1906
J . Phys. Chem. 1992,96, 1906-1915
Atomic Distribution in Crystalline I I-IV-V, Semiconductor Alloys. 31Pand '13Cd Magic Angle Spinning, Spin-Echo, and 31P-113CdSpin-Echo Double-Resonance NMR Studies of the Systems ZnGeAs, P, and CdGeAs,, P, Deanna Franke, Kesha Banks, Robert Maxwell, and Hellmut Eckert* Department of Chemistry, University of California, Santa Barbara, California 93106 (Received: July 29, 1991)
Anion substitution in the ternary II-IV-V2 semiconductor systems ZnGeAs2-,P, and CdGeAsz-,P, results in homogeneous solid solutions, which crystallize in the chalcopyrite structure. The distribution of local (nearest neighbor) environments in these materials has been examined by a number of complementary solid-state NMR techniques, including IL3Cdand 31P magic angle spinning (MAS), II3Cd and 31Pspin-echo, and 113Cd-3'P spin-echo double-resonance (SEDOR) experiments. II3Cd and 31P MAS-NMR spectra show single resonances, whose chemical shifts change monotonically as a function of x. Contrary to the situation in 11-VI semiconductor alloys, no chemical shift discrimination attributable to different local nearest neighbor configurations is apparent in either MAS or locally selective SEDOR experiments. These results suggest that the chemical shifts in these compounds are not determined by local bonding configurations and, rather, that a description involving spin-echoand slP-l13Cd SEDOR decay data reveal that the 31P-31Phomonuclear charge delocalizationis more appropriate. dipole coupling is stronger, and that the 31P-113Cddipolar coupling is weaker than expected for a random distribution of P and As over the anionic sublattice of the chalcopyrite structure. These results are qualitatively consistent with a structure in which both the cation and the anion sublattices are disordered, and where the average number of P-Ge and As-Cd bonds is somewhat higher than that of the P-Cd and As-Ge bonds.
Introduction The distribution of local environments in binary and ternary semiconductor alloys has become an important issue in the attempt to understand their physicochemical properties. In this connection, experimental and theoretical evidence for bimodal bond length distributions, clustering, and long-range-order phenomena has been presented for several 111-V systems, suggesting that such configurations are more stable than corresponding random arrangements.',2 To test such hypotheses, locally selective spectroscopic techniques are needed, and it has been previously demonstrated that nuclear magnetic resonance is a well-suited technique for this purpose. For instance, in CdTebased semiconductor alloys, the lz5Techemical shifts, measured by the technique of magic angle spinning (MAS-NMR), were shown to differentiate sensitively between the various possible nearest neighbor codigurations present. In the system Zn,Cd,-,Te, these site populations were found to be distributed in a close to random fashion,j whereas in Hg,Cd,,Te alloys, a certain amount of chemical ordering can be The only 111-V semiconductor in which this issue has been addressed previously by N M R is Ga,-,In,P.' However, in this case 31PNMR proved to be rather insensitive to disordering phenomena, since the chemical shifts of the end members GaP and InP lie very close to each other. The present study is devoted to substitution effects in a closely related family of compounds, based on the 11-IV-V2 semiconductors, which crystallize in the tetragonal chalcopyrite structure.*-" Specifically, we discuss here the effect of anion substitution in the semiconductor alloys CdGeAs,-,P, and ZnGeAs_,P, on the basis of detailed composition-dependent 'IP ( I ) Martins, J . L.; Zunger, A. Phys. Reu. 1984, B30, 6217. (2) Akimoto, K.; Mori, Y.; Kojiman, C. Phys. Reu. 1987, 835, 3799. (3) Beshah, K.; Zamir, D.; Becla, P.; Wolff, P. A.; Griffin, R. G. Phys. Rev. 1987, 836, 6420. (4) Zax, D.; Vega, S.; Yellin, N.; Zamir, D. Chem. Phys. Lett. 1987, 138, 105.
( 5 ) Zamir, D.; Beshah, K.; Becla, P.; Wolff, P. A.; Griffin, R. G.; Zax, D.; Vega, S.; Yellin, N. J. Vac. Sci. Technol. 1988, A6, 2612. (6) Vieth, H. M.; Vega, S.; Yellin, N.; Zamir, D. J . Phys. Chem. 1991, 95, 1420. ( 7 ) Duncan, T. M.; Karlicek, R. F.; Bonner, W. A,; Thiel, F. A. J . Phys. Chem. Solids 1984, 45, 389. (8) Vaipolin, A. A. Izv. Akad. Nauk SSSR Neorg. Mater. 1967, 3, 260. (9) Pfister, H. Acta Crystallogr. 1958, 1 1 , 221. (10) Girault, B.; Mater. Res. Bull. 1978, 13, 457. ( 1 1 ) Shah, S.; Green, J. Mater. Left. 1983, 2, 115.
and 13Cd single- and double-resonance experiments. NMR spectroscopy offers a variety of experimental approaches for studying the issue of clustered (chemically ordered) versus random distributions. In the present systems, the detailed atomic distribution will affect 31Pand Il3Cd chemical shifts, the strength of homonuclear 31P-31Pdipole-dipole couplings, and the strength of heteronuclear 31P-113Cdinteractions. Since the compounds under study here are also known to form glasses under fastquenching conditions, the results of the present study provide important benchmark data for the N M R characterization of the glassy phases.
