New technique for determining the diffusion mechanism by NMR: application to Cl35 diffusion in TlCl

A major problem in the study of atomic motions is the determination of the dominant mechanism responsible for translational diffusion. Recently Ailion and Ho predicted that the rotating-frame spin-lattice relaxation time T l p would have an angular dependence which depends on the diffusion mechanism in the ultraslow-motion region. In this paper we have extended these ideas to T1C1 for which the dominant mechanism is known by other experiments to be CI vacancy diffusion. We observed experimentally an angular dependence consistent with the dominant mechanism being CI vacancy diffusion, thereby corroborating the basic ideas of Ailion and Ho. We have also been able to eliminate CI interstitialcy diffusion as a possible dominant mechanism. We measured an activation energy of (0o 733 ± 0 . 012) eV, in agreement with results of other experiments.

2 4 8 8 E . V O N M E E R W A L L A N D T . J . R O W L A N D 5 14D. O. Van Ostenburg, H. D. Trapp, and D. J. Lam, Phys. Rev. 126, 938 (1962). 15D. O. Van Ostenburg, J. J. Spokas, and D. J. Lam, Phys. Rev. 139, 713 (1965). 16M. Bernasson, P. Descouts, P. Donz6, and A. Treyvaud, J. Phys. Chem. Solids 3£, 2453 (1969). 17D. O. Van Ostenburg, D. J. Lam, H. D. Trapp, D. W. Pracht, and T. J. Rowland, Phys. Rev. 135, 455 (1964). 18A. J. Maeland, J. Phys. Chem, 68_, 2197 (1964). 19E. von Meerwall and D. S. Schreiber, Phys. Letters 27, 574 (1968). L. E. Drain, Proceedings of the XIII Colloque Ampere Amsterdam, 1964 (North-HoHand, Amsterdam, 1965), p. 181. 21E. von Meerwall and T. J. Rowland, Solid State Commun. 9, 305 (1971). 22G. Horz, Z. Metallk. 61, 371 (1970); 60, 50 (1969). 23R. H. Geils, Rev. Sci. Instr. 42, 266 (1971). 24T. G. Berlincourt and R. R. Hake, Phys. Rev. 131, 140 (1963). 25For all work on transition-metal alloys in this study, the same specimen of Leico V has been used as intensity reference. 28B. T. Feld and W. E. Lamb, Jr., Phys. Rev. 67, 15 (1945); R. V. Pound, ibid. 79, 685 (1950). 27K. Kambe and J. F. Ollom, J. Phys. Soc. Japan 11, 50 (1956). 28W. B, Pearson, A Handbook of Lattice Spacings and Structures of Metals and Alloys (Pergamon, New York, 1958), Vol. I; ibid. (Pergamon, New York, 1967), Vol. II. 29W. Kohn and S. H. Vosko, Phys. Rev. 119, 912 (I960). 30R. E. Watson, A. C. Gossard, and Y. Yafet, Phys. Rev. 140, A375 (1965). 31R. M. Sternheimer, Phys. Rev. 96, 951 (1954). 32C. H. Cheng, C. T. Wei, and P. A. Beck, Phys. Rev. 120, 426 (1960). 33A. J. Freeman and R. E. Watson, in Treatise of Magnetism, edited by R. H. Suhl and G. T. Rado (Aca­demic, New York, 1965), Vol. IIA. 34A. M. Clogston, A. C. Gossard, V. Jaccarino, and Y. Yafet, Phys. Rev. Letters 9, 262 (1962). PHYSICAL REVIEW B VOLUME 5, NUMBER 7 1 APRIL 1972 New Technique for Determining the Diffusion Mechanism by NMR: Application to Cl35 Diffusion in TlCl^ Gary L. Samuelson* and David C. Ailion Department of Physics, University of Utah, Salt Lake City, Utah 84112 (Received 16 August 1971) A major problem in the study of atomic motions is the determination of the dominant mech­anism responsible for translational diffusion. Recently Ailion and Ho predicted that the ro- tating-frame spin-lattice relaxation time Tlp would have an angular dependence which depends on the diffusion mechanism in the ultraslow-motion region. In this paper we have extended these ideas to T1C1 for which the dominant mechanism is known by other experiments to be Cl vacancy diffusion. We observed experimentally an angular dependence consistent with the dominant mechanism being Cl vacancy diffusion, thereby corroborating the basic ideas of Ailion and Ho. We have also been able to eliminate Cl interstitialcy diffusion as a possible dominant mechanism. We measured an activation energy of (0o 733 ±0. 012) eV, in agree­ment with results of other experiments. I. INTRODUCTION In this paper we report the experimental verifica­tion of a new technique for determining diffusion mechanisms in solids.1 This technique was pro­posed theoretically by Ailion and Ho,2 who pre­dicted that in the ultras low-motion region3 the ro­tating- frame spin-lattice relaxation time Tlp would depend upon the angle which the external magnetic field Ho makes with respect to the crystal axes. In particular they predicted that this angular depen­dence could vary significantly with diffusion mech­anism and could, for example, be measurably dif­ferent for vacancy, interstitialcy, and interstitial diffusion. Direct comparison of the experimental data with the results of calculations could then be used to distinguish the dominant mechanism from alternate possibilities. We chose to perform our experimental study on a solid for which the diffusion mechanism has been determined to a high degree of probability by other techniques. In particular we studied T1C1 for which the dominant mechanism, as determined by Friauf, 4 is chlorine vacancy diffusion. Our results are in excellent agreement with the results of Friauf and strongly corroborate the ideas of Ailion and Ho. II. BACKGROUND A. Strong-Collision Theory The theoretical calculations for TXp in T1C1 uti­lize the strong-collision approach, originally de­veloped by Slichter and Ailion5 (SA). We now re­view the strong-collision theory and the conditions under which it is valid. The motivation for a strong-collision theory arises when we attempt to study the spin-lattice relaxation time in very weak Zeeman fields. Under conditions such that the ex­ternal magnetic field is less than or comparable to the local field, weak-collision6"8 theories which5 N E W T E C H N I Q U E F O R D E T E R M I N I N G T H E D I F F U S I O N . . 