Strain fluctuations and elastic constants

M. Parrinello and A. Rahman

Citation: J. Chem. Phys. 76, 2662 (1982); doi: 10.1063/1.443248

View online: http://dx.doi.org/10.1063/1.443248
View Table of Contents: http://jcp.aip.org/resource/1/JCPSA6/v76/i5
Published by the American Institute of Physics.

Additional information on J. Chem. Phys.

Journal Homepage: http://jcp.aip.org/
Journal Information: http://jcp.aip.org/about/about_the_journal
Top downloads: http://jcp.aip.org/features/most_downloaded
Information for Authors: http://jcp.aip.org/authors

Downloaded 09 Dec 2012 to 139.184.30.132. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/about/rights_and_permissions
Strain fluctuations and elastic constantsa)
M. Parrinello
University of Trieste. Italy

A. Rahman
Argonne National Laboratory. Argonne. Illionis 60439
(Received 20 August 1981; accepted 23 November 1981)

It is shown that the elastic strain fluctuations are a direct measure of elastic compliances in a general
anisotropic medium; depending on the ensemble in which the fluctuation is measured either the isothermal or
the adiabatic compliances are obtained. These fluctuations can now be calculated in a constant enthalpy and
pressure, and hence, constant entropy, ensemble due to recent developments in the molecular dynamics
techniques. A calculation for a Ni single crystal under uniform uniaxial 100 tensile or compressive load is
presented as an illustration of the relationships derived between various strain fluctuations and the elastic
modulii. The Born stability criteria and the behavior of strain fluctuations are shown to be related.

I. INTRODUCTION where XB, the isothermal compressibility

=-(1/V)(av/ap)T; using Eqs. (A1) and (A3) and com-
It was in a paper by Lebowitz, Percus, and Verletl
bining with Eq. (1)
that a study was made of the ensemble dependence of
fluctuations; in the same paper it was shown that fluc- «~V)2)P,B,N+(a{3/aH)p(av/a{3); =(V/{3)X s • (2)
tuations of the kinetic energy in a molecular dynamics
calculation (which generates members of a micro- From Lebowitz et al. 1 [their Eq. (2.12)], we see that
canonical ensemble of states) is related to the constant the left-hand side of the above equation is precisely
volume heat capacity. It has now become standard «~ V)2)p, H, N' Hence (as has already been shown by Haile
practice in molecular dynamiCS and Monte Carlo cal- and Graben 5 )
culations to use fluctuations of various phase space func- (3)
tions to determine the thermodynamic properties of the
system. Thus, volume fluctuations in a system with constant
pressure, constant enthalpy, and constant number of
Recently, we have 2 presented a Lagrangian formula-
particles are just a measure of the adiabatic compres-
tion for molecular dynamics calculations in which the
sibility. (See the last remark in the Appendix. )
ensemble of states corresponds to constant stress 5,
enthalpy H (heat function), and number of particles N.
A special case of this is where the constant stress is III. THE (S,H,N) ENSEMBLE
a hydrostatic pressure, thus giving a (p, H, N) ensem-
In dealing with anisotropic media under the influence
ble. This was presented in another paper. 3
of a general stress tensor the elastic energy, V. I , is
In the following paper we show how the fluctuations of usually written as
strain in an (5, H, N) ensemble are related to adiabatic
(4)
(constant entropy 5) compliances and hence to the
elastic modulii of the system. In the case of the (p, H, N) where no is the unstrained volume and E: the strain
ensemble the relevant compliance is the adiabatic com- tensor, assumed small enough for the above expression
pressibility. to be a valid approximation. (The sign convention is
usually taken to be 5 = - p when 5 is isotropic.)
II. THE (p,H,N) ENSEMBLE
With the above V. I , the expression for the probability
There is no difference in prinCiple between the treat- of a fluctuation takes the form 4
ment of the (p, H, N) and the (5, H, N) ensemble for the
purpose in hand. Hence, first because one is more (5)
familiar with the former and second, because of the But
simplicity in writing equations involving variables with-
out indices, we shall first deal briefly with the (p, H, N)
ensemble. where C. is the heat capacity at constant strain. Hence
The thermodynamic relations are stated in the Appen- w O! exp[ - {3/2 {(C./T)(AT)2 + no(clSi/ OE kl ) T~Eij~Ekl r I , (6)
dix for completeness. As shown in Landau and Lif-
shitz,4 in a (p, {3, N) ensemble the volume fluctuations where summation over repeated indices is understood.
are From this it follows that in a (5, {3. N) ensemble
(1) (~E ij~Ekl)S, B,N = (kT /no)( oE i/ as kl ) T • (7)

The equation corresponding to Eq. (2.12) of Ref. (1) is

a)Work supported by the U. S. Department of Energy. now

2662 J. Chern. Phys. 76(5),1 Mar. 1982 0021-9606/82/052662-05$02.10 © 1982 American Institute of Physics

Downloaded 09 Dec 2012 to 139.184.30.132. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/about/rights_and_permissions
 M. Parrinello and A. Rahman: Strain fluctuations and elastic constants 2663

<~E/i ~Ekl)S, H, N = <~E/i~Ekl)S,8, N £ 2 = 0 2/ 3 !.