Experimental Section Sample Preparation md Characterization. Crystalline samples in the systems CdGeAs2-,P, and ZnGeAs,-S, were prepared from the elements (Aldrich; Zn,99.99%; Cd, 99.5%; Ge, 99.99%, P, 99.9%; As, 99.999%) in evacuated (lr3Torr) silica glass ampules. The temperature was increased slowly to 450 "C,kept there for 3-5 h, then increased to 600 OC for 3-5 h, and subsequently increased to 800 O C . After 24 h or more at this temperature, the samples were cooled slowly (30-60 OC/h) to room temperature. The resting periods at 450 and 600 OC are crucial in order to avoid too rapid sublimation of phosphorus and arsenic prior to reaction. Violent explosions and fires (due to elemental phosphorus) resulted if the heating rates were too fast. X-ray powder diffraction data confirmed that phase-pure materials were formed, crystallizing in the chalcopyrite structure (space group 142d).8-" Samples in the system CdGeAs2-,P, with high As contents showed two additional very weak diffraction peaks, which coincide with the two most intense peaks for Cd2P207(JCPDS card 17-635). Although the samples are not air-sensitive, sample preparation steps and subsequent manipulations were generally carried out in a nitrogen glovebox, in order to minimize the exposure to arsenic, which is known to be extremely toxic and carcinogenic. The furnaces in which these samples were prepared were initially placed in hoods to minimize fire hazards and laboratory contamination in cases of explosions during the determination of optimal synthesis conditions. Carbonizing the silica ampules proved to eliminate any reaction between the sample and ampule materials and provided a nonstick surface for ease of sample removal. Table I lists the best-fit lattice parameters using the LATCON fitting procedure. In both cases, linear Vegard plots are observed (see Figures 1 and 2), indicating that homogeneous solid solutions,
0022-365419212096- 1906%03.00/0 0 1992 American Chemical Society
The Journal of Physical Chemistry, Vol. 96, No. 4, 1992 1907
Atomic Distribution in II-IV-V2 Semiconductor Alloys
TABLE I: Compositions, Lattice Parameters, Average Chemical Shifts (hl ppm vs 85%HQO, and Liquid (CH3)2Cd, Respectively), and Calculated and Measured Average Dipolar Second Moments, MUu
(a) Crystalline Samples in the System CdGeAs2,P,
0.00 0.25 0.50 0.75 1.00 1.25 1.50 1.75 2.00
5.955 5.956 5.917 5.877 5.855 5.811 5.803 5.760 5.744
11.224 11.195 33.8 11.138 23.2 1 1.047 14.0 1 1.043 2.7 10.966 -8.2 10.875 -15.3 10.807 -24.0 10.771 -32.0
151 158 161 172 186 190 209 220 237
1 .o 2.1 3.1 5.0 5.8 7.6 9.1 11.2
3.0 3.1 4.2 6.2 8.1 10.4 12.7 15.0
1.5 3.8 5.7 7.5 12.0
1.9 4.0 6.2 8.4 11.0 13.5 16.3 19.2
0.066-0.118 0.062-0.129 0.076-0.087 0.075-0.126 0.128-0.156 0.132-0.188 0.150.332 0.422-0.426
0.0374 0.0376 0.0390 0.0407 0.0414 0.0433 0.0442 0.0456 0.0470
23.7 20.8 18.5 16.1 13.1 10.3 7.0 3.6
0.0015 0.0020 0.0020 0.0025 0.0050 0.0070 * 50 ‘25
12.3 10.5 9.5 7.6 6.1 4.0 2.1
(b) Crvstalline SamDles in the Svstem ZnGeAs,-,P,
0.25 0.50 0.75 1.00
5.634 5.621 5.571 5.559
11.059 11.031 10.924 10.901
1 .o -9.5 -20.0 -28.7
2.4 3.4 6.1 9.1
1.4 2.9 4.6 6.2
1.50 1.75 2.00
5.519 5.479 5.473
10.813 10.733 10.720
-46.7 -52.4 -60.3
9.8 11.9 13.7
13.1 15.8 19.6
“ I n units of lo6 rad2/s2;estimated error * 5 % for Mu(31P-31P),*lo% for M2d(113Cd-113Cd) and Mzd(113Cd-31P). bAsteriskdenotes that Tz,s(31P)is given; all other values refer to T2,s(75As). 11.31
11.27
0 0
0
0
10.91
4
0
0
10.8
0
0
10.81
O
O
0
0
10.
o + l.+
$
{
5.7
6.0
0 0
a,,
!
0.0
0.5
1 .o
1.5
2.0
X
Figure 1. Lattice parameters ( a and c) as a function of x in crystalline semiconductor alloys of composition CdGcAs2-,P,.
rather than phase-segregated materials, are formed over the entire composition range. solidstrte NMR Shdies. Fielddependent MAS-NMR studies were carried out on General Electric GN-300, Nicolet NT-300, and General Electric GN-500 spectrometers, at approximate frequencies of 12 1.65 and 66.70 MHz at 7.05 T and 202.4 and 111.0 MHz at 11.7 T, for 31Pand I l 3 C d MAS-NMR, respectively. These measurements employed a 5-mm high-speed MAS-NMR probe from Doty Scientific. Typical conditions were as follows: pulse length 4 ps and spinning speed 9.0 kHz. The spectra were obtained under representative conditions, i.e., with relaxation delays such that in compounds giving rise to multiple peaks, the relative peak areas were independent of relaxation delays. Isotropic chemical shifts are reported relative to 85% H3P04and liquid (CH3),Cd, respectively. 3’P-decoupled Il3Cd MAS studies (at 5-kHz spinning speed) and lI3Cd and 3’Pspin-echo and 113Cd_31P SEDOR NMR experiments were undertaken on a General Electric GN-300 spectrometer, equipped with a 2-MHz fast di-
5.4
I
0.0
0.5
1 .o
1.5
2.0
X
Figure 2. Lattice parameters ( a and c) as a function of x in crystalline semiconductor alloys of composition ZnGeAs2-,P,.
gitizer and a 7-mm doubly broad band tuned MAS-NMR probe from Doty Scientific. Typical 90° pulse lengths were 5-8 (31P) and 7-10 cis (Il3Cd). For each sample, the rmnance frequencies for lI3Cd and 31Pwere carefully adjusted to minimize resonance offset effects. The spin-echo and SEDOR experiments were conducted on nonspinning samples, under rigorously quantitative conditions, necessitating delays from 15 to 60 min for 31Pand 60 to 120 min for l13Cd. Due to excellent signal to noise ratio the 31Pspin-echo heights could be measured directly from the computer screen, whereas the II3Cd spin-echo heights were obtained by fitting the relatively noisy echoes to a Gaussian function. (Lorentzian fits carried out for comparison yielded virtually identical results.) Io, the spin-echo height at zero evolution time, was obtained from a polynomial fit to the experimental data. This value was then used to normalize the data. Usually at least two independently prepared samples of the same composition were studied and every experiment was repeated at least once with a
--
1908 The Journal of Physical Chemistry, Vol. 96, No. 4 , 1992
x
X
= 2.0
=
Franke et al.
X
0.5
n A
0.5
-
1.5 x 40 0
300
200
100
1
I
-130
=mu
2.0
, 100
0.0
1.0
x
n
0.0
* I
,
r
I
,
A,
,
I
50
,/!-
,
-50
0
PPM
2.0
Figure 3. 66.70-MHz Ii3Cd MAS-NMR spectra (top) of crystalline
semiconductor alloys with composition C ~ G ~ A S ~(representative -~P, samples): spinning speed ca. 5 kHz. All the spectra were obtained with CW decoupling. The compositional dependenceof the Il3Cd chemical shift is plotted in the bottom part of this figure. The solid line is a linear least-squares fit to the data. relaxation delay twice as long as that previously used. Results, Data Analysis, and Interpretation Il3CdMASNMR. In the chalcopyrite lattice, the metal cations (Zn, Cd, Ge) are coordinated to four pnictogens (P or As) and all pnictogen atoms are coordinated to two Ge and two metal atoms (Cd, Zn).*-II Partial substitution of P by As in ZnGeP, and CdGeP, is expected to generate five distinct Cd (or Zn) environments, having zero, one, two, three, or four phosphorus nearest neighbors. The statistical probability P(n,x) of a Cd (or Zn) atom to be surrounded by n phosphorus atoms and 4 - n arsenic atoms is given by
In principle one expects that these sites have distinct chemical shifts. Thus, as the P/As ratio is varied, multiple-peak spectra are expected, whose area ratios should reflect site populations. This expectation can be tested by high-resolution solid-state NMR studies, where all the anisotropic line-broadening mechanisms affecting NMR spectra in solids are eliminated by fast magic angle spinning. Figure 3 shows representative I13Cd MAS-NMR spectra of various crystalline CdGeAsz-xPx samples. In contrast to the expected site resolution, only single broad peaks flanked by weak spinning side band shoulders are observed. These peaks shift
1
0.0
1.0
x
2.0
Figure 4. 121.65-MHz 31PMAS-NMR spectra (top) of crystalline semiconductor alloys with composition CdGeAs2-xP, (representative samples). Spinning side bands are labeled by asterisks. The two sharp lines observed in the spectra are attributed to an impurity of crystalline Cd2P20,. The compositional dependence of the ,IP chemical shift is plotted in the bottom part of this figure. The solid line is a linear least-squares fit to the data.
