Approaching Dissolved Species in Ammonoacidic GaN Crystal Growth: A Combined Solution NMR and Computational Study

Abstract Solutions of gallium trihalides GaX 3 (X=F, Cl, Br, I) and their ammoniates in liquid ammonia were studied at ambient temperature under autogenous pressure by multinuclear (71Ga, 35Cl, 81Br) NMR spectroscopy. To unravel the role of pH, the analyses were done both in absence and in presence of ammonium halides, which are employed as mineralizers during ammonoacidic gallium nitride crystal growth. While gallium trifluoride and its ammoniate were found to be too sparingly soluble to give rise to a NMR signal, the spectra of solutions of the heavier halides reveal the presence of a single gallium‐containing species in all cases. DFT calculations and molecular dynamics simulations suggest the identification of this species as consisting of a [Ga(NH3)6]3+ cation and up to six surrounding halide anions, resulting in an overall trend towards negative complex charge. Quantitative 71Ga NMR studies on saturated solutions of GaCl3 containing various amounts of additional NH4Cl revealed a near linear increase of GaCl3 solubility with mineralizer concentration of about 0.023 mol GaCl3 per mol NH4Cl at room temperature. These findings reflect the importance of Coulombic shielding for the inhibition of oligomerization and precipitation processes and help to rationalize both the low solubility of gallium halides in neutral ammonia solution and, in turn, the proliferating effect of the mineralizer during ammonoacidic gallium nitride formation.


Introduction
Ammonothermal synthesis is an emerging technique for preparation and crystal growth of many materials difficult to obtain. Pioneering work by RobertJ uza and Herbert Jacobs startingi nt he 1960s already revealed the potential of the methodf or various material classes. [1] In the past two decades, the technique gainedg reat importance for growing high-quality GaN crystalsw ith particularly low defectr ates, [2] and it was recently furtherd evelopedt op roduce variousn ovel nitridebased materials. [3] For GaN crystal growth, two generalroutes have been established:T he ammonobasic method typicallye mploys alkali metal amides, or suitable precursors like alkali metals or alkali metal azides, as so-called mineralizers to maintain ah igh concentrationo fa mide (NH 2 À )i ons, while the ammonoacidic methodm akes use of ammonium halidestob oost the concentration of the ammonoacid NH 4 + . [2b] As uccessful ammonothermal synthesis depends crucially on the choice of these mineralizers to control the mobilization of startinga nd target materials as well as the solubility of any dissolved intermediates involved in the process. For ad eeper understanding of the chemicalp rocesses, it is furthere ssential to obtain concise knowledge on the speciesp resenti ns olution. Recently,w e have been able to isolate several solid intermediates of ammonobasic GaN syntheses with different mineralizers, all of which contain isolated tetraamidogallate ions, [Ga(NH 2 ) 4 ] À . [4] Furthermore,w eh avee stablished the tetraamidogallate ion as the dominating dissolved species in liquid ammonia at ambient temperature and under autogenous pressure via solution NMR studies. [5] Applying the heavier alkali metal amides as mineralizers may eventually afford ionic liquids with high gallium concentrations. Depending on the ammonia content,t hese liquids contain condensed m-imido amidogallate ions, which can be regarded as deprotonated intermediates towards crystalline GaN. [5] In addition, molecular dynamics calculations have provent he complex ion [Ga(NH 2 ) 4 ] À to represent the predominant species in liquid ammonia over aw idet emperature and pH-range. [6] In ammonoacidic systems with ammonium halide mineralizers, we were similarly able to isolate solid intermediates of compositions [Ga(NH 3 ) 6 ]X 3 ·NH 3 (X = Br,I ), [Ga(NH 3 ) 5 Cl]Cl 2 ,a nd [Ga(NH 3 ) 4 F 2 ][Ga(NH 3 ) 2 F 4 ] = Ga(NH 3 ) 3 F 3 ,r espectively. [7] The heavier halide anions (bromide and iodide) give rise to salts containing hexaammineg allium ions, while the lighter halides (chloride and fluoride) are able to enter the first coordination sphere of gallium.R emarkably,t he crystalline ammoniate Ga(NH 3 ) 3 F 3 features both isolatedc ationic and anionic complexes.M olecular dynamics calculations suggestt hat such ions may persist in liquid ammonia at ambient pressure, although with an increased number of ammonia ligands in the first coordination sphere of gallium, and agglomerateu pon heating and pressurizing, due to the diminishing permittivity of ammonia under these conditions. [8] Unfortunately,t he extremelyl ow solubility of this complex fluoride in ammonia at ambient temperaturethwarted as yet any experimental studies.
