Structural and Topological Control on Physical Properties of Arsenic

Feb 3, 2014 - The structures of Ge-doped arsenic selenide glasses with Se contents varying between 25 and 90 at. % are studied using a combination of ...
0 downloads 10 Views 2MB Size
Article pubs.acs.org/JPCB

Structural and Topological Control on Physical Properties of Arsenic Selenide Glasses Derrick C. Kaseman,† Ivan Hung,‡ Zhehong Gan,‡ Bruce Aitken,§ Steven Currie,§ and Sabyasachi Sen*,† †

Division of Materials Science, University of California at Davis, Davis, California 95616, United States Center of Interdisciplinary Magnetic Resonance, National High Magnetic Field Laboratory, 1800 East Paul Dirac Drive, Tallahassee, Florida 32310, United States § Glass Research Division, Corning Incorporated, Corning, New York 14831, United States ‡

S Supporting Information *

ABSTRACT: The structures of Ge-doped arsenic selenide glasses with Se contents varying between 25 and 90 at. % are studied using a combination of high-resolution, twodimensional 77Se nuclear magnetic resonance (NMR) and Raman spectroscopy. The results indicate that, in contrast to the conventional wisdom, the compositional evolution of the structural connectivity in Se-excess glasses does not follow the chain-crossing model, and chemical order is likely violated with the formation of a small but significant fraction of As−As bonds. The addition of As to Se results in a nearly random cross-linking of Se chains by AsSe3 pyramids, and a highly chemically ordered network consisting primarily of corner-shared AsSe3 pyramids is formed at the stoichiometric composition. Further increase in As content, up to 40 at. % Se, results in the formation of a significant fraction of As4Se3 molecules with As−As homopolar bonds, and consequently the connectivity and packing efficiency of the network decrease and anharmonic interactions increase. Finally, in the highly As-rich region with 2.4 it becomes rigid. However, these topological arguments do not take into account the effects of the degree of chemical order in the short-range structure or of the formation of structural units at the intermediate-range length scale such as molecular entities on the compositional dependence of physical properties in chalcogenide glasses. Previous studies have Received: December 19, 2013 Revised: January 31, 2014 Published: February 3, 2014 2284

dx.doi.org/10.1021/jp412451h | J. Phys. Chem. B 2014, 118, 2284−2293

The Journal of Physical Chemistry B

Article

demonstrated that the elastic moduli, density and Tg of AsxSe1−x glasses indeed display an extremum at ⟨r⟩ = 2.4, consistent with a rigidity percolation type transition. On the other hand, ⟨r⟩ = 2.4 in these glasses also coincides with the chemical threshold where heteropolar As−Se bonding is known to be maximized, leaving the chemical vs topological origin of the extremum in properties open to interpretation. The structures of simple binary AsxSe1−x chalcogenide glasses have been studied extensively over the past decade using a wide range of spectroscopic and diffraction techniques.9,13,14,18,21,23 These structural studies have primarily addressed the average short-range structure of these glasses. As and Se K-edge extended X-ray absorption fine structure (EXAFS) spectroscopic results have demonstrated that As and Se atoms are predominantly 3- and 2-coordinated, respectively, in AsxSe1−x glasses, and the 8−N coordination rule is largely obeyed.18 However, the relative extent of violation of chemical order and formation of “wrong” bonds remains unclear. Furthermore, suggestions have been made in the literature of the violation of 8−N rule and formation of significant concentrations (5−10%) of defects such 4-coordinated As, AsSe double bonds, and singly coordinated Se atoms.29,30 77Se magic-angle-spinning nuclear magnetic resonance (MAS NMR) spectroscopic results on Se-rich AsxSe1−x glasses (i.e., x < 40 at. %) are interpreted to suggest a chain-crossing model where addition of As to Se results in cross-linking of Se chains via formation of AsSe3 pyramids such that the connectivity between these pyramids via corner-sharing in the network is avoided for as long as possible.9,21,23,24 This increasing connectivity results in increasing dimensionality of the network. The stoichiometric composition As2Se3 at the chemical threshold is characterized by a network of corner-shared AsSe3 pyramids where all Se atoms exist in As−Se−As environments. Recent 77Se MAS NMR and Raman spectroscopic studies of Se-deficient glasses (i.e., x > 40 at. %) have suggested that further addition of As results in the formation of As4Se3, As4Se4, and As4 molecular entities, and Se−Se−As linkages are reintroduced into the network, leading to a structure with lower degree of connectivity and dimensionality compared to that near the chemical threshold.21,23 This reversal in the connectivity and dimensionality of the network across the chemical threshold has been argued by Yang et al. to be responsible for the various property extrema observed near ⟨r⟩ = 2.4.21 The validity of the chain-crossing model for the structure of AsxSe1−x glasses rests largely on the interpretation of the corresponding77Se MAS NMR spectra reported in the literature. Such spectra, even when collected at a relatively fast spinning speed of 15 kHz, may suffer from lack of resolution since the 77Se NMR line shapes are broadened not only by structural disorder but also by the chemical shift anisotropy (CSA) which, for Se sites with noncubic symmetry, can easily be too large to be averaged by regular MAS.15 This lack of spectral resolution, in combination with unconstrained line shape simulation, may lead to somewhat inaccurate structural models. For example, the 77Se MAS NMR spectral line shape for As2Se3 glass, reported in recent works by Yang et al., substantially overlaps with that of AsSe9 glass.21,23 Yet, the structure of the former is suggested to contain only As−Se−As environments while that of the latter contains only Se−Se−Se and Se−Se−As environments! While this structural conclusion may well be correct, the interpretation of the 77Se MAS NMR spectra remains subjective and equivocal.