2 4 8 9 treat the fluctuating dipolar Hamiltonian as a per­turbation on the Zeeman system break down and must be replaced with a nonperturbation (strong- collision) theory. The strong-collision theory was originally developed in order to relate the experimentally measured spin-lattice relaxation time in the ultra- slow-motion region to the atomic-diffusion corre­lation time r. According to Bloembergen, Pur­cell, and Pound (BPP), 6 atomic diffusion will re­sult in a maximum in the spin-lattice relaxation rate (i. e., a Tx minimum) when r~ l/co0, where o;0 is the Larmor frequency. If we perform our experiment in weak applied field, the Tx minimum will correspond to a longer value of r occurring at lower temperatures. (In zero field, SA9 showed that the Tx minimum would occur at the onset of motional narrowing; i. e., at t~Tz. ) We thus see that an ability to study spin-lattice relaxation in weak fields is essential to the observation of "ultra- slow motions" (i. e., motions for which r» T2 which occur in the rigid lattice at temperatures below the onset of motional narrowing). A commonly used method for studying weak-field relaxation is to demagnetize the field adiabatically and then to remagnetize adiabatically. By waiting a variable time in the demagnetized state and then plotting the signal in the remagnetized state, we can determine the weak-field relaxation time.10 In order to satisfy easily the somewhat conflicting re­quirements that the demagnetization occur sufficient­ly slowly to satisfy the adiabatic condition11 and yet occur in a time short compared to the relaxation time, it is usually convenient to perform the demag­netization in the rotating frame.9,12,13 The theory of SA is based on two assumptions. The first is that sufficient time elapses between dif­fusion jumps for the dipolar and Zeeman systems to come to thermal equilibrium between jumps. This assumption allows us to assign a temperature to the spin system prior to each jump. We can then as­sign to the spin system a density operator14 p = e-x/M/Z . (1) (If we perform our experiment in the rotating frame, we must replace by 3C°, the rotating-frame Ham­iltonian. 12) The use of this equilibrium density ma­trix enables us to calculate the expectation values of quantities like the magnetization and energy without first determining the wave functions, since expecta­tion values can be expressed as traces which are in­dependent of representation: (M >= Tr(pM) (2a) and systems to come to thermal equlibrium is given by the cross-relaxation time Tcr (which is approxi­mately equal to Tz for fields comparable to or less than the local field), and since the mean time an atom sits between jumps is the correlation time r, this assumption will be valid provided r» Tz. Thus, the strong-collision theory will complement the weak-collision theories in that the former applies in the rigid-lattice region on the low-temperature side of the weak-field Tx minimum, whereas the latter applies to the motionally narrow region above the minimum. The second assumption of the strong-collision theory is that the same equilibrium density matrix can be used to calculate the spin energy immediate­ly before and immediately after the atomic jump. This assumption is equivalent to the sudden approx­imation in that the effect of diffusion is to cause, from the point of view of the jumping nucleus, a sudden change in the dipolar Hamiltonian but not in the density matrix. In other words we are assum­ing that the time the nucleus spends in the actual process of jumping is sufficiently short that, im­mediately after a diffusion jump, the spin will have the same orientation as it had immediately before the jump. If this assumption were not satisfied be­cause the nucleus jumped so slowly that it had time to align itself along the new local field, the jumping would not result in a change in dipolar order and would not be observable by magnetic resonance. Using the sudden approximation we can calculate the energy change resulting from a jump: <AE>= (5C?>-<3C?)=Tr[p(3C"-3C?)] , (3) where 3C° and X° are the initial and final values of the spin Hamiltonian, respectively. B. Calculation of (T'j ^iff The considerations leading to an expression for the diffusion contribution to Tlp, have been described in detail elsewhere. 2,3,5,15 Accord­ingly, we shall outline them only briefly here. The original treatment of SA applied to the case where only dipolar interactions contribute to the energy change resulting from a jump. More re­cently, Rowland and Fradin15 (RF) and Wagner and Moran16 extended their treatment to the case where quadrupolar interactions are also present. For a rotating-frame Hamiltonian consisting of Zeeman, dipolar, and quadrupolar terms, we ob­tain tpO_ i %pO , rtP® (A\ UV- - (JV jg' -f- oVj jrj + 'J'-'Q • \ / Using the high-temperature approximation, the to­tal average spin energy can be calculated and writ­ten in the form (3C°) = Tr(p3C°) . (2b) Since the time required for the dipolar and Zeeman (3C°> (5)2 4 9 0 G . L . S A M U E L S O N A N D D . C . A I L I O N 5 HD is the dipolar contribution to the local field and Hq is the equivalent local field arising from the quadrupolar interactions. Heff is the rotating-frame Zeeman field and equals + (H0- o>/y)k; C/ is the Curie constant for / spins NIyIMI(I+ l)/3kB; and 0 is the prejump equilibrium temperature. Since (A£) is also proportional to 1/0, the fractional energy change (AE)/(3€°> is independent of the equilibrium temperature. From this consideration, we can now relate (Tlp)diff to the jump correlation time r. We let Nj be the total number of I spins each of which jumps on the average once in a time r. We can write the rate of energy change as 8<K*> d(W°D) d(<) d(X°z) dt dt + a t + dt r i ({*E)d o v <AE>0 ,™o fn ~\W£T( d}' (6) SA defined the dipolar-order parameter^ by letting Nl (A22 )D/(X°D) equal - 2(1 -pD) and RF used15 a similar definition for the quadrupolar-order pa­rameter: Ni(AE)q/(Kq)~ - (1 -pQ). We can thus think of pD and pQ as parameters which describe, respectively, the loss of dipolar and quadrupolar order resulting from a jump. It is easy to show3 that these considerations lead to the following re­laxation equation for the magnetization (M) in the rotating frame: 3(M) (Meq - (M )) dt i p (?) where Meq is the thermal equilibrium magnetization. The diffusion contribution (Tlp)diff can be shown3 to be given by (Tie) diff- 2(1 ■pD)HDz+{l-pQ)HQz (8) This formula reduces to the SA result5 when HQ = 0 III. THEORY A. Calculation of HD 2 In this section, we calculate the quantities HDz and (1 ~pD) in order to derive a theoretical expres sion for (Tlp)diff in T1C1 for chlorine diffusion. Toward the end of this section we discuss r, and (1 -Pq)Hq2 which have been experimentally de ter mined. Before going from Eq. (2b) to Eq. (8), we first define the dipolar local field HD as follows: V, TABLE I. T1C1 data. Isotope % abundance y/2ir MHz/104 G Spin Quadrupole moment Cl35 75.4 4. 172 3 2 - 0. 079 x 10"24 cm2 Cl37 24.6 3.472 3 2 - 0. 