Herein, we report on NMR studies of ammonacidic solutions of galliums alts in ammonia at ambient temperature and under autogenous pressure, whicha im at pinpointing the solute speciesp resent.T he resultsa re interpretedb ased on DFT-calculations of the spectroscopic data and discussed in the context of molecular dynamics simulations in aw ide temperatureand pressure range.
All gallium trihalides are known to exhibit quite low solubilities in liquid ammonia aroundr oom temperature. [9] Similarly, we found that ammine complexes [Ga(NH 3 ) 6 ]Br 3 ·NH 3 (2)a nd [Ga(NH 3 ) 6 ]I 3 ·NH 3 (3)a re only sparingly soluble in pure liquid ammonia,a nd most of the samples prepared contained as olid residue next to the saturated solution.N onetheless, 71 Ga NMR signals of these solutions were readily observable, and recording of a 71 Ga NMR spectrumo ft he ammonia solution of GaCl 3 (4)w as likewise feasible after prolonged acquisition.I nc ase of Ga(NH 3 ) 3 F 3 (1), neither 71 Ga nor 19 FNMR signals were detectable. Interpretationo ft his last finding requires some concern about the effect of spin coupling on the spectra.R eported values of 1 J 71Ga,19F range from 264 Hz in (NH 4 ) 3 [GaF 6 ]t o4 45-490 Hz in [(NNN)GaF 3 ]( NNN = triazacyclononane ligand), [10] and observable splitting of signals into multipletsi st hus to be expected when intermolecularf luoride exchange is slow and a small electric field gradienta tt he metal center precludes rapid relaxation induced by the quadrupolar nature of the two naturally occurring galliumi sotopes 69, 71 Ga (both I = 3/2). NMR studies on fluoro-gallium complexes are scarce, but indicate that 19 FNMR signals seem readily detectable even if linewidths may in unfavorable cases exceed1kHz. [11] Against this background, we conclude that the failure to detect any 19 FNMR signals from 1 is most likely an indicationt hat the complex is essentially insoluble.
The 1 Ha nd 14 NN MR spectra of solutions of 2-4 (as well as all other samples included in this study) display as ingle resonance representing ad ynamic average of the signals of bound and free NH 3 molecules. Exchange of this type is known for aluminumc omplexes, and the lifetimes observed for bound ligands in that case ( % 1s at À37 8C) [12] are in accord with the assumption that the reaction has reached the fast exchange regime at ambient temperature.
The 71 Ga NMR spectra of solutions of 2-4 display single lines ( Figure 1) with very similar chemical shifts around 74 ppm and moderate line widths, indicating that all three solutions contain ac ommon,h ighly symmetrical gallium-containing species. Based on previousr eports on the presence of six-coordinate complexes in solutionso fa luminum(III) halides in liq. NH 3 [13] and isolated gallium halide ammoniates [7a, 9b] as well as aqueous solutionso fg allium(III) salts, [7a, 14] we tentatively appoint this speciesa sah exa-coordinated complex [Ga(NH 3 ) 6 ] 3 + .Afurther discussion of this assignment in the light of computational studies will be given furtherbelow.
The 71 Ga NMR spectrum of as olution of 4 that contains a sediment of undissolved material features, in addition to the signal already observedf or the clear solution, as econd resonance with as imilar chemical shift but as ignificantly larger line width (Figure 2). We presume that this signal arises from a speciesp resent at the interface between the solid and the supernatant solution.I nv iew of the relationb etween the linewidths in NMR spectra of quadrupolar nuclei like 71 Ga (I = 3/2, Q = 0.106 barn) and both local electric field gradients andr otational correlation times, [15] the increased linewidth of the addi-  tional resonance can in principle be explained as arising from the reducedm obility of as urface species, or the presence of a less symmetrical coordination sphere than in a[ Ga(NH 3 ) 6 ] 3 + complex in the bulk solution, respectively.