Recently we have shown the application of a relatively novel two-dimensional (2D) NMR technique that combines the idea of magic-angle turning and phase-adjusted sideband separation (MATPASS) to separate isotropic/anisotropic chemical shift to yield high-resolution 77Se NMR spectra free of CSA-related broadening in GexSe100−x glasses.15 The Carr−Purcell− Meiboom−Gill (CPMG) echo train acquisition can be employed in combination with the 2D MATPASS technique to enhance the sensitivity for nuclides such as 77Se. Here, we present the results of a combined high-resolution 77Se 2D MATPASS/CPMG NMR and Raman spectroscopic study of the structure of As−Se glasses. These glasses were doped with a small amount of Ge (Ge:As = 1:18) in order to extend the glass formation range to 25 at. % Se with the aim to explore the composition dependent evolution of structure and topology over the widest composition range reported to date in the literature. The resulting structural model is used to establish structure−property relationships based on measurements of density, Tg, softening point Ts, thermal expansion coefficient α, and fragility m in these glasses and corresponding supercooled liquids.

2. EXPERIMENTAL SECTION 2.1. Synthesis. The Ge-doped As−Se glasses of composition (GeAs18)xSe100−xwith Se contents varying between 25 and 90 at. % in steps of 5% were synthesized in 20 g batches by melting mixtures of the constituent elements Ge, As, and Se with ≥99.995% purity (metals basis) in evacuated (10−6 Torr) and flame-sealed fused silica ampules at 1000 K for 24 h in a rocking furnace. The ampules were subsequently quenched in water. All samples were found to be X-ray amorphous. Physical property and Raman spectroscopic measurements were carried out on all glass samples while 77Se NMR spectroscopic data were collected on a subset of samples with Se contents of 25, 30, 40, 50, 60, 70, 80, and 90 at. %. 2.2. 77Se NMR Spectroscopy. All 77Se NMR measurements were carried out at the National High Magnetic Field Laboratory (NHMFL) using a 19.6 T narrow bore magnet equipped with a Bruker DRX console operating at a resonance frequency of 158.8 MHz for 77Se. A home-built 4 mm probe with a Samoson MAS stator was used. Crushed glass samples were packed into ZrO2 rotors and spun at 10 kHz. For each 2D experiment, 16 hypercomplex t1 points were acquired with 36 transients per point and 64 CPMG echoes per transient in ∼24 h.31 Each transient was obtained using the MAT sequence of five π-pulses with interpulse delays that were incremented according to the timings detailed by Hung et al.31 The MAT sequence is followed by CPMG pulses for multiple-echo acquisition. Hypercomplex data acquisition was employed using the method of States et al.32 to the phases of the CPMG pulses and the receiver. The π/2- and π-pulse lengths were 2.0 and 4.0 μs, respectively, and a 75 s recycle delay was used. All spectra were referenced to neat (CH3)2Se at δiso = 0 ppm by recording the 17O signal of neat, natural abundance H2O and then using the frequency ratios reported in the IUPAC recommendations, i.e., ν( 77Se, (CH3)2Se) = ν( 17O, H2O) × 19.071513/ 13.556457.33 Summation of the CPMG echoes and application of a shear transformation on the MAT/CPMG data yield the resulting spectra, an example of which is shown in Figure 1a. A further shear along the direct dimension and summation along the indirect dimension results in an infinite spinning speed isotropic spectrum, shown as the projection in Figure 1b. At a given isotropic shift, a slice along the indirect dimension results 2285

