You are on page 1of 10

Quantum diffusion in solid HZ:a short review

H. MEYER

Can. J. Phys. Downloaded from www.nrcresearchpress.com by University of Wyoming Lib on 12/16/15


For personal use only.

Department of'Physics, Duke Ur~iversity,Durharn, NC 27706, U.S.A


Received October 13. 1986
A review is presented on the experimental observations and the theoretical progress made in investigating the quantum
diffusion in solid HI. The mechanisms of resonant ortho-para conversion and of quantum tunneling are discussed. It is pointed
out that the ortho-HI clustering reaction process, by which quantum diffusion is detected, necds to be better understood to
account for the dependence of the observed reaction-time constant Ton temperature and on ortho-H? concentration. The gcneral
trends in the experimental results are reviewed, and the discrepancies bctween various determinations of T are discussed. Further
experiments are suggested as well as computer simulations for the clustering procedure. A review is also given on the status of
quantum diffusion of HD in solid HI. Recent observations on recombination rates of H and on depolarization rates of muons in HI
that show evidence of quantum effects are briefly mentioned.
On prksente une revue des observations expirimentales et des progres theoriques effectuks dans I'investigation de la diffusion
quantique dans Hz solide. Les mkcanismes de conversion resonante ortho-para et d'effet tunnel quantique sont discutes. On
signale que le processus de rkaction de formation d'amas d'ortho-Hz, au moyen duquel la diffusion quantique est dktectie, doit
&tremieux compris, pour expliquer les observations concernant I'influence de la tempirature et de la concentration ortho-H2 sur
la constante de temps T de la reaction. Les tendances gkntrales des rksultats expkrimentaux sont passies en revue, et les CcartCs
qui existent entre diffkrentes dkterminations de T sont discutkes. On suggkre de nouvelles experiences et des simulations i
I'ordinateur pour le processus de formation d'amas. On prisente aussi une revue de Ia situation concernant la diffusion quantique
de HD dans HI solide. Des observations recentes sur les taux de recombinaison de H et sur les taux de dkpolarisation des muons
dans Hz, qui mettent en Cvidence des effets quantiques, sont mentionnkes brikvement.
[Traduit par la revue]
Can. J . Phys. 65. 1453 (1987)

1. Introduction
Solid hydrogen is a remarkable medium where the angular
momentum J of the ortho-Hz impurities in para Hz can diffuse
through the lattice (1). It is now almost 20 years since this
phenomenon was discovered, and therefore it seems appropriate
to present a review on the status of knowledge on this
intriguing topic and to suggest further research.
The quantum diffusion rate is too slow to be observed
directly. This is to be contrasted with classical diffusion, which
above a minimum hopping frequency, v, from one site to a
neighboring one, can be detected by nuclear magnetic resonance (nmr) techniques such as rotating-frame pulse sequences
(see ref. 2 and references cited therein) or with pulsed magnetic
field gradients. Instead, quantum diffusion in Hz is detected via
the clustering (or unclustering) of the diffusing excitations as a
function of time after a rapid change in temperature. This
redistribution results from the anisotropic electric quadmpolequadmpole (EQQ) interaction between the 0-H2 particles,
leading to changes with temperature in the equilibrium proportions of the clusters of various sizes. Hence the motion of the
excitations permits the free energy F to be minimized. At any
temperature there is such motion, and at equilibrium, on
average as many pairs are formed as are broken up. For kBT >>
A E , the energy released through formation of a nearest
neighboring pair or larger 0-H2 cluster, the distribution is
random. As will be shown later, A E / k B = 3 K . As T is
decreased, the proportion of pairs and larger clusters in
equilibrium increases, and conversely as the temperature
increases, the distribution becomes more randomized.
Classical thermally activated diffusion in Hz can be expressed
by the average time T~ between successive jumps from one site
to another. This time, obtained from nmr relaxation measurements (see ref. 3 and refs. therein to earlier work), is given by

For T > 7 K, where

T! 5

1 s, the 0-H2 distribution is nearly

random, but as T decreases below 7 K, this diffusion mechanism becomes too slow to permit rearrangement of the o-Hz
particles in an acceptable time, say a few hours. Hence other
mechanisms based on quantum effects have to be invoked to
account for the observed diffusion below 4 K.
The original technique leading to the accidental discovery of
this motion was continuous-wave proton nmr (4). Here, the
characteristic spectrum intensity of the isolated o-Hz impurities,
henceforth called "singles," is found to decrease with time after
a rapid lowering of the temperature. Nuclear magnetic resonance
in hexagonal close-packed (hcp) H, single crystals makes it
possible, in fact, to monitor the spectrum intensity, and hence
the density of the singles and of nearest neighboring (nn) 0-HZ
pairs. Of the latter there are two kinds: the "in-plane" (IP) and
the "out-of-plane" (OP) pairs having their axis in and out of the
basal plane. Hence, one can check whether or not a decrease of
one species, the singles for instance, happens simultaneously
with an increase of pairs ( 5 ) .
After the first observations, other methods led to the detection
of quantum diffusion. The thermodynamic methods (6, 7)
exploit the dependence of the free energy, F, of the solid on the
statistical distribution of o-H, in the lattice. This potential can
be calculated in terms of the EQQ interaction between the o-H,
molecules. For small clusters of three and less, the energy states
have been calculated (8).' Derivatives of F, namely, the
specific heat, Cv,and the pressure, Pv,at,constant volume then
show a time dependence during the clustering (9). The optical
methods (10-12) are based on the time dependence of the
transition intensity between the energy states of the pairs.
Therefore, just like the nmr spectra, they measure the change
with time in the population of pairs.
A further method of clustering detection is the thermal
conductivity (13), which is a sensitive measure of phononscattering processes, and it is affected by the cluster sizes.
'See also A. B. Harris, unpublished results. 1968.

1454

CAN. J. PHYS. VOL. 65, 1987

Can. J. Phys. Downloaded from www.nrcresearchpress.com by University of Wyoming Lib on 12/16/15


For personal use only.

Finally, the ortho-para conversion rate, and hence the evolved


conversion heat, depends on the cluster size and is a function of
time during the particle redistribution (7).
In all these experiments, a transient exhibiting a nearly
exponential decay with time to an equilibrium value is observed, from which a characteristic reaction time T is observed.
This time is, in general, a function of T and of the ortho-Hz
concentration, X. The process works in both directions:
clustering after rapid lowering of T by an amount A T , and
unclustering after raising T . For a given sample of composition
X, T is a function of T , independent of the direction or size
of AT.
2. Theory
Van Kranendonk (1) presented in 1983 a detailed review of
the theoretical aspects of quantum diffusion. This section can,
therefore, be kept short, its purpose being to draw attention to
important unresolved problems.