Al ikely origin of al ower local symmetry is the incorporation of chloride ligands in the metal coordination sphere. Af irst attempt to confirmt his hypothesis by characterizing the chemical environmento ft he chlorine atoms by 35 Cl NMR spectroscopy failed, however,a sl ow solute concentration and/or coordination-induced line broadening precluded the observation of any signals above noise level. Assuming that the formation of heteroleptic complex ions like [Ga(NH 3 ) 5 Cl] 2 + as found in solid [Ga(NH 3 ) 5 Cl]Cl 2 can be promoted by the addition of excess ligand,w ee xtended our studies to solutions of mixtureso f GaCl 3 and an ammonium halide (NH 4 Cl) in liquid ammonia. Moreover, since addition of NH 4 Cl lowers the pH value, this approach offers an opportunity to study the behavior of the gallium salt not only in neutral, but also in ammonoacidic milieu.
Considering that salt addition affects also the ion strength of the solutions,w ec hose to use both NH 4 Cl and NH 4 Br as mineralizer (the absence of detectable amounts of bromo-complexes in solutions of 2 led us to consider bromide as "innocent" with respectt oc omplex formation) in order to separate both effects.
Somewhat unexpected, the 71 Ga NMR spectra of saturated solutionso fG aCl 3 in liquid ammonia containing varying quantities of NH 4 Cl gave no evidence for the formation of any new species, but revealed merely ar ise in signal intensity with increasinga mmonium halide concentration ( Figure 3a;n ote that the spectrao fs amples prepared by incomplete dissolution of 4 in the presenceo fN H 4 Cl displayed as imilar additional broad resonance as samples obtained without mineralizer,s ee Supporting Information). Variation of the nature of the anion in solutionsc ontaining similar total concentrations of NH 4 X (X = Cl, Br) hadn oi mpact on chemicals hifts or linewidths ( Figure 3b), suggesting that the halide ions do not enter the first metal coordination sphere.T his conjecture is further corroborated by the observation of narrow lines indicatingt he pres-ence of non-coordinated halide anions in the 35 Cl and 81 Br NMR spectra of these solutions.
Ap lot of gallium concentrations determined from quantitative 71 Ga NMR measurements on saturated solutionsv ersus the concentrationo fN H 4 Cl added ( Figure 4) suggestsalinear correlationb etween both quantities which implies as imilar dependence of gallium solubility on mineralizer concentration as had previously been established for GaN in supercritical ammonia at 490 8C. [16] The molar amount of GaCl 3 dissolved per mole of mineralizer can be computed from the slope of a linear regression curve as 0.023 (14)mol GaCl 3 per mol NH 4 Cl. This result comes close to ar eported value of 0.05 mol GaCl 3 per mol NH 4 Cl for the solubility of GaNi nN H 4 Cl-containings upercritical ammonia at 550 8Cd etermined by in situ X-ray imaging, [8,17] but is far from figures of up to 0.43 mol-% GaN per mol-% NH 4 Cl at temperatures between 400 and 600 8Ct hat had been obtained by ag ravimetric method. [16] While any quantitative comparison of the data from the diverse sources seems problematic in view of the differences in methods as well as concentration and temperature regimes involved, our results nonetheless further corroborate the previous findings on the importance of mineralizer concentration as ad ecisive factor for the solubility of galliums altsi na mmonoacidic milieu. [8,16,17] For closer comparison of the variousf igures, it is noteworthy that the solubility of GaN in presence of ammonium halides will strongly depend on the properties of ammonia, which are known to significantly change with temperature and pressure, particularly in vicinity of the critical point.N ext to a significant decomposition of ammonia to form hydrogen and nitrogen, [18] thus reducing the solventc oncentration,t he de-  creasingp ermittivity of ammonia with increasing temperature has to be taken into account.
