Вы находитесь на странице: 1из 8

Primitive model electrolytes. I.

Grand canonical Monte Carlo computations


John P. Valleau and L. Kenneth Cohen

Citation: The Journal of Chemical Physics 72, 5935 (1980); doi: 10.1063/1.439092
View online: http://dx.doi.org/10.1063/1.439092
View Table of Contents: http://scitation.aip.org/content/aip/journal/jcp/72/11?ver=pdfcov
Published by the AIP Publishing

Articles you may be interested in


An efficient iterative grand canonical Monte Carlo algorithm to determine individual ionic chemical potentials in
electrolytes
J. Chem. Phys. 132, 244103 (2010); 10.1063/1.3443558

Boltzmann bias grand canonical Monte Carlo


J. Chem. Phys. 128, 134109 (2008); 10.1063/1.2883683

Phase behavior of the restricted primitive model and square-well fluids from Monte Carlo simulations in the grand
canonical ensemble
J. Chem. Phys. 110, 1581 (1999); 10.1063/1.477798

Monte Carlo study of mixed electrolytes in the primitive model


J. Chem. Phys. 96, 7656 (1992); 10.1063/1.462366

Grand canonical Monte Carlo simulation of liquid argon


J. Chem. Phys. 85, 2169 (1986); 10.1063/1.451110

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
137.112.236.68 On: Thu, 18 Dec 2014 19:04:03
Primitive model electrolytes. I. Grand canonical Monte
Carlo computations
John P. Valleau and L. Kenneth Cohen
Lash Miller Chemical Laboratories. University of Toronto. Toronto. Ontario. Conada M5S IAI
(Received 20 December 1979; accepted 21 February 1980)

Monte Carlo calculations in the grand canonical ensemble are described for coulombic systems. and
carried out for 1:1. 2:2, 2:1. and 3:1 aqueous electrolytes in the primitive model with equal ion sizes.
Energies and activity coefficients are obtained, and the scope and reliability of the method is discussed.

I. INTRODUCTION often,7 but it is only recently, with the work of Norman


and Filinov,8 Adams, 9,10 and Nicholson, Rowley, and
Once upon a time this laboratory published1 • 2 Monte
Parsonage, 11,12 that attempts have been made to exploit
Carlo results for the restricted primitive model of 1 : 1
the method.
aqueous electrolytes. The results were useful in eval-
uating various theoretical treatments of the electrolyte The GCMC method appears not to be useful at high
problem. At the same time similar Monte Carlo (MC) densities (e. g., for liquids). This is because steps
investigations were carried out for the 3: 1 and 2: 2 in the Markov chain which lead to changes of con-
cases. 3 These data are of still greater interest, but centration are then exceedingly rare: one is unlike-
they were not published: many of them are reported in ly to find room to insert particles at random positions
later papers of this series. in a dense system. As a consequence the concentration
does not fluctuate adequately and convergence is prohib-
We were unwilling to publish those results earlier be-
itively slow. There are no such problems at low den-
cause we were not confident of the accuracy of the os-
sities, however, and this makes GCMC ideal for study-
motic coefficients, and hence of the free energies, that
ing dilute electrolyte solutions.
were obtained. The experiments were carried out in the
canonical ensemble, as is most Monte Carlo work. Like The present article (I) describes the application of
the pressure in uncharged systems, the osmotic coeffi- GCMC methods to electrolyte solutions, and reports results
cient requires the values of the pair correlation func- for 1: 1, 2: 2, 2: 1, and 3: 1 aqueous electrolytes in the
tions at contact of the particles, and these are obtained primitive model with equal ion sizes. The following pa-
by extrapolation of the pair function data. In the ionic per (II) reports some of the earlier canonical Monte Car-
case, however, the pair functions for unlike ions vary lo (CMC) data for 2: 2 electrolytes, and uses both sets of
extremely rapidly near contact, and the extrapolation data to examine the (by now many) theoretical approxi-
is correspondingly dubious even with quite precise pair mations for the problem and to comment on the structure
function data. USing conventional techniques the free of such ionic solutions. The third paper (m) does the
energy and activity coefficients would be obtained by in- same thing for the unsymmetrical 2: 1 and 3: 1 cases.
tegration of the osmotic coefficient (with respect to Inc) A subsequent paper (IV) will report GCMC results for
and so our values of those quantities were also in doubt, the 1: 1 aqueous primitive model with unequal ion sizes.
although the internal energy and the structural data were
This article (I) is therefore primarily methodological,
both good in themselves.
and begins (Sec. II) by discussing the theory of GCMC
In the case of the 1 : 1 electrolytes 1 • 2 the problem is for the electrolytic case and choosing an appropriate
not so severe, and the extrapolations seemed adequate. transition matrix for the MC Markov chain. After other
Furthermore, those free energies were subsequently details of the computations are described (Sec. m), the
obtained by a second and independent method· which results are reported (Sec. IV). Most of the scientific
confirmed the conventional results. Application of that discussion of these results is left for later papers; here
technique (multistage sampling) to the other systems we are interested in the efficiency and reliability of the
would have been somewhat tedious, but a later develop- GCMC results.
ment of the idea using non-Boltzmann sampling tech-
niques 5 • 6-so-caUed "umbrella sampling" -would be II. GRAND CANONICAL MONTE CARLO THEORY
quite efficient.
To specify a configuration in the grand canonical en-
A still better approach is available, however, for the semble requires the number and the locations of the par-
low ionic densities in which we are interested: the ticles of each species present. Suppose 1Ti is the proba-
problem is a natural one for the grand canonical Monte bility of a configuration i in the grand canonical en-
Carlo methods 8 • 7 (GCMC). In GCMC work one fixes the semble. We require a convenient MarkOV chain among
cbemical potential, along with the temperature and the such states having a limiting distribution proportional
volume. The Markov chain allows fluctuation of the to the distribution hi}' The Markov chain is defined by
concentration, and one finds eventually the mean con- the stochastic matrix I IPiil I of the probabilities Pi} of
centration (and energy, etc.) corresponding to the par- transitions from state i to state j. A sufficient (though
ticular values chosen for the fixed parameters. The not a necessary13,14) condition for the irreducible chain
possibility of such calculations has been pointed out to have the correct limiting distribution is