dx.doi.org/10.1021/jp412451h | J. Phys. Chem. B 2014, 118, 2284−2293

The Journal of Physical Chemistry B

Article

where δxx, δyy, and δzz are the principal components of the CSA tensor. The magnitude of the CSA is Δ is while the asymmetry of the CSA is denoted by η. 2.3. Raman Spectroscopy. Unpolarized Raman spectra were collected at ambient temperature and in backscattering geometry using a Bruker RFS 100/S Fourier transform (FT) Raman spectrometer equipped with a Nd:YAG laser operating at 1064 nm and a liquid nitrogen cooled solid-state Ge detector. A laser power level of 50 mW was used for all samples, and spectra were collected using a resolution of 2 cm−1. Approximately 32 scans were collected and averaged to obtain each Raman spectrum. 2.4. Physical Property Measurements. The Tg was measured using a differential scanning calorimeter at a heating rate of 10 K/min, and Ts (defined as the temperature where the viscosity is 106.6 Pa·s) was measured by the parallel plate technique using fused silica plates, both with a precision of ±2 K. Thermal expansion coefficient was measured with a precision of ±0.1 ppm/K by dilatometry from room temperature to Tg using Al2O3 as a standard. Density was measured in water to ±0.001 g/cm3 using the Archimedes’ method.

3. RESULTS 3.1. 77Se NMR Spectra. The projections of the 77Se MATPASS/CPMG spectra in the isotropic dimension for glasses with Se contents of 25, 30, 40, 50, 60, 70, 80, and 90 at. % are shown in Figure 2. All but one of these spectral line shapes can be simulated well with three Gaussian peaks centered around (i) 810−850 ppm, (ii) 610−620 ppm, and (iii) 425−465 ppm, with full widths at half-maximum (fwhm) of ∼180, 250, and 240 (±10) ppm, respectively (Figure 2). The exception is the near-stoichiometric glass with 60 at. % Se where the isotropic 77Se line shape can be simulated well with just peaks ii and iii. The simulation parameters and the peak areas for all glasses are listed in Table 1. The results of previous 77 Se NMR spectroscopic studies of binary Ge−Se and As−Se glasses indicate that the peaks ii and iii in Figure 2 can be assigned to Se atoms bonded to 1 Se and 1 As/Ge (i.e., Se− Se−As/Ge sites) and to two As/Ge atoms (i.e., As/Ge−Se− As/Ge sites), respectively.9,15,16,21,23,24 Extraction of the CSA tensor from simulation of the sideband patterns in the second dimension (Figure 3b,c) yields CSAs of 255 and 300 ppm for peaks ii and iii, respectively. These CSAs are consistent with the values reported in a previous study for Se−Se−Ge and Ge− Se−Ge sites in binary Ge−Se glasses.15 On the other hand, the CSA for peak i shows a discontinuous behavior as a function of glass composition. The CSA for this peak is ∼150 ppm in glasses with Se in excess of stoichiometry (Figure 3a) while for Se-deficient compositions the CSA abruptly increases to ∼875 ppm (Figure 3d). This result indicates that peak i corresponds to two different Se sites in Se-excess and Se-deficient glasses. The 77Se isotropic chemical shift and the relatively small CSA of peak i in Se-excess glasses clearly correspond to Se−Se−Se environments, in agreement with previous 77Se NMR studies on amorphous and crystalline Se, which consist exclusively of Se−Se−Se environments.8,24 On the other hand, the unusually large CSA of peak i in Se-deficient glasses allows for its unique assignment to As−Se−As sites in As4Se3 cage molecules.34 Recent 77Se NMR spectroscopic measurements in combination with density functional theory based calculations34 have shown that the77Se isotropic shifts of the As−Se−As sites in As4Se3 and As4Se4 molecules in their corresponding crystal structures are very similar and range between ∼760 and 835 ppm. The