To substantiatet he nature of the complexesp resent in solution and assess the effects of chloride coordination and protolysis on 71 Ga NMR chemical shifts, we performed DFT calculations on as eries of tentative structural candidates, namely [Ga(NH 3 Considering that the low dielectric permittivity of liquid ammonia might favor ion pairing, we furthere valuated d 71 Ga fori on 6 ]Cl n } (nÀ3)À (n = 4-6) and {[Ga(NH 3 ) 5 (NH 2 )]Cl 2 }, respectively.C omplexes with higher metal coordinationn umbersw ere neglected, since attempts to locate energy optimized geometries for such species converged inevitably to the structures of van der Waals complexes between hexa-coordinate complexes and additional, loosely interacting,a mmonia molecules.
The molecular structures used for these studies wereo btained by performing first energy optimizations on isolated complexes or ion assemblies (in the "gas phase'), and then reoptimizing the initial structures under application of the PCM (polarizablec ontinuum model) [19] approach for modelling the effect of solvation.C omparison of "gas phase" and "solute" structures (see Figure 5f or an example) discloses that embedding in the dielectric mediumt ends to lengthen Ga-Cl and shorten Ga-N distances, which implies as ignificant electrostatic contribution to the metal-ligandb onding. The chloride anions in ion assemblies {[Ga(NH 3 ) 6 ]Cl n } (3Àn) + (n = 1-3) and {[Ga(NH 3 ) 6 ]Cl n } (nÀ3)À (n = 4-6) display each three close contacts to NH-bonds of coordinateda mmine ligands (Cl···H 222-249 pm, N···Cl 322-336 pm) with an essentially linear arrangement of the NÀH···Cl moieties. The distances match those reported for NH···Clh ydrogen bonds in organic crystals( 320-334 pm for NH···Cl and NH + ···Cl units based on the sum of hydrogen bond radii [20] and the same structuralm otif with N···Cl distances ranging from 326.7(2)-352.4(3) pm was also found in crystal structures of some hexammine cobalt complexes. [21] Attachment of chlorides induces structurald istortions, which are evidenced by al arger varianceo fi ndividual Ga-N distances in some assemblies and ag eneral contraction of the coordination polyhedra with increasing number of anions (cf. Figure S4  Subsequentmagnetic shieldingc alculations were carried out using the zero order regular approximation( ZORA) [22] in order to account in ac onsistentm anner for spin-orbit induced effects associatedw itht he presence of heavy atom substituents [23] (X = Br,I ). Further details on the computations are given in the experimental section, and al isting of calculated 71 Ga chemicals hifts and the squareo ft he largestc omponent of the electric field gradient (efg) tensor, q zz 2 ,w hich has been shownt oc orrelate in favorable cases with observed spectral linewidths, [24] is presentedinT able 1.
The trend in calculated 71 Ga chemical shifts of ammine complexes [Ga(NH 3 ) n ] 3 + (n = 4-6) reflectst he expected [25] strong influence of the coordination number on the magnetic shielding of the metal atom, but the dependence of d 71 Ga calcd on the number and nature of the halide anions in the aggregates {[Ga(NH 3 ) 6 ]X n }e mphasizes that changes in the second coordination sphere have as well av isible impact.W ithouti ntending aq uantitative assessment (whichw ould presumably require averaging over al argen umber of structurali somers with different arrangement of the surrounding anions [26] ), we note that d 71 Ga calcd of the ion clusters increases generally-although not linearly-with the number of surrounding halide anions, and that the magnitude of the deshielding effect grows in the order Br < Cl < I. These trends are presumably related to the structurald istortions of the cationic core arising from the Hbondinga nd electrostatic interactions with the surrounding anions.
