150 D ISP LACE M ENT SP I KES IN CUBI C M ETAI. S.

I
J. A. Mehne. The author is greatly indebted to
C. Erginsoy and G. H. Vineyard for their provision of
GRApE program decks and to R. A. Johnson for his 104
provision of "Johnson's Program" decks. Technical
discussions with J. W. Corbett, C. Erginsoy, R. A.
Johnson, B. Ralph, W. F. Schilling, G. H. Vineyard, o 103
and T. O. Ziebold were particularly rewarding.
C
Z 10~
APPENDIX
An interatomic potential for tungsten was con-
10
structed which had the same form as does Potential III
of Erginsoy et u/. ' for o.-iron. It consisted of an expo-
nentially screened Coulomb potential for r(1
A joined
to a simple exponential for r)1 A
1

ATOMIC SEPARATION, A
2

P(r) = (24500/r) exp( — 4.55r), r(1 A; FIG. 21. Tungsten potential used in the present study compared
P(r) =24500 exp( — 4.55r), r) 1 A. with two
tungsten.
high-energy and two low-energy potentials for

This potential is compared with the Bohr, Girifalco- Fig. 21. At small separations distances (r(0.4 A)g(r)
Weizer, 4' Johnson, "
and Thomas-Fermi potentials in fits the Thomas-Fermi potential for tungsten. It is
equal to the Morse potential of Girifalco and %eizer
"L. A. Girifalco and V. G. Weizer, Phys. Rev. 114, 687 (1959). a, t r= 1 A and to Johnson's potential at 2.2 A.

P I YSI CAL REVI EW VOLUME 150, NUM B ER 2 14 OCTO B ER 1966

Crystal Dynamics of Potassium. I. Pseudopotential Analysis

of Phonon Dispersion Curves at O'K
R. A. CowLEY, A. D. B. WooDs, AM) G. DQLLING
Chalk River Nuclear Laboratories, Chalk River, Ontario, Canada
(Received 23 May 1966)

The frequencies of normal modes of vibration of potassium at 9'K have been measured by inelastic-
neutron-scattering techniques. Certain selected frequencies are (units 10" cps): H&5 2.21+0.02, P4 1.78&
&0.02, N1' 2.40&0.04, N3' 1.50%0.025, and N4' 0.53~0.02. The results are very similar, apart from a scale
factor 1.65, to those for sodium. Analysis of the results has been carried out in terms both of conventional
Born-von Karman models and of potential functions defined in reciprocal space. A fifth-neighbor, axially
symmetric force model has been used to compute the frequency distribution function for the normal modes
and the associated heat capacity. The reciprocal-space analysis was performed in two ways: (a) in terms of a
total potential function, whose Fourier transform is the effective interatomic potential between "neutral
pseudo-atoms" of potassium, and (b) in terms of the screened pseudopotential for the conduction-electron-
ion interaction. Analysis (a) shows that a wide variety of interatomic potentials, both with and without
long-range oscillatory character, can be found which give a satisfactory fit to the results. These potentials
are compared with those obtained from an analysis of x-ray scattering data for liquid potassium. The
pseudopotentials obtained from analysis (b) are in good agreement with that derived by Bortolani from the
Heine-Abarenkov model. A reanalysis of the phonon dispersion curves for sodium leads to very similar con-
clusions, confirming earlier work by Cochran.

I. INTRODUCTION pally on the alkali metals. ' Measurements of v(q) for

sodium' showed that while quite long-range effective
'HE crystaldynamics of metals has received con- forces were present, they were not very strong, and no
siderable experimental and theoretical attention
over the past decade. A large amount of experimental
' For a general review of phonon dispersion curves in metals see
G. Dolling and A. D. B. Woods, in Therma/ Neltron Scattering,
information on dispersion relations v(q) of frequency v edited by P. A. Egelstaff (Academic Press Inc. , New York,
versus wave vector q exists for a number of metals, 1965), Chap. V.
~A. D. B. Woods, B. N. Brockhouse, R. H. March, A. T.
while theoretical attention has been concentrated princi- Stewart, and R. Bowers& Phys. Rev. 128, 1112 (1962).
 CORLE Y, WOODS, AND DOLL IN C