2.1. Resonant ortho-para conversion


Oyarzun and Van Kranendonk (14, 15) have shown that the
mechanism of resonant ortho-para conversion between nearest
0-HZand p-Hz neighbors accounts for quantum diffusion. In this
process, the nuclear dipolar field from the two parallel I = 112
proton spins of 0-Hz produces simultaneous spin flips, giving a
parallel spin alignment in one of the neighboring p-Hz and an
anti-parallel one in 0-Hz AS the nuclear spin I = 1 is transferred
from 0-H2 to p-Hz, so is the rotational angular momentum
J = 1, and hence the original 0-Hz and p-H2 are interchanged.
For an isolated 0-H2 in a pure p-H2 lattice, the jump frequency,
vo, for this conversion is predicted to be (1)

where [ is a measure of electrostatic screening and is in the order


of unity; yph/2 = pp = 1.4 X 1OPz3erg/G is the proton
magnetic moment (1 erg = 0.1 pJ, 1 G 0.1 mT); re = 0.74A
is the interproton distance in HZ;and Ro = 3.79 A is the lattice
constant. In this ideal process, the propagation of this excitation
has a wavelike nature. The uncertainty in the energy of an ortho
excitation on a given site is SO/kB= 12 voh/kB 6 X l o p 7K,
where 12 is the number of nearest neighbors. Any mismatch in
energy between the initial and final stage of the jump that is
larger than So results in a decrease of the jumping frequency.
In a real crystal at a finite temperature, at least two effects lead
to a broadening of the energy band of the ortho excitations.
These are (i) the coupling of the angular momenta Z and J to the
lattice vibrations, and (ii) the EQQ interaction between the o-Hz
impurities. These interactions give rise to relaxation processes
with certain characteristic times that are short in comparison
with the time T( = (12 vo)-' between jumps. This limits the
lifetime of the orientation of I and J and renders the jumps to
different sites incoherent.
The effect of lattice vibrations that are influenced by strains
and dislocations is particularly difficult to estimate, but a
calculation of the impact of J = 1 impurities on the jumping
frequency has been carried out. This problem is also very
complicated, because the EQQ field, which is inversely
proportional to R~ (where R is the distance between two 0-HZ
particles) changes after each jump.
The effect of orientational fluctuations is now discussed
briefly. The effective EQQ interaction is characterized by an
energy T, which is proportional to (eQ)2/R;, where (eQ) is the
electric-quadrupole moment of the o-H2 molecule. Various

FIG. 1. (a) Energy diagram of a pair of O-H2particles separated by


a distance R. (b) Schematic energy release in the hopping process as
the particles approach each other during the clustering process for
T < T/kB.

experiments have determined (for a review, see ref. 16) this


energy as r / k B = 0.83 K. Figure l a shows the energy levels
of an 0-H2 pair where Ro, the nearest neighbor (nn) distance, is
Ro, d?Ro, and 2Ro for next (nnn), third, and
replaced by
fourth nearest neighbor shells.
In a crystal with a ( J = 1) impurity concentration, X, the
average orientational fluctuation rate, WQ,can be calculated
from the assumed form of an appropriate orientational correlation function, G(t). For X 2 0.4, a Gaussian form (17) appears
to be consistent with nmr spin-lattice relaxation time measurements (see Amstutz el al. (1 8) and Hardy and Gaines as quoted
by Sung (19)). Van Kranendonk has used this Gaussian form for
lower concentrations as well and calculated WQ for the process
that excluded jumps with formation of (nn) pairs of ( J = 1)
molecules. Rewriting [9.46] of ref. 1, one obtains

&

[3]

WQ

= 2.5

x 10" ( r / k B ) X1I2radvs-l,

(Gaussian)

Oyarzun and Van Kranendonk (15) have discussed this


Gaussian assumption for G(t) and concluded that it was better
adapted to the situation of hopping impurities than a statistical
form developed for low impurity concentrations (X 5 0.1). This
latter form (see ref. 17; see also ref. 20 and references therein to
previous works) predicts nmr relaxation times in good agreement with experiments (19,21). The average fluctuation rate for
this statistical model is given by
[4] WQ = 2.6 x 10" (T/ks) X5I3 rad-s-' (statistical)
Clearly in the presence of (J.= 1) impurities, only a fraction
(vo/oQ)2n of transitions consewe energy within So. Then, the
average jump frequency in a crystal with impurity concentration
X is calculated to be (1)
Assuming a Gaussian approximation for G(t), or alternatively
using the statistical model, one obtains
[6]

v = 1.3 X

[7]

v = 1 . 2 x I O - ~ X - ~s ' ~

For X = 1

X-I'z

(Gaussian)
(statistical)

lop2, the time between jumps is then given by

1455

MEYER

= ( 1 2 ~ ) and
~ ' one has TC = 700 s (Gaussian) and TC = 32 s
(statistical).
The same calculation of the resonant ortho-para conversion
frequency vo of an isolated ( J = 1) p-D2 impurity in ( J = 0)
o-D2 has been carried out by Van Kranendonk (1). Here the
nuclear spins are I = 1 for ( J = 1)D2 and I = 0 and 2 for
,
into an expression
( J = O)D2. The nuclear g factor, y ~entering
similar to [I], is only O.15yP, and it is chiefly responsible for the
much smaller jump frequency in D2 (v (D2) = 5 X l o p 3 v
(Hz)). This implies that the redistribution times of ( J = 1)D2
impurities would be about two orders of magnitude larger than
in Hz and thus not observable on a realistic experimental time
scale.

Can. J. Phys. Downloaded from www.nrcresearchpress.com by University of Wyoming Lib on 12/16/15


For personal use only.

2.2. Quantum tunneling


Oyarzun and Van Kranendonk (15) have also estimated the
exchange energy, J = hvo, between two nearest neighbors in
solid Hz. They used the expression from the overlap of a
single-particle Gaussian wave function, as developed by Guyer
and Zane (22), and obtained vo between lop4 S - I and 10-I s f 1 .
They pointed out that this interchange process is also inhibited
by the EQQ interaction, as energy is not conserved. Assuming
again a Gaussian shape for the orientational correlation function
G(t), they obtained the same form as [3] for the fluctuation rate
but with a prefactor of 1.4 X 10" instead of 2.5 X 10" rad.s-I.
In the presence of ( J = 1 ) impurities, the hopping frequency,
vo, was reduced then again by the factor (vo/wQ)2.rr, as in [5].
Hence, Oyarzun and Van Kranendonk (15) concluded that the
interchange mechanism is much slower than that of resonant
ortho-para conversion.
However, recently Delrieu and Sullivan (23) have shown
that three-body cyclic permutations dominate the tunneling
processes in hcp quantum crystals and that the Gaussian overlap
function has to be replaced by an exponential one. Having been
successful in the agreement between the calculated frequency
vo = J l h and the experimental one in 3He, they have extended
their method to Hz and predicted vo = 9 X lo3 S- at P = 0 and
1.3 x 10' s- at P = 100 bar ( 1 bar = 100 kPa). This frequency
is higher by roughly five orders of magnitude than vo for
tunneling, as calculated by Oyarzun and Van Kranendonk (15).
Also, it is roughly nine times larger than vo from resonant
ortho-para conversion, [ 2 ] . Of course, as before, the hopping
frequency v is affected by the presence of J = 1 impurities and is
calculated via [5] (Delrieu and Sullivan used the statistical
model, [5], for w Q ) Given the larger uncertainties involved in
the calculation of quantum tunneling frequencies than in those
of ortho-para conversion, one is probably being realistic to
estimate both processes as having a comparable frequency.
From the experiments, all performed so far at P 5 30 bar, it is
not possible to decide which of the two processes dominates.
However, the tunneling frequency is calculated to be more
sensitive to density, and future experiments at higher pressures
should be able to clarify this question.
Finally, it should be mentioned that the tunneling frequency
20cm3.mol-') is estimated (23) to be
in D2 at P = 0 ( V
vo
10-'s-', six orders of magnitude smaller than in HZ.
Hence, quantum diffusion in D2 due to the tunneling process
should not be observable.