062 x lO'24 cm2 rp|203 29.52 24.33 * 0 *p|205 70.48 24.57 * 0 aThese data were obtained from the Varian NMR Table (fourth edition). Tr«)2 CtHDl (3C»)=- kez e (10) From this relation and the previous definition for (1 ~Pd)> we obtain Nl6(AE)D (1 -Pd) = (11) Table I lists isotopic data on T1C1 which show that both chlorine and thallium consist of two abun­dant isotopes. We thus have a four-spin system with two-spin species on each sublattice. The di­polar Hamiltonian 30^ can then be broken down into ten interactions. In order to describe these inter­actions, the following definitions are made: Spin numbers I and /' will refer to the Cl35 and Cl37 iso­topes, respectively, and S and S' to the TI205 and TI203 isotopes, respectively. Indices i and j refer to chlorine spins and s and t to the thallium spins. If N represents the total number of chlorine spins in our sample, we can define the fractions fj-Nj/N, fr = Nr/N, fs = NS/N, and fs> = Ns,/N, where Nj + Nr = NS + NS, = N. There are several different types of terms in the rotating-frame dipolar Hamiltonian. The like -like contribution is 3Cz>//= rj2^2 2a3 1 2 (12) where _(1 - Scos2©,,) ' ti (13) and where is the angle between the internuclear vector Tu and H0. The lattice parameter a has been factored out so that the internuclear distance r{j is in units of a. Equation (12) is a like-like spin interaction on the same sublattice. We also have two interaction terms, and oc5>ss», which involve interac­tions between unlike spins on the same sublattice, and have the form 0 \2 2 _ Tr(3C p) H D - CfiZ (9) a (14) itj where Z is the partition function. Using Eq. (2b), we then see that Finally we have a term describing interactions be­tween unlike spins on different sub lattices:5 N E W T E C H N I Q U E F O R D E T E R M I N I N G T H E D I F F U S I O N . . 2 4 9 1 o rfys & j) DIS- 3 a i,s C is^ iz&sz (15) Using Eq. (11) we can obtain the ten contributions to by performing standard trace calculations involving squares of the terms in the Hamiltonian. From3C/)j7, we obtain (16) In the above we have replaced the sum over j by a factor Nj representing the number of equivalent Cl35 sites. We introduce the factor fT in order for the index i to range over all N lattice sites rather than Hd2 = ±/V0, *W/(r+1) those belonging to I spins. We assume that the I spins are randomly distributed over the chlorine sublattice. Using similar arguments,17 we obtain Kw)= - ~r-±1 *rfr S C„* (17) and ,1(,0 v ^y/rs2/(/+l)S(S+i) „fTr 2 i^Dis)-------------------------QkBcP N,fslj • (18) Combining all ten contributions and grouping terms, we have 3f,y/ f/r, '+' fsyss(.s+1)' 77V/2(/+ if 4fsyszS(S+1) + 3fIyIiI(I+ 1) (‘ (‘♦w-x 4/s<yS'2 / + / ;¥-)]? c«: fs ,yS'* \j) Q 2 f7^)sCsi (19) In order to perform the lattice sums and £SCS/, we employed a procedure similar to that used in the Appendix of the paper by Ailion and Ho.2 The other data needed to evaluate Eq. (19) appear in Table I. If we take the lattice parameter a to be 3. 846 xlO"8 cm, we obtain for HD2, Hdz= (1. 666 - 0. 548 sin22<p) G1 (20) where the angle cp is measured between the crystal (100) axis and the static field H0. B. Calculation of (1 - pD) In the calculation of (1 - p D), we must derive ex­pressions for the initial and final Hamiltonians which describe the specific jump mechanism under study. Then, by using Eqs. (3), (11), and (20), we can ob­tain the value of (1 - p D) related to that particular jump mechanism.18 As we observed in Sec. II, an essential require­ment of the theory of SA is that the atoms jump suf­ficiently infrequently that r»T2 in order for the spin-temperature assumption to be satisfied prior to each jump. Since typically diffusion results from the motion of a small concentration of defects (such as vacancies or interstitials), we must consider two special cases. First we observe that the mean time rd that a defect spends between jumps is re­lated to the mean time r that an atom spends be­tween its jumps by means of the following relation: Ta = (Nd/N) 7 , (21) where Na/N represents the defect (e. g., vacancy or interstitial) concentration. We thus have the possibility that rd « T2 even though r» Tz, This means that the defect produces along its path a trail of neighboring spins which individually have not ap­proached semiequilibrium and are thus "hot. " Since the time between encounters19 is of order r, the ef­fect of the encounter is to cause a displacement of the spins in the trail. By calculating the energy change of the entire hot-spin trail and then dividing by the number of spins involved, we can obtain the average energy change per jumping spin.20 The second case occurs when rd» T2 (this is the case originally considered by SA). In this case every spin on the defect's path will attain equilibrium prior to the next jump of the defect. For typical low defect concentrations this second condition can be achieved only at low temperatures. Under typi­cal experimental conditions, rd will be less than Tz. Therefore we shall now calculate 1 - pD for the case Ta K< T2 « r for vacancy diffusion in T1C1. (We shall calculate the results for the case rd» Tz in the Ap­pendix). In calculating the energy change (AE)D resulting from the jumping of an I spin (i. e., Cl35), we must consider four kinds of terms: (AEn)D, (AEir)D, (AEIS)Dy and (&EIS')D‘ Evaluating Eq. (3) for the dipolar contribution to (&E)D for the II interaction, we obtain 1)2 6 kda 6 2 \ifJ c 2 ECiACu)')- i,j / (22) The factor (C,/ comes from 3C°DIIf and is the value of Ci} immediately after an encounter. We can define g and Mo be a subset of the indices i and j, respectively, which refers only to the jumping spins along the path of the vacancy. Since only the terms2 4 9 2 G . L . S A M U E L S O N A N D D . C . A I L I O N 5 for which either i or j corresponds to g or h can change as a result of a particular encounter, we can write these terms explicitly and cancel the rest. If we do so and combine equivalent terms involving i and j (which refer to the same terms), we obtain 6 k6a teg ' h g,h gh i*6 h g,h (23) ffi- (x © n j® cr © @ (a) (b) FIG, 1. Jump models describing diffusion in thallium chloride: (a) vacancy diffusion and (b) interstitialcy dif­fusion. If we assume the vacancy motion to be random, then any h spin can be treated as a typical spin. Then each of the h spins will on the average con­tribute equally to the sum. We can thus replace our sum over h by a factor n representing