2 Lm I /
82I - V + 1.2 CO,2 - pO ,
(13)
+ (a{3/aH)s (aE I/a{3)s (aE kl /a{3)s. (8)

Analogous to Eq. (A3) we have (a{3/aH)s=-k{32/Cs, where 51 =r/0 1/ 3 • Periodic boundary conditions of the
where C s is the constant stress heat capacity; the equa- usual kind give a pulsating cubic box which changes in
tion analogous to Eq. (A1) is time according to a Lagrangian equation of motion. 7 The
role of the constant C in Eq. (13) is discussed by Ander-
(aE I/ as kl )8 = (aE I/ as kl ) S
sen. 7
+ Oo(kl:Nc s )( atu/ a(3)s (aE kl/ a(3)s • (9) (iii) The introduction of a time dependent shape by
Parrinello and Rahman3 who used vectors a(t), b(t), and
Using Eq. (7) and Eq. (9) we get from Eq. (8)
c(t) to define the molecular dynamics cell and used (a
<~E/i~Ekl)S, H,N = (kT /00 )( aE I/ askl)S • (10) prime indicating the transpose)
It is customary to use elastic modulii e:" Ii
£3 -
_ 1 "'"' . ,.
2" L..J m / Sl GS I -
1 ., '
V + "2 W Tr h h - pO , (14)
;: (askl/aEli)S rather than the compliances. Let us denote
by c S the 9 x 9 matrix of the adiabatic elastic modulii so
whererl=hs l , h(t)={a,b,c}, G=h'h, andO=lIhll and
that the modulus Cfi,kl is the (ij, kZ) element of C s• Then
we can go from the modulii to the compliances by a ma- W is a mass associated with the coordinates h~JJ.. Peri-
trix inversion, to get odic boundary conditions of the usual kind give a pulsat-
ing molecular dynamics cell of arbitrary shape which
(11) changes according to Lagrangian equations of motion. 3
The final step in this development is the introduction
IV. BORN STABILITY CRITERIA of an anisotropic stress tensor S in place of p in £3.
We have shown in Sec. III that at constant stress, This was done by the authors 2 in the following manner.
constant enthalpy, and constant number, (in other words The elements that are used in constructing £3 naturally
at constant entropy) the strain-strain correlation func- lend themselves to the introduction of the notion of
tion is given by the adiabatic elastic compliance. The strain. The concept of strain and that of the metric ten-
matrix of the adiabatic elastic modulii efi, kl = (as/j/aE kl )s, sor are intimately connected and, as surely has been
was denoted by C s . In other words, S = C s( and hence noticed, G the metric tensor is an integral part of the
(=(CSrIS; the compliance (aEI/askl)S = (CS);},kl [Eq. Lagrangian £3.
(11)]. This element of (CSr l is the algebraic comple-
In describing a strained state of a system one needs a
ment of the element C:',Ii of C S divided by the determi-
so-called reference state; for this we shall use the ma-
nant of C s . [Similar statements of course can be made
trix ho and the corresponding volume 0 0 = II ho II. The
for the (5,13, N) ensemble as well except that instead of
matrix ho can be used to set up a mapping between space
the superscript 5 for constant entropy one will show T
points r and a dimensionless vector ~, i. e., r = ho~. A
to indicate isothermal elastic moduliL]
homogeneous distortion changes ho to h, moving r to d,
The stability criteria of Born state that a necessary where d = h~ =h hOi r. Hence the displacement u (at r)
condition for crystal stability is that the quadratic form due to the distortion is d - r or
(C(;: CU,kIE/jEkl be positive definite. 8 Instability can
occur with the vanishing of a prinCipal minor of the de- u = (h hOi - 1) r • (15 )
terminant of C. This implies that a divergence in some The strain tensor E is defined ass
of the correlations <~EiJ~Ekl) will occur when the Born
stability criteria are violated. E
~JJ.
= !.(au~
2 arJJ.
+ ~
ar~
+ L• arJJ.
au. au.)
ar~
(16)