Figure 1. (a) Representative 2D 77Se MATPASS/CPMG spectrum for Ge1.05As18.95Se80 glass. (b) Isotropic projection of the spectrum in (a) after shearing along the direct dimension, representing an MAS spectrum corresponding to infinite spinning speed. (c) Projection along the anisotropic dimension at the positions of the isotropic peak maxima in (b) showing the corresponding spinning sideband patterms for the Se−Se−Se and Se−Se−As sites.

in a sideband pattern, shown in Figure 1c. The sideband pattern was simulated using the program Dmfit to obtain the CSA tensor. Here the CSA tensor is reported following the Haeberlen convention defined as δzz − δiso ≥ δxx − δyy ≥ δyy − δiso

δiso =

1 (δzz + δxx + δyy) 3

Δ = δzz − δiso η=

δxx − δyy Δ 2286

dx.doi.org/10.1021/jp412451h | J. Phys. Chem. B 2014, 118, 2284−2293

The Journal of Physical Chemistry B

Article

excitation wavelength and resolution, these spectra are quite similar to those recently reported by Yang et al.21 for binary As−Se glasses (see Figure S1 in Supporting Information). Here we focus on the compositional evolution of the high frequency modes in the region between 175 and 325 cm−1. For glasses with Se contents ≥60 at. %, i.e., up to the point of the chemical threshold, two primary Raman bands are observed: one at ∼252 cm−1 corresponding to Se chain mode and the other at ∼225 cm−1 corresponding to As−Se stretching modes in AsSe3 pyramids. Increasing As concentration in this composition range results in progressive lowering of the intensity of the 252 cm−1 band and concomitant increase in the intensity of the 225 cm−1 band (Figure 4). In addition, a weak band is observed at ∼197 cm−1 that can be assigned to the Ge−Se stretching mode of GeSe4 tetrahedra, consistent with the low level of Ge doping. In Se-deficient glasses beyond the chemical threshold, two new Raman bands appear near 203 and 237 cm−1. These bands are relatively sharp, which are often characteristic of molecular units or breathing modes of large and well-defined atomic clusters in glasses. An example of the latter is the breathing mode of the boroxol rings in borate glasses.35 As mentioned above, the 77Se NMR results indicate the presence of As4Se3 molecular units in these glasses, and such molecules are indeed characterized by a breathing mode of the basal As3 ring near 237 cm−1.36 Therefore, the appearance of a sharp Raman band in this location in the glass spectra is consistent with the 77Se NMR results and indicates the presence of As4Se3 molecules. This peak maximizes in intensity near the composition with ∼45 at. % Se and practically disappears in the most Se-deficient composition, again consistent with the observed compositional variation of the corresponding 77Se NMR signal (Figure 1 and Table 1). In contrary to the suggestions made in a recent study by Yang et al. on binary As−Se glasses, no evidence for the vibrational modes corresponding to As4S4 molecular units is observed in the Raman spectra in Figure 4. As discussed above, a similar conclusion can also be drawn from the 77Se NMR results on these glasses. The band at 203 cm−1 remains strong in the most As-rich compositions and was assigned by Yang et al. to molecular As4 units in binary As−Se glasses with comparable Se contents.21 Previous investigations into amorphous elemental As also reported a similarly narrow, highly polarized band at 203 cm−1 in the Raman spectra and tentatively assigned it to “quasimolecular” units.37 It may be noted that subsequent diffraction and 75As nuclear quadrupole resonance studies have shown the absence of any significant fraction of As4 molecules in the structure of amorphous As.38−41 Rather, the structure of amorphous As is believed to consist of randomly corrugated two-dimensional sheets of 3-fold-coordinated As atoms.38,40 Furthermore, this sharp Raman band is not observed as a distinct peak in the vibrational density of states of amorphous As, obtained using inelastic neutron scattering.37 Therefore, it is most likely that this sharp and polarized Raman band is associated with breathing modes of large, low-dimensional As clusters in these As-selenide glasses. 3.3. Physical Properties. The compositional variation of the density (ρ), thermal expansion coefficient (α), glass transition temperature (Tg), and softening point (Ts) for these glasses are shown in Figure 5 and are compared in Figure S2 (see Supporting Information) with the corresponding data for binary As−Se glasses reported in the literature for up to 40 at. % Se. The remarkable correspondence in the trends of the compositional variation of these properties between the Ge-