the num­ber of spins in the trail. We will replace the run­ning index h by a fixed index r which refers to a particular spin of interest; the g coordinates will then refer to other atoms on the trail. We can now replace the quantity in square brackets in Eq. (23) by j±g g itg g (24) We shall now replace the spin sum with a lattice sum letting i range over all AT sublattice sites. In order to account for the /' isotopes, we shall mul­tiply the sum over lattice sites by fu the fraction of the spins which are I spins. The restriction i can be removed by subtracting explicitly from the sum over all sites the terms for which i=g. If we wish to consider only jumps which contribute to the dipolar relaxation, we must account for the possibility that one out of six jumps in the T1C1 lat­tice will return a spin to its previous site. For Td « T2, this second jump effectively cancels the re­laxation contribution of the previous jump. In gen­eral, if there are G nearest neighbors to a vacancy, then the contributing jumps are reduced by the fac­tor (G - 2)/G. Out of the five remaining jump di­rections which do contribute (ignoring the less like­ly possibility that a higher-order path may return the spin to its original site18) there is a i probabil­ity that the vacancy will continue in the forward direction and a f probability that it will move at right angles relative to the previous jump. If r is the initial site of the jumping atom of interest and q is its final site, we can designate nearest-neigh­bor sites to r in the following way. The site in line with r and q is labeled I and the four equivalent sites at right angles to the line of r and q are labeled t. Corresponding sites located about q will be m and n, respectively (see Fig. 1). The vacancy movement will result in the atom originally at sites I or t being transferred to site r, the atom at r moving to site q, and the atom at q going to sites n or m. A reasonable approximation is to consider the summation over g as including only terms arising from atoms in a region around r and q which con­tains only sites t, I, ra, or n. Since the C*/s are proportional to we can neglect contributions from atoms on the "hot"-spin trail which are not nearest neighbors to r or q. We shall average over the four equivalent t and n sites. The IV interaction will have the same lattice sums as the II interaction since the /' spins occupy the same sublattice as the I spins. The difference is that the coefficient before the sum contains yv and fr as well as a slightly different numerical factor. The energy change per spin on the chlorine sublat­tice is then obtained by dividing by the factor n. The IS and IS' interactions will not involve the vacancy trail since the thallium atoms are fixed in position. The isotope concentrations are accounted for as before with the factors fs and fs» when going from a spin to a lattice sum. Evaluating (1 -pD) for the case rd«Tz from Eqs. (3), (11), and (20), we obtain (l + yJ2 " CqrZ - CirCiq- 2{^Gmr2‘+ |^nr2)+ 2( 5C lr Clq+ f-C tr ^ tq) c . 4 r c , 2S(5+ 1) (f s'Ks2 j fs'Ys* \(r.c - (, c;r Cra+rC* c„)j + -- + It C (25)5 N E W T E C H N I Q U E F O R D E T E R M I N I N G T H E D I F F U S I O N . . . 2 4 9 3 _T\_ Wr FM PULSE SHAPER ALTERNATOR FIELD-SHIFT CIRCUIT VARIAN "FIELDIAL" CONTROL FM XTAL OSCILLATOR POLARITY rf GATE & AMPLIFIER ; H, MODULATOR H, GATE PULSE REFERENCE AMPLIFIER S PHASE SHIFTER FABRITEK SIGNAL AVERAGER COLLINS rf POWER AMPLIFIER COUPLING NETWORK PHASE- SENSITIVE DETECTOR - BROAD-BAND AMPLIFIER _>------- AUDIO­FREQUENCY AMPLIFIER ELECTRONIC FILTER DEWAR ASSEMBLY J SIGNAL MONITOR SCOPE FIG. 2. Block diagram of pulse NMR apparatus. If we evaluate all lattice sums for T1C1 and use the parameters in Table I, we find the value of (1 -PD) in Eq. (25) to be n h \ °- Q255+ 1936sin22<? 1 Pd) 1.666-0. 548 sin22<p ' > IV. APPARATUS The apparatus21 used in gathering data on the angular dependence of (Tlp)diff is a cross-coil NMR spectrometer which provides a pulsed Hx field of up to 7 G at a frequency of 4 MHz over a 12-cm3 T1C1 sample. Figure 2 comprises a block diagram of the basic components utilized in the system. In order to measure Tlp it is necessary first to orient the magnetization parallel to Hx in the rotating frame in the "spin-locked" configuration. 22 This process causes the temperature of the spin system to be reduced to a value lower than the lattice tem­perature. Under these conditions jumping results in a decrease in the magnetization as the spin tem­perature rises towards the lattice temperature. Tlp is then measured by plotting the magnetization as a function of time in the spin-locked state. The magnetization, as measured by the amplitude of the free-induction decay following the turn-off of the rf pulse, will decrease with a time constant Tlp. A. Description of Block Diagram It is clear that, in order to measure Tlp, one must have a transmitter which can spin lock the magnetization and can also provide rf pulses which are long compared to Tlp (which may be several seconds). In our system spin locking is achieved by means of adiabatic demagnetization in the rotating frame (ADRF). However, in contrast to the meth­ods9' 13 which pulse the static field H0y we demag­netize the rotating-frame effective field by modulat­ing the rf frequency. 23 In order to obtain large shifts in He{i by frequency modulation, 16- and 20- MHz crystal oscillators are "pulled" off frequency in opposite directions using voltage-controlled capacitative loading. The basic 4-MHz Hx frequen­cy is then obtained by rf mixing. The modulating voltage is an upward step followed by a downward ramp. This ramp allows the system to be returned to exact resonance adiabatically so that the magne­tization will then be spin locked along H^ The cw output of the FM XTAL OSCILLATOR circuit is fed to a gating circuit similar to the one developed by Blume,24 and then to a modified Heath- kit 90-W amplifier. The Heathkit has been exten­sively modified17 to provide the capability of de­magnetizing Hx to zero (or to any level between zero and maximum) and of re magnetization back to the maximum. This second demagnetization provides a complete transfer of Zeeman order to the dipolar system. The rf output is further amplified by a Collins 30-S 1-kW linear amplifier before being ap­plied to the sample's transmitter coil. A convenient method has been devised for match­ing the transmitter coil (a cylindrical Helmholtz pair25) to the Collins output. The tuning and match­ing procedures have been "orthogonalized"17 by series tuning and parallel matching with a 77-cou­pling network. The coil's Q is controlled with an external noninductive series resistance which is placed outside the Dewar. By then using a coil of2 4 9 4 G . L . S A M U E L S O N A N D D . 