 4 " graphite cathode : on the right is a l / 8 " graphite anode shrouded with fused quartz tubing . Strea mers ha v e begun to emanate from the cathode .
Ch l orine gas bubbles can be seen on the anode. Figure 5 ( b ) shows the same cell as Figure 5 ( a ) after approximately 5 minutes of e l~ctro l yc ic at a current de~s ity of 100 mJ.. . err. .
lt is e vi de nt that the streamers emanating from the cathode have grown over essentia l ly the entire breadth of the cell.
The ele ct rolyte has become cloudy . Figure   40C 30: _ l 2 0~  (a) after l ainute; (b) after 5 minutes. 51b ) also shows the ap pe a~ance of tiny droplets of magnes ium on the tip of the cathode.
Chlor in e bubbles are seen to continue evol vl ng on the anode.
Attempts to take Raman spectra of the streamers have failed . There is a good poss ibili ty that the streamers are not Raman a ctiv e . Other techniques , both spectral and electroche mical , will be used to study streamers further .

Conclusion
Some preliminar y results of Raman scattering studies of laboratory -scale magnesium chloride electrolys is cells have been presented.
Metal fog . streamers, and melt coloration are all observed in these cells which are 2 operated at current densities of up to 2 A/ em . On the basis of the Raman scattering data it appears that it should be possible to measure the concentrat ion of MgCl in the electrolyt e as a function of time 2 ~fforts are underway to improve the spatial resolution of the system to permit the measureme nt of electrolyt e concentrat ion profiles .     Following is the Raman spectroscopy technical discussion relevant to cryolite based electrolytes:

Cryolite
The single crystal sapphire tube, closed at one end, was tested in a spectrocell with molten cryolite. It proved to be able to survive the thermal cycle of heating to 1015°C, containing molten cryolite, and cooling to room temperature with the solidification of molten cryolites. When the solidified cryolite was dissolved in an aqueous solution of aluminum chloride, the sapphire showed no evidence of etching or loss of transparency.
Cryolite spectra were taken in the sapphire spectrocell. Three stokes (red Raman) shifts were observed: 180, 390, and 533 cm-1. The 533 cm-1 line is strongest and polarized, while the one at 390 cm~l is weak and depolarized. The 180 cm-1 line is in the anomalous Rayleigh shoulder and is depolarized. ·The sp~ctrometer was also scanned to deeper red to search for higher wave number Raman shifts. None were detected.
Electrolysis was attempted in a single crystal sapphire tube, open at both ends. The bottom of the tube was plugged with graphite, which also served as the cathode. Unfortunately the seal was poor, and the electrolyte leaked out upon melting. The cell has been redesigned to incorporate a longer cathode plug so that the freeze line forms midway along the plug, ..

EXCERPT FROM QUARTERLY PROJECT REPORT DOE/CE/40545-23
Following is a discussion of research accomplishments from the quarterly project report DOE/CE/40545-23: This quarte: was marked by continued attempts to obtain high quality spectra from th~ cryolite-based melts. Several designs of spectrocell were constructed and tested ~ith the final result being the invention of a what we believe to be a truly remarkable piece of laboratory apparatus: namely, a spectrocell capable of providing research grade spectra under extreme condi· tions. The cell ~ill be described later. But first it is instructive to recount the othe: less successful designs that were tested and to report upon their attributes and shortco~ings. All of the comments are in reference to measuring the Raman spectra of cryolite-based melts, i.e., fluoride che~istries.
Tne first cell conSisted of single crystal sapphire tubing. Tne though~ ~as that although sapphire is attacked by cryolite, the rate of attack would be slo~ enough to allo~ the measurement of the spectra of oxide· free fluoride melts. Tnis proved not to be the case. Eventually we would learn that the rate of attack and the geometry of the cell would combine so as to ~agnify the presence of oxide in the melt even at very short times.
A second cell consisted of a crucible anci hanging rod both of which were co~tained in an outer closed end fused quartz tube the top of which was !ittec ~ith the stainless steel cap used in the previous chloride studies. Tne crucible and rod were made of inert materials such as boron nitride or graphite. The idea was to hold the melt in the crucible which acted simply as a reservoir. For taking spectra the rod was immersed in the melt to form a pendant drop which was held above the crucible in sight of the window of the furnace for ill~ination by the incident laser. This arrangement had ewo problems. First, cryolite would not wet some of the rods. Secondly, the droplet's curved external boundary had a lens effect which thre~ the laser be~ in unwanted directions uncontrollably and made focussing impossible.