The computed 71 Ga chemical shifts forc hlorogallium complexes[ Ga(NH 3 ) n Cl 6Àn ] (nÀ3) + (n = 3-5) are quite similaro r-in case of trans-[Ga(NH 3 ) 4 Cl 2 ] + and fac/mer-[Ga(NH 3 ) 3 Cl 3 ]-slightly larger than that of the "naked" hexaamminec omplex, whereas the chemical shifts of the homologous bromo-and iodo-complexes are substantially lower and tend to become increasingly more negative with growing atomics ize (thus showing a normalh alogen dependence and number of halide ligands). [27]  An inspectiono fi ndividual shielding contributions reveals that the differentialc hanges in d 71 Ga calcd for complexes containing exclusively amine-/amide-and chloride-basedl igands depend mainly on variations in the paramagnetic shielding term, [28] whereas the trends in complexes with metal-bound bromide and iodide ions are dominated by spin-orbit effects. [23] Quite surprising, incorporation of halide ligands into the first coordination sphere induces in most cases no sharp increasei nt he magnitude of q zz 2 ,w hichi mplies that the effective local elec-tron density distribution around the metal ion remains quite symmetrical. In view of this finding, it is tempting to interpret the discrepancyb etweenc omputed "gas phase" and "solute" geometries ( Figure 5) as the consequence of as tructurala daptation to the polarizable continuumt hat is driven by the demandt oe stablish am aximum symmetrical charged istribution. The spread of computed chemical shifts fora ll complexes considered in Ta ble 1c orroborates our initial hypothesis of interpreting the coincident values of d 71 Ga for solutions of 2-4 in liquid ammonia (Figure 1) as an indication for the presence of the same homoleptic complex in all cases.Anumerical comparisons uggests further that structures with tetra-or pentacoordinate gallium ions can definitelyb er uled out and the experimental dataa re best compatible with ah exa-coordinate species. Even so, the assignment of the solutes pecies to a "naked" [Ga(NH 3 ) 6 ] 3 + ion implies an unusually large deviation from an empiricalr egression curve connecting observed and calculated chemical shifts (Figure 6a nd ESI). In principle, two straightforward explanations of this discrepancy can be conceived, viz. the sensitivity of d 71 Ga calcd to changes in the second coordination sphere, or acid-base reactions implying deprotonation of one (or more) of the NH 3 ligands, respectively.C onsidering the presence of ionic aggregates {[Ga(NH 3 ) 6 ]X n } (3Àn) + (n = 1-3; X = Cl, Br,I )a nd {[Ga(NH 3 ) 6 ]Cl n } (nÀ3)À (n = 4-6, see also Figure6 for X = Cl) is indeed suited to improvet he conformity with the correlation, with the best match with as ingle species achievable in case of 4 for ion clusters of compositions À69.20 .0 {[Ga(NH 3 ) 6 ]Cl 3 }a nd {[Ga(NH 3 ) 6 ]Cl 5 } 2À ,r espectively.Asimilari mprovement is feasibleb ya ssuming that the solutionsc ontain rapidly exchanging dynamic mixtures consisting of substantial amountso fb oth the hexaammine complex and its conjugate base, [Ga(NH 3 ) 5 (NH 2 )] 2 + .H owever,t his hypothesis is in conflict with the expectation that the addition of varying amounts of NH 4 X should then induce changesi nt he equilibrium composition andt hus the observable average chemical shift, whichi s actually not observed. We conclude therefore that postulating the presence of ion clusters provides the most likely scenario for solutions containing additional ammonium halide as mineralizer,although it cannotbee xcluded that formation of amidocomplexes may competew ith ion pairing in solutionso fg allium halides in pure ammonia.The modulation of the local environmento ft he metal center by dynamic exchange of anions in the second coordination spherem ay well be considered to contribute to the observed variation in 71 Ga linewidths. It must be admitted, however,t hat the limited accuracy of the computationalm odel precludes ar eliable evaluation of the number of attached anions and that, in view of the rather minute impact of the variation of the nature and number of halide anions surrounding ac omplex cation on the resulting chemical shift, we cannot exclude that the real solution may not contain as ingle species but rather ad ynamically equilibratingm ixture in which variations in d 71 Ga are levelled beyond the limit of experimental observability. While the DFTcalculations provide av ery accurate account of the interactions within model complexes, the