definite Kohn anomalies, ' such as had been observed action in metals. It is further intended that these experi-
in lead, ' and other metals, ' were present. Analysis of the mental measurements and theoretical calculations will
dispersion curves in terms of a pseudopotential was serve as the basis for more extensive studies of the effect
carried out by Cochran' ' and resulted in an effective of temperature on the phonon spectrum.
interatomic potential. At large distances this potential
has an oscillatory character, and Koenig' showed that 2. EXPERIMENT
this could arise from the discontinuity in the slope of
The phonon dispersion curves have been measured
the dielectric screening function, i.e., another aspect of
along the high symmetry lines A, Z, A(F), D, and 6
the Kohn effect. Cochran' suggested, however, that the
oscillations might result from his method of calculating
(in previous notation' [00$], [g0], [-', —',f], and [gl],
the effective potential. TABLE I. Normal-mode frequencies (units 10's cps) in potassium
Attempts to improve the accuracy of the sodium v(tI) at 9'K. The reduced wave-vector coordinates g are expressed
in units of (a/2v), where a is the lattice constant.
and to measure neutron-scattering line shapes as a (001')tery
function of temperature' ' were made dificult by the
large incoherent neutron-scattering cross section for
LOOI/A, pg'Igh. 3, 1'3
sodium. In addition, the occurrence of the martensitic
transformation in sodium made it necessary to carry 0.15 0.66 ~0.04 0.52~0.03 ~ ~ ~

0.20 0.89 &0.02' 0.68~0.025 0.77 +0.05

out the experiments above 40'K, where it is possible 0.25 1.08 ~0.05 0.82+0.03 1.00 ~0.06
that line shapes and frequencies are already affected 0.30 1.21 +0.03 0.99%0.02 1,20 ~0.06
by anharmonic interactions. Because it was felt to be 0.35 1.35 +0.04 1.15+0.03 ~ ~ ~

0.40 1.53 ~0.03 1.28' 0.02; 1.53 ~0.05

desirable to have a large body of precise information 0.45 1.63 +0.03 1.44~0.03 1.68 +0.04
about all aspects of the lattice dynamics of an alkali 0.50 1.74 ~0.035 1.57a0.02 1.78g~0.02
0.55 1.81 ~0.04 1.69+0.02 1.91 ~0.04
metal, it was decided to pursue the measurements on 0.60 1.89 +0.03 1.79+0.02' 2.00 +0,04
a more favorable material; potassium was chosen for 0.65 1.96 +0.06 1.89~0.02 2.05 ~0.06
several reasons: (1) The ratio of coherent to incoherent 0.70 1.965+0.04 1.99~0.02 2.09 ~0.03
0.75 2.08g+0.03 2.07+0.03 2.12 ~0.04
scattering cross sections is about two and a half times 0.80 2.11 ~0.025 2.11+0.025 2.16 +0.03
larger than for sodium. (2) Its electronic properties are 0.85 2.15 ~0.03 2.19~0.04 2.19 ~0.03
nearly ideally free-electron-like, in common with sodium 0.90 2.19 +0.025 2.20+0.03 2. 18 ~0.02'
0.95 2.21 ~0.025 2.21+0.04 2, 22 ~0.03
hut in contrast with many other metals. (3) No marten- 1.00 2.21 +0.02 ~ ~ ~