downfield with increasing phosphorus content. The compositional dependence of the isotropic chemical shifts extracted from these spectra is plotted at the bottom of Figure 3, revealing a more or less monotonic trend with increasing x. Field-dependent N M R studies reveal that the width is governed by a distribution of isotropic chemical shifts. MAS-NMR spectra obtained with and without 31Pdecoupling show further that the dipolar broadening by adjacent 31Pspins is negligible. The sharp resonance observed for crystalline CdGeAsz suggests that the MAS line shapes are largely unaffected by the dipolar interaction to the quadrupolar 75Asnuclei. This conclusion, which is further substantiated by the spin-echo N M R results (see below), indicates that the 75As spins fluctuate rapidly on the N M R time scale due to fast spinlattice relaxation, hence resulting in efficient self-decoupling. 31PMAS-NMR. Figures 4 and 5 show results from 31P MAS-NMR on representative samples in the CdGeAs2_,P, and Z ~ G ~ A S ~phase _ ~ P fields, , respectively. These spectra contain a broad line, comprising the majority of the P a t o m in the sample, whose chemical shift is systematically dependent on x. In addition, there are minority components, consisting of unusually sharp 31P MAS-NMR signals, at -2.5 and -4.5 ppm, and at 3.9 ppm for
The Journal of Physical Chemistry. Vol. 96, No. 4, 1992 1909
Atomic Distribution in 11-IV-V2 Semiconductor Alloys
of P and As atoms within the second coordination sphere, and whose intensity is again governed by a binomial distribution law. The observation of a single broad line in the jlP MAS experiment is, however, not too surprising in view of the Il3Cd MAS-NMR results. Nevertheless, the monotonic dependences of the chemical shifts on x confirm, in agreement with the X-ray diffraction data, that ZnGeAs2-,P, and CdGeAs2-,P, samples are homogeneous solid solutions rather than phaseseparated materials. jlP Spin-Echo NMR. Dipolar spectroscopy offers a particularly powerful approach to these systems, because the spectroscopic information can often be calculated from first principles. As discussed in detail previ~usly,'~-'~ 90°-tl-1800 spin-echo experiments with incrementation of the pulse delay tl offer, in prinpicle, selective information concerning homonuclear dipole-dipole couplings, e.g., among jlP or II3Cd spins. For multispin systems, a Gaussian decay of spin-echo intensity with time is expected, yielding a homodipolar second moment hf2d(31P-31P)from the semilogarithmic plot defined by I(2tl)/IO = eXP-((2tl)2hf2d/2)
1
"
"
/
50
100
"
"
"
"
"
"
- .
"
0
-50
-100
(2)
The M2d values characterize the average strength of the homonuclear interactions under consideration. Theoretical second moment values can be calculated from the van Vleck equation?
PPM
hfZd
= (4/15)(~0/4*)~1(1 + i)y4h2N-1zddijd i#j
20 1
a
1
L
0.0
1.0
x
2.0
Figure 5. 121.65-MHz 31P MAS-NMR spectra (top) of crystalline semiconductor alloys with composition ZnGeAs2-,PX (representative samples). Spinning side bands are labeled by asterisks. The sharp line observed in the spectra is attributed to an impurity of crystalline aZn3(P04)2. The compositional dependence of the I'P chemical shift is plotted in the bottom part of this figure. The solid line is a linear least-squares fit to the data.
the CdGeAs,-,P, and ZnGeAs2-,P, systems, respectively. The positions of these sharp peaks are independent of the composition, but their intensities relative to the main resonances increase consistently with increasing As content. The jlP chemical shifts coincide with those reported for ( Y - Z ~ ~ ( Pand O ~Cd2P207, )~ respectively.I2 While replicate sample preparations show little variation in the intensities of these peaks, their intensity can be reduced if the Zn and Cd metals are pretreated with hydrogen a t elevated temperatures. All this evidence suggests that these extra raonances are attributable to the above-mentioned metal phosphates, which appear to form from metal oxide impurities in our starting materials. Note that there is no evidence for Cd2P207in the Il3Cd MAS-NMR spectra, but this may simply be due to a very long spin-lattice relaxation time of the impurity phase. Since only the spectral parameters of the dominant broad lines are sensitive to the composition, the discussion to follow deals primarily with this spectral component, which in most cases comprises a t least 90%of the total phosphorus content, as measured by peak integration. Fielddependent NMR studies confirm that, as in the case of the Il3Cd MAS resonances, the width is governed by a distribution of isotropic chemical shifts. In principle, substitution of P by As in CdGePz and ZnGeP2 is expected to generate 13 distinct P environments, which differ in the number (12) Cheetham, A. K.; Clayden, N . J.; Dobson, C. M.; Jakeman, R. J. J . Chem. Soc., Chem. Commun. 1986, 195.