consideration of the embedding solventl acks explicit atomic detail. To permitu nbiasede xchange of ammonia molecules between solvents hells and the bulk solution, we hence employed molecular mechanics based molecular dynamics (MD) simulations. On this basis, bulk ammonia solutionsa re mimicked by periodic simulation cells of 3000 NH 3 molecules at 300 Ka nd 8atm. Followingarecently introduced 'local pK'm odelling scheme, [6] we use combined quantum/molecular mechanicst oa ccount for ammonium/amidea cid/base reactions as functions of complex formationa nd pH. On this basis, we recently calculated the local pK of the [Ga(NH 2 ) 4 ] À /[ Ga(NH 2 ) 3 (NH 3 )] 0 reaction,a nd clearly established [Ga(NH 2 ) 4 ] À as the preferred complex by 44 kcal mol À1 .T his implied shiftingt he pK of ammonia from 32 to 1w hen contrasting autoprotolysis in bulk ammonia to Ga 3 + assisted NH 3 protolysis. [6] To elucidate the role of Cl À ions, it is educativet od irectly compare[ Ga(NH 2 ) 4 ] À with the analogous [GaCl 4 ] À complex in ammonia solution. Strikingly,w ef ind the chloride complex to quickly evolve from nearest-neighborG a-Cl contacts towards {[Ga(NH 3 ) 6 ]Cl 4 } À species. The latter type of complexes predominantly exhibit Cl À ions in the second coordination shell of Ga (with an average Ga-Cld istance of 350 pm) whereas nearest-neighborG a-Cl contacts (Ga-Cld istance of 280-300 pm) are only observeda st emporary fluctuations. The stabilityo ft he hexaammine motif, [Ga(NH 3 ) 6 ] 3 + drasticallyd ependso nt he presenceo fC l À in the second coordination shell. Contrasting the two speculative systems "[Ga(NH 2 ) 4 ] À + n Cl À "a nd "{[Ga(NH 3 ) 6 ]Cl n } (nÀ3)À + 4( NH 2 ) À "( with n ! 3), we find the amminec omplex to be favored by more than 100 kcal mol À1 , hence fully overcompensating the beforehand discussed promotion of amide formationn ext to dispersed Ga 3 + ions.
We then placed two {[Ga(NH 3 ) 6 ]Cl n } (nÀ3)À type complexes in our modelso fa mmonia solution (withoutc harge compensation) to explore trends towards agglomeration. Indeed, in our small simulation cell of approx.5 5 5nm 3 dimensions, we find the association of {[Ga(NH 3 ) 6 ]Cl n }{[Ga(NH 3 ) 6 ]Cl m } (6Àn-m) + dimer complexes for n, m < 4w ithin af ew 100 ps. On the other hand, for n ! 4t he {[Ga(NH 3 ) 6 ]Cl n } (nÀ3)À type species displayed stable dispersions during the entire length of the MD runs (5 ns). To account for the range of ammonium chloride concentrations added to the GaCl 3 solution, we probedt he fate of {[Ga(NH 3 ) 6 ]Cl n } (nÀ3)À complexes up to n = 7. From this, n = 6, that is the {[Ga(NH 3 ) 6 ]Cl 6 } 3À complex (Figure 7), was identified as the most negatively charged species to whichn of urther Cl À could be associated. Based on the MD simulations, we hence argue that {[Ga(NH 3 ) 6 ]Cl 3 } 0 complexes tend to form oligomers because of energetic favoring. To get stable dispersion, entropic favoring stemming from strong dilution (much more than the modelled ratio of 1500 NH 3 per Ga) is required. This explainst he comparably low solubility of GaCl 3 in the absence of NH 4 Cl. In turn, already an equimolar solution of GaCl 3 and NH 4 Cl leads to {[Ga(NH 3 ) 6 ]Cl 4 } À complexes that show Coulombic shielding against dimerization. Such stabilization may be further boosted by increasing the NH 4 Cl content-with the maxi-  6 ] 3 + octahedraa nd engage, in additiont ot he Coulombic interactionw ith the Ga 3 + ion, in hydrogen bondingwith ammonia molecules in boththe metalc omplex and the embedding solvent (H···Cl 240-250pm). While full dissociation from the complex is not observed within the ns scale of the moleculardynamics runs, residencet imes above specific triangular motifs are only around 3ps. Both, associationoffurther chloride ions and complex dimerization is disfavoredbyr epulsive Coulomb interactions. Colors:Ga( yellow),C l(green), NH 3 (blue/white). solventp roperties,p articularly decreasing permittivity with increasingt emperature and decomposition of ammonia, reducing the available concentration of solvent, if conclusions for ammonothermal GaNc rystal growth are derived. Still, we strongly believe that these investigations open the door to a deeper understanding of the chemical processes involved.I t, however,r emains am ajor challenge to carry the NMR investigationst os upercritical fluids for ac loser approacht ot he chemicalconditions relevant for GaN crystal growth.