sitic transformation exists; thus experiments can be

Lk'sl 1Dr I:l l 1'1D4
done at low temperatures. (4) Large single crystals
were available. " 0.10
0.20
1.52 &0.04
1.62 ~0.04
2.35~0.06
2.30~0.035
0.70 ~0.03
0.97 &0.03
Accordingly, the dispersion curves have been measured 0.30 1.69 +0.04 2.20~0.05 1.27 ~0.03
at a temperature of 9'K and the results interpreted on 0.40 1.72 ~0.04 2.01+0,04 1.55 ~0.04
the basis of a pseudopotential similar to that derived LÃ0j»
for sodium. The results have also been fitted to a 0.05 ~ ~ ~
0.53 +0.03
Born — von Karman model which was then used to 0.10 0.74 ~0.02 ~ ~ ~
0.95 ~0.02
calculate the density of phonon states, g(v), and hence 0.15 1.08 ~0.03 0.67~0.03 1.34 ~0.04
0.20 1.41 ~0.02 0.93~0.03 1.695+0.02
thermodynamic properties such as the heat capacity. 0.25 1.69 ~0.05 1.11~0.03 1.95 +0.04
A large part of this paper describes the methods used 0.30 1.94 ~0.03 1.23&0,03 2. 10 &0.02'
in this analysis, and the results of several different 0.35 2.08 ~0.05 1.36~0.03 2.19 ~0.05
0.40 2.25 +0.035 1.44+0.025 2.15 +0.025
approaches to calculating an effective interatomic po- 0.45 2.33 +0.06 1.49~0.05 2.03 ~0.06
tential as well as the pseudopotential for electron- 0.50 2.40 ~0.04 1.50~0.025
phonon scattering. It is hoped that these results will
assist in the understanding of electron-phonon inter- Lrl 17Gr LZIJGs [fR7P&'
0.05 2.22~0.05 2.18 ~0.05 0.55 1.55 ~0.04
e W. Kohn, Phys. Rev. Letters 2, 393 (1959). 0.10 2.14~0.04 2.13 ~0.04 0.58 1.38 ~0.05
4 B. N. Brockhouse, T. Arase, G. Caglioti, K. R. Rao, and A. D. 0.15 2.01+0.06 2. 10 +0.035 0.60 1.24 ~0.03
B. Woods, Phys. Rev. 128, 1099 (1962). 0.20 1.78w0. 03 2.02 ~0.035 0.62 1.17 ~0.03
' W. Cochran, Proc. Roy. Soc. (London) A276, 308 (1963).
W. Cochran, in Irtelastic Scatterirtg of ItIeutroIts in Solids and 0.25 1.55~0.04 1.895&0.03 0.65 1.04 ~0.025
Liglids (International Atomic Energy Agency, Vienna, 1965),
0.30 1.31~0.04 1.78 ~0.02' 0.68 1.005~0.02'
Vol. I, p. 3.
0.35 1.07+0.04 1.70 ~0.04 0.70 1.02 a0.03
' S. H. Koenig, Phys. Rev. 135, A1693 (1964). 0.40 0.81~0.03 1.57 ~0.03 0.72 1.06 ~0.03
SA. D. B. Woods, B. N. Brockhouse, R. H. March, and R. 0.45 0.61+0.03 1.53 ~0.04 0.75 1.20 ~0.03
Bowers, Bull. Am. Phys. Soc. 6, 261 (1961).
0.50 0.53+0.02 0.78 1.36 +0.03
' See B. N. Brockhouse, A. D. B. Woods, G. Dolling, and I. M. 0.80 1.47 ~0.02
Thorson, in I'roceedings of the Third United Sutions International
0.85 1.77 ~0.03
0.90 1.99 ~0.03
Coeferertce oN the Peaceful Uses of Atomic Ertergy (United Nations, 0.95 2.19 ~0.04
New York, 1965), Vol. I, Paper P/12, p. 419.
' We are extremely grateful to P. A. Penz for growing two
excellent specimens for the purpose of these experiments. a COntinuatiOn of A branCh.
 CRYSTAL DYNAM I CS OF K 489

F
H ~cp & 0
2.0- 2.0

0—
1,
g o I.O
Fro. 1. Measured dis-
persion curves for po-
tassium at 9'K. i'(2s/u) EO
is a reduced wave-vector CL
coordinate. In the L00i ), CP

Pi 10), and (iTI') direc- O I I I I

oo
tions, transverse (longi- 0 0.2 0.4 0.6 0.8 I.O I.O 0,8 0,6 0.4 0.2
=
tudinal) modes of vibra- lOOh i llRl
tion are denoted by Is
I I I
I I I I
circles (triangles). The I

solid curves represent I

the best 6t to the re- N,
sults on the basis of H
a Born —von Karma, n
model with axially sym-
.0 I5
metric forces extending
to fifth-nearest-neighbor
atoms.

/ G,
—I.O
I.o—

N N H

0 ! I 1 I l I I I I l

0 08 0 0.2 0.4 0.6 0.8 I.O 0.5 0

iffol = i~, ~ ll—
REDUCED WAVE VECTOR COORDINATE

L|$1$, respectively) at 9'K, using the Chalk River shapes of the curves in this region; they are very close
triple-axis crystal spectrometer in its constant-mo- together, and the possibility of double-scattering proc-
mentum-transfer (constant Q) mode of operation. All " esses" introduces an uncertainty in the identification of
of the measurements were carried out by observation observed peaks with particular modes. With the possible
of neutron-energy-loss processes with variable incident exception of these branches near )=0.75, the curves are
energy; the (aluminum) analyzer was set to observe quite smooth and show no evidence of any Kohn
several neutron energies in the region of 0.012 eV. anomalies.
The results are shown in Fig. 1 and Table I. The fitted
curves shown in I'ig. 1 will be discussed in Sec. 3. In-
3. ANALYSIS OF RESULTS
spection reveals that, apart from a scale factor of The measured normal-mode frequencies have been
approximately 1.65, they are very similar in shape to analyzed in two ways: (a) using the method of Born
the corresponding curves for sodium' a not unexpected— and von Karman'4 to obtain interatomic force constants
result. The same kind of correspondence, though not (in real space) between near-neighbor atoms, and (b) in
quite so close, exists between these results and those terms of potentials defined in reciprocal space, in a
for the alloy P-brass (CuZn). " similar manner to that employed by Cochran' for
In common with the sodium dispersion curves the sodium.
hr and Ds branches appear to cross near t =0 75 It
difFicult to confirm this point, however, and we are
is. .
A. Born —
von Karman Analysis
hesitant about drawing any conclusions about the The processes of setting up equations of motion for
the atoms in a monatomic bcc crystal, and of the re-
"B. N. Brockhouse, in Inelastic Scattering of Eeltrons in Solids
and Liglids (International Atomic Energy Agency, Vienna, "' Reference
1, p. 196, and Ref. 4.
1961), p. 113. M. Born and K. Huang, Dynamical Theory of Crystal Lattices
"G. Gilat and G. Doiling, Phys. Rev. 138, A1053 (1965). (Clarendon Press, Oxford, England, 1954).
490 COWLEY, WOODS
DS, AND
A DOLLING
TABLE II.
orce constan K
5 - e ghb
5th nearest-ne' H y symmetric force
t s a;, P; is that of Ref 2 ~ ~
IO0