V. MOLECULAR DYNAMICS AT CONSTANT Hence, we find

APPLIED STRESS (=t(h~-IGh(jI_1) . (17)
The three stages prior to the formulation of new mo- If S denotes the external stress the elastic con-
lecular dynamics equations of relevance here were the tribution to the energy will be, from Eq. (4),
following.
Vol = t 0 0 Tr S(h~-l Gh(jl -0 . (18)
(i) The traditional case of generating a microcanonical
ensemble of states using the Lagrangian The new Lagrangian £., which takes account of the
anisotropic strain will then be, on leaving out incon-
(12) sequential constant energy terms,

Periodic boundary conditions are applied, most often, (19)

in the form of a repeating cubic cell of volume L 3 • The
point to note is that L is a constant and can be used as where
the unit of length.
(20)
(ii) The introduction of a time dependent volume O(t)
by Andersen 7 who used The equations of motion ariSing out of ,c. are simple

J. Chern. Phys., Vol. 76, No.5, 1 March 1982

Downloaded 09 Dec 2012 to 139.184.30.132. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/about/rights_and_permissions
2664 M. Parrinello and A. Rahman: Strain fluctuations and elastic constants

to write down and have been given elsewhere. The point VII. CONCLUDING REMARKS AND AN
of interest here is that the above Lagrangian gives a ILLUSTRATION
constant of motion JC which is
The general result given in Eq. (11) and hence also the
JC= L tmlr;+ V+ tw Tr h'h -no Tr5( ,
I
(21)
special case of tetragonal symmetry shown in Eq. (24),
implies that certain strain fluctuations will be enhanced
as a result of a reduction in the value of one or more of
with v I indicating h 8i. the prinCipal minors of the matrix of elastic modulii.
In equilibrium, at temperature T, the term in W con- In this context of enhanced strain fluctuations we recall
tributes (9/2)k B T while the term with m;'s contributes the Lindemann criterion for mechanical instability in
(3N/2)k B T. Hence to an accuracy of 3: N, the enthalpy crystals and assert that mechanical failure will occur
when these fluctuations are so large that the underlying
H=E-V. 1 , (22) atomic displacements become a sizable fraction of the
atomic spacing.
(23)
However, one should not overlook the possible pres-
ence of short wavelength phonons which due to the vanish-
is a constant of the motion.
ing of the vibration frequency can also lead to instabil-
The formal development of Sec. III and the fact that ity even when none of the minors of the matrix of elastic
under £. one generates an (5, H, N) ensemble allows us modulii is small enough to produce large strain fluc-
to conclude that in a molecular dynamics calculation tuations that are being discussed here.
using £. the elastic constants of the system can be de-
The possible divergent behavior of strain fluctuations
termined from Eq. (10) of Sec. Ill.
is similar to what occurs at the liquid-gas phase tran-
sition where the bulk modulus goes to zero and volume
VI. SPECIAL CASE OF TETRAGONAL SYMMETRY fluctuations diverge. Hence if there is an elastically
driven second order transition between different poly-
As an application of the above general development
morphic crystalline phases then there will occur diver-
we shall consider a system having tetragonal symmetry
gences in the strain fluctuations for the appropriate val-
with [100 I as the direction of tetragonal symmetry. We
ues of temperature and stress.
can write the following set of nonredundant equations
[using for notation D =C n II(C 22 22 + ('2233) - 2cfl 22 and From Eq. (11) we expect that one or more of the
E' = (E22 ± E33)/v'Z1, , , , , strain fluctuations will be enhanced if, under suitable
conditions of temperature and external stress, the elas-
(.6.E u.6.E u) = (kT / nO)(C 22 , 22 + C 22 ,33)/D , (24.1) tic constants are brought close to the Born condition
det C = O. Light scattering experiments using a crystal
(.6.Eu.6.E+)=(kt/n o)v'Z(-C ll ,22)/D, (24.2)
at a suitable temperature and/or suitable conditions of
(.6.E+.6.E+) = (kT/ no) Cn,n/D , (24.3) external stress might indeed confirm this phenomenon.
(.6.E- .6.E-) = (kt / Si O)/(C 22 ,22 - ('22,33) , (24.4) As regards the special case treated in Sec. VI, there
(24.5) are certain aspects of Eqs. (24) and Eqs. (25) which are
(.6.E23.6.E23) = (kT/n o)/C 23 ,23 ,
worth dwelling upon. The four conditions, Eq. (25.1)
(.6.E I2 .6.E I2 ) = (kT / noll C 12 , 12 • (24.6) to Eq. (25.4), are independent, and hence when one
holds the other three need not be satisfied.
The Born conditions for this case are
When Eq. (25.1) is satisfied, i. e., when D defined in
Cll,u >0; Cll,ll C 22 ,22 - C~I,22 > 0 ;
Sec. VI is zero, then Eqs. (24.1), (24.2), and (24.3)
show a divergence whereas Eqo (24.4), (24.5), and
(C 22 ,22- C 22,33)D>0; C 23 ,23 >0; C I2 ,12>0
(24.6) remain finite. Thus, when the condition D=O
These are equivalent to is approached, E22 and E33 must fluctuate in phase so as
to prevent E- = (E 22 - E33 )/v'Z from having large fluctua-
C uou > 0; C 22 ,22 - C 22 ,33 >0; D> 0; C 23 ,23 > 0; C 12 ,12 >0
tions. On the other hand, when the condition C 22 ,22
Thus the four denominators in Eqs. (24. 1) -(24.6) are - C 22 ,33 = 0 is approached we must have E22 and E33 fluc-
the quantities whose vanishing defines the boundary of tuating out of phase.
the elastic stability region
Thus, even if the theoretical stability limits [Eqs.
(25. 1) (25.1) to (25.4)1 are not reaChed, and this will be so in
most practical situations, we expect that the trends de-
C 22 ,22 = C 22 ,33 , (25. 2)
scribed above may be observable under suitable tem-
C 23 ,23 =0, (25.3) perature and stress conditions. We have recently re-
ported on a molecular dynamics study2 of the effect of
C 12 ,12 = 0 . (25.4)
uniaxial stress on a single crystal of Ni. This data is
The condition C n, n = 0 need not be included in the list being analyzed from the point of view of the behavior of
of conditions that define the boundary of the region of strain fluctuations dealt with in the present paper. To
stability since for a sufficiently small and positive illustrate the observation made above about the phase re-
Cu,n the condition in Eq. (25.1) is already satisfied. lation between the fluctuations of E22 and E33 we will