2.3. The reaction-time constant


In this subsection, a connection is attempted between the
microscopic process of hopping and the macroscopic observation of a time constant, T, for the excitation distribution
throughout the H2 lattice.

On a macroscopic scale, one can describe (6) the particleredistribution process by a rate equation, where for simplicity
only singles and pairs are included and larger clusters are
neglected. The single $ pair rate equation is then
where s = N,/N is the fraction of singles and where the terms on
the right-hand side refer, respectively, to the formation and
breaking up of pairs. Here the reaction parameters A and B are
taken to be functions of the average (time independent)
composition and the temperature. The solution of this equation
can be written as
[9]

s - s, =

(SO- s,)(s, - s,) exp -t/7


(SO- s,) - (SO- s,) exp - t / ~

where T, the reaction time, is given by T = [A(s, - s,)]-I; s, is


the equilibrium fraction when d s l d t = 0 and is a function of
(AIB). Furthermore, -s, = s,/(l - s,) is the ratio at
equilibrium of the number of molecules forming singles and
pairs, and so is the fraction at t = 0. Because in practice the first
term (so - s,) in the denominator of [9] is larger than the second
one, which furthermore decays with time, the equilibrium is
reached in practice in an exponential way. Hence, T can be
determined from the decay of ( s - s,) versus time, when the
system is not too far from equilibrium.
The problem of calculating the reaction-rate constant T-'
from the hopping frequency v has not yet been completely
solved, but Van Kranendonk (1) has described two models that
treat the random motion of the single excitations. In the
diffusion model (DM), the probability distribution for the location of each J = l impurity at time t in relation to its position
at t = 0 is calculated. Initially, impurities are distributed
, they
randomly with an average separation of d = R ~ / x " ~and
move through the lattice with a diffusion coefficient found to be
D = 2R&. The time constant T is then estimated by setting it
equal to the average time it takes a particle to approach another
to a distance of d/2. This time is found to be (1)
[lo]

= 0.05 x-'I3 v-

(DM)

Together with [6], one finally has


[ll]

= 3.8 X

lo3 x-'I6s ,

(DM, Gaussian)

The number of steps during the time T is then given by ri = 12 vr;


and for X = 0.01, one finds T = 2.4 h and n = 13 steps.
The kinetic model (KM) calculates the probability of successive steps and the average number of steps until pair formation;
it gives ( 1)
[12] (

(KM)

Therefore, from this model,


[13] T = (n) ( 1 2 ~ ) - '= 9.5 X lo2 x-'j2, s, (KM, Gaussian)
which is to be compared with the prediction from the diffusion
model.
In both models, it has been assumed that the energy
exchange, A -- hwQ, involved in the hopping was much smaller
than kBT, and this accounts for the temperature independence of
the predicted T in [lo]-[13]. (As an example, A/kB = 0.1 K for
X = 0.01.) These expressions are then suitable for the
high-temperature limit (T '> 3 K). However, theory should also
be made to account for the probability of transition from the
(nnn) to the (nn) pair state, where the energy released, AE =

1456

CAN. J . PHYS. 1IOL. 65, 1987

Can. J. Phys. Downloaded from www.nrcresearchpress.com by University of Wyoming Lib on 12/16/15


For personal use only.

4 r = 3kB,is comparable to kBT(see Fig. 1 b). The reaction rate


should, therefore, be temperature dependent and this has been
briefly discussed by Oyarzun and Van Kranendonk (14, 15).
Estimations made by these authors, but not presented in detail,
appear to give qualitative agreement of the calculated temperature dependence of T with the experimental one where T is
observed to increase by a factor of two as the temperature is
decreased from 4 to 1 K.
It is useful to discuss the predictions for T-' if the statistical
model for the EQQ correlation function is used. Then [7],
combined with [lo] or [13], gives
[I41

= 4.2

[15]

x lo4 X s,

(DM, stat.)

9.9 x 1 0 ' ~ S,
~ ' ~ (KM, stat.)

Clearly, the concentration dependence of T ([I 11 and [13]-[15])


is strongly dependent on the models for both the orientational
fluctuation spectrum and the random motion of the single
excitations.
A very different approach to calculate the reaction rate T - '
was made by Kokshenev and Strzhemechnyi (24). They
assumed that the pair-formation time consisted mainly of the
process where a (nn) pair was formed from (nnn) excitations in a
single transition. The excitations separated by distances equal to
or larger than those for (nnn) were assumed to be "in
equilibrium" (which implies that the hopping times between
these sites were assumed to be very short in comparison with
the (nnn) to (nn) pair formation time), and hence no time
was allotted for their diffusion from initial positions. The
calculation of the EQQ fluctuation rate used the statistical model
x513. The transition probability from a (nnn) to a
where WQ
(nn) pair (or vice-versa) was calculated to be

[16] W a v exp -(4r/kBT)


where v 4 / o Q , as in [ 5 ] .Finally, the reaction rate was found
to be
which is to be compared with the predictions above.
The following flow diagram, suggested by R. G. Palmer and
representing the various steps in the clustering process, might be
useful to the reader.
resonant
OP conversion

Gaussian

17

quantum
tunneling

diffusion model

vcc~-112v< y;Kx-213~-'

rv\

v -kinet~c

statistical
v cc x-513
v;

model
v-1

7 " x-1

(nnn) to (nn) HOP

a v-I exp (AE/kBT)

2.4. Pair hopping


Not only single excitations are found to diffuse but also (nn)
pairs. This phenomenon has been discovered again through the
decay of the pair signal intensity with time in the nmr spectrum
(5). Based on the EQQ interaction, the calculated equilibrium
proportion of clusters with three or more 0-H2 particles
increases with respect to the proportion of pairs as T decreases.
Hence the number of pairs increases at first through formation
from "singles" and then decays with time as T decreases, with
formation of larger clusters.
Van Kranendonk (1 ,25) has carried out a detailed calculation
of the pair-hopping frequency. In this process, only one particle

t Coxis

FIG.2. ( a )The axial directions of (nn) in-plane (IP) and out-of-plane


(OP) pairs in the Hz hcp crystal. (b) The lowest energy states of the 1P
and OP (nn) pairs.