A third cell consisted of a suspended platforu onto which a droplet of melt sat. ~nile wettability was not a problem the lens effect was.
A fourth cell was designed by experimenting with room temperature spectra of carbon tetrachloride. lt was discovered that by flattening the droplet through compression i.e., by essentially squeezing the droplet be:ween ewo plates, the droplet took the shape of a cylinder and the laser be~ could be focused. Much experimentation resulted in the discovery of the optimal spacing berween the plates to get the best shaped droplet.
The f:f~h anc final cell is that sho~~ in Figure 1. T.~e main features are (1) the crucible into which a cross cut has been macie half way through so as ~o foru a horizo~:al gap and (2) the roc ~hich acts as a ran to force the melt to rise to the poi!1": -·here i: is visible in the gap. This explo-its the features cf the fo~rtt cell bu: does so in a much more convenient and elegant manner. ~e have tested such spectrocells constructed of graphite and of boron ni:ride. The principal a~t:ribute of this cell is that we are able to see the me::.: \."ithout allo~·ing it to ~take physical contact \.'ith any window r..c.:.e:-ial such as fused quartz or sapphirt, both of which we no~o· kno~· will raridl~ conta~inate the melt ~ith oxygen. Fi£ure 2 is the spectrUJt of pure cryolite taken in the spectrocell of Fifure 1. The results are far superior to any achieved to date. Polarization settings are vertical in and vertical out. Of interest is the strong peal: way out at 1463 cm· 1 • Figure 3 is the same as Figure 2 with the excep.tion that the polarization set:tings are horizontal in and vertical out. Note that the strong peak at around 600 c~·l has completely disappeared which demons:rate~ full depolarization, something that should occur but could not b~ observeo in the spectra obtained previously. There was no explanat~on for the peak at 1463 Cit ·l at the time. Since then we have learned that this must have been a reaction product on the outer fused quartz container, probable due to attack by AlF 8 bearing vapor. we ha~ never eneountered this before because our cells had never been robust enough to pe~it us to conduct a experiment for such long times, except in the case of the sapphire monocr;·s:al -·hich siJ:t?lY c!issolvec as A! 2 0 3 in the melt.  Figure 1 shows the Raman spectrum of a melt containing AlF 3 and Na 3 AlF 6 in a 1:1 mole ratio. The temperature of the melt was 870°C, and the excitation radiation had a wavelength of 488 nm. The spectrum is of superb quality. There are several reasons for this. First, the melts were of high purity due to our careful preparation procedures which involved premelting. Secondly, the spectrocell allowed us to avoid contaminating the melts dur~ng the course of t~king the data. Thirdly, with the equipment modifications described previously, data processing was greatly enhanced in terms of our ability to do curve smoothing, plotting, analysis, etc. There are five peaks in Figure 1: 210 c~·l dp, 330 cm·l dp, 625 cm·l p, 760 cm·l dp, and 580 cm· 1 p. The first four fit the pattern for a tetrahedrally coordinated complex, which in this case is obviously AlF 4 -. The peak at 580 cm• 1 is on the shoulder of the major peak at 625 cm• 1 and represents the major peak of pure cryolite. As such it is a representation of octahedral coordination as expected in AlF,-. Figure 2 shows the Raman spectrum of a melt containing AlF 1 and Na 1 A1Fe in a 3:4 mole ratio. The temperature of the melt was 89o•c, and the excitation radiation had a wavelength of 488 nm. The same five peaks as those present in Figure 1 appear in Figure 2. However, in Figure 2 the peaks at 625 cm· 1 and 580 cm· 1 have nearly the same intensity. This is an indication that the change in bath chemistry had changed the relative populations of the AlF 4 -and Alr,s-species according to the reaction cited above. As the amount of cryolite in the bath increases, the line associated with Alr,aintensifies. · Figure 3 shows the Raman spect~ of a melt containing Alfs and Na 3 Alf 6 in a 1:2 mole ratio. The te~perature of the melt was 890°C, and the excita· tion radiation had a , • .-aveleng:h of 488 nm. The results are much the same as those in Figure 2 ~ith appropriate changes in the relative intensities of the peaks at 625 cm·l and 580 cm· 1 • ln addition, there seems to be some broadening of the peak at 330 cm· 1 • Figure 4 shows the Raman spectrum of a melt containing Alfs and Na,AlF 6 in a 1:4 mole ratio. The temperature of the melt was 99o•c, and the excitation radiation had a wavelength of 488 nm. At this composition substantial changes are evident when the spectrum is compared to those in the previous figures. The peak at 625 cm· 1 is barely visible, appearing in the shoulder of the peak at 580 cm· 1 • Despite excellent curve smoothing it is evident that this spectrum is much noisier than the previous ones. This is in part due to the higher temperature of the measurement, 99o•c, as compared to 870° -sgo•c. · Finally, just for reference, Figure 5 shows the Raman spectrum of a melt of pure eryolite, Na,AlF 6 • The temperature of the melt was 103o•c, and the excitation radiation had a wavelength of 488 nm. The results are superb. Tne peak at 580 cm· 1 is sharp. The peak at 380 cm· 1 is clear. The depolarization is complete. Signal-to-noise is remarkable for a spectrum takefi at a temperature of l030°C in a 1:1 mole ratio.