J. Chern. Phys., Vol. 76, No.5, 1 March 1982

Downloaded 09 Dec 2012 to 139.184.30.132. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/about/rights_and_permissions
 M. Parrinello and A. Rahman: Strain fluctuations and elastic constants 2665

0<:(

'" FIG. 1. Fluctuation of the three lat-

V>
f- tice constants of a tetragonal struc-
Z
<:( ture generated by applying a tensile
f-
V>
Z load to an fcc crystal of Ni. Strain
0 components can be calculated from
U
W
u
thelj shown above; (Jj-(lj»!(lj) is
f- a good approximation (Ref. 2) for Ell'
f-
<:( The condition C 22 22 =C 22 33 is close to
-'
0 23456789 being satisfied at' this lo~d; see Eqs.
(24.3) and (24.4) for consequences.
TIME in 10-12 s

briefly mention one of the several calculations on Ni On applying a compressive [100] load the static calcu-
already reported. 2 lations of Milstein 10 show that at a load of about 7 x 10 10
dyn cm- 2 the instability criterion D = 0 [Eq. (25.0] is
On applying a [100] tensile load to a single crystal of
satisfied. (Our dynamical calculations 2 showed a poly-
fcc Ni one obtains a face -centered tetragonal structure.
morphic transition to occur at a value slightly lower
Under conditions of zero load and a temperature of 356
than 7 x 10 10 dyn cm- 1 • )
K the model 9 of Ni we have used2 gave a stable perfect
fcc lattice in thermal motion, the lattice constant being The time behavior of the three lattice constants for a
3.55 ± O. 09 A. As the load was increased the lattice stable tetragonal structure at 356 K and compressive
became tetragonal and at a tensile load of 8.6 x 1010 [100jload of 5.25xl010 dyn cm- 2 is shown in Fig. 2.
dyn cm- 2 and 330 K, the two lattice constants became In spite of the noise we can see (i) that E22 and E33 are in
3. 76±0.01 and 3.49±0.01 A. This state of the system phase, (ii) that Ell is out of phase with those two, and
was perfectly stable over a long MD calculation of (iii) that the fluctuation Ell is larger than that of the
1. 7 x 10- 11 s. Further details about many such calcula- other two. Statement (i) is in accord with Eqs. (24.3)
tions are in Ref. 2. One more detail from Ref. 2 is and (24.4), (ii) is a consequence of Eq. (24.2) since ll
relevant. According to static calculations of Milstein 10 C U ,22 >0, and (iii) is in accord with Eq. (24.1).
at about 16 x 10 10 dyn cm- 2 , the Born condition [Eq.
(25.2) j C 22 22 =C 22 33 is satisfied. (Our dynamical cal- The time scale of the fluctuations shown in Figs. 1
cUlation 2 showed ~ystem failure to occur already at and 2 is dependent on the choice of W, see Eq. (19).
- 11 x 10 10 dyn/ cm 2 • ) The statistical averages we have dealt with here do not
depend on this choice. This has been discussed by
The time behavior of the three lattice constants for Andersen,7 Haile and Graben, 5 and Parrinello and
the system described above, i. e., for a stable tetrago- Rahman. 2 ,3
nal structure at 330 K and tensile [100 jload of 8.6 X 10 10
dyn cm- 2 is shown in Fig. 1. In spite of the noisy fea- We finally note that a different method for the numer-
tures we see from this figure that E22 and E33 are out of ical calculation of the isothermal elastic constants of
phase, in complete accord with the prediction of Eqs. crystalline systems has been proposed by Squire et ai, 12
(24.3) and (24.4L An extension of their method to noncrystalline systems