J. Chem. Phys. 72(11). 1 June 1980 0021·9606/80/115935·07$01.00 © 1980 American Institute of Physics 5935

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
137.112.236.68 On: Thu, 18 Dec 2014 19:04:03
5936 J. P. Valleau and L. K. Cohen: Primitive model electrolytes. I

E.uo:!!.L (1)
state. Combining Eqs. (1)-(5), we obtain (still with
Pjj 1T I N;=~+v., Nj=Nj-+vJ
and we seek such transition probabilities. As usual we
attempt, at each step of the chain, a "trial" change of (6)
state, and this trial move is then accepted or rejected
in such a way as to lead to (1). Suppose the probability where Il =:; V.(.1.. + v_f.1._ is the chemical potential of the elec-
of a trial step i to j is qlj' and the probability of ac- trolyte. Extending a definition due to Adams, 2 we in-
ceptance of that trial step isflj for i*j. Then troduce
Pi};' qjjfiJ> (ii-j) , (2) B'" !3«(.1. - Iltdeal) + lnyu+N-u.
V"
(7)
PH;' 1- LP jj
j~i

which leads to
There is of course wide latitude in the choice of I I qjJII ,
and then also of I I ftJl I, satisfying (1). We examine h , -N:'!'''''
_ N."! Ni! [
expB-!3Ui -Uj
( )] (8)
simple choices of I I qjjll 'and I I fjjll, of the sort pro- f Ji j.lV j '