(3)
Here I and y are the spin quantum number and gyromagnetic ratio of the nucleus under consideration, h is Planck's constant, N is the number of nuclei for which M M is calculated, and d..are the distances between the nuclei under consideration a n d the surrounding nuclei generating dipolar fields. Equation 3 assumes that the chemical shift dispersion quenches the flip-flop term in the homonuclear dipolar Hamiltonian, which is, in fact, a necessary condition for the complete refocusing of chemical shift by the spin-echo sequence. In simple cases, eq 3 provides an excellent opportunity for testing possible atomic distribution models against experimental data. This principle has been previously exploited for investigating the phosphorus distribution in silica glass,17 amorphous hydrogenated silicon,lsJ8 and a variety of non-oxide g l a s s e ~ , *including ~ - ~ ~ glassy CdGeP2.22 Preliminary results on the application of this technique to the present system have been communciated by us p r e v i o u ~ l y . ~ ~ Figures 6 and 7 show representative jlP spin-echo NMR decays for the two systems under study. Table I compares experimental M2d(31p-31P)values (eq 2) with calculated average values. The latter were obtained from eq 3 by assuming a statistical distribution of P atoms (according to individual compositions) over the anionic sites of the chalcopyrite lattice. These calculations are based on P-P distances derived from the experimentally determined lattice constants. Note that for all compositions considered the experimental M2d(31P-31P)values are generally ca. 20-30% higher than expected from random distribution over the anionic sublattice. Possible reasons for this observation will be discussed below. II3Cd Spin-Echo NMR in Crystalline CdCeAs2-,P, Alloys. Figure 8 shows representative spin-echo decays for the II3Cd resonance in crystalline CdGeAs2-,P, alloys. The analysis is carried out in the same fashion as for the jlP results. Since there are in certain cases significant deviations from a Gaussian decay (13) Engelsberg, M.; Norberg, R. E. Phys. Reo. 1972, 85, 3395. (14) Boden, N.; Gibb, M.; Levine, Y. K.; Mortimer, M. J . Magn. Reson. 1974, 16, 471. (15) Reimer, J. A.; Duncan, T. M. Phys. Reu. 1983,827, 4895. (16) Van Vleck, J. H. Phys. Reu. 1948, 74, 1168. (17) Douglas, D. C.; Duncan, T. M.; Walker, K. L.; Csencsits, R. J. Appl. Phys. 1985, 58, 197. (18) Boyce, J. B.; Ready, S. E. Phys. Reu. 1988, 838, 11008. (19) Lathrop, D.; Eckert, H. J. Am. Chem. Soe. 1989, I l l , 3536. (20) Tullius, M.; Lathrop, D.; Eckert, H. J . Phys. Chem. 1990, 94, 2145. (21) Lathrop, D.; Eckert, H. Phys. Reu. 1991, B43, 7279. (22) Franke, D. R.; Maxwell, R.; Lathrop, D. A.; Eckert, H. J . Am. Chem. SOC.1991, 113,4822. (23) Eckert, H.; Franke, D.; Lathrop, D.; Maxwell, R.; Tullius, M.; Mater. Res. SOC.Symp. Proc. 1990, 172, 193.
1910 The Journal of Physical Chemistry, Vol. 96, No. 4, 1992
Franke et al.
llla
0.0
0.5
2t1 [ms]
1.0
20
2.0
1
0
'
2
x
Figure 6. spin-echo decays (top) for crystalline semiconductor alloys and (c) with compositions (a) C ~ G ~ A S , , ~ , (b) P ~ ,CdGeAsl,oPl,o, ~~, CdGeAsl,7sPo,2s.Experimental results are shown as symbols; solid curves represent simulated decays using calculated average Mu(31PIP) values. The compositional dependence of Mu(31P-31P) is plotted in the bottom part of this figure. Squares denote experimental values and filled diamonds denote calculated values (see text). The solid curve is a secondorder polynomial least-squares fit to the calculated values. Ill0
0.0
cl
0.5
Q
00
I 0
1
x
2
Figure 8. I W d spin-echo decays (top) for crystalline semiconductor alloys with composition CdGeAsz-xPx (representative samples). The compositional dependence of M2d(l13Cd-113Cd)is plotted in the bottom part of this figure. Squares denote the range of experimental values for each compound and filled squares denote calculated values (see text).
to which non-Gaussian behavior is observed. Whichever value is used, however, it is clear from Table I that there are significant discrepancies between the experiment and the calculation based on direct 113Cd-113Cd dipolar interactions only. These results are consistent with previous l13Cd spin-echo decay data obtained by us on crystalline and glassy cadmium arsenides and c d G e A ~ ~ .These * ~ discrepancies will be discussed in more detail below. 113cb.31P Spin-ECho Double-ReSOanneeNMR in C ~ C ~ A S , - ~ P ~ Alloys. As previously discussed, the spin-echo double-resonance (SEDOR) experiment, developed by Slichter and ~ e w o r k e r s , 2 ~ ~ ' provides an opportunity to measure the strength of heterodipolar interactions. In our present application, we conduct a II3Cd spin-echo experiment with simultaneous 180° pulse irradiation of the 31P resonance at the time of the *I3Cdrefocusing pulse. In principle, the II3Cd SEDOR decay is then governed by the combined effect of 113Cd-113Cd and 113Cd-31Pinteractions: z(2tl)/zO =
exp((2tl)2M2Cd-Cd/2) eX&(2h)*M2Cd-P/2~ (4)
2 g
As will be illustrated later with experimental results, the h f 2 d 10 relating to the 113Cd-113Cd interaction is usually not predictable from eq 3. Thus, when the determination of the 113Cd-31Pinteraction is of interest, it is more practical to analyze and simulate the experimental SEDOR decay data according to 0 'I I 1(2tl)/IO = F(2tl)/FOC eXp((2tl)2M2Cd-P/2) (5) 0 2 ' x where F(2tl)/F0 is the Il3Cd spin-echo decay function measured Figure 7. 31Pspin-echo decays (top) for crystalline semiconductor alloys in the absence of the 31Ppulse. with compositions (a) Z ~ G ~ A S , , ~ ~ P (b) , , ~ ~Z, ~ G ~ A S , , ~and ~ P(c) ~,~~, It is important to realize that SEDOR results can be seriously ZnGeAs,,75Po,2S. Experimental results are shown as symbols; solid curves affected by systematic experimental errors, Le., if the 180° (31P) represent simulated decays using calculated average values of M&lPpulse is imperfect and/or applied off-resonance. In such cases, 31P). The compositional dependence of M2d(31P-31P)is plotted in the only a fraction of 31Pspins are flipped and the Il3Cd spin-echo bottom part of this figure. Squares denote experimental values and filled 7-
Y
diamonds denote calculated values (see text). The solid curve is a second-order polynomial least-squares fit to the calculated data.
curve, Table I lists two M2d(113Cd-113Cd) values, obtained from linear least-squares fits to the decay at short evolution times and to the decay over the entire set of evolution times, respectively. The difference between these two numbers characterizes the degree
(24) Franke, D. R.;Eckert, H. J . Phys. Chem. 1991, 95, 331. (25) Shore, S.E.; Ansermet, J. P.; Slichter, C. P.; Sinfelt, J. H. Phys. Rev. Lett. 1987, 58, 953. ( 2 6 ) Wang, P. K.; Slichter, C. P.; Sinfelt, J. H. Phys. Rev. Lett. 1984, 53, 82. (27) Makowka, C. D.; Slichter, C. P.; Sinfelt, J. H. Phys. Rev. Lett. 1982, 49, 319.