0 . A I L I O N 5 FIG. 3. Relation of transmitter and receiver coils to sample. The sample is mounted in a rotatable holder for angular studies. fairly high Q, most of the heat energy is expended in the series resistance outside the Dewar thereby permitting more uniform temperature control. A self-switching solid-state damping circuit26 has been installed in the matching network to reduce the transmitter ringdown time and to attenuate shot noise between pulses. We see from Fig. 3 the arrangement of the trans­mitter and receiver coils. Ceramic forms are used for support and spacing. The receiver-coil assembly with sample can be removed by using a long stainless-steel tube extending down into the cryostat. An inner concentric tube is used to cross the receiver coil for minimum coupling with the transmitter, whereas a third is used to rotate the sample holder in order to do the angular studies. A calibrated dial attached to the inner tube is used to record the crystal orientation. The preamplifier for the receiver consists main­ly of a dual-gate FET cascode circuit.17 Provision is made to clamp the input to ground for a few jusec after the Hi pulse in order to damp the receiver, thereby improving the recovery. This becomes es­sential in light of another feature of the preamplifier; by using positive feedback we are able to multiply the effective value of the receiver Q to several hundred for improved signal to noise. In order to observe the free-induction-decay shape, the Q is reduced for several jusec following the rf pulse turnoff thereby improving the receiver recovery. The magnetization signal is amplified by a wide­band amplifier and is then phase-sensitive detected. High-frequency noise is attenuated by passing the signal through an active low-pass filter. In order to improve further total system recovery we in­stalled series FET gates before the broad-band amplifier and the low-pass filter for the purpose of blocking the Hi pulse. The filtered-signal data are then processed by a Fabri-Tek 1074 multichannel averager.27 The Fabri-Tek has been modified to subtract auto­matically the off-resonant signal from the resonant signal in an alternate manner so that only real resonance data is accumulated for multiple aver­aging. By this means we were able to eliminate nonresonant noise sources and offsets. This fea­ture has been extremely useful in attenuating the effect of a noise source thought to be an electro­mechanical acoustic oscillation in the brass sup­ports and shield near the transmitter coil. Our Varian 12-in. magnet Fieldial control has been modified to alternate the field between two levels separated by an adjustable amount, typically about 100 G. The repeat time is adjusted to be longer than the settling-time response of the mag­net controller to this step impulse. One of the levels is the on-resonance level, and it is synchro­nized with the Fabri-Tek polarity control so that a signal obtained during this time will be added to the accumulated data. The alternate magnetic field level is off resonance and is synchronized with the subtract mode of the Fabri-Tek so that everything but the resonance signal is removed from the mem­ory. B. Temperature-Control System The cryostat has been designed to allow continuous control of temperatures from 90 to better than 400 °K by connecting to the sample head a cold tank and heater arrangement. The heater is placed close to the sample and is wound on the neck of the rf head. The heat leak to the cold tank can be ad­justed by changing the number and composition of connecting rods connecting the sample head to the cold tank. The rods were composed of either cop­per or stainless steel depending on whether a strong or weak thermal coupling was desired. In order to achieve better thermal stability, the entire cryostat was mounted inside a Dewar. The sample temperature was monitored with a calibrated platinum-resistance thermometer placed in direct contact with the sample. In order to im­prove the temperature stability, we employed a Leeds-and-Northrup proportional controller in a feedback loop between the heater and the thermom­eter. We were then easily able to hold the temper­ature variations to within ±0. 2 °K at each tempera­ture. V. EXPERIMENTAL RESULTS We shall now describe the results of a number of experiments which we performed on T1C1. In the Tlp measurements, our technique is sim­ilar to that employed by SA except that, after adia­batically demagnetizing H0, we performed an adia-5 N E W T E C H N I Q U E F O R D E T E R M I N I N G T H E D I F F U S I O N . . 2 4 9 5 io3/e (°k_i) FIG. 4. Tlp vs reciprocal temperature. T2 vs recip­rocal temperature is also plotted and illustrates the tech­nique used to determine the value of t at the temperature corresponding to the onset of motional narrowing. The T2 data were taken in a powder. batic demagnetization of Hx. Our pulse sequence is as follows. At t- 0, we set H0 exactly to resonance with Hx = 0. We then turned on our Hx at a frequency different from the resonant value and then performed an adiabatic demagnetization28 in the rotating frame by letting the rf frequency (rather than the external field) slowly return to the resonant value. 23 Our off-resonance frequency corresponds to an effective field of approximately 50 G. We then adiabatically demagnetized Hx to a constant value, waited a vari­able time, and then increased Hx to its original value. We then turned off Ht sharply and, by ob­serving the amplitude of the free-induction decay as a function of time in the demagnetized state, were able to measure Tlp. A. Temperature Dependence of Figure 4 shows a plot of the temperature depen­dence of Tlp in the ultras low-motion region for a single crystal29 of T1C1 along with a plot of Tz for a pressed powder sample of T1C1. Figure 5 shows the temperature dependence of (Tlp)diff, the diffusion contribution to Tlp, which is obtained by subtracting the reciprocal of the low-temperature (or asymptot­ic) Tlp from the measured relaxation rate. There are several significant features of the Tlp data. , First, for the temperature-dependent linear portion (on the semilog plot) of (Tlp)diff we measure an ac­tivation energy of (0. 