posed by Adams, 9 but elaborated to deal with the multi-


Evidently [cf. Eq. (7)] fixing B fixes fl at fixed tempera-
component ionic solutions.
ture and volume. If the acceptance probabilities fii> fji
Monte Carlo experiments can be regarded as describ- are made to conform to Eq. (8) with some particular
ing an ensemble either of labeled particles or of un- value of B, the corresponding deviation from ideality,
labeled particles. It turns out to be most convenient to f.1. - f.1.Sdeal' may be obtained from it once the expectation
regard the particles as unlabeled, and this is done in values of the concentrations Y and N" are known. Of
what follows. (An alternative formulation in terms of course
labeled particles is of course possible. 15)
!3( Il - Ilfd.al) = v In)' * , (9)
We now describe a scheme for carrying out a grand
where 'Y z is the mean ionic activity coefficient in the
canonical Monte Carlo computation. Consider an elec-
trolyte which on solution dissociates into v.Lo cations McMillan-Mayer system.
of charge Z. and v.Lo anions of charge Z. per mole A simple way of realizing the result (8) is to set
(where Lo is Avogadro's nwnber). Evidently v.Z. + v.Z.
;, 0, and v=:; V. + v_ ions can form an electrically neutral fjj =min{l, Ii) fji} for addition, (10)
combination. At each step of the chain either (i) one fii =min{1, fu! ii}} for deletion
tries to add or to delete an electrically neutral combi-
nation of v ions, or (ii) one tries to move one ion to a of particles, where on the right hand sides of the ex-
new location. The steps are done as follows: pressiOns the ratio f i ) fii is given by Eq. (8) after find-
ing the value of {UJ - U j ).
(i) Additions or deletions are attempted with equal
(ii) Particle moves are attempted with a probability
probability P; v. cations and v. anions are added or de-
(1- 2P) at each step, and are carried out exactly as in
leted in a Single step. In a trial addition each of the v
the canonical case. That is, a particle is chosen at
ions is inserted anywhere in the box with equal probabil-
random and moved to a random position in a volume ele-
ity. It is convenient (and valid) to regard the box as
ment surrounding its original position. The volume ele-
consisting of V discrete sites (V extremely large).
ment is chosen so that the part of the transition matrix
Then if state j is obtained from state i by addition, Nj
describing these trial moves is symmetric: qiJ = qJI for
= Nj + v+ and Nj 0: N; + v. in an obvious notation, and evi-
moves. It is usual to choose a position within a cube or
dently
sphere centered on the original position of the particle;
P
(3) a cube of side 6 was used in the computations described
qjJ 0: V"v+ I v.1 here. The move is accepted with a probability fii given
by
In a trial deletion any set of v. cations and v. anions is
fiJ =min{1, exp[ - /3(Ui - Uj )]} • (11)
removed, with equal probability, so (with i obtained
from j by deletion)
III. COMPUTATIONAL DETAILS
_ PNjlNil The calculations are for the primitive model with ions
'" + 1''-'
(4)
qjj - lVJ lVi·v•., v_., of equal size. Thus the potential energy of interaction
U"b between two ions a, b separated by a distance rd is
It is interesting that the "underlying matrix" II qui I is
given by
not symmetric. The probability of a configuration i in
an ensemble of unlabeled particles will be (12)
1 1
1Tt = IT A!Nl A:N'i exp[,9(Il.N'i + Il·Nt) - tm.] (5) =00 if r,,&<R ,
where Il. and fl. are the chemical potentials of the ions, where Zoe, Z6e are the charges on the ions (with - e the
n is the grand canonical partition function, A. =:; hi electronic charge), E: is the dielectric constant chosen to
(2mn.kT)1/2, and U j is the configurational energy of the represent the screening effect of the solvent, and R is

J. Chem. Phys., Vol. 72, No. 11, 1 June 1980


This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
137.112.236.68 On: Thu, 18 Dec 2014 19:04:03
J. P. Valleau and L. K. Cohen: Primitive model electrolytes. 5937

4
TABLE I. Results for 1: 1 electrolyte with R = 4.25 A and e:T= 2.345 x 10 • The columns are as follows: L is the length of the
cubical periodic box; B the chemical potential input parameter [Eq. (7)]; No. Steps refers to the length of the Me run, while No.
Disc. shows how much of the run was regarded as aging and ignored in the averaging; 2P is the proportion of steps in which addi-
tions or deletions were attempted, and the next column gives the acceptance rate for such steps; the average number of particles
N and the resulting molarity M are followed by the data for the activity coefficients and the reduced configurational energies.

1O-3 x 10- 3 x Acc. rates


L(A) B No. Steps No. Disc. 2P Add/Del (N) (M) - (In)' ±) - (U)/(N)kT
165.21 6.722 40 16 0.20 0.76 62.9 ± 0.8 O. 01158± 0.00015 0.087 ± 0.013 0.116 ± O. 004
80.99 6.471 100 10 0.20 0.64 64.1±0.4 0.1002 ± 0.0006 0.232 ± 0.006 0.277 ± 0.004
47.36 6.415 100 10 0.20 0.47 63. 4± O. 6 0.495±0.004 0.249 ± 0.009 0.460 ± O. 007
37.59 6.679 100 20 0.20 63.4± 1. 0 0.991 ± 0.016 0.117±0.016 0.556 ± O. 014
37.59 6.683 312 26 0.20 0.33 64.2± O. 4 1. 003 ± O. 006 O. 127 ± O. 006 O. 558± O. 006
29.84 7.468 100 20 0.20 0.14 63. 8± 1.1 1.994 ± 0.034 -0.271±0.017 0.672 ± O. 021
26.06 8.660 720 120 0.20 0.04 65.l± O. 4 3.052 ± O. 019 -0.847±0.006 O. 740±0. 008
32.84 10.046 720 60 0.20 0.04 129. 9±0. 7 3. 045±0. 016 -0. 849± O. 005 0.731 ± 0.007
23.68 10.300 3024 504 0.20 0.008 66.1±0.3 4.13± O. 02 -1. 652 ± O. 005 0.800 ± 0.006
29.84 11.686 1680 336 0.20 0.008 131.0±0.6 4.09 ± O. 02 -1. 661 ± O. 005 0.786 ± O. 006
22.238 11.366 2400 200 0.67 0.002 63.05±0.24 4. 76± 0.02 - 2. 232 ± 0.004 0.830 ± 0.005
28.02 12.752 1560 156 0.67 0.002 125.75 ± O. 57 4. 75±0. 02 - 2.235 ± O. 005 0.813 ± 0.008