Atomic Distribution in 11-IV-V2 Semiconductor Alloys decays more slowly with 2tl than expected. This situation has been discussed by Slichter and co-workers.2s,26To account for this experimental artifact, eq 5 then has to be replaced by
The Journal of Physical Chemistry, Vol. 96, No. 4, 1992 1911
A
1/10
IWI)/IO = F(2t,)/FoK1 - a ) -t aC exP-K2t1)2M2Cd-P/2H (6) where n (0 Ia I 1 ) denotes the degree to which the 31P magnetization is inverted by the 180' pulse; it characterizes the fraction of 31Pspins actually flipped by the pulse. For very broad lines and in the absence of spin diffusion, the inversion is incomplete for those nuclei contributing to the off-resonance edges of the wide-line N M R spectrum, resulting in a nonunity value for a.
Theoretically, the 113Cd-31Pheterodipolar second moment can be calculated according to M2Cd-P = ( 4 / 1 5 ) ( ~ 0 / 4 ~ ) ~ + I ( Il)yCd2yp2h2N-1CdCd-p-6(7) where I = and all the symbols have the same meanings as in eq 3, but now the Cd-P distance is of crucial importance. Again, eq 7 neglects contributions to the 113Cd-31Psecond moment from indirect spinspin couplings. Parts A-C of Figure 9 show typical results illustrating the expected increase of the 113Cd-31P dipolar coupling strength with increasing x. In all cases, application of the 31P180' pulse during the dipolar evolution time of the Il3Cd spin-echo experiment results in the expected drastically accelerated decay of the I I 3 C d spin echo. Application of eq 2 to the SEDOR data of Figure 9A-C yields average second moments characterizing the strength of the heteronuclear 113Cd-31Pinteractions. These values are significantly smaller, however, than expected from calculations based on a random distribution of P and As atoms over the anionic sites of the chalcopyrite structure. It appears that the 31P-113Cddipolar interaction in the alloys is significantly weaker as compared to a statistical distribution of P and As over the anionic sites. This conclusion remains even if one allows the possibility that a sizable fraction (20%) of the 31Pspins are not inverted due to pulse imperfections. At low phosphorus concentrations ( x = 0.25, O S ) , a bimodal distribution of SEDOR decay rates is clearly evident. We attribute the fast decaying component to phosphorus-bonded Cd atoms, and the very slowly decaying component evident at dipolar evolution times 2t, > 1 ms to Cd atoms that are coordinated exclusively to arsenic atoms (CdAs, sites). The decay rate observed for these cadmium sites is essentially the same as that observed for all the Cd atoms in the absence of the 31Ppulse.
Discussion Charge Delocalization Effects in Crystalline Zn(Cd)GeAs2-,P, Alloys. The Inabmty of Chemical Shift Spectroscopy To Provide Information on Local Environments. The Ik3CdMAS-NMR spectra reveal that there is no site resolution for individual cadmium nearest neighbor configurations. This observation is surprising in view of the large chemical shift difference between pure CdGeP2 and pure CdGeAs2 (87 ppm). One might argue that the chemical shift differences between the individual C ~ A S ~ - sites ~P, still might not be large enough compared to the line width to resolve separate peaks. To test this possibility, Figure 10 shows simulated spectra, calculated via eq 1 for a statistical distribution of these sites. These simulations are based on two assumptions made based on the experimental results obtained here: (1) In accord with the linear Il3Cd chemical shift dependence on x , it is assumed that the chemical shifts for the individual CdAs4-,P, sites can be interpolated from the values of the two end members CdGeP, and CdGeAsp. (2) The line width for each site is assumed to be identical with the experimental line width for that CdGeAs2-,P, sample for which eq 1 predicts this site to be most prevalent (excluding the pure end members). Accordingly, the line widths used in these simulations are 18.9 ppm for the sites with x = 0 and 4, 29.7 ppm for the sites with x = 1 and 3, and 31.5 ppm for the site with x = 2. The comparison between experiments and simulations shows clearly that the expected site resolution is not seen in the spectra and that the simulated spectra
1/10
, B
a
* O
0.4
1
k:
0.2
00 0.0
1
0.5
1 .o
1.5
2.0
2t1 ( m s )
04] 0.2
0.0 0.0
\ e 0.5
1 .o
I
1.5
2.0
2 t 1 [ms] Figure 9. 113Cd-31P SEDOR results on crystalline (A) CdGeAsl,sPo,s, (B) CdGeAsl,&,o,and (C) CdGeAs,,5PI,s. Part a of each plot shows the I1'Cd spin-echo heights in the absence of the 31P180' pulse, together with a polynomial fit to the data. Part b shows the II3Cd spin-echo heights with the 31P 180' pulse present. The solid curve corresponds to the simulated SEDOR decay, assuming a random distribution of P and As
on the anionic sublattice in the chalcopyritestructure and includes also the polynomial fit to the data in part a. In panels A and B are shown experimental SEDOR results for two independent data sets (filled and open circles, respectively). are also much wider than what is observed experimentally. These results suggest very strongly that the chemical shift in these systems is not simply dominated by local coordination spheres. Rather, it appears that there is extensive charge delocalization, resulting in an average shielding constant for all of the distinct cadmium sites present. The same conclusion must be drawn from the MAS-NMR shifts, whose compositional trend probably reflects the change in the band gap. More evidence for the nonlocal contributions to the chemical shifts in these compounds comes from Figure 11, which compara the Fourier-transformed l13Cd spin-echo spectra of CdGeAsl,5Po.5 and CdGeAsl,$l,o with the corresponding SEDOR spectra. The evolution times in both experiments are identical, corresponding
1912 The Journal of Physical Chemistry, Vol. 96, No. 4, 1992
Franke et al.
A
400
. 300
.
2 50
200
150
100
300
200
100
0
-100
PPM
PPM
Spin Echo
SEDOR 400
300
200
100
0
PPM
Figure 11. I13Cd spin-echo and 113Cd-31PSEDOR Fourier transforms for crystalline C ~ G ~ A S (top) ~ , ~ and P ~ CdGeAsl,oP,,o ,~ (bottom). The experiments were carried out with long evolution times (2000 and 800 ps for the two samples, respectively) to emphasize the contributions from As-only bonded Cd atoms to the SEDOR spectra. Note that for a given sample the spectra are essentially identical.