Neighbor (f) Radial Tangential

1 2576 —109 786 895 0
2 432 29 432 29
3 95 12 —41 12 90
3 4 2
5 15 2 6 4
0
~ ~

duction of these
ese equations
e to a 3 matrix 80—
I I I I I
'
e 1 0 IO 20
TE MP ERAT URE eK
norTI1
IG. 3. Equivalent De ure s or potass
tt db th oo0
thod o ) Fi . mpare
pproxsmation from the d'
- '
imental heat-capacity
with experimental
cmental daata
lilvo vln ax
A-S~" f '
atoms. The e general tensor n ing to 5th-nearest-n s. e best-fit
-S forces cant pro veme tl values for the model (one
g
h" 'd w'th one "tangenti 1"
r are given in T bl g er with the
'
3rd-n }1bors. Inclusion of
or force
bor f constants t"n'h 'ps between these
tl li d R f. 5. Th o the
d t i ill t t d b y th e curves in Fiig.. 1.. requenc
a modes has
s opes pu e by means of th e extra p ol
of th a wave vectors Th u t of sampling
'
vau i canty(8 y0 fo
orC~~, forexam le
those obtained from illo i zone is shown in Fi ls unction lt
e , sa is actor fi st ightfd e e variation of th e eat
ccapacity with tern
emperature in tehrmo c
emperature for which
l i
mo el is appropriate g t
hhh
I

p "yo
e equiva-
I. 5—
emperatures obbtained from th e ca1culated
e h eat capaciti y, asshownin Fig.. 3 , agiee
and observed
to better than 2 g
I.O—

mpare theoretical and e

'va ent
'
s v erived" from
0.5— Thea r urve o
crepancy larger than
perimental ror. Fiigure
error. 4 also show
te onl y to
0 2.0 2.5 t e elastic constants. The
0 0.5 I.O I.5 g
l2
FREQUENCY
d consistencies betwee
requency distnbution
ot s t9'K , calculated
c from the mo c nt t Thi do
~ 'y' py"
466 176 ~ormsa mo d es was empl oye d to n is in error, since the
he ultrasonic
isto gram.

T. Wolfram, G. W. L
15
Rev 129 2483 (196
arquardt
L,
and J. Trivisonno,
DeWames,

J. Ph ys.s. Chem. Solids

Ph ys.
"G. Gilat
'