moves at a time but in such a way that the (nn) condition is not
broken. As this process involves motion where the neighbors
are coupled by EQQ interaction, we can anticipate that it is
much slower than the hopping of single excitations. The
position of the IP and OP pairs in the lattice and their energy
states are shown in Fig. 2. For the IP pairs, the lowest two states
are separated by A/kB 6 5 mK; whereas for the OP pairs,
A(OP)/kB = 64 mK. Hence the energy involved, if an IP pair is
transformed into an OP pair by a jump of one particle, is in the
order of A(0P) and much larger than the width 6 of the rational
states. This process is, therefore, unlikely.
Van Kranendonk's result for IP pairs, [27] of ref. 25, can be
rewritten as (1)
with vo given by [2]. Unfortunately, 6 is quite uncertain, and a
value 6 / h = 30 MHz, as measured in microwave pair spectra
(12), is adopted. This leads to
which has to be compared with [2] for single excitations.
Furthermore, it has been noted that a single impurity can jump to
12 nn sites, but a pair can only make four different jumps.
Hence, the ratio of the times spent on a given site is tlP/ts =
3(vs/vlp). As for the OP pairs, a calculation for kBT >> A(OP)
shows that vOp/vlp = 0.3. Each hopping of an OP pair involves
the probability p = exp (-AE/kBT) of an excited state with
energy A(0P). As T is decreased, p decreases and so does vop.
It is calculated that for kBT < A(OP), vop/vIp is SO small that
comparatively speaking, the OP pair can be considered immobile. Then, if we assume that each encounter of two pairs leads
to a cluster formation and hence to disappearance of the pairs,
Van Kranendonk's rate equations show that the disappearance
time, T(O,), is three times slower than T ( ~ PThis
) . result can be
directly compared with experimental data (5). The theory has
not yet yielded a prediction on the temperature dependence of
the observed pair-decay rate. Based on geometrical arguments,
Van Kranendonk (25) has concluded that clusters of three (nn)
0-HZor more cannot displace themselves.

3.. Experiments
3.1. Redistribution of single excitations
In general, the experimental procedure has consisted of
monitoring an appropriate physical quantity (such as optical
signal intensity, specific heat, thermal conductivity, etc.) versus
time at an initial temperature. After the H2 sample was rapidly
cooled (or warmed) to a new temperature and a thermal
equilibrium was reached, measurements were carried out versus
time at this temperature until they indicated an equilibrium state
has been reached. From the change with time, usually represented by an exponential decay, an experimental time constant T

1457

Can. J. Phys. Downloaded from www.nrcresearchpress.com by University of Wyoming Lib on 12/16/15


For personal use only.

MEYER

was obtained, which corresponded to the motion tacitly assumed to be that of the "singles." (As was discussed above, the
hopping motion of pairs is slower than that of singles.)
The data obtained so far show clear trends, but there has been
a surprising amount of inconsistencies and scatter among the
various results and much remains to be done to clarify the
experimental situation. The reason for the scatter among the
different sets of experiments is not understood. However, it is
possible that the crystallization procedure, which is no doubt
nearly the same for each sample in a given series of experiments but probably different among the various research
groups, plays a role in influencing the clustering reaction time.
This is consistent with Van Kranendonk's predictions (1) that
the hopping frequency depends on the interaction between
lattice vibrations and the angular momenta, and hence on the
strains and defects in the crystals.
Probably the most direct- information on the clustering has
been provided by nrnr and spectroscopic methods, where a
signal intensity proportional to the number of singles and (or)
pairs is monitored as a function of time. As the IP and OP pairs
have different energy-level spectra because of differences in
crystalline-field interactions, the microwave experiment by
Hardy et al. (12) that measured transitions between these levels
was able to give different signals for the two species. This
experiment required a low concentration, X < 2 x l o p 3 , to
produce sufficiently sharp lines. After the temperature was
rapidly decreased and stabilized at a certain value, the intensity
of these pairs was found to increase but with different relaxation
~
7 ~1 p , ~,where
,
~ the subscript O P or IP defines the
times, T
pair in question and G means "growth" from singles. This
transient is shown in Fig. 3.
From nmr experiments, the spectrum of "singles", OP, and IP
pairs could be observed in single crystals at appropriate angles
with respect to the magnetic-field direction (5). After rapidly
decreasing T and stabilizing it, the intensity of the singles was
found to decrease with time while that of both O P and IP pairs
were found to increase, all with the same time constant, as
shown in Fig. 4. Here T,(S+ OP, IP) = 7 0 p . ~= 7 1 p , ~and this
result is therefore inconsistent with that by Hardy er al. (12).
However unpublished measurements' of the intensity of microwave absorption lines for X > 0.002 as a function of time seem
to indicate that the difference between the IP and the O P growth
rates is not well established, and more experiments are needed.
In Fig. 5 , the transient in three other measurements (pressure,
ortho-to-para conversion heat, thermal conductivity) is shown
versus time, where it can be seen that a simple exponential law is
followed from which a time 7 can be determined. One can
estimate that. on average. 7 can be determined to about 10% or
even better, and therefore the differences among the data are a
real effect and are not caused by poor precision or ambiguity in
the measurement of 7 .
To discuss the various results, we find it useful to show in Fig.
6 a general trend of the concentration dependence of 7 at two
reference temperatures, 1.2 and 2.1 K , with data from several
research groups. Generally, 7 increases with decreasing X but
tends to saturate below X = 0.01 and for T S 1.5K. Very
roughly, most of the experimental evidence indicates a trend
expressed by

( a ) - GROWTH

OF PAIR INTENSITIES

X=O.Zo%;

T = 1.2K

3.0
0

GI

OUT OF PLANE
13.65 GHz

A,

64.15 GHz

:2.0
a

--

20

10

TlME

k
V)
z

1.0,

(b)-

IN PLANE
84.1 GHr

30

40

(h)

OF 13.65 GHz LINE


X = 0 . 2 0 % : T.2.I
K

DECAY

+
Z

a
0
W

0.5-

----

------------------

0 - -

TlME ( h )

FIG.'^. The intensity of certain IP and OP pair microwave transitions


versus time after cooling and after warming. From Hardy et 01. (12).

2W. N. Hardy, private communication.

TIME

(h)

FIG.4. The nmr spectrum intensity of singles decay (a),IP pair


growth (t),and OP pair growth ( x ) versus time at T = 0.62 K ,
showing them to be equal within experimental scatter. Sample # 1 with
X 0.01. From Washbum et al. (5).
This result indicates that the choice of the Gaussian correlation
function for the orientational fluctuations (1, 15) is more
realistic than that of the statistical model. This is in contradiction with the evidence from nmr relaxation experiments
(19,21), where the statistical model is shown to be valid for low
enough values of X.
Figure 7 shows the temperature dependence of 7 for data of

1458

CAN. J . PHYS. V 'OL. 6 5 , 1987

Can. J. Phys. Downloaded from www.nrcresearchpress.com by University of Wyoming Lib on 12/16/15


For personal use only.