733 ±0. 012) eV which agrees quite well with Friauf's results, 4 obtained using other methods. A second feature of Fig. 4 is that the Tlp mini­mum occurs considerably above the temperature corresponding to the onset of motional narrowing. This phenomenon is^due largely to the existence of static quadrupole interactions which, according to Eq. (8), will cause the measured Tlp to be shifted to values considerably higher than those resulting from dipolar interactions alone. As we shall see, further evidence for the existence of these static quadrupolar interactions is indicated by measure­ments of the dependence of Tlp on Hx. A third feature is the slight dip30 at the shoulder separating the diffusion region from the asymptotic region. This dip appears to be real and has two possible explanations. The first possibility is that the dip is the diffusion minimum due to the quadru­pole interaction between the chlorine atoms and vacancies (which should occur when the vacancy jump time rv is of the order of the mean quadrupole splitting). Since tv«t, the quadrupole minimum would undoubtedly occur at a much lower tempera­ture than the dipolar minimum. At all tempera­tures above this minimum the Cl-vacancy-quadru- pole interaction should be motionally narrowed and should not contribute appreciably to the relaxation time. A second possible explanation is that the dip is I05 I04 to3 a> </> 6 H- H- T> ^ I02 10' 10° 2.4 2.6 2.8 3.0 3.2 3.4 3.6 3.8 I03/ © (°K ) FIG. 5. (TiP)^ff vs reciprocal temperature. These data are taken from the Tlp data in Fig. 4, with the non­diffusion asymptotic contribution subtracted off, thereby giving the diffusion contribution to T1p / / Hi * O(DEMAG) H0iKioo> /y~ Jf / / / * / ¥ / * / J________L / / f 0 EQ= .733+ ,OI2eV2 4 9 6 G . L . S A M U E L S O N A N D D . C . A I L I O N 5 really a shift in Tlp caused by going from a high- temperature region for which rv < Tz to a lower- temperature region for which rv > Tz. The main effect should be that, at the lower temperatures, (^ip)diff will be reduced by a factor of order (G - 2)/G relative to its value at higher tem­peratures. In T1C1, for which G = 6, this factor should be It is interesting to note that the reduc­tion observed in Fig. 5 is of the order of f within experimental error. If this interpretation is cor­rect, rv at the temperature of the dip should be ~240 fJisec (since it would equal the rigid lattice T2). Using Eq. (8), we would obtain r at this same tem­perature to be approximately \ sec. We then find that the vacancy concentration can be determined, since Nv/N~ rv/r= 10"3. If the vacancies are pro­duced thermally, we could then estimate the forma­tion energy Ef of a vacancy, since Nv/N=e‘Ef/ke . (27) Using Eq. (27) we obtain EfZ 0.16 eV. This expla­nation of the source of the dip is really only tenta­tive. However, it fits the observations and sug­gests tantalizing possibilities for using NMR to measure the formation energy independent of the total activation energy. In order to measure independently all the param­eters in Eq. (8), it was necessary to determine r. Since motional narrowing begins when r~ T2, we could use our Tz data31 from Fig. 4 to estimate r. We somewhat arbitrarily assumed the onset of mo­tional narrowing to occur at the intersection of the asymptotes to the high- and low-temperature Tz data. Since a fraction 2/G of all jumps do not con­tribute to relaxation in the case rv « TZi r in Eq. (8) should be replaced by an effective-jump-corre­lation time32 equal to tG/(G-2). Thus, for rv«Tz, we replace Eq. (8) by (T \ r(H^+HDz+HQ2) ( G \ ( lp)diff 2(l-pD)HDi+{l-pQ)H<? \G - 2 / • ( ' B. Hx Dependence of (Tlp )diff Since (Tlp)diff in Eq. (28) has a linear dependence on Htzf a plot at a constant temperature in the ultra- slow region of (TipJdiff vs Hz will intersect the Hz axis at - (HDZ+HQZ). We can thus measure the to­tal local field. Figure 6 shows the results of one of four (Tlp)diff vs Hz runs. The average of three runs with the sample oriented with H0 parallel to the (100) direction gives HDZ + HQZ= [(9. 57 ±0. 5) G]2. Since the dipolar contribution HDz was calculated in Sec. Ill A to be (1. 666 - 0. 548sin22<p) G2, these ex­perimental results indicate that the local field arises largely from static quadrupole interactions33 due probably to strains, dislocations, and similar de­fects. In order to see whether the sources of quad­rupole interaction were randomly oriented in our sample, we performed a fourth (Tlp)diff vs Hz run with the sample oriented such that H0 was parallel to the (110) direction. We obtained the same local field, within experimental error, as in the (100) runs. In order to relate (Tlp)diff to (1 - pD), we must first determine the quantity (1 -Pq)Hqz. Since the static quadrupolar interaction gives rise to a broadening which is independent of angle, we shall assume that (1 -Pq)Hqz is also independent of angle. By using Eq. (28), we can relate the y intercept of Fig. 6 to (1 -Pq)Hqz1 since all the other parameters have been either independently measured or cal­culated theoretically. We then obtain the value (1 -PqWq2= (2. 6 ±0. 3) G2. The experimental technique used in measuring TiP vs Hz is to perform the demagnetization of the ^ field in a large Hx field followed by a partial de­magnetization of Hi to a reduced (though not neces­sarily zero) value.34 After a suitable time interval Hi is remagnetized to its original value and is then turned off sharply. By plotting for different #i's, the amplitude of the subsequent free-induction de­cay as a function of time in the state of reduced FIG. 6. vs data showing the intersection of the extrapolated best fit with the H\ axis where H* = - (HD2 +Hq2). H0 is parallel to the (100) direc­tion and the temperature is (350. 0 ±0. 2) °K. Also shown is the pulse se­quence: Hi refers to the rf field and ho is the off-resonant z field.5 N E W T E C H N I Q U E F O R D E T E R M I N I N G T H E D I F F U S I O N . . . 