the diameter of each ion. The properties of the system Hoskins and Smith16 on two-component plasmas confirm
depend on R and on the product ET, as well as on the that for low r the methods agree well while for higher r
ionic charge types. Rand E T were chosen to match they give totally different results. Up to r - 10 the dis-
closely those used earlier in the CMC calculations 3 : for crepancy is comparable to the statistical uncertainty.
the 1: 1 case R=4.25 A and ET=2.345xl0· deg (cf. R For all of the calculations reported here r is substan-
=4.25 A and ET=2.339xl0· for CMC), and for the 2:2, tially less than 10 (e. g., for the 2: 2 system at 2 M, r
2:1, and 3:1 cases R=4.2 A and ET=2.342xl0· (cf. R = 4. 9), so the energy approximation should be trust-
=4. 2 A and E T =2. 336 X 10·). A t room temperature the worthy. For the more concentrated solutions this was
parameter ET corresponds to the dielectric constant of tested, however, by carrying out parallel runs with sys-
water. Of course the results can equally well be inter- tems of double the usual size. One expects the results
preted as corresponding to a lower E at a higher tem- to vary with the system's size if the energy approxima-
perature, for example, a plasma at about 2.34 x lOS OK. tion is becoming inadequate; the results are discussed
In a similar way the results we describe as those of 2: 2 below.
electrolyte may be interpreted as a 1 : 1 electrolyte at a
value of ET reduced by a factor of 4 (e.g., an E of 19.6 Two "experimental" parameters will affect the conver-
at 25°C). gence rate of the Markov chain. These are the "step
size" 6 allowed when a particle move is attempted, and
Periodic boundary conditions were used, and the re- the ratio 2P/(l- 2P) of attempted additions and deletions
peating box was cubical. The energy calculations were to attempted particle moves. The former was chosen to
done in the minimum image apprOXimation (MI): that is make the acceptance rate for moves well behaved; for
the potential energy of a particle is evaluated by sum- the denser systems it was adjusted to give an acceptance
ming the pair interactions with only one image of each rate close to 50%. Little is known about the best way to
of the other particles, the image closest to the particle choose P, and we made no systematic study of this. At
in question. This corresponds to a cubical cutoff having low densities the acceptance rates for addition and de-
the dimensions and orientation of the repeating box. It letion are substantial, and we chose P =0.1. A high
has been showni • 16 that a spherical cutoff is totally un- densities the acceptance rate becomes small, however,
acceptable for coulombic systems, and this is readily and it was necessary to increase P in order to get ade-
understood. 14 Many calculations on Coulombic systems quate variation of concentration during the runs. Data
have approximated the truly periodic boundary condi- on P and the addition/deletion acceptance rates are re-
tions by doing Ewald summations, although it seems ported. Since pair correlation function data for most of
that this will introduce very nonphysical forces and cor- these systems had been obtained in the earlier CMC
relations. 14 Fortunately at low ionic densities the calculations,3 it was not collected in the runs reported
Ewald and MI approximations are expected to agree with here. (For systems with unequal ion Sizes, reported in
each other and to be physically meaningful. Brush, Paper IV, CMC results were not available and pair cor-
Sahlin, and Teller 17 compared the two approximations relations were studied in the GCMC runs. )
for the one-component plasma and found that they agreed
as long as the dimensionless parameter
IV. RESULTS AND DISCUSSIONS
_/4rr1J'\1/3 Z2 e 2
r -\- 6Y) EkT (13) Tables I-IV refer to the 1: 1, 2: 2, 2: 1, and 3: 1 sys-
tems, in that order. In each table the last three columns
was below about 10. (Here N/V is the number density report the physical results: activity coefficients y,. and
and Ze the ionic charge.) Our own tests 1 ,14 and those of configurational energies U at various concentrations.

J. Chem. Phys., Vol. 72, No. 11, 1 June 1980


This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
137.112.236.68 On: Thu, 18 Dec 2014 19:04:03
5938 J. P. Valleau and L. K. Cohen: Primitive model electrolytes. I

TABLE n. Results for 2: 2 electrolyte with R=4. 2 A and E:T=2. 342 x104. The columns are described in the legend of Table I.