TIME (h)

FIG. 5. The transients of pressure at constant volume (6), A P d T ) ;


ortho-para conversion heat release (7), A WO+p (I); and normalized
thermal conductivity, AX(r)/AYC(O) (13), versus time, where A indicates the change between t = and t.

FIG. 8. Dependence of T on T for nmr data (5) below 1 K and


comparison with results from the ortho-para conversion heat (6) and
thermal conductivity (13). The solid line is an arbitrary fit to the nmr
data given by [22].

Figure 8 shows the temperature dependence of T as obtained


from nrnr absorption data for two samples, #1 and #2, with
compositions X = 0.01 and 0.002. In spite of the scatter, a clear
trend emerges; as T decreases below -0.3 K , a sharp acceleration of the clustering rate is observed. An empirical expression

FIG.6. Dependence of the reaction time T on X at two temperatures.


Here the results of Ramm and Meyer (6) (Pv), (e),Amstutz et al. (4)
(nmr) ( X ) , Roffey et al. (1 1) (infrared) (A), Hardy et al. (12)
(microwave) (V), Washburn et al. (5) (nmr) (U), and Minchina et al.
(7) (C,,) ( 0 )are shown.

FIG. 7. Dependence of 7 on T for X = 0.01 and 0.3 < T < 4 K. The


data are from Ramm and Meyer (6) (Pv) (dashed line, e ) ,Amstutz et
al. (nmr) (4) (solid line, U), Roffey et 01. (1 1) (infrared) (dotted line,
x), and Minchina er al. (7) (C, and Whp) (0).
TO avoid overcrowding, no line was drawn through the data points of ref. 7.

several groups for X


0.01, and clearly T increases with
decreasing T, at least as low as 0.3 K. Again, very roughly, the
temperature dependence can be expressed by
[21]

,T-'" mm ==10.7,
,

refs. 4 , 5 , 6 , 11
ref.7

which is a considerably less steep function than the prediction


T " exp(3.2/T) of ref. 24 (see [17]).

that reflects the "high" and the "low" temperature behavior can
express the data quite well and is shown by a solid line in Fig. 8.
For T 5 0.05K, the rate was found to be too fast for
measurement with the procedure outlined above, where the H2
sample was cooled together with the dilution refrigerator from
1.5 K, a process requiring approximately 1.5 h. This fast
clustering rate was found on a number of other crystals
investigated via the nmr spectrum. A few hours after cooling
below -0.07 K , the nmr signal of the singles near the center of
the H2 spectrum could not-be detected any more. Ain obvious
next step will be keeping the dilution refrigerator operating at
T 5 0.07 K and rapidly establishing or breaking the thermal
contact with the sample by means of a heat switch. In that
modification, the sample can be warmed up to -2 K to uncluster
for a period of several hours before the next rapid cooling for a
further clustering experiment.
However, Fig. 8 shows a discrepancy with the recent
Kharkov data (7). In these latter experiments, T is obtained from
the time-dependent specific heat for T > 1 K and from the
energy release due to ortho-para conversion for T < 1 K for
0.1% IX I1.1%. These data show a T-I dependence of T.
Also nmr data by Kohl (27) show an increase of T to 19 h at
0.12 K and no evidence of a maximum.
From the variation of the thermal conductivity with time
during clustering (13), it has been concluded that this method is
a very suitable one for a systematic study of T over a wide range
of X and T, and such measurements are planned for the near
future.
Minchina et al. (26) have- extended their calorimetric
experiments to solid H2-D2 mixtui-es and measured the reaction
time T for several mole fractions X(D2) of D2, where the initial
impurity concentration X of 0-H2 was between 0.5% and 1%.
As in pure HZ,they found T to rise with decreasing temperature.
A surprising result, shown in Fig. 9, is that the lowest value of T
along an isotherm was measured not for pure H2 but for a
concentration of 2% D2. Second, even high concentrations of
30% D2 led to an increase of T by a factor of less than 2. These
findings contradict earlier calorimetric experiments by Roberts
and Daunt (28), who reported that for a mixture of 15% D2 in
Hz, quantum diffusion is suppressed. Finally, Minchina et al.

1459

MEYER

#2, X=0.002
T =0.025K 1

--

Can. J. Phys. Downloaded from www.nrcresearchpress.com by University of Wyoming Lib on 12/16/15


For personal use only.

O P Poir decoy
+.v I P Poir decoy

0
I

0.1

60

40

TlME

20

80

(h)

FIG.11. Decay of the IP (+ , V) and O P ( 0 ) pair intensities with time


at T = 0.025 K. From Washburn et al. (5). The crosses and inverted
triangles indicate two different experiments.

0.2 0.3

both types of pairs has to be in a crablike fashion along


directions parallel to the basal plane, and the IP and OP pair do
not interchange, as their decay rates are different. This is
consistent with Van Kranendonk's statement that IP and OP
pairs cannot interchange for energetic reasons (25).
Hence, the pair-decay rate appears to accelerate as the
temperature decreases, in a way that is similar to the single
excitation decay. For none of these surprising results has there
been a theoretical explanation so far. The idea that the motion of
both single 0-H2 impurities and pairs can be related to that of
lmpuritons," similar to 3He impurities in 4He (29-31), has
been briefly discussed elsewhere (5). More experiments at still
lower temperatures and at different values of X are necessary to
produce a more complete picture of the diffusion of both singles
and pairs in solid p-Hz.

FIG.9. The reaction time in H2-D2 mixtures versus D7 concentration, X(D2), at two temperatures. From Minchina et 01. (26).

"'

Singles

Y
o0 ~

"

OP Pairs

"

I0

"

~15 "

'

20

"

"

25

TlME ( h )

FIG.10. The nuclear magnetic resonance intensity of singles ( 0 ) and


pairs (+, O P pairs; x , IP pairs), after rapid cooling to T = 0.30 K ,
plotted versus time. Reprinted from Washbum et al. ( 5 ) with permission.
(26) observed that the variation of
independent of the D2 composition.

with temperature was

3 . 2 . Pair hopping and clustering


Nuclear magnetic resonance experiments have also demonstrated (5) that the stability of the OP and IP pairs is different.
This is apparent from the variation of the signal intensity with
time after a rapid cooling of the sample from 1 K. After an
initial intensity rise for both types of pairs, the IP intensity
decreases with time at constant temperature, while that of the
OP pairs just levels off over a period of 100 h. This observation
is shown in Fig. 10. The pair-decay rate accelerates as the
temperature is decreased, from T = 30 h at T = 0.30K to T
10 h at 0.025 K for the IP pairs. For the OP pairs, T decreases
from >50 h to -30 hat 0.025 K, as shown in Fig. 11. It appears
that in turn, pairs can propagate through the lattice but at a
considerably slower rate than the singles, and they become part
of larger clusters. Arguments (5) show that the propagation for