Figure 2. Simulations of the isotropic 77Se NMR spectra. The dashed lines show the average 77Se isotropic chemical shifts for Se−Se−Se (purple) and molecular As−Se−As (yellow) units at ∼820 ppm, Se− Se−As (pink) units at 620 ppm, and As−Se−As (green) units at 420 ppm.

presence of both molecules was proposed in Se-deficient arsenic selenide glasses on the basis of this chemical shift assignment.21,23 However, these sites in the As4Se3 and As4Se4 molecular crystals were found to differ significantly in their CSAs.34 While the CSA ranges between 125 and −261 ppm for the four As−Se−As sites in As4Se4, it is 783 and 799 ppm for the two As−Se−As sites in As4Se3. Therefore, the large CSA observed for this site in the Se-deficient glasses in the present study clearly indicates the absence of As4Se4 molecules in the structure. It may be noted here that such exceptionally large CSAs (∼125 kHz) cannot be averaged by regular MAS, and hence, accurate Se speciation for these glasses will be practically impossible from 1D 77Se MAS NMR experiments. 3.2. Raman Spectra. The unpolarized Raman spectra of all glasses are shown in Figure 4. Although acquired using different 2287

dx.doi.org/10.1021/jp412451h | J. Phys. Chem. B 2014, 118, 2284−2293

The Journal of Physical Chemistry B

Article

Table 1. Simulation Parameters for the 77Se MATPASS/CPMG Isotropic NMR Line Shapes of the Glasses Studied Here, Corresponding to the Four Se Sites: Se−Se−Se, Se−Se−As, As−Se−As, and As−Se−As (As4Se3) glass composition Se−Se−Se shift (ppm) (±10 ppm) width rel fraction (±5%) Se−Se−As shift (ppm) (±10 ppm) width rel fraction (±5%) As−Se−As shift (ppm) (±10 ppm) width rel fraction (±5%) As−Se−As (in As4Se3) shift (ppm) (±10 ppm) width rel fraction (±5%)

Ge0.5As10Se90

Ge1.1As18.9Se80

Ge1.6As28.4Se70

Ge2.2As37.8Se60

Ge2.7As47.3Se50

Ge3.3As56.7Se40

Ge3.8As66.2Se30

Ge4As71Se25

851 170 63.15

832 188 38.01

813 188 15.91

620 243 34.35

610 252 49.67

606 252 50.52

610 262 13.37

620 262 32.39

620 262 38.2

620 262 41.56

620 262 29.18

464 243 2.5

464 243 12.32

465 243 33.57

450 243 86.07

456 238 56.67

441 241 52.24

426 241 54.15

426 241 66.46

818 175 11.02

821 177 9.56

821 177 4.29

821 177 4.36

Figure 3. CSA spinning sideband patterns for the Se environments (a) Se−Se−Se, (b) Se−Se−As, (c) network As−Se−As, and (d) molecular As− Se−As (in As4Se3 molecular units).