10- 3 x 10-3 x Acc. rates


B No. Steps No. Disc. 2P Add/Del (N) (M) - (lrry.) - (U)/(N)kT
174.48 5.190 1248 104 0.20 0.36 65.63 ± 0.50 0.01025 ± O. 00008 0.896 ± O. 008 1. 208 ± O. 019
102. 04 3. 872 630 60 0.20 0.25 58.36±0.56 0.0456±0.0004 1.437±0.010 1. 774± O. 031
59.675 2. 632 540 60 0.20 0.16 65.07 ± 0.52 0.254 ± O. 002 2.166 ± 0.008 2.465 ± O. 032
75.186 4.018 630 60 0.20 0.15 127.47 ± 1. 43 0.249 ± 0.003 2. 146 ± 0.011 2.450 ± 0.044
44. 571 1. 952 540 60 0.20 0.12 63.11 ± O. 97 0.592 ± 0.009 2.476 ± O. 015 2. 841 ± 0.043
37.590 1. 603 630 60 0.20 0.099 62.16 ± O. 82 0.971 ± O. 013 2.635 ± O. 013 3.102 ± 0.065
32.139 1. 570 1456 208 0.20 0.065 63.89±1.121.597±0.028 2.679±0.017 3.383 ± 0.095
29.837 1.592 630 60 0.20 0.051 63.41±0.54 1. 982±0. 017 2.660±0.009 3. 528 ± O. 046
37. 592 2. 986 630 60 0.20 0.045 123.48 ± 1. 33 1. 929 ± 0.021 2.630 ± O. 011 3. 484± O. 059
26. 065 1. 946 2520 280 0.20 0.024 61. 01 ± 0.37 2.860 ± 0.017 2.445 ± 0.006 3.771 ± O. 038
26.065 1.946 2464 154 0.20 0.024 61. 96 ± 0.39 2.904 ± 0.018 2.460 ± 0.006 3.785 ± 0.038
32.840 3.332 1512 126 0.20 0.018 127.30±1.31 2.984±0.031 2.487±0.010 3.878 ± 0.070
23. 682 1. 946 2464 308 0.40 0.019 51.40±0.31 3.212±0.019 2.273±0.006 3.815 ± O. 037
29.837 3.332 1464 244 0.40 0.008 118.13 ± 2. 38 3.691 ± O. 074 2.413 ± O. 020 4.265±0.171
23.682 3. 060 2400 200 0.20 0.008 62. 84± O. 42 3. 927± 0.026 1. 917± O. 007 3.984 ± O. 044

TABLE m. Results for 2: 1 electrolyte with R=4. 2 A and E:T=2. 342 x 104• The columns are described in the legend of Table I.

10-3 x 10- 3 x Acc. rates


B No. Steps No. Disc. 2P Add/Del (N) (M) - (lny.) - (U}/(N)kT
328.4 10. 111 280 40 0.20 0.71 62. 65 ± O. 26 0.000979 ± 0.000004 0.131 ± 0.004 0.129 ± O. 002
151.6 9.518 280 40 0.20 0.54 63.60 ± O. 38 0.01009 ± 0.00006 0.343 ± 0.006 0.362 ± 0.005
88.68 8.630 380 40 0.20 0.40 60.07 ± O. 50 0.0477 ± 0.0004 0.582 ± O. 008 0.637 ± 0.010
51. 86 7.992 280 40 0.20 0.24 65.30 ± 0.42 0.259 ± 0.002 0.878 ± 0.006 1. 035 ± O. 011
38.94 7.720 280 80 0.20 O. 15 61. 71 ± O. 68 0.578 ± O. 006 0.913 ± O. 011 1.255 ± 0.022
32.84 8.120 280 40 0.20 0.078 62.68±0.47 0.979 ± 0.007 0.795 ± 0.007 1. 416± O. 018
39.37 10.359 672 96 0.20 0.057 124.39 ± 1. 08 1. 128 ± O. 010 O. 734± 0.009 1. 45l± 0.020
31.24 8.280 672 96 0.20 0.059 62. 93 ± O. 41 1. 141 ± 0.007 0.745 ± O. 007 1. 463 ± 0.016
29.20 8.720 720 60 0.20 0.036 64.04 ± O. 40 1. 423 ± 0.009 0.616±0.006 1.540±0.017
29.20 8.720 840 60 0.20 O. 036 64. 23 ± O. 40 1. 427 ± O. 009 0.619 ± 0.006 1. 542 ± 0.016
36.79 10.799 720 60 0.20 0.035 126.27 ± 0.95 1. 403 ± 0.011 O. 602 ± O. 008 1. 523 ± O. 019
27.38 9.180 1152 96 0.20 O. 020 62. 83 ± O. 35 1. 694 ± O. 009 0.444 ± O. 006 1. 600 ± 0.014
34.49 11.259 960 96 0.20 O. 019 125. 53 ± O. 55 1. 692 ± O. 007 0.443 ± 0.004 1. 589 ± 0.013
25.93 9.849 1584 132 0.32 O. 010 62.54 ± 0.30 1. 984 ± 0.010 0.216 ± O. 005 1. 654 ± O. 013
32.67 11.928 1200 240 0.32 0.010 123. 80± O. 73 1. 964± 0.012 0.206 ± O. 006 1. 640 ± O. 015
24.75 10.906 1584 132 0.60 0.0036 64.68 ± O. 26 2.360 ± 0.009 - O. 102 ± O. 004 1. 718 ± O. 012
31.18 12.985 1200 240 0.60 0.0033 131. 35 ± 1. 27 2.396 ± 0.023 - 0.087 ± O. 010 1. 726 ± O. 029
23.76 12.226 2000 200 0.80 0.00089 66. 91 ± O. 39 2.760 ± 0.016 - O. 508 ± 0.006 1. 779 ± 0.018
29.94 14.305 1440 180 0.80 O. 00093 134.8 ± 1. 4 2.780 ± 0.029 - 0.501 ± O. 010 1. 777 ± O. 026