2 4 9 7 Angle (p (degrees) FIG. 7. vs crystal orientation withtf^O. The points represent experimental data and the solid line is a theoretical angular dependence determined from cal­culated and independently measured parameters. Hu we can measure Tlp vs Hi. A second method for determining the local field involves the direct measurement of Mx as a function of Hx following an ADRF sequence. Slichter and Holton have shown that the equilibrium magnetiza­tion along Hu after adiabatic demagnetization, is given by Mx=M0iJ1/(i/12+//x2)1/2, (29) where HL2 is the square of the total local field (i.e., H^+Hq2). Because of the large sample volume over which Hx was applied, we were limited to fields for which Hx< HL. Nevertheless, our data were consistent with a local field greater than 7 G which is in approximate agreement with the pre­vious results. C. Angular Dependence of (7\p)diff The primary contributor to the angular depen­dence of Tlp should be the term 1 -pD and this is the quantity which, according to the calculations of Ailion and Ho,2 would be most sensitive to the de­tails of the motion responsible for the relaxation. In the absence of quadrupolar effects, (Tlp)dlff in zero Hx will be simply proportional to (1 -PD)~l and so the angular dependence of (Tlp)din would then be identical to that of (1 -p D)~1. The extra quadrupolar terms in the denominator of Eqs. (8) and (28) will have the principal effect of reducing the angular dependence of (Tlp)diff. Since we have independently measured all the parameters other than pD and HDz on the right side of Eq. (28), we can use their observed values along with the theoretical values of pD and HDZ to obtain "theoreti­cal" values for the angular dependence of (Tlp)dlff for different diffusion mechanisms. By comparing these curves with experiment we can distinguish between rival mechanisms if their angular depen­dences are sufficiently different. Figures 7 and 8 show comparisons of experimental and theoretical angular dependences for Hx = 0 and =5.8 G, re­spectively. For Hx = 0, the theoretical angular dependence predicted by Eq. (28) for vacancy dif­fusion is 13.4% as compared to our observed re­sults of (15.0±2)%. For #!= 5. 8-G, the agree­ment is even better, with a theoretical angular de­pendence of 14. 3% and experimental of (15. 5± 2)%. The very slight displacement of our experimental curves to the left is probably due to a slight mis­alignment of our crystal. The entire curves could be displaced vertically by a small error32 in our determination of r. It should be emphasized that our measurements of r and HQ2 are independent of our measurements of (Tlp)diff. The value of (1 -pQ) x HQ2 is determined by our measurements of (Tlp)dlff at cp- 0 and H1- 0. Our measurements of (Tlp)diff at <p=45° then provide an independent check on our theoretical angular dependence. It is instructive to compare our observed angular dependence with the results obtained from an alter­native mechanism. If we consider the admittedly unlikely possibility of Cl interstitialcy diffusion in which the Cl interstitial atom is centered between two T1 sites, we can then calculate the angular de­pendence of (1 - p D) for interstitialcy jumps (see Fig. 1). The details of this calculation are worked out in Ref. 17. For the Hx - 0 case, these calcula­tions predict a 9. 8% angular dependence, whereas, for the Ht = 5. 8 G, they predict a 9. 0% dependence. Since both of these results deviate considerably from the experimental values of (15.0 ±2)% and (15. 5 ±2)% for the two Hxs, we can exclude Cl interstitialcy diffusion as a dominant mechanism in T1C1.35 Angle (p (degrees) FIG. 8. vs crystal orientation as in Fig. 7 but with i?! = 5. 8 G.2 4 9 8 G . L . S A M U E L S O N A N D D . C . A I L I O N 5 VI. CONCLUSIONS In this paper we have described the experimental verification of the prediction of Ailion and Ho that the rotating-frame spin-lattice relaxation time Tlp would have an angular dependence in the ultraslow- motion region which is sensitive to diffusion mech­anism. In particular we performed our experi­ments in T1C1, a substance in which the diffusion mechanism has been independently determined by Friauf to be Cl vacancy diffusion. We obtained excellent agreement between our experimental re­sults and our theory which assumes diffusion to result from the motion of Cl vacancies, whereas our data disagreed with the results of calculations based on an interstitialcy model for diffusion. Sim­ilar experiments and comparisions may be used to determine the dominant diffusion in other substances, ACKNOWLEDGMENTS The authors wish to thank Robert Morse for his assistance in writing the computer programs used in calculating the lattice sums. They are grateful to Dr. Franz Rosenberger of the University of Utah /, * i K2fi7iZI(I+1) (1 Pd)~ 4a6HB2 Crystal Growth Laboratory for growing the T1C1 crystal. Discussions with D. Wolf, Dr. D. Alder­man, Professor T. J. Rowland, Professor P. R. Moran, Professor I. J. Lowe, Professor J. S. Waugh, and especially Professor C. P. Slichter have been particularly valuable. One of the authors (G. L. S.) is grateful to the federal government for providing a 3-year NDEA IV fellowship. APPENDIX: CALCULATION OF 1 -pn FOR THE CASE tv » T2 This calculation is similar to the one originally performed by SA in lithium.5 It can be handled considerably more easily than the rv« Tz case since now the vacancy waits until all spins near it are in equilibrium before jumping again. Even return jumps are considered as independent contributions to the relaxation; therefore, there is no spin trail to consider. In going from a spin sum to a lattice sum, the site of the neighboring vacancy must be excluded explicitly from the sum in the II and IV contributions to (AE)D. We then have17 for rv» Tz that c't-cJ-Z c i ir iot + 2S(S+ 1 )/fs ys* si(i+i)\Tr7~ + fs-rs;2 \ fry? / 0240+ 0.1948 sin22<p 666 - 0. 548 sin22(p (Al) ^ This research has been supported by the National Science Foundation under Grant No. GP-17412 and in part by the Research Corporation. *This paper is based in part on the Ph. D. thesis pre­sented by G. L. Samuelson, at the University of Utah, Salt Lake City (unpublished). *A preliminary version of this work is to appear in an article by the above authors, in Proceedings of the XVIth Colloquium AMPERE, Bucharest, 1970 (unpublished). 2D. C. Ailion and P. Ho, Phys. Rev. 167, 662 (1968). 3A review of the ultraslow-motion techniques can be found in the article by D. C. Ailion, Advan. Magnetic Resonance 5, 177 (1971). 4R. J. Friauf, J. Phys. Chem. Solids 18, 203 (1961); also private communication. 