"Th-.i.
nd G. Boiling, Phys.
and
. Martin ' Phys. . Re .
TB8
. H. K. Barron,nn, W.. T.
'
'. 'sI
8,
I.ett
o( ), and ref erences
orrison, Proc. Roy
6, 273 (1965). So . (Lo do ) A242
 CRYSTAL D YNAM I CS OF K
I I I I I I I I I .
I I I I I I I
lattice of the crystal, and q+~= Q. Following Cochran
2.4 we may subdivide this expression for the dynamical
matrix element into three contributions
D s(q) =Dan+Do+DE. (6)
o 22
C4
O
These terms represent contributions arising from short-
I- 2.I—
range overlap forces, long-range electrostatic forces be-
tween the bare ions, and interactions involving the
2.0— conduction electrons and the ions, respectively. Recent
work by Vosko" has strongly suggested that D~ is very
I~
9- &, --&'-
small indeed for both Na and K, and in most of our
I.s I I I I I I I I I I I calculations we have neglected this contribution. (Intro-
6 9 I2 duction of a small overlap force produces no significant
change in any of the results discussed below. ) The
F&G. 4. Equivalent Debye frequencies derived from the moments
of the distribution of Fig. 2, dashed curve, compared with experi- "Coulomb" contribution D~ is easily calculated with
mental heat-capacity data (Ref. 18). The solid curve shows the the help of the Ewald 8 transformation, leaving only
eBect of utilizing a force model fitted only to the results in Fig. 1,
ignoring the measured elastic constants.
the conduction-electron term DE to be dealt with. Two
different methods of analysis suggest themselves:
measurements refer to vibration frequencies &10' cps, (a) To subtract from the observed phonon frequencies
far below the lowest value (5&(10"cps) obtained in the Li.e., the D s(q)$ the known contribution Do, as was
done by Cochran for sodium, ' and to fit the remainder
neutron-scattering experiments. At 5'K, the heat
to some function V~(Q) corresponding to the contribu-
capacity is dominated by modes of intermediate fre-
tion DE. A refinement of this procedure is to assume a
quencies (&10"cps).
local pseudopotential model for the conduction electron-
B. Reciprocal-Space Analysis ion interaction: If V„(Q) is the Fourier transform of the
local pseudopotential (we omit the vector sign to
The normal-mode frequencies v are given by the
emphasize the axial symmetry of the potentials), and
equation'4
e(Q) is the dielectric function, we have
4m'mv'U. (q) = Q D. Us(q),
(qs)
V (Q) = (e/4~e')Q'(1 —e '(Q)) V. (Q) I'
I

where
D s(q)=p g s(O, t) exp(iq r~); (2) The experimental results are then fitted to the screened
potential V~(Q)/e(Q). The advantage of this is that,
as shown by recent calculations of Animalu' and by
& (q) specifies the polarization of the mode, p &(O, t)
Bortolani s' V„(Q)/e(Q) is of oscillating sign. Thus,
represents force constant linking the atom at an origin
while V (Q) is always non-negative, it displays cusp-
(0) with that at position r~. We assume an axially like singularities at Q values for which V„(Q) =0. It is
symmetric two-body interaction V'(r) between the
therefore much easier and more satisfactory to 6t the
atoms, dependent only on the distance between them:
experimental data with the fairly well-behaved screened

y., (o, t) =
d2V r() potential than with VE(Q) itself.
(b) To analyze the total D, (qs) in terms of a potential
du. (0)dgp(t) function V (Q), using Eq. (5). This total potential is
(3)
y., (0,0) = —p y., (o,t), two the I'ourier transform of the effective ion-ion potential
in real space, whose first and second derivatives at the
where I(t) is the displacement from equilibrium of the interatomic distances give the conventional Born —von
atom t. We now express V(r) in terms of its Fourier Karman force constants. We might refer to this as the
transform potential between "neutral pseudo-atoms". '4
Both methods of analysis have been carried. out in
V(r) = (2rr) 'v V(Q) exp( iQ r)dQ—, (4) the same manner, as follows: It is convenient to define
new functions GE(Q) and Gr(Q):
where v is the volume of the unit cell and V(Q) depends
tp2 e2
only on Q I. Substituting Eqs. (3) and (4) into (2), and =
I

taking into account the term P„s(0,0) we obtain"

V'(Q) '(Q); V'(Q)= '(Q)' (g)
Q2e Qse
D.s(q) = Z L(q-+r-) (qs+re) V (q+~) r-ruV (~)3,—
(5) In the first instance the functions G~(Q) and Gr(Q)

where ~= (r, rp, rr) represents a vector of the reciprocal

"S.
" H. Vosko, Phys. Letters 13, 97 (1964).
A. O. E. Animalu, Technical Report No. 3, Solid State Theory
Group, Cavendish Laboratory, Cambridge, 1965 (unpublished).
20 D. 2' V. Bortolani (private communication).
Pines, E/ementary L&'recitations its Sohds (W. A. Benjamin. ,
Inc. , New York, 1963), p. 26. '4
J, M. Ziman, Advan. Phys. 13, 89 (1964).
492 COWLEY, WOODS, AND DOLLI N6 l$0

were specified by a Fourier series of up to 30 terms: Ryd

.-( -- (-Q/Q ..)&(-+ Z

I I I I I I I I
r 1 1 1 I ERRORS IN CURVES
C and D
0.02—
(Q)=
A odd V (Q)