composition intervals. Consequently, all three physical properties show two extrema: one near the chemical threshold (60 at. % Se) which also coincides with ⟨r⟩ = 2.4 and the other near the composition with 45 at. % Se corresponding to ⟨r⟩ = 2.58. The first extremum is the only one reported to date in the literature for binary As−Se glasses due to the limited composition ranges explored in previous studies.21 The origin of this first extremum has remained controversial for binary As−Se glasses as its location coincides with the chemical threshold where the bonding is (nearly) solely heteropolar (i.e., As−Se bonds) as well as with the rigidity percolation threshold where the number of degrees of freedom equals the number of

doped and Ge-free As−Se glasses, even for the Se-deficient compositions, is noteworthy (Figure S2). This observation strongly suggests that the compositional evolution of the microstructure of As−Se glasses, and consequently the structure−property relationships remain essentially unperturbed by the low levels of Ge doping employed in the present study. Both ρ and Tg show monotonic increase with decreasing Se content from 90 to 60 at. % followed by a decrease upon further lowering of the Se content up to 45 at. %, and subsequently both properties increase again up to the lowest Se content of 25 at. % (Figure 5). On the other hand, α shows an opposite variation with respect to ρ and Tg, in these 2288

dx.doi.org/10.1021/jp412451h | J. Phys. Chem. B 2014, 118, 2284−2293

The Journal of Physical Chemistry B

Article

constraints and the structure becomes optimally rigid.21,26−28 The compositional variation of Ts, which is a property of the supercooled liquid state, is compared with that of Tg in Figure 5c. It is clear that the strongly defined extrema in the variation of Tg are significantly weaker in the nearly monotonic variation of Ts in the supercooled liquid state.

4. DISCUSSION 4.1. Atomic Structure: Connectivity and Topology. The compositional evolution of the 77Se NMR isotropic NMR and Raman spectra, when taken together, yields the following general structural scenario for Se speciation in these Ge-doped As−Se glasses. Amorphous Se consists of Se−Se−Se chain environments, and the addition of As and Ge cross-links the Se chains, resulting in a decrease in the relative fraction of the Se− Se−Se environments while increasing those of the Se−Se−As/ Ge and As/Ge−Se−As/Ge environments. Near the chemical threshold, in the slightly Se-deficient Ge2.2As37.8Se60 glass, the Se−Se−Se environments disappear completely, and a crosslinked network of corner-shared AsSe3 pyramids (and a minor fraction of GeSe4 tetrahedra) forms, which is comprised entirely (or nearly entirely) of As/Ge−Se−As/Ge environments. This result is similar to previous reports of the observation of complete chemical order in the stoichiometric As40Se60 glass where all Se atoms are present in As−Se−As environments.9,13,21,23 Increasing deficiency in Se results in the formation of As4Se3 molecules and Se−Se−As/Ge environments, besides the most abundant As/Ge−Se−As/Ge environments. The maximum contribution of the As4Se3 molecules is observed in the glass with 45 at. % Se in which more than 11% of the Se atoms, and hence, nearly 15% of the As atoms are present in this molecular unit. The concentration of As4Se3 molecules drops rapidly with further decrease in Se content, and in the most Se-deficient composition with 25 at. % Se, only ∼4.4% of the Se atoms and hence ∼2% of the As atoms are associated with these molecules. The reappearance of the Se− Se−As/Ge sites in the Se-deficient glasses, maximization in their relative fraction near the glass composition with 30 at. % Se, and their existence even in the most Se-deficient glass studied here with 25 at. % Se are rather intriguing results. The lack of any correlation between the relative concentrations of these sites and those of the As4Se3 molecules in these Sedeficient glasses indicates a corresponding absence in a causeand-effect relationship in the formation of these structural moieties. However, a plausible structural model that can explain the reappearance of the Se−Se−As/Ge sites in the Se-deficient glasses is the formation of As-rich clusters with a connective “tissue” that contains excess Se, thereby necessitating the formation of Se−Se−As/Ge sites. Such clustering and molecule formation are consistent with the recent observation of bimodal relaxation in As-rich binary As−Se glasses by heat capacity spectroscopy.23 The relative fraction of the Se−Se−Se sites in binary AsxSe100−x glasses, as obtained in previous studies from unconstrained simulation of low-resolution 77Se MAS NMR spectra was shown to follow the chain-crossing model.9,21,24 Deschamps et al.,42 on the other hand, tentatively suggested a violation of the chain-crossing model in binary As−Se glasses on the basis of 77Se 2D-CPMG MAS NMR results, but unfortunately, these experiments were nonquantitative and the 77 Se NMR line shapes were affected by incomplete averaging of CSA. In contrast, the relative fraction of the Se−Se−Se sites in Ge-doped As-selenide glasses obtained in the present study