TABLE IV. Results for 3: 1 electrolytes with R = 4. 2 A and E:T= 2. 342 x 104• The columns are described in the legend of Table I.

10-3 x 10- 3 X Acc. rates


B No. Steps No. Disc. 2P Add/Del (N) (M) - (lrry.) (U)/(N)kT

298.4 13.410 420 60 0.20 0.53 64.54 ± 0.31 0.001008 ± 0.000005 0.252 ± 0.005 0.298 ± 0.006
375.9 16.182 480 60 0.20 0.53 129.0 ± 1. 3 0.001008 ± O. 000010 0.252 ± 0.010 0.282 ± O. 005
138.5 11.770 420 60 0.20 0.30 64.00 ± O. 48 0.01000 ± O. 00007 0.654 ± 0.008 0.756 ± O. 012
80.99 9.890 420 60 0.20 O. 18 59.88 ± O. 34 0.0468 ± O. 0003 1. 058 ± 0.006 1. 202 ± O. 013
80.99 10.110 420 60 0.20 O. 18 65. 23 ± O. 48 0.0510 ± 0.0004 1. 088 ± O. 007 1. 236 ± 0.016
47.36 8.046 420 60 0.20 0.087 61.17±0.64 O. 2390 ± O. 0025 1. 540 ± 0.010 1. 759 ± 0.030
35.376 7.560 420 60 0.20 0.036 60. 80±0. 53 0.570 ± 0.005 1.655 ± O. 009 2.101 ± O. 030
29.835 8. 120 1440 120 0.20 0.011 62.38 ± 0.46 0.975 ± 0.007 1. 541 ± O. 007 2.344±0.028
34.62 10.466 1200 60 0.50 0.0079 106.32 ± O. 83 1. 06~ ± O. 008 1. 488 ± O. 008 2.378±0.031
27.48 9.080 1680 120 0.60 0.0035 65. 04± O. 49 1. 30J ± 0.010 1. 343 ± 0.008 2.489 ± 0.031
34.62 11.852 960 60 0.50 0.0029 131. 7 ± 1.1 1.317± O. 011 1.355±0.009 2. 489± 0.034
25.51 10.390 2400 240 0.80 0.00091 64.65 ± O. 46 1.617±0.012 1. 009 ± O. 007 2.605 ± 0.028
32.14 13.162 1800 200 0.80 0.00078 130.68 ± 0.94 1. 634± O. 012 1.020 ± 0.007 2.610 ± 0.032

J. Chern. Phys., Vol. 72, No. 11,1 June 1980


This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
137.112.236.68 On: Thu, 18 Dec 2014 19:04:03
J. P. Valleau and L. K. Cohen: Primitive model electrolytes. I 5939

The other columns report technical information on the


GCMC runs. The physical results are also shown in I
F.igs. 1 and 2. I
2.0 I
I
The scientific implications of these results are dis- I
cussed in Papers II and Ill, in conjunction with the CMC
results reported in those papers. Here we examine
~
I
methodological matters relating to the GCMC method. 1.5 I
I
I
At very low densities the acceptance rates for addition I
and deletion steps are of course high, and convergence I
of the chains is rapid. However, as the density in- I
1.0 I
creases that acceptance rate falls off rather quickly.
In order to sample adequately one then needs to go to I
d
very long runs, and at a quite modest density GCMC cal- 1: 1 /1
culations become impractical. This can be examined I I

for each of our systems in the tables. Of course in this 0.5 I ~


I I
electrolyte work· we are adding or deleting more than I I
~ I
one particle at a time, so the acceptance rate is lower / I
/ !/).
than it would be for a simpler system. The runs report- / I
ed for the higher concentrations pushed the GCMC tech-
nique pretty hard: they managed to use addition/deletion
t o ,
b~,
\\ 'CSt- __ -1(0........
,~
//
"
I
I

acceptance rates as low as - O. 001. In order to do this /


it was necessary to increase the proportion 2P /(1 - 2P) /
c
2:1/~
of attempted additions or deletions drastically, and also
\ [({
ft\ , f;/
/