2 ~. 1 —--(. Q/Q, . )j,

%max

+ n=l t .. & (Q) -0.02

-0.04
-0.06
ANI ET AL
U-HEINE
CURVE(eARDEENI
CURVE (H-a)
where Q, is some upper limit beyond which the two -0.0B
functions G(Q) are assumed to be negligible. is a2, -O. 0
I Y CUT-OFF-
constant, equal to 47r for G~(Q) and zero for Gr(Q). The URVES Cando

form of the first part of this expression was chosen so -0,12 = I I

I.O 2,0 3.0 4.0

that the G(Q) would tend smoothly to their respective WAVE VECTOR aQ/2n.
limits at Q=O and Q, „.The coeKcients A„were then I I
—I.O
obtained by the method of linear least-squares fitting to
either D s(=Dc+Dan) or D~, as the case may be. This 10 0.8
G(Q)
procedure, although most efficient, was found to be un- -0.6
satisfactory owing to the existence of "ripples" in the -0.4
fitted G functions arising from the large e terms in the -0.2
Fourier series. These fitted G functions could, however,
be used as the first stage of a nonlinear least-squares I I I I

fitting procedure in which the function is specified by 0 I.O 20 3,0 4.0

a table of values at between 10 and 30 selected Q values. WAVE VECTOR aQ /2m.

Values of G at intermediate Q=Q, are found by Fio. 5. The screened pseudopotential V~(Q) je(Q) for the con-
duction electron-ion interaction in potassium at 9'K (upper
Lagrange (cubic) interpolation, using the two nearest
diagram). A and 8 are theoretical curves based on the Heine-
tabulated values on either side of Q;. The same method Abarenkov model, while C and D are 6tted to the results of Fig. 1,
was used in the process of fitting D~ by means of a tabu- as described in the text. The lower diagram shows the function
G'-"(Q) derived from curves A (solid) and C (dotted) above. The
lated V„(Q)/e(Q) function. By definition, G~(0)=4~ large wave-vector region is shown on a 10)& scale on the right.
and Gr (0) = 0, and the screened pseudopotential
V~(Q)/e(Q) tends to a finite value at Q=O. We assume a The results of these least-squares fitting calculations
value Q=Q, , beyond which GE(Q) =Gr(Q) = V„(Q)/ for potassium, with (a/2')Q, chosen to be 2.4, are
e(Q) =0. The fitting process is initiated by specifying a shown in Fig. 5 (upper diagram) both for the Bardeen
table of values for the function concerned at Q values
(curve C) and Heine-Abarenkov (curve D) dielectric
between 0 and Q, „, and imposing the condition that functions. Two screened pseudopotentials calculated on
the function tend smoothly to zero at Q,
. These the basis of the Heine-Abarenkov modeP2" are also
tabular values constitute the variable parameters of the shown in curves A and B, respectively. The downward
least-squares fit. The phonon frequencies for any given
reduced wave vector q may now be computed from
kink in the curves C and D near Q,
results from the
condition that V~/e goes smoothly to zero at Q„„.As
Eqs. (1) and (5) through (8). The computation is
simplified if q lies along a direction of high symmetry,
Q, „ is changed, the quality of the fit to the phonon

since the dynamical matrix factorizes into three equa-

frequencies increases as Q,„
increases, but the same
tions involving linear combinations of the D p(q). The
general shape of curve results. If (a/27r)Q, is increa, sed „
beyond 2.4, then the calculated errors of the tabulated
number of lattice points g involved in the summation
values of the curve increases, showing that there are
of Eq. (5) becomes very large as Q, increases, as does Insufhcient data to determine them.
the number of tabular values required for adequate The phonon frequencies calculated from curves C and
specification of the function. These factors effectively D are an excellent fit to the measurements, similar to
„
restrict the choice of (a/2~)Q, to less than about 2.8; that shown in Fig. 1. As would be expected from the
indeed, there seem to be insufhcient data at present to close similarity of our curve, D, with that of Bortolani, "
specify the G functions much beyond 2.4. (For com-
parison, the calculations previously carried out' for Na
A, and the contrast with that of Animalu, B, the fit "
to experiment given by use of A is substantially better
assumed a cutoff at aQ/2~= v2. ) Calculation of phonon
than that obtained with B. The lower part of Fig. 5
frequencies from the screened pseudopotential involves shows the function G~(Q) derived from curves A and C
a knowledge of the dielectric function e(Q). We have above. Note the enlarged scale for aQ/2ir) 1 to show the
employed both the expression given by Bardeen" and
also that proposed by Heine and Abarenkov, which " details more clearly.
In Fig. 6 we show the results of calculations of the
incorporates certain corrections (e.g. , for exchange
effects) to the simple Bardeen formula.
function Gr(Q) for various Q,
. All these curves
provide a satisfactory fit to the experimental data, but
the increase in size of error bars as Q, increases is much ,„
"J.V.Bardeen,
s5 Phys. Rev. 52, 689 (1937).
Heine and I. Abarenkov, Phil. Mag. 9, 451 (1964). more pronounced than in the screened-pseudopotential
 CRYSTAL D YNAM I CS OF K
I
I
I
at the atomic positions are simply related to the force
CUT-OFF VALUES
constants of a conventional Born — von Karman analysis.
A I60
FIG. 6. The total po- l6— 8 200
The derivatives shown in Fig. 7 are in good agreement
tential function G~(Q) O
C 2.40 with the force constants obtained in the Born — von
derived from the results
of Fig. 1, for various
C3 Karman analysis of Sec. 3 A. After the initial oscillation
values of the cutoB wave U
l2— at the near-neighbor distance, V(r) falls smoothly and
vector Q ME for potas- monotonically to zero, in marked contrast to the inter-
sium at 9'K. Typical I-
error bars are indicated UJ 8— atomic potential deduced" for liquid potassium from
at arbitrary intervals. I-
O x-ray scattering data. Transformation of curve C of
For (o/2s)Q~)2. 4, the CL