1 ' _ " ' 1 " " 1 " " I " " I ' 250 200
300
300
250
200
150
100
PPM
150
100
PPM
Figure 10. Comparison of experimental "'Cd MAS-NMR spectra (top in each panel) and spectra simulated on the basis of site distributions according to eq 1, using the assumptions made in the text: (A) CdGeA S , , ~ P ~(B) , ~ CdGeAsl,oPl,o, , (C) C~G~AS,,~~P~.,~. Note that the experimental and the simulated spectra do not agree.
to 2.0 and 0.8 ms for CdGeAsl,5Po,5and CdGeAsI,oPl,o,respectively. While the spin-echo spectra reflect a superposition of the resonances arising from the various CdAs,_,P, sites, the SEDOR spectra under the conditions shown only detect the CdAsl sites. (For this reason, the signal to noise ratio in the SEDOR spectra is appreciably lower.) Most notably, there is essentially no chemical shift difference between both spectra. Again, this finding is nicely consistent with the MAS-NMR results, emphasizing that
in the CdGeAs2_,P, system the chemical shift is a nonlocal property. This behavior contrasts sharply with the previously mentioned results on the systems Zn,Cdl-,Te and Hg,Cd,,Te, for which the chemical shifts appear to be dominated by the local ligand environments. It also contrasts with the 31PMAS-NMR spectrum of pure crystalline ZnSnP2, which shows separate peaks for the 3'P resonance of the main P coordination environment (PSn,Zn,) and for the 31Presonance of disordering defects (PSn3Zn and PSnZn3).2s Although a number of 31PMAS-NMR spectra of crystalline phosphides have been reported in the literat~re,~*"O the structural significance of these data has remained an open question. The results of the present study suggest that such difficulties may arise from nonlocal chemical shift contributions. Thus, unlike the situation with phosphates, it may not be possible for phosphides to develop satisfactory correlations of 3'P chemical shifts with local geometric parameters. From the above discussion it is clear that due to charge delocalization effects no chemical shift discrimination of local cadmium sites can be attained in the present materials. Consequently, the '13CdMAS-NMR spectra are unable to provide detailed structural insights about possible atomic distributions in these alloys, and one has to turn to the measurement of internuclear dipoledipole couplings for such information. Ability of Dipolar Spectroscopy To Provide Information about Local Environments in Zn(Cd)GeAsz_,P,, The success of the experimental dipolar spin-echo approaches used here for the deduction of realistic atomic distribution models for these materials rests with two important questions: (a) Do the simple spin-echo and SEDOR sequences produce sufficiently accurate dipolar second moments in these systems and (b) can such experimental second moments be attributed solely to direct dipoledipole couplings, calculable from eq 3? The first question concerns the valid time domain during which the refocusing of chemical shifts (28) Ryan, M. A.; Peterson, M. W.; Williamson, D. L.; Frey, J. S.;Maciel, G. E.; Parkinson, B. A. J. Mater. Res. 1987, 2, 528. (29) Nissan, R.A.; Vanderah, T. A. J. Phys. Chem. Solids 1989,50,341. (30) Vanderah, T. A.; Nissan,R.A. J . Phys. Chem. Solids 1988.49, 1335.
Atomic Distribution in IT-IV-V2 Semiconductor Alloys can be considered complete so that the decay is entirely due to dipolar interactions. Previous experiments on crystalline model compounds have shown that the first question is of no concern as long as the frequency dispersion due to dipole-dipole couplings is small compared to the chemical shift dispersion.22 This is certainly the case for the materials under study here, except for pure CdGeP, and ZnGeP, (see below). The second question concerns possible contributions to the )IP and II3Cd spin-echo decays from various different sources: (a) indirect 31P-31P, 113Cd-113Cd, or 3'P-113Cdinteractions, (b) rapid fluctuations of the z components of the 31Pspins due to 31P-31P flipflop processes, and (c) rapid fluctuations of the z components of the 75Asspins due to effective spin-lattice interactions. To address the second issue, we will at first discuss the spin dynamics in these materials as inferred from the experiments reported here. Following this discussion, we will offer possible structural conclusions. Spin Dynamics in Crystalline Zn(Cd)CeAs,-,P, Alloys. Let us at first discuss why the Il3Cd spin-echo decays in the cadmium-containing samples deviate so strongly from the calculated behavior according to eq 3. Besides the possibility that a strong indirect component to the I3Cd-Il3Cd spin-spin coupling could contribute to this deviation, it is also important to realize that the heterodipolar 113Cd-75Asand 113Cd-31Pinteractions may not be refocused completely and may contribute to the spin-echo decay. In a spin-echo experiment, the dipolar couplings between the observed nuclei I and the nonresonant nuclei S are refocused only if the eigenstates (S,) of these nonresonant spins remain constant during the range of dipolar evolution times 2t1 studied. If the nonresonant spins are strongly coupled to each other and/or fluctuate rapidly with a characteristic rate T2,s, the transverse magnetization of the observed I spins will be refocused incompletely. In such cases, the dipolar spin-echo decay can be approximated by 1)
/ I o = e x p ((2t I )2M2d/ 2) exPid21
The factor exp(r$2] in eq 8a then reflects the heteronuclear contribution to the spin-echo decay. A phenomenological expression for this effect has been given by Reimer and Duncan:I5
r$2 = 2(T2,s/ T2,1s)2f(2tl/ 7-2,s) + (4 exP-t, / T2.s) (exp2tdT2,S) - 31 ( W ( Tz,Is)2(= 1/M2,1s) characterizes the strength of the heteronuclear I-S dipoldipole interactions. Thus, for the present system, the I13Cd spin-echo decays may be influenced by 75As spin-state fluctuations via the 1'3Cd-75Asdipoledipole coupling and by 31Pspin-state fluctuations via the 113Cd-31Pdipole-dipole coupling. Application of eqs 8a,b to this situation yields
Z(2tl)/Zo(113Cd)= e x p i (2t I ) 'M2Cd-Cd /21
exp(d2Cd-As1
exp(r$2Cd-€') (9)
If we neglect indirect dipolar contributions and assume that the spin-echo decay in excess of the decay due to the 113Cd-1'3Cd dipole-dipole coupling is entirely due to these heterodipolar contributions, the experimental data can be used to determine the appropriate value of T2,s. An example for such a fit is shown in Figure 12 for CdGeAsl,oPl.o.In carrying out such fits, we use average calculated values of T 2 , ~ & d and T2,p-cdbased on an assumed random distribution of P and As over the anionic sites of the chalcopyrite structure. Furthermore, for calculating T,a modified version of eq 3 has to be used, which takes into account that the 75Asnuclei are in non-Zeeman states.jl Table I summarizes 75Asand 31Pspin fluctuation rates T2s(75As)and T2s(31P), extracted from the experimental Il3Cd spin-echo data with the (31) As previously shown (VanderHart, D. L.; Gutowsky, H. S.;Farrar, T. C.; J. Am. Chem. SOC.1967,89,5056. Sheldrick, G. M.J . Chem. SOC., Chem. Commun. 1967, 751), this correction can be made universally by multiplying the van Vleck value with a scaling factor that depends on the ratio of Zeeman and quadrupolar interaction energies. In the absence of experimental NQR data on analogous compounds, we assume here that the scaling factor in our case corresponds to the asymptotic value (1.76 for I = 3/2) for dominant quadrupole interactions.
The Journal of Physical Chemistry, Vol. 96, No. 4, 1992 1913 1/10
-
0.8
0.6-
0.20.4
0.04 0
'
1
.
I
2
.