3.70
..... 0·:" ........ .." ........
3.65 ~'-~:...-. .
0<2: '. '.' .;..-.....
~ '.- """;T c' .,' ....'?s. ··!."'··.;fl/ --';-".......,,-----'-,-"".,-::'.~<~3>
c: 3.60 '.'
'.'

FIG, 2. System under a compressive

load. The condition D =C \1 \1 (C 22 22
2 . ' ~
+C 22 33) - 2 C \1 22 = 0 1S close to bemg
sati;fied; see Eqs. (24.1), (24.2),
(24.3), and (24.4) for consequences.
The shading is to show the phase re-
lation between ~E\1' ~E22' and ~E33'

3.30 ~----7------=----=---L-------=----'-----'-----'
o
TIME in 10- 12 5

J. Chern. Phys., Vol. 76, No.5, 1 March 1982

Downloaded 09 Dec 2012 to 139.184.30.132. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/about/rights_and_permissions
2666 M. Parrinello and A. Rahman: Strain fluctuations and elastic constants

does not seem to be straightforward. In contrast, the dH= dS/k(3+ Vdp , (A2)
computational scheme described above seems to be
suitable for all systems, crystalline or otherwise. (:~)p =_k(32/C p. (A3)
This remark is of particular significance in the study
of superionic conductors. These have normal solid- We note finally from Eq. (A2) that for P and H constant
like elastic behavior but one or more of the constituents dS=O.
have large liquid -like constants of se If -diffusion. How -
ever, extensive calculations are necessary before one
can assess the practical merits and limitations of the
method. We plan to undertake such calculations in the lJ. L. Lebowitz, J. K. Percus, and L. Verlet, Phys. Rev.
near future. 153, 250 (1967l.
2M. Parrinello and A. Rahman, J. Appl. Phys. 52, 7182 (1981l.
3M. Parrinello and A. Rahman, Phys. Rev. Lett. 45, 1196
APPENDIX (1980l.
Using p, T as variables, 4L. D. Landau and E. M. Lifshitz, Statistical Physics (Addi-
son-Wesley, Reading, Mass., 1958l.

(av) _(aMP,ap T)) s

ap s - (p,
_(av)
T) -
(av) (aT)
aPT + aT &p s p
5J. M. Haile and H. W. Graben, Mol. Phys. 40, 1433 (1980).
6M. Born and K. Huang Dynamical Theory of Crystal Lattices
(Clarendon, Oxford, 1954).

= (:~t -(::)p (~) p(~ t lH. C. Andersen, J. Chern. Phys. 72, 2384 (1980).
8L . S. Landau and E. M. Lifshitz, Theory of Elasticity (Addi-
son-Wesley, Reading, Mass., 1959l.
= (av)
ap T
_(T /C (~)
aT p)(av)
aT ] p
[_
p,
(At) 9 F . Milstein, J. Appl. Phys. 44, 3825 (1973).
1°F. Milstein, J. Appl. Phys. 44, 3833 (1973).
11 As is shown by Milstein (Ref. 10), for central pairwise inter-
or
actions, the conditions C 23 23 and C 22 22 > 0 are equivalent to
C 11,22> O. In the same pa~r Milstei~ has given a plot of
( apav)a = (av)
ap s
_k(33/C (aV)2 .
a(3
p
p
(Al) C 11 22 against stress.
12D. 'R. Squire, A. C. Holt, and W. G. Hoover, Phys. 42, 388
We also have the enthalpy H =U + PV, and (1969).

J. Chern. Phys., Vol. 76, No.5, 1 March 1982

Downloaded 09 Dec 2012 to 139.184.30.132. Redistribution subject to AIP license or copyright; see http://jcp.aip.org/about/rights_and_permissions