error bars become very Fig. 6, however, leads to a long-range oscillatory po-
large, indicating insufB- I 4 tential V(r), similar in character to the dashed curve
cient data to specify the in Fig. 7. Thus, in the case of solid potassium, it is
function.
0 I
possible to construct long-range oscillatory potentials
0 IO 2.0 3.0 which are consistent with the measured phonon fre-
WAVE VECTOR aQ/2v quencies. However, such potentials are certainly not
necessary and seem less physically plausible than the
calculations. As mentioned earlier, the Fourier trans- potential shown as a solid line in Fig. 7.
forms of these curves and their derivatives represent
effective potentials (and derivatives thereof) in real Ry
'. x. l i I' ~
I
ERRORS
I
I N
I
CURVE D
I

space between the potassium atoms, including con- 0.02-

8/'
duction-electron effects. The existence of many different / /
functions Gr(Q), all of which fit the experimental data, //
-0.02— /9'
merely rejects the fact that the interatomic potential /, 8 ————ANIMALU —HEINE
V(r) for all r cannot be determined from a knowledge (( )-0.
04— / /
D— ——— FITTED CURVE (H —AI
V
P I'/ ~

of phonon frequencies alone. Nevertheless, it seems of 'Q) -o.o6- /i

interest to compute V(r) for the various Gr(Q). The -0.08- /o

/

results for Q,
(a/2')=1. 6 (curve A of Fig. 6) are -O Io- /P
/i

shown in Fig. 7. The values of (1/r) d V/dr and d'V/dr' /j,

CUT OFFS FOR CURVES

-O. I2— AND D

$0
-O. 14 =)' o
.006- NEAR NEIGHBOUR POSITIONS oE
2kF
-0. 6ciao
.005- I

I
I
2 45 6 78
1

I I I I I

.004- IO 2.0 5.0 4.0

NAVE VECTOR a Q/2m.
I- .0(g—
O Fio. 8. The screened pseudopotentiai V~(Q)/a(Q) for the con-
.002- IO POTASSIUM
duction electron-ion interaction in sodium. The labels 3 and D
HNSON ETAL3 are the same as for Fig. 5. Curve E (open circles) is that derived
O
.OOI- by Cochran (Ref. 5) from the function Gs(Q).
I- ~~
J dyn/crn
-4000
001—
—.
f/ We have also performed the above analysis for the
-3000 2 case of sodium, using the normal-mode frequencies ob-
tained by Woods et a/. ' As in the case of potassium, we
dV
-2000 dr&
—1000 again assumed the short-range interaction D~ to be
negligible. The results are generally similar to those for
dyn/cm potassium, particularly in respect of the total potential
200- —-IOOO function Vr (Q) and its Fourier transform in real space.

r r
-200-
The screened pseudopotential
Heine-Abarenkov
derived assuming the
dielectric constant, is shown in "
-400- Fig. II (curve D), together with that computed by
Animalu" (curve II). The fmal oscillation in curve D
MOO—
near Q, is not significant; the experimental data for
-800- sodium are barely sufhcient to specify the pseudo-
-IOOO
I I I potential at this range. Indeed, a rather satisfying Gt to
p 4 5 6 7 8 9 IO 12
the data can be obtained with (a/27r)Q, „as low as 1.3,
II