I
*
I
'
#
'
I
4 ztltms] 6
'
I
' I 8
Analysis of the 'I3Cd spin-echo data in crystalline CdGeAs,,$,,,,: dashed curve, calculated spin-echo decay, assuming only direct 113Cd-113Cd dipolar interactions; solid curve, fit of the experimental data to eq 8a,b. T2,s(75As)= 0.005 ms. Figure 12.
help of these equations. The analysis assumes further that at most compositions (x S 1.25) the 75A~-113Cd dipolar term in eq 9 dominates, whereas at the highest P contents ( x = 1.5 and 1.75), the 31P-113Cdterm dominates. These assumptions are validated by the compitional dependence of the II3Cd spin-echo decay rate. The T2,s(75As)values extracted for samples in the arsenic-rich region are determined by the 75Asspin fluctuations caused by rapid nuclear electric quadrupolar spin-lattice relaxation. For this reason, little compositional dependence is expected (or seen). In contrast, the Ts(~IP) values are the result of 31P-31P spin flipflops whose rate is expected to increase with increasing 31P-31Pdipolar coupling strength and, hence, phosphorus contents. The comparison of Tzs(31P) for the samples with x = 1.5 and 1.75 reveals that this expectation is borne out in the experiments. While these estimates of T2,s(31P)and T2,s(75As)do not bear structural information by themselves, they become important in the interpretation of the 31Pspin-echo results. In all of the samples under study, the experimentally determined 31Psecond moments are about 20-30% higher than expected on the basis of eq 3 for a random distribution of P atoms over the anionic sites of the chalcopyrite lattice. In elucidating the reasons for these systematic deviations, an inspection of the compositional dependence is helpful. The largest deviations are observed for very high and very low x values. For the samples with x = 1.5 and 1.75, the deviation is most likely attributed to the partly homogeneous character of the 31P-31Pdipole coupling (i.e., 31Pspin fluctuations due to dipolar couplings), as also indicated by the II3Cd spin-echo data. All of the other effects are expected to be small. For x I 0.75, the discrepancy between calculated and predicted behavior becomes increasingly large. In this compositional regime, we have to be especially concemed with the effect of the quadrupolar 75As nuclei. As discussed above, the eigenstates of this isotope could easily fluctuate due to spin-lattice relaxation via nuclear electric quadrupolar couplings. If this is the case, the 31Pspin-echo decays need to be analyzed according to the expression instead of expression 2. The 31P-75A~ dipolar contribution to the 31Pspin-echo decay can be estimated from eqs 8a,b, using Tz,ls(75As-J1P) = {Mzd(75As21P)~'/2, as calculated from the known internuclear distances in the chalcopyrite structure, assuming a random distribution of P and As over the anionic sites. When the value of T2,~(7sAs) extracted from the previously discussed II3Cd spin-echo decay data is inserted into eqs 8a,b, the contribution of the 75Asspin flips to the 31Pspin-echo decay can be calculated with eq 10. Figure 13 is representative for the results obtained here. It illustrates that fits to the experimental )lP spin-echo data can be improved when this effect is taken into account. However, the effect is relatively small and cannot account completely for the systematically increased kf2d values. In fact, the correction of the simulated 31Pspin-echo decay for the 75As spin-flip effects shown in this figure is the maximum possible correction. As previously stated, the above analysis assumes that the discrepancy between the experimental 'I3Cd spin-echo decays
1914 The Journal of Physical Chemistry, Vol. 96, No. 4, 1992
Franke et al.
1/10
0.201
'\ 0
o . o o l . . I . . , . . I . . , . . , 0.00
0.20
0.40
0.60
0.80
1.00
211 (ms)
Figure 13. Analysis of the spin-echo data in crystalline CdGeAsl,oPI,o: solid curve, calculated spin-echo decay assuming only direct "P-"P dipolar interactions; dashed curve, calculated spin-echo decay, corrected for the effect of 75Asspin fluctuations. The T2(75As) value used for this correction was taken from the fit obtained in Figure
and the ones simulated via eq 2 is exclusively due to the effect of 75Asspin flips. If we drop this assumption by considering the Random Distribution With preferential Cd fact that indirect I13Cd-113Cdinteractions contribute to the lI3Cd Bonding: A S C d , P-Ge spin-echo decays as well, the 75Asspin flips would have an even @ =e 0 As lesser influence on the 31Pspin-echo decays. Thus, the Il3Cd spin-echo decays, in conjunction with the accompanying simulaQ P tions, suggest that the 75Asspins fluctuate so rapidly on the time scale of the 31Pspin-echo experiments, that their influence on the 31Pspin echoes is weak. We conclude, therefore, that the systematically increased apparent values of M2d(31P-31P)as extracted from the experimental spin-echo decays reflect a 31P-31Pdipole-dipole coupling that is consistently stronger than would be found for a random distribution of P atoms over the anionic sites in the chalcopyrite structure. Spin Dynamics in Crystalline Zn(Cd)GeP, and Zn(Cd)CeAs,. Note that the experimental 31PMld values reported for CdGeP2 and ZnGeP2 deviate from the values calculated with eq 3. The P atoms in both compounds are all chemically equivalent, and the chemically shift anisotropy is relatively small. This situation increases the likelihood that neighboring phosphorus atoms are effectively isochromatic a t many crystal orientations. If this is so, the flip-flop term in the Hamiltonian representing the 31P-31P dipoldipole coupling is not quenched and therefore the refocusing of chemical shift evolution during 2 t , is incomplete. For more quantitative knowledge on this subject, single-crystal NMR data are necessary in order to localize the 31Pshielding tensor for neighboring P atoms relative to the crystal axes, and the orientational statistics in the magnetic field need to be considered. The strong 31P-31Pinteraction, communicated via 113Cd-31Pdipolar couplings, is also responsible for the fact that no well-defined Il3Cd spin echo is observed in crystalline CdGeP,. In contrast t o t h e end members, all of the phases containing both P and As produce well-defined spin echoes, because the site disordering generates a wide chemical shift distribution which dominates the rate of Random Distribution With No Preferential Cd the free induction decay. Bonding Atomic Distribution Models for Crystalline Z ~ ( C ~ ) G ~ A S , - ~ P , Alloys. Both the findings from 31Pspin-echo and from 31P-113Cd SEDOR NMR indicate that the atomic distribution in these alloys W ' deviates from simple P, As scrambling on the anionic sublattice Figure 14. Atomic distribution model of crystalline Cd(Zn)GeAs2,P, of the chalcopyrite structure. it is worth noting that alloys qualitatively consistent with all of the experimental results reported the results presented here Cannot Simply be explain4 in terms here. This model assumes both cation and anion disordering, with a of a partial segregation into more phosphorus-rich and more preference for P-&. and Cd-As bond formation as compared to the arsenic-rich microphases or Cluster domains O n Smaller (10 A) formation of p q d and GeAs bonds. Shown for comparison is a model length scales. While such clustering and/or partial phase sepaof crystalline CdGeAsl.oP,,oin which only the anionic sublattice is disration would result in stronger 31P-3'Pdipole-dipole couplings ordered.