3 . 3 . Injluence of clustering on macroscopic properties


So far, the reaction time T , observed for various properties,
' has" been
' ~discussed. In this subsection, a short description is
presented on what happens to these properties as clustering
proceeds.
For the static properties, the free energy, FQ, of the rotational
motion can be calculated from the partition functions for
fractions s of isolated o-Hz, p of molecules forming (nn) pairs,
and t of 0-H2 (nn) triangles (9). For these configurations, the
energy levels are known as a function of the quadrupolar
interaction parameter T, which is proportional to vP5l3,where V
is the molar volume. There are six cluster configurations of three
(nn) o - H ~ ,and for simplicity, the larger clusters are ignored;
therefore,

) ~the, specific heat, C V , ~


The pressure, PQ = - ( ~ F ~ I ~ vand
- T ( ~ * F / ~ T ~are
) ~ then
,
obtained. Because the splitting
between the three rotational states of the singles is much smaller
than r, it can be neglected, and then the contribution from the
singles to PQand Cv,Qis zero. Quantum diffusion permits FQto
tend towards a minimum at any temperature, and the equilibrium values of p and ti increases with decreasing values of T.
Their values for a composition X = 5 x l o p 3 have been
presented by Minchina et a l . (7) and are shown in Fig. 12,
where for simplicity only three types of triangular clusters are
=

Can. J. Phys. Downloaded from www.nrcresearchpress.com by University of Wyoming Lib on 12/16/15


For personal use only.

1460

CAN J . PHYS. VOL. 65. 1987

function of time at 0.2 K after rapid cooling show an increase


A K / K by approximately 10%. This implies that for a given
value of X, inelastic phonon scattering by clusters of 0-H2 is less
than by a random distribution of "singles." The average phonon
wavelengLh at 0.2 K in H, is calculated to be in the order of
2 X 1 0 3 ~which
,
is much larger than the dimensions of the
scattering centers and the distance between them. In optics, this
situation would correspond to Rayleigh scattering of a light
beam where the intensity loss is proportional to the number of
scattering centers. An argument, based on conservation of the
number of molecules, that assumes the scattering amplitude to
be proportional to the volume of the scattering element, shows
that the intensity loss should increase as clustering occurs
(example, formation of fog from water-saturated air). The
argument assumes that the cluster size is still much smaller than
the wavelength of the light. This increase in turbidity with
clustering is the opposite of what is observed in the conductivity
experiment in solid H, (13). A better understanding of the effect
of clustering on K is therefore needed and more experiments
should be performed.
NOTEADDED I N PROOF: Very recently, Gorodilov et nl. (37)
have studied the effect of 0-Hz clustering on the thermal conductivity of Hz over the temperature range 1.5 5 T 5 3 K and
the concentration range 0.2% 5 X 5 4.4%. Under these conditions they found the conductivity to decrease upon clustering,
which is in the opposite direction from that reported in ref. 13.
The clustering reaction times were found to be consistent with
previous experiments.
FIG. 12. The concentration of single 0-HZ. (nn) pairs, and three
configurationsof clusters of three (nn) as a function of temperature for
X = 0.005. From Minchina et al. (7).

shown. As equilibrium is approached at a given T following


rapid cooling, CQ increases and Po decreases with time, and
this is consistent with the experiments (6, 7).
During the irreversible ortho-to-para (0-p) H, conversion
(16), the nuclear dipolar interaction between two (nn) o-Hz
molecules induces a proton spin flip, I = 1 + I = 0 , a transfer of
an angular momentum J = 1 to the lattice, and a release of
conversion energy, E
165 k B . The ortho-para conversion
rate, calculated for a random distribution o-H,, is a function of
the concentration of o-Hz in the nearest neighbor shell around a
central particle, and the rate equation is given by

where k is the rate constant, 0.019 h-I. In a "frozen" lattice, the


lost o-Hz in the cluster is not replaced, and as the number of
nearest neighbors decreases with time, the o-p conversion rate
gradually drops to zero (32). However, in a lattice where
particles can diffuse, an equilibrium distribution can be maintained, and the o-p conversion rate decreases with time
according to [23]. As T decreases, and clustering is favored, the
o-p conversion rate is increased above its value for a random
distribution. This was first shown for T = 1.5 K by Schmidt
(32), who presented a detailed discussion of the effect of
diffusion on the conversion rate. Still larger enhancements are
observed at lower temperatures (33).
Thermal conductivity in Hz is highest for pure p-Hz. The
presence of rotational excitations from the o-H2 impurities
introduces inelastic phonon scattering (34), and this has been
discussed by several authors (35,36). Measurements (13) of the
thermal conductivity K , for a sample with X = 0.03, as a

3.4. other examples ofquantum difluSion in H~


3.4a. Turzneling of HD in hexagonal close-packed HZ
Delrieu and Sullivan (23)
.
. have extended their theory on

three-body cyclic permutation processes in quantum crystals to


the quantum tunneling of HD impurities in hcp Hz. They have
estimated the tunneling frequency vo = J / h to be in the order of
1 kHz. Provided the concentration of the o-Hz impurities is
small enough, the motion of the HD impurities is that of
wavelike excitations that travel freely through the crystal and
are only scattered by the o-H, impurities. The displacement of
the HD is expected to lead to a motional narrowing of the HD
nmr line shape owing to the modulation of the nuclear magnetic
dipole-dipole interaction between HD and 0-HZ. Delrieu and
Sullivan (23)
. , have derived an exmession for the line width of the
HD proton resonance as a function of vn and of the o-H7
impurity concentration X. The situation for X << X,, the
impuriton regime, is particularly interesting. Here X, is a
critical concentration, estimated to be in the order of 0.03,
below which the HD impurities can travel quasi-freely and the
scattering probability by o-Hz is low. In this regime, the nmr
line shape is predicted to be Lorentzian. The line width is to be
proportional to X and lower than the width for the rigid particles. From the proportionality factor, the tunneling frequency
can be obtained.
The experimental situation has not been finalized yet. The
nmr absorption spectrum of Hz with 300ppm HD abundance
and X(o-H2) 5 0.02 showed (33) a very sharp line in the center,
which was attributed to HD. Spin-echo measurements (37)
showed the line shape to be Lorentzian and determined the line
width FWHM (full width at half maximum) from the transverse
relaxation time. However, the apparent proton line width of HD
was found to depend on the length of rf pulses used in the
spin-echo technique. For long pulses (-200 ks), the width
(open circles in Fig. 13) was much smaller than the statistical

1461

Can. J. Phys. Downloaded from www.nrcresearchpress.com by University of Wyoming Lib on 12/16/15


For personal use only.