DISTANCE r (A')
in con6rmation of the earlier work, ' shown in Fig. 8 as
FIG. 7. The eGective interatomic potential and its derivatives
for potassium at 9'K, obtained by Fourier transformation of
curve A, Fig. 6. The dashed curve is one of several similar oscil- M. D. Johnson, P. Hutchinson, and N. H. March, Proc.
latory potentials derived (Ref. 2/) from x-ray scattering data Roy. Soc. (I,ondon) A282, 283 (1964). See also J. E. Enderby and
f or liquid potassium. N. H. March, Advan. Phys. 14, 455 (1965).
 COWLEY, WOODS, AND DOLLI NG
Karman model which, with forces extending to fifth
nearest neighbors, gives an excellent account of the
IO'
—0.8 experimental results. This model has been then used to
calculate the frequency distribution and specific heat of
0.6 potassium. At low temperatures the model gives a small
G (Q)
but probably significant discrepancy with the measure-
6— 0.4 ments of the specific heat"; this may be associated with
the elastic-constant behavior.
—0.2 The experimental results have also been used to
deduce pseudopotentials for the electron-phonon inter-
action, following the work of Cochran. ''
These po-
tentials are in quite close agreement with that found
by Bortolani" using the method of Heine and Abaren-
0.5
WAVE
I.O
VECTOR
I, 5
a Q /27r
2.0
"
kov. The shapes of these potentials are largely inde-
Fro. 9. The function G (Q) for sodium derived from curve D pendent of their range in reciprocal space, so thatwe may
of Fig. 8. The large wave-vector region is shown on a 10)& scale conclude they are physically quite well defined, and
on the right. The cutoff is at 2.0.
that the measurements lead fairly unambiguously to
the pseudopotentials. It is probably worth commenting
open circles. The function G~(Q) derived from curve D
that the screened pseudopotential changes sign very
of Fig. 8 is shown in Fig. 9. Up to a wave vector
close to 2k~ (Fig. 5), and so the magnitude of the Kohn
(aQ/2s. ) = 1.2, this function is almost indistinguishable
effect is expected to be very small in both potassium
from the analogous curve obtained by Cochran. (Beyond
and sodium, as found experimentally. At the smallest
this point, Cochran's G function is virtually zero. )
reciprocal lattice vectors, (110) and (002), the curve
Dispersion curves for potassium have also been calcu-
does have nonzero though small values, and so the
lated by Krebs" on the basis of a model in which (a)
band structure is not given exactly by the free-electron
short-range forces between an atomic and its first and
approximation.
second nearest neighbors are described by only two
empirical "force constants, "
and (b) an arbitrary
An alternative method of analysis was also employed
to find the interatomic potential between neutral
analytic form is used to describe the screened pseudo-
pseudo-atoms. This approach gave results which were
potential. The three disposable parameters of the model
are fitted to the measured elastic constants, and the very dependent on the range chosen for the potential.
When transformed to real space, some of these po-
calculated normal-mode frequencies are in surprisingly
tentials were of comparatively short range, whereas
good agreement with our measurements, the average
others had a very long-range oscillatory character. Since
(largest) discrepancy being about 5% (14%). From the
both types of potential gave excellent accounts of the
foregoing discussion, however, it is clear that there is
experimental results, we must conclude that there is
little physical justification for assuming a contribution
considerable arbitrariness in the choice of an interatomic
(a), particularly in the two-constant form adopted by
Krebs. Nor can the use of the "Wigner-Seitz" function potential. However, it is difIicult to understand the
in the specification of the screened electron-ion inter-
physical origin of these long-range oscillations, since the
Kohn effect is so small in these materials. Rather similar
action be justified in detail. Its analytic form merely
considerations result from an analysis of the phonon
has qualitatively the desired behavior. The Krebs model
dispersion curves for sodium. Long-range oscillatory
would nevertheless appear to be a quite convenient
potentials have also been found from an analysis of
alternative to the conventional Born —
analysis.
von Karman
liquid metal structure factor curves. ""
It is perhaps of
interest to speculate whether there is also a possibility
4. CONCLUSIONS of fitting these curves by a comparatively short-range
potential. This arbitrary behavior did not occur in the
The frequencies of the normal modes of vibration of pseudopotential analysis, which therefore appears to
potassium have been measured at 9'K for propagation be physica. lly the more sensible approach at present.
directions along sylninetry lines to an accuracy of about
2%%u~. The results are significantly different from those ACKNOWLEDGMENTS
given by ultrasonic measurements of the elastic con-
"
stants. Possibly this arises from different modes of We are grateful to Professor R. Bowers and P. A.
propagation of sound at frequencies &10' cps and Penz for supplying us with excellent single-crystal
)5&(10" cps.
The measurements have been used to 6nd the param-
specimens. We have benefited from discussion with
Professor W. Cochran. Technical assistance was pro-
eters of a conventional axially symmetric Born — von vided by E. A. Glaser and A. Bell.

~8 K. Krebs, Phys. Rev. 138, A143 (1965). ' A. Pasquin and A. Rahman, Phys. Rev. Letters 16, 30 (1966).