\
q,
, ......0- __ - /
0
5)'

\
-1.0 ,\ a \ ~
I
I

,b \
,
)(
\ I
I

-0.5 ~\ \x\
\ , 3: 1 I~
I
1:1
\
b\ ' \
b
---_A .,--~-
-1.5 q, ',CSt /'
-1.0
\
\
, '0
\
, . . " -2-"
, X
I \ "
\
o, " '0, -2.0 \ °
\
\
" 'o"f!l \
\ x
x
0,
-1.5 \ .......~
\ . ..c.2:1
i \
~
'~
'a... -2.5
-2.0
2
o
'"'
"-
'"::;) 3:1 FIG. 2. Mean ionic activity coefficients (as l.rry • .J as a function
-2.5 of concentration. Open circles refer to GCMC results with
N ~ 64, open triangles to systems with N ~ 128. Crosses give
some apparent values obtained from CMC osmotic coefficients:
except for the 1 : 1 system they are dubious.
-3.0

to do very long runs, often of a few million configura-


tions. (CMC runs for these systems require only
-3.5
- 100 000 configurations.)
" ,Ql These techniques enabled one to go to 4. 75 M, 2. 78 M,
o
6 and 1. 63 M for the 1: 1, 2: 1, and 3: 1 systems, respec-
tively (or p/Po=0.31, 0.26, and 0.21, where Po is the
close-packed density). The upper concentration depends
FIG. 1. Reduced configurational energies as a function of con-
strongly on the valence type, of course, since one is
centration. Open circles refer to GCMC results with N ~ 64, adding 2, 3, or 4 particles, respectively, in the three
open triangles to systems about twice as large. The crosses types. The upper concentration limit for the 2: 2 case
and solid curves show the very precise results from CMC work. seems to depend on other factors which are discussed
A few typical error bars are shown. below.

J. Chern. Phys., Vol. 72, No. 11, 1 June 1980

This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
137.112.236.68 On: Thu, 18 Dec 2014 19:04:03
5940 J. P. Valleau and L. K. Cohen: Primitive model electrolytes.