A second set of experiments sought to determine whether Raman spectroscopy could reveal the state of chemical coordination in the fluoroaluminate complexes when oxygen was added to the melt. Actually, the work was prompted by the observation that with the new spectrocell, the main peak in pure cryolite was nearer 580 cm· 1 than 535 cm· 1 as had been previously measured in this laboratory (DE/CE/40545-15). lt was thought that this was due to more accurate calibration of the spectrometer which in fact records intensiey as a function of channel number rather than as a function of wavenumber. The mapping from channel nUmber to wavenumber is done by calibration with a standard light ~our~e, typi.cally the plasma lines of a neon light or the plasma lines of the Raman laser which has been purposely detunecL Mowevesr, nr.alioraticm is unlikely to have been the cause for a shift of 45 cm· 1 for this cryolite peak. A more plausible explanation is that the spectra obtained in the new spectrocell were truly tho5e of th*.! pure fluoride, while the spectra obtained earlier in the monocrystal sapphire cells had been contaminated with oxygen. To test this hypothesis a set of experiments was conducted in which the amount of alumina dissolved in cryolite was varied. Figure 6 shows the Raman spectrum of a melt containing 11 weight percent Al 2 0s in cryolite. The t.emperat\lre of the melt was 103o•c, and the excitation radiation had a wavelength of 488 nm. Just as is the case with the fluoride spectra ifi Figut'~s 1 through 5, t:hfl spectrum is of superb quali--At first glance, Figure 6 seems identical to Figure 5. However, in Figure 6 the main peak, which in Figure 5 appears at =580 cm· 1 , has shifted to .. 530 cm•1. As mcn;~one.d above, the spectra of pure cryolite measured in sapphire cells looked very similar to Figure 6 except: that the ~ign&J.-to noise ratio of the earlier spectra was much poorer. Thus, it is clear that the presence of oxide in the melt is shown by the change in the position of the main peak. The next question was, "Is there a functional relationship between oxygen content and wavenumber shift?" To answer this question a set of spectra was measured for a variety of melts of different oxygen content.
In all cases the temperature of the melt was 1030•c .and the excitation radiation had a wavelength of 488 nm. Figure 7 shows the Raman spectrum of a melt containing 3 weight percen~ Al 2 0 8 in cryolite. The same peaks as those present in Figure 6 appear in Figure 7. However, in Figure 7 the main peak is found at c552 cm· 1 • Figure 8, 9, 10, and 11 show the Raman spectra melts containing respectively, 5, B, 13, and 15 weight percent Al 2 0 8 in cryolite. The main peak shifts to lower values of wavenumber shift as the alumina concentration increases. However, once the alumina concentration reaches a certain value, there is no further change in wavenumber shift. This is an indication that the bath has become saturated -ith alumina, a fact detected in the Raman spectrum. Figure 12 is a plot summar~z~ng the data and showing the relationship between alumina concentration and wavenumber shift. It appears that there is a linear relationship between alurr:ins concentration and wavenumber shift. Furthermore, the intersection of the two line segments occurs at a composition consistent with the known liquidus at the temperature of the measurements. These findings are very exciting as there are commercial implications of this discovery as well as implications for scientific research. Commercially, we are talking about an alumina sensor. Academically, do we no-have a technique to measure noninvasively the solubility of metal oxides in cryolite? Imagine the consequences for inert anode studies. What about phase diagram determination?