MEY ER

FIG. 14. The normalized recombination rate K of H in solid HI


versus T - I . From Katunin c.1 nl. (4 1).

X
FIG. 13. The line width of H(HD) with -300ppm HD in H, crystal
IV versus the concentration X(o-Hz). The open circles are the results
for long rfpulses, while the closed circles are those for short rfpulses.
From Washburn el nl. (38) and Yu (39).

one calculated (33) for the rigid lattice (Fig. 13, curve labeled
ST) and the one calculated (33) from nmr data (3) of 0-HZ
impurities in p-Hz (line labeled B). Hence, it was believed that
motional narrowing was detected with a hopping frequency in
the order of 2kHz, which is consistent with the estimations.
However, Yu (39) has pointed out that the true line width must
be obtained from short-pulse sequences (< 10 p s , solid circles
in Fig. 13) to reflect the true dipolar interactions with the
surrounding HZ.For short pulses, experiments at Duke University have indicated only little exchange narrowing, as shown in
Fig. 13. Hence, Yu concludes that the hopping frequency is
insignificant. More experiments are necessary to resolve this
discrepancy with theory.
3.46. Recombination of H atoms in hexagonal close-packed

HZ
For a number of years, there has been an interest in studying
the interactions between atomic H embedded in a solid H,
matrix and the surrounding molecules. The first studies of the
recombination rate of H in this matrix as a function of
temperature were made by Leach (40). He produced H atoms in
HZby using pulsed electron injection from an accelerator into a
microwave cavity of an electron paramagnetic resonance (EPR)
spectrometer. The intensity of the EPR signal was then
measured as a function of time. It was found that in the
temperature range between 6 and 8 K, the recombination rate
was proportional to exp -(E,/kBT) with an activation energy
E,/kB = 200 K, which is the same as that for the diffusion of H2
in solid Hz (3). However, he found that at 4.2 K, the rate was
lo4 times faster than the extrapolation from higher temperature
indicated. He ascribed this to phonon-assisted tunneling characterized by a much lower activation energy. However, he did not
report on recombination measurements below 4.2 K.
A more complete study of the recombination rate was
reported by Katunin et al. (41). They condensed a beam,
consisting of a mixture of hydrogen atoms and molecules
produced in a radio-frequency dissociator, into a microwave
cavity of an EPR spectrometer. They found that above 4.5 K ,
the recombination rate was again proportional to exp -(E,/kBT)
but with E,/kB = 1 10 K, roughly half the value found by Leach
(39). At temperatures below 4.5 K, the recombination rate was
much larger than expected from such an activation energy, and
the authors concluded that diffusion of the H atoms had to

involve tunneling without activation energy, namely, quantum


tunneling. Their results for the recombination rate K, normalized to unity at 4.3 K, are shown in Fig. 14. The influence of
the quantum diffusion on the rate K has been discussed by
Kagan and Maksimov (31), who found that at low enough
temperatures, K is to be proportional to T, which is not
incdnsistent with the data o f ~ a i u n i net al. (41).
3 . 4 ~Depolarizatiotz
.
of p+ mesons in solid H2
Finally, attention is briefly drawn to an intriguing experiment
on the muon depolarization rate A in solid H2 with X 5 0.02 and
over the range of 1.5 to 2 0 K (42). In this experiment, p+
mesons from a beam are injected into a sample of solid HZ.
When captured, they most likely form H,p+ ions, and the
spin precession amplitude in a magnetic field is recorded as a
function of time. The decay rate is -0.04 ms-' in the liquid
state, where diffusion is highest, leading to a partial averaging
out of the proton nuclear magnetic fields that cause depolarization. Below the melting point, as T decreases, A increases and
passes through a maximum at T 4 K. This increase is possibly
caused by a slowing down of the diffusion. As the temperature is
further decreased, A sharply decreases. In this regime, the
H2p+ particle might possibly show the effects of impurity
motion by quantum diffusion, which would reduce the effects of
local fields from 0-Hz and decrease the depolarization rate.
Obviously experiments at lower temperatures would be very
desirable, where they can be correlated with the experiments
on quantum diffusion discussed above.

4. Summary and conclusions


Although a solid basis has been laid (1) for understanding the
quantum diffusion phenomenon in Hz, much needs to be done
both theoretically and experimentally to understand the most
important observable characteristics. The experimental situation can be summarized as follows.
Quantum diffusion is observed via the redistribution of single
o-Hz excitations or 0-Hz (nn) pairs in the lattice after a change in
temperature. Experiments of various kinds record an exponential transient to equilibrium characterized by a time constant, T.
There is a fair amount of disparity among the determinations of
the various research groups. These differences are much larger
than the uncertainty in the individual experiments, and are
possibly caused by different strains in the crystal that might
influence the hopping rate.
The observations show the following:
(i) At temperatures above 4 K , equilibration should be
reached in less than -2 h. At least for T 2 0 . 3 K, T increases
with decreasing temperature, roughly proportional to T-" with
0.7 5 n 5 1 .
(ii) This temperature dependence is probably determined

Can. J. Phys. Downloaded from www.nrcresearchpress.com by University of Wyoming Lib on 12/16/15


For personal use only.

1462

CAN. J. PHYS. VOL. 65, 1987

mainly by the energy exchanges needed to pass from (nnn) to


(nn) pair states and vice versa.
(iii) However, near 0 . 3 K , 7 is found to pass through a broad
maximum and to decrease with T. This reaction-time decrease is
observed for both the decay of the single O-Hzexcitations and of
the pairs. There is no theory for these puzzling observations that
indicate a new mechanism permitting a faster clustering,
different from the one chiefly responsible for the behavior above
0.3 K .
(iv) In spite of inconsistencies between various experiments,
it appears that at least at temperatures above 1 K , 7 decreases
with increasing X for X > 1%, in agreement with Van
Kranendonk's models. Hence, it appears that even at values of
X greater than a few percent, the EQQ interaction does not
quench the diffusive process. The crystal does not remain a
"frozen alloy" but is permitted to tend towards thermodynamic
equilibrium-i.e.,
minimization of the free energy-within
a
few hours. This clustering produces an observable increase of
the ortho-para conversion rate.
(v) For T 5 1.5 K , it appears that the clustering rate becomes
independent of X. at least for X < 0.0 1.
More experiments to determine 7 at intermediate concentrations, say X > 0.02, are needed over the range T > 0.05 K . The
easiest experiments are those of pressure at constant volume and
of thermal conductivity, both carried out as a function of time
after a rapid temperature change. In resonance experiments, the
line widths become too broad in this concentration range for
unambiguous measurements. Experiments where hydrostatic
pressure is applied would give the dependence of 7 on density
for a given Hz sample-for
nrnr measurements, single crystals
are necessary.
It is also suggested that nrnr experiments be performed on
single crystals with X < 5 x
using a dilution refrigerator
capable of cooling the sample to 1 0 m K . A heat-switch
arrangement is necessary to permit rapid cooling of the sample if
the short clustering times below T = 0.08 K are to b e measured
more accurately.
On the theoretical front, more progress is needed to understand quantitatively the various features of the redistribution
reaction time. The system of very dilute 0-H2 impurities in a
p-Hz crystal (say, X < 5 x
seems an attractive subject
for computer-simulation studies, where only the E Q Q interactions need to be considered. As these are kno-wn, the progress of
two excitations diffusing through the lattice could be followed
and an average of the reaction time calculated over a number of
such processes.