Experimental Section Synthesis
The gallium halides were purchased from Sigma Aldrich (purity ! 99,999 %t race metals basis, anhydrous). All manipulations were carried out under argon (glove-box:M Braun, Garching, Germany, p(O 2 ) < 0.1 ppm). The ammoniates were synthesized as described by Zhang et al. [7a] [Ga(NH 3 ) 6 ]X 3 with X = Br,Iwere synthesized by reaction of the respective gallium halides in liquid ammonia at room temperature. Ga(NH 3 ) 3 F 3 was obtained from elemental gallium (purity ! 99.9999 %t race metal basis) and NH 4 Fi na mmonia at 753 Ka nd 150 MPa in a9 7mLa utoclave made from the nickelbased alloy Inconel 718 and equipped with as ilver liner. [29] The synthesis was carried out in an one-sided closed tubular furnace LOBA 1200-60-400-1 OW (HTM Reetz GmbH, Berlin, Germany), which produces at emperature gradient in the reaction vessel. Ga(NH 3 ) 3 F 3 crystallizes in the colder temperature zone of the autoclave. The pressure was monitored with ap ressure transmitter and ad igital analyzer (P2VA1/5000 bar and 1DA2510 by HBM, Darmstadt, Germany). Samples for NMR measurements were prepared according to two different protocols. In case of samples intended for the identification of dissolved species, weighed amounts of GaCl 3 (between 1 and 10 mg) and NH 4 Cl were filled in a5mm medium walled NMR tube and approximately 0.5 go fa mmonia (Linde, purity ! 99.999, and further purified with aM icroTorr MC400-720F gas purifier, SAES Pure Gas, which reduces H 2 O, O 2 and CO 2 to < 1ppb) was condensed into the tubes via as elf-made Te nsi-Eudiometer after Hüttig. [30] After filling, the ammonia was solidified by cooling with liquid nitrogen, the NMR tubes were evacuated and flame sealed. In most samples as olid residue remained, due to the low solubility of gallium halides in liquid ammonia. The volume of the resulting solution was determined from the measured fill height and the known inner diameter of the NMR tubes and used to calculate the concentration of NH 4 Cl in the sample. Samples to be used for quantization of dissolved gallium species were prepared in Hshaped glass vessels allowing for direct decantation of as ample prepared in one leg into an NMR tube attached as second leg. Solutions were prepared by charging the vessels with known amounts of GaCl 3 ,N H 4 Cl and NH 3 (between 5a nd 7g). The mixture was allowed to equilibrate at room temperature and stirred for several minutes before part of the resulting saturated solution was decanted into the NMR tube. The liquids in both legs were then solidified by cooling with liquid nitrogen, and the NMR tube was flame sealed.

NMR measurements
NMR spectra were recorded on aBruker AvanceA V400 spectrometer ( 1 H4 00.1 MHz, 71 Ga 122.0 MHz, 81 Br 108.0 MHz, 35 Cl 39.2 MHz, 14 N2 8.9 MHz) at ambient temperature (296-299 K) in unlocked mode if not stated otherwise. Chemical shifts were calibrated using the 15 Ns ignal of liquid ammonia (d = À381.7 ppm at 298 K with at emperature dependence of 40 ppb [31] as external standard and are referenced to external TMS using the X-scale [32]  Measurements aiming at quantification of dissolved species were carried out using the FIXPUL method. [33] All spectra were recorded using the same number of transients and processed by applying an exponential apodization function with al ine broadening factor of 50 Hz prior to Fourier transformation. The signal strength was evaluated by both numerical integration and spectral deconvolution (with both methods yielding consistent results), and gallium concentrations were calculated as described [33] from the measured signal integrals of the samples and ar eference sample of known concentration (14.17(5) mm aqueous GaCl 3 ).