J. Phys. Chem. 1992, 96, 1915-1921
(as observed), it cannot account for the experimentally observed SEDOR decays. Partially segregated samples would result for all compositions in bimodal 31P-113CdSEDOR decays, the faster decaying component of which should reveal 31P-113Cddipoledipole couplings that are stronger than the calculated average. This kind of behavior is not a t all observed in the experiments. An alternative possibility, which is qualitatively consistent with all of the findings here, is shown in Figure 14. This model invokes disorder on both the anion and the cation sublattices. It is then assumed that the phosphorus atoms are, on average, coordinated to more than two germanium atoms and less than two cadmium atoms, whereas the As atoms, conversely, prefer cadmium over germanium coordination. While this tendency could be quantified in principle, we will refrain from doing so here, in view of the simplifying inherent assumptions made in the analysis. Specifically, the contributions from indirect 31P-113Cdspin-spin interactions have been neglected in the SEDOR analysis. Previous 203*205Tl NMR work has shown that, in covalently bonded systems involving isotopes heavier than 13Cdnuclei, the strength of the indirect interactions can exceed that of the direct dipolar couling.^^ Justification for our 113Cd-31P SEDOR approach comes from the experimental observation that this method produces the correct 113Cd-31Pdipolar second moment in crystalline CdP2.24 Furthermore, the isotropic IJ(31P-1I3Cd) value as extracted from the 31PMAS-NMR spectra of the corresponding isotopomers is only 330 Hz. Magnitude and orientation of the anisotropic component of the indirect 31P-113Cd interaction are currently not known. Such knowledge requires the measurement of singlecrystal rotation patterns, which are planned for the future. (32) Bloembergen, N.; Rowland, T. J. Phys. Reu. 1955, 97, 1679.
1915
Conclusions The atomic distribution in the 11-IV-V2 semiconductor alloys Z I I G ~ A ~ ~ and - ~ PCdGeAs2-,Px , has been studied by X-ray powder diffraction and l13Cd and 31PMAS and spin-echo and 31P-113Cd spin-echo double-resonance NMR. In contrast to the results obtained for 11-VI semiconductors, the MAS-NMR spectra of the present systems show a monotonic chemical shift trend and do not permit resolution of individual peaks for specific nearest neighbor environments. Site-selective 13Cd-31Pspin-echo double-resonance experiments confirm that the chemical shifts in these samples are dominated by nonlocal contributions. The 31Pspin echoes decay somewhat more rapidly than calculated from the crystal structure with randomly occupied sites in the anionic sublattice. This result cannot solely be attributed to un-refocusable 75As31P interactions because the I13Cd spin-echo decays, in conjunction with accompanying simulations, suggest that the 75Asspins fluctuate so rapidly on the time scale of the 31Pspin-echo experiments, that their influence on the 31Pspin echoes is too weak. Conversely, the 31P-113CdSEDOR results in the CdGeAs,P, alloys suggest that the 31P-113Cdinteractions are substantially weaker than calculated from the above model. A possible explanation for these results is a structure involving both cation and anion disorder with a stronger tendency of P-Ge and Cd-As bond formation as compared to P-Cd and As-Ge bond formation. Acknowledgment. Thanks are due to David A. Lathrop and Francis A. Dempsey for assistance with computer program development. Financial support of this research by N S F Grant DMR-89-13738 is gratefully acknowledged. Acknowledgment is also made to the donors of the Petroleum Research Fund, administered by the American Chemical Society.
Molecular Dynamics Simulations of the Conformational Dynamics of Tryptophan H. L. Gordon,*'+ H. C. Jarrell, A. G. Szabo, K. J. Willis, and R. L. Somorjai Institute for Biological Sciences, National Research Council of Canada, Ottawa, Ontario, Canada Kl A OR6 (Received: July 3, 1991; In Final Form: October 22, 1991)
Molecular dynamics simulations of a single tryptophan molecule were performed using CHARMm-parametrized potential functions, where both explicit molecular and dielectric continuum models were used for the solvent. When all hydrogens were modeled explicitly, rotations about the x2 dihedral were more frequent than about the x1dihedral. The presence of a molecular solvent damped librational motion about both dihedral angles. Conversely, when a united atom representation for hydrogen was used, rotations about the xIdihedral were more frequent than about the x2 dihedral. We show that conflicting rotamer models for the biexponential fluorescence of tryptophan zwitterion can be supported, depending on the model for hydrogen representation in the simulation. We also show that the predicted relative populations of tryptophan rotamers are not consistent with the available experimental data. Simulators are thus warned not to impute universal applicability to empirical potential energy functions found in biomacromolecular molecular mechanics packages.
Introduction A great deal of effort has gone into the development of computational methodologies for molecules of biological interest, such as proteins. Well-known packages, such as CHARMM,] ECEPP,2-4 AMBER,S,6and GROMOS,' have been developed specifically for the computational investigation of biomacromolecules. All of these programs rely on the use of empirical intermolecular potential energy functions. These have been parametrized to reproduce experimental results of selective structural and thermodynamic properties of small molecules and subsequently adjusted for the description of macromolecules. These potential functions and parameters have been designed with the intent that they be widely transferable, so that the energy of any arbitrary *To whom correspondence should be addressed. Canadian Government Laboratory Visiting Fellow, 1990-1991.
protein, for example, can be described given the atom types, their connectivities, and their coordinates. However, it would be unreasonable to expect that these empirically determined potentials (1) Brooks, B. R.; Bruccoleri, R. E.; Olafson, B. D.; States, D. J.; Swaminathan, S.; Karplus, M. J . Compur. Chem. 1983, 4 , 187-217. (2) Momany, F. A.; McGuire, R. F.; Burgess, A. W.; Scheraga, H. A. J .
Phys. Chem. 1975, 79, 2361-2381. (3) Dunfield, L. G.; Burgess, A. W.; Scheraga, H. A. J . Phys. Chem. 1978, 82, 2609-26 16. (4) VBsquez, M.; NCmethy, G.; Scheraga, H. A. Macromolecules 1983, 16, 1043-1049.
( 5 ) Weiner, S. J.; Kollman, P. A.; Case, D. A,; Singh, U. C.; Ghio, C.; Alagona, G.; Profeta, S., Jr.; Weiner, P. J . Am. Chem. Soc. 1984, 106, 765-784. (6) Weiner, S. J.; Kollman, P. A.; Nguyen, D. T.; Case, D. A. J . Comput. Chem. 1986, 7 , 23C-252. (7) van Gunsteren, W. F.; Berendsen, H. J. C. Groningen Molecular Simulation (GROMOS), University of Groningen, Groningen, The Netherlands.
0022-365419212096-1915%03.00/0 0 1992 American Chemical Society