Acknowledgments
Support of this research by the National Science Foundation
through Grant D M R 8516156 is gratefully acknowledged. The
author has greatly appreciated detailed comments and suggestions on this manuscript by R. G . Palmer and J. van Kranendonk. Stimulating conversations with W. N . Hardy, N . S.
Sullivan, and H. G. Robinson are also acknowledged as well as
R. P. Behringer's comments o n an early draft. H . Ishimoto
attracted the author's attention to ref. 41.
1. J. VANKRANENDONK.
Solid hydrogen. Plenum Press, New York,
NY. 1983. Chap. 9.
2. D. C. AILION.I n Advances in magnetic resonance. Vol. 5. Edited
by J. S. Waugh. Pergamon Press Ltd., London, England. 197 1.
3. R. F. BUZERAK,
M. CHAN,and H. MEYER.J. LOWTemp. Phys.
28, 415 (1977).

4. L. I. AMSTUTZ,
J. R. THOMPSON,
and H. MEYER.
Phys. Rev. Lett.
21, 1175 (1968).
5. S. WASHBURN,
R. SCHWEIZER,
and H. MEYER.J. LOWTemp.
Phys. 40, 187 (1980).
6. D. RAMMand H. MEYER.J. LOWTemp. Phys. 40, 173 (1980).
7. I. YA.MINCHINA,
M. I. BAGATSKII,
V. G. MANZHELII,
and A. I.
KRIVCHIKOV.
Fiz. Nizk. Temp. (Kiev), 10, 1051 (1984); Sov. J.
Low Temp. Phys. (Engl. Transl.), 10, 549 ( 1984).
8. H. MIYAGI.Prog. Theor. Phys. 40, 1448 (1968).
9. H. MEYER.Phys. Rev. 187, 1173 (1969).
10. S. A. BOGGSand H. L. WELSH.Can. J. Phys. 51, 1910 (1973).
11. B. J. ROFFEY,S. A. BOGGS,and H. L. WELSH.Can. J. Phys. 52,
2451 (1974).
and A. B. HARRIS.
Can. J. Phys.
12. W. N. HARDY,
A. J . BERLINSKY,
55, 1150 (1977).
and H. MEYER.J. Low Temp. Phys. 57,265 (1984).
13. M. CALKINS
14. R. OYARZUN
and J. VANKRANENDONK.
Phys. Rev. Lett. 26,646
(1971).
15. R. OYARZUN
and J. VANKRANENDONK.
Can. J. Phys. 50, 1494
(1972).
Rev. Mod. Phys. 52, 393 (1980).
16. I. F. SILVERA.
17. A. B. HARRIS.Phys. Rev. B, 2, 3495 (1970).
H. MEYER,S. M. MYERS,and R. L. MILLS.J.
18. L. I. AMSTUTZ,
Phys. Chem. Solids, 30, 2693 (1969). C. C. SUNG.Phys. Rev.
167, 271 (1968).
19. C. C. SUNG.Phys. Rev. 167, 271 (1968).
Prog. Theor. Phys. 54,
20. M. FUJIO,J. HAMA,and T. NAKAMURA.
and T. NAKAMURA.
Prog.
293 (1975); J. HAMA,M. INUZUKA,
Theor. Phys. 44, 303 (1970).
and H. MEYER.Phys. Rev. B, 7, 2974 (1973).
21. F. WEINHAUS
22. R. A. GUYERand L. I. ZANE.Phys Rev. 188,445 (1969).
23. J. M. DELRIEUand N. S. SULLIVAN.
Phys. Rev. B, 23, 3197
(1981).
and M. A. STRZHEMECHNYI.
SOV.J. LOW
24. V. B. KOKSHENEV
Temp. Phys. 2, 529 (1976).
25. J. VAN.KRANENDONK.
J. LOWTemp. Phys. 39, 689 (1980).
M. I. BAGATSKII,
V. G. MANZHELII,
and A. I.
26. I. YA MINCHINA,
Fiz. Nizk. Temp. (Kiev) 11, 665 (1985); Sov. J.
KRIVCHIKOV.
Low Temp. Phys. (Engl. Transl.), 11, 366 (1985).
27. J. KOHL.Ph.D. thesis. Ohio State University, Columbus, OH.
1976.
28. R. G. ROBERTSand J. G. DAUNT.J. LOWTemp. Phys. 6, 97
(1972).
29. A. F. ANDREEV
and I. M. LIFSHITZ.
Zh. Eksp. Teor. Fiz. 56,2057
(1969).
J. Phys. Colloq. 39, C6-1257 (1978).
30. A. F. ANDREEV.
Zh. Eksp. Teor. Fiz. 84, 792
31. Yu KAGANand L. A. MAKSIMOV.
(1983); Sov. Phys. JETP (Engl. Transl.), 57, 459 (1983).
32. F. SCHMIDT.
Phys. Ref. B, 10, 4480 (1974).
33. R. SCHWEIZER,
S. WASHBURN,
and H. MEYER.J. LOWTemp.
Phys. 37, 290 (1979).
34. R. G. BOHNand C. F. MATE.Phys. Rev. B, 2, 2121 (1970).
35. C. EBNERand C. C. SUNG.Phys. Rev. B, 2, 21 15 (1970).
36. J. H. CONSTABLE
and J . R. GAINES.Phys. Rev. B, 8,3966 (1973).
I. N. KRUPSKII,V. G. MANZHELII,and
37. B. YA. GORODILOV,
0 . A. KOROLYUK.
Fiz. Nizk. Temp. (Kiev), 12,326 (1986); Sov.
J. Low Temp. Phys. (Engl. Transl.), 12, 186 (1986).
and H. MEYER.J. LOWTemp.
38. S. WASHBURN,
R. SCHWEIZER;
Phys. 45, 167 (1981). .
39. I. Yu. J. Low Temp. Phys. 60, 425 (1985).
40. R. K. LEACH.Ph.D. thesis. University of Wisconsin, Madison,
WI. 1972.
et al. Pis'ma Zh. Eksp. Teor. Fiz. 34, 375
41. A. YA. KATUNIN
(1981). JETP Lett. (Engl. Transl.), 34, 357 (1981).
et al. Pis'ma Zh. Eksp. Teor. Fiz. 41,275 (1985);
42. S. G. BARSOV
JETP Lett. (Engl. Transl.). 41, 339 (1985); Hyperfine Interact.
32, 557 (1986).

You might also like