Figure 4. Unpolarized Raman spectra of the Ge-doped arsenic selenide glasses.

Figure 5. Compositional varitaion of (a) denstiy ρ, (b) thermal expansion coefficient α, and (c) Tg (black squares) and Ts (red circles).

2289

dx.doi.org/10.1021/jp412451h | J. Phys. Chem. B 2014, 118, 2284−2293

The Journal of Physical Chemistry B

Article

from constrained fits of the high-resolution 77Se isotropic spectra provides quantitative evidence for significant deviation from the chain-crossing model in glasses with ≤80 at. % Se (Figure 6). This deviation cannot be explained to be due to the

bonded to 1 As and 12.3% are in As/Ge−Se−As/Ge sites, bonded to 2 As or Ge atoms. Therefore, 39.8 (9.8) Se atoms in the formula unit are in the Se−Se−As (As/Ge−Se−As/Ge) environment. It is reasonable to assume that the larger electronegativity difference between Ge and Se atoms compared to that between As and Se atoms ensures Ge atoms to be preferentially heteropolar bonded to Se. Hence, 2.2 Se atoms need to be assigned for bonding to 1.1 Ge atoms in the formula unit in this glass to satisfy the GeSe2 stoichiometry which leaves (9.8 − 2.2) = 7.6 Se atoms to form As−Se−As linkages. For each Se in the Se−Se−As (As−Se−As) environment, one can count one-third (two-thirds) of an As atom to be heteropolar bonded. Therefore, the total number of heteropolar bonded As atoms in this glass is [(39.8/3) + (7.6 × 2/3)] = 18.3. This number, within the error bars of ±1.5 resulting from the corresponding error bars of ±5−7% for the estimation of the relative fractions of various Se-species from NMR data, is almost the same as the total number of As atoms (18.9) in the formula unit. Thus, the Se speciation results indicate the absence of any significant homopolar As−As bonding in this glass. In contrast, for the Ge1.6As28.4Se70 glass 50.5% of the Se atoms are bonded to 1 As and 33.6% are bonded to 2 As or Ge atoms, and following the same calculation scheme as above, one obtains 25.3 ± 1.5 As atoms per formula unit that are heteropolar bonded in this glass, a number that is significantly lower than the total number of As atoms (28.4) per formula unit. This result therefore implies violation of chemical order. It is interesting to note that such violation of chemical order has not been observed in binary Ge−Se glasses with excess Se, consistent with the larger electronegativity difference between Ge and Se atoms (0.6) compared to that between As and Se atoms (0.4). Similar calculations for the slightly Se-deficient Ge 2.2 As 37.8 Se 60 composition indicate that ∼34.5 As atoms are involved in heteropolar bonding, again significantly lower than the number 37.1 expected from the chemical composition and order, under the assumption that all Ge atoms are heteropolar bonded. 4.3. Structure−Property Relationships in the Glassy State. Consistent with the structural information obtained from the NMR and Raman spectroscopic results, the compositional variation of the properties in Figure 5 can be divided into three regions: (i) the Se-excess region with ≥60 at. % Se, (ii) the region of significant As4Se3 molecule formation, up to 40 at. % Se, and (iii) the As-rich region with