5C. P. Slichter and D. C. Ailion, Phys. Rev. 135, A1099 (1964). 6N. Bloembergen, E. M. Purcell, and R. V. Pound, Phys. Rev. 73, 679 (1948). 7H. C. Torrey, Phys. Rev. 92, 962 (1953); 96, 690 (1954). - 8D. C. Look and I. J. Lowe, J. Chem. Phys. 44, 2995 (1965); 44, 3437 (1965). 9D. C. Ailion and C. P. Slichter, Phys. Rev. 137, A235 (1965). 10L. C. Hebei and C. P. Slichter, Phys. Rev. 113, 1504 (1959). ^A. Abragam, The Principles of Nuclear Magnetism (Clarendon, Oxford, England, 1961), p. 65. 12A. G. Redfield, Phys. Rev. 98, 1787 (1955). 13C. P. Slichter and W. C. Holton, Phys. Rev. 122, 1701 (1961). 14John Waugh and co-workers have recently shown that the existence of a spin temperature does not require the vanishing of all the off-diagonal density-matrix elements but rather only the vanishing of operators like pMx. More­over, they showed that the density operator is not correct­ly described by Eq. (1); Eq. (1) would be correct, how­ever, for a density matrix averaged over a coarse region of phase space. In order to observe the nonrandomness of the offdiagonal elements they applied a pulse sequence which effectively reverses time (by changing the sign of the Hamiltonian). [See W. K. Rhim, A. Pines, and J. S. Waugh, Phys. Rev. Letters 25, 220 (1970); Phys. Rev. 3_, 684 (1971).] Since nowhere in the ultraslow-motion experiments is the Hamiltonian reversed, a treatment based on Eq. (1) will yield the correct answer. 15T. J. Rowland and F. Y. Fradin, Phys. Rev. 182, 760 (1969). 1GJ. Wagner, Ph. D. thesis (University of Wisconsin 1970 (unpublished). 17G. L. Samuelson, Ph. D. thesis (University of Utah, 1971 (unpublished). 18The detailed steps in this calculation are given in Ref. 17. We will just outline the essential features here. 19Effects of encounters have been considered in detail5 N E W T E C H N I Q U E F O R D E T E R M I N I N G T H E D I F F U S I O N . 2 4 9 9 by Redfield and Eisenstadt [Phys. Rev. 132, 635 (1963)]. 20Our treatment for this case differs from that used in Ref. 2, since the authors of that article calculated the energy change resulting from one jump. Since the spin system has not achieved a temperature prior to the next vacancy jump, they had to estimate an equivalent temper­ature in order to calculate the final energy. Our treat­ment avoids this difficulty and also includes the contri­bution to the energy change resulting from the vacancy's approach from an infinite distance to a site neighboring our atom of interest as well as the contribution resulting from the vacancy's departure to infinity after the jump. 21 The apparatus is described in considerably more de­tail in Ref. 17. 22S. R. Hartmann and E. L. Hahn, Phys. Rev. 128, 2042 (1962). 23G. L. Samuelson and D. C. Ailion, Rev. Sci. Instr. 41, 743 (1970). 24R. J. Blume, Rev. Sci. Instr. 32., 1016 (1961). 25F. M. Lurie, Ph. D. thesis (University of Illinois, 1963 (unpublished). 26G. L. Samuelson and D. C. Ailion, Rev. Sci. Instr. 41, 1601 (1970). 27The company manufacturing these multichannel analyz­ers is now referred to as Nicolet Instrument Corp. and is no longer a subsidiary of Fabri-Tek, Inc. 28It is critical that the demagnetization be adiabatic. Because of the small y of Cl35, the adiabatic condition puts stringent requirements on the minimum demagnetiza­tion time. If the demagnetization is too rapid for the magnetization to follow, the magnetization will be left in the xz plane pointing in a direction other than H\* A ^two-component^ relaxation will then be observed (see Ref. 17). The first component results from the trans­verse decay of the magnetization component perpendicular to Hi in a time of the order of T2, whereas the second component will be the normal Tip decay of the magnetiza­tion component parallel to Hit 29The crystal was grown in the University of Utah Crys­tal Growth Laboratory from high-purity powdered T1C1 obtained from Alfa Inorganics, Inc. Even though the starting material was stated to have a purity with respect to heavy-metal Impurities of 10 ppm, we further purified it by means of a vacuum distillation prior to crystal growth. 30Similar effects have been observed in doped LiF sam­ples by Wagner and Moran [J. Wagner, Ph. D. thesis (University of Wisconsin, 1970) (unpublished)]. 31 It would have been preferable to perform the T2 ex­periments on our single crystal rather than on a powder sample. However, these measurements were performed prior to the T,p measurements and our apparatus was subsequently modified. Because of the over-all consis­tency of our results, it appears that small errors in the determination of r are not signif icant in our (1-Ad)"1 de­terminations. 32We have neglected the fact that the mean time between atomic jumps is not precisely the mean time between en­counters since a vacancy will have a relatively high prob­ability of causing a nearby atom to undergo additional jumps before the vacancy departs (see Ref. 19). When rv« T2, such extra jumps will not contribute to the re­laxation and thus we are really observing the mean time between encounters. These effects are only partially in­cluded in our factor G/(G-2). D. Wolf (private communi­cation) has calculated that a determination of r which ne­glects the effects of encounters could be in error by as much as 30%. 33We can not absolutely conclude that the extra broaden­ing is due to quadrupolar effects. Another possible source of broadening could arise from the indirect couplings be­tween the Cl and T1 nuclei. However, because of the large antishielding factor of Cl35 along with its relatively smaller atomic number, we felt that the broadening is more likely to be due to quadrupolar effects. Moreover, the free-induction decay appears to be Gaussian which is consistent with the idea that it is due to random defects. 34It is necessary to perform the z demagnetization in a large Ht in order to ensure that it will be adiabatic, since the smallness of y for Cl35 makes the adiabatic con­dition quite stringent for small H{ s. 35Xt is interesting to note that G=4 for interstitialcy diffusion in T1C1 since there are four nearest-neighbor sites to which the Cl interstitial may jump. Thus, for the case Ttf«T2, the factor (G-2)/G will be different for the two mechanisms. Accordingly, the absolute value of (Tlp)diff could also be used to distinguish between rival mechanisms provided r can be determined with sufficient accuracy.