It is well known that the nonideality !J lny" =/3(1J. - lJ.ideal) different ways, in fact, with perfect consistency. Most
of electrolytes goes through a minimum in the concentra- of the 3: 1 CMC results for lny. turn out to be fairly
tion ranges studied here. It follows (if the number of par- good, in spite of our misgivings about them. For the
ticles N is to be kept roughly constant) that the parame- 2: 2 case, however, the CMC results are quite wrong.
ter B will also go through a minimum as the concentra- This is undoubtedly due mostly to lack of success in ex-
tion is increased for approximately constant N, and this trapolating g._ (r) back to r:R. The extrapolation was
can be seen in the tables. One might imagine that this done by using a quadratic fit to the small-r values of
would lead to ambiguous MC results, two concentrations g._ (r); the errors correspond to too-high values of g._
corresponding to a single value of B. This is not the (R). It is possible to get lower (or higher) values of g._
case, however, since for fixed volume the two concen- (R) by extrapolating differently, but there is no a Priori
trations would have to be such that [cf. Eqs. (7) and (9)] way to choose the best extrapolation. These results
the difference in lny. was given by justify our reluctance to publish CMC results for the os-
motic coefficients of the higher-valence electrolytes.
~lny = -1. ~ln(N·"·N-"-) = - ~lnM
• !J ' The GCMC results appear totally satisfactory ex-
but lny. always decreases much less rapidly than -lnM; cept for the 2: 2 system at concentrations> 2 M. There
hence the results are single valued. the number dependence appears to be large and erratic.
In particular the result at 3.69 M (with N= 118) should
For most of the runs the volumes and the values of B coincide with that at 3.21 M (with N= 51), instead of
were chosen to give roughly N- 64 particles. The num- which they are totally discrepant in energy and density
ber dependence of the thermodynamic properties for di- and thus activity. The apparent number dependencies
lute solutions was studied previously, 1,3 and it is known at 1. 9 M and 2.9 M are just believable as random error,
that a system of this size gives adequate scientific re- but also look really to be discrepant. Although the
sults. For the denser systems studied here, where one small-system results (N- 64) fall on a credible curve for
might worry about the adequacy of the energy approxi- both energy and activity, the peculiar behaviour of the
mation, parallel runs were performed with double the large-system (N- 128) results puts all of the results into
volume and an appropriately adjusted B [cf. Eq. (7)]. question.
These results appear in the figures as open triangles to
distinguish them from the smaller systems. It is evi- It is natural to seek an explanation in terms of slow
dent that, except for the 2: 2 systems, the agreement is convergence due to low addition/deletion acceptance
entirely satisfactory, in most cases well within the sta- rates. In fact, however, these 2: 2 runs show accept-
tistical uncertainty of the results. In going to a larger ance rates very much higher than those which were easi-
system one expects to find a very slightly less negative ly handled for the other systems. Furthermore, control
energy (since the cancelling charge surrounding each charts of Nand U show neither long correlation times
ion is allowed to spread over a larger volume); in the nor any tendency to drift, so simple slow convergence seems
CMC work this was found to raise the energy by - 1% on to be ruled out. The other obvious explanation is an enhanced
doubling the size from N =64 (and by - 2% in extrapolat- number dependence due to breakdown of the minimum-
ing to an infinite system). The N dependence of the image energy approximation. All the evidence 14 ,16 sug-
GCMC results is not inconsistent with that, although it gests that in this case the larger system should have a
may be even weaker for the 2: 1 and 3: 1 systems. The less negative energy, however, while these results show
very large 2: 2 discrepancies are discussed separately. a lower energy. Furthermore, the fact that the density
discrepancies are in different directions seems puzzling
Comparison with the CMC results l - 3 on the same sys- on this hypotheSiS, while previous tests 3 ,16,17 indicate
tems provides a useful check on the GCMC results. One that the minimum image approximation remains quite ac-
expects the precision of the GCMC energy estimates to curate at the values of r in question here.
be less than those of the CMC, due to the density fluctu-
ations. This is borne out by the results, in which the So (always supposing that no error crept into the pro-
standard deviations of the energies are several times gramme!) the problem looks to be one of quasiergodic-
larger than in the CMC case, in spite of the longer runs. ity. In this connection it is of interest that there is be-
The accuracy is all one could hope for, however: in lieved to be a phase change in Coulombic systems with
Fig. 1 the solid lines show the very precise CMC results a critical point at a temperature somewhat lower than
for N=64, and the agreement of the GCMC results is ours and at a somewhat uncertain density. 18,19 [Some
everywhere complete. results of Vorontsov-Vel'yaminov and Chasovskikh,18
based on MC computations, suggest that there might be
In the case of the activity coefficients, on the other a critical point in a 2-2 "aqueous" electrolyte (i. e. ,
hand, it is the GCMC results which are expected to be with E = 78.4) near 200 A and 3.6 M.] It is natural to ask
more precise, as discussed above. Figure 2 shows (as whether the apparent quasiergodicity of the 2-2 results
crosses) some CMC results obtained by Gibbs-Duhem is associated with such a phase transition; that is the
integration of the osmotic coefficients. 3 For the 1: 1 subject of a separate study now underway. Meanwhile
case the extrapolations of g._ (r) to r= R were not diffi- it seems prudent not to use the results for the 2-2 sys-
cult, and the results were checked by using multistage tem at high concentrations.
sampling,4 so it is not surprising to find the total agree-
ment with the new GCMC results. The free energy re- The scientific implications of the results are discussed
sults for the 1 : 1 system have now been obtained in three in the following two papers 20 ,21 of the series.

J. Chern. Phys., Vol. 72, No. 11. 1 June 1980


This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
137.112.236.68 On: Thu, 18 Dec 2014 19:04:03
J. P. Valleau and L K. Cohen: Primitive model electrolytes. I 5941

ACKNOWLEDGMENTS Methods (Methuen, London, 1964).


14J. P. Valleau and S. G. Whittington, "A Guide to Monte
The authors acknowledge with thanks the financial sup- Carlo for Statistical Mechanics. 1. Highways," in Modern
port of the National Research Council of Canada. They Theoretical ChemiStry, Vol. 5A: Statistical Mechanics:
have enjoyed useful discussions with Glenn Torrie and Techniques, edited by B. Berne (Plenum, New York, 1977),
Stuart Whittington, and the helping hand of Mike McNeal. p. 137.
Jay Rasaiah pointed out long ago the problems of esti- ISIf one wishes to regard the particles as labeled, one must

mating osmotic coefficients from CMC data. imagine a process of fixing the labels to the particles in a
consistent way in spite of the additions and deletions: every
system with N+ cations and N- cations should involve a particu-
I D• N. Card and J. P. Valleau, J. Chern. Phys. 52, 6232 lar set of labels, e. g., integers from 1 to N+ for the cations
(1970). and from 1 to N- for the anions. One way of achieving this
2J. C. Rasaiah, D. N. Card, and J. P. Valleau, J. Chern. would be to assign those labels to the particles, at random,
Phys. 56, 248 (1972). after each attempted addition or deletion. We could then
3D. N. Card, thesis, University of Toronto, 1972. impute, for the labeled states [cf. Eqs. (3) and (4)]
4J. P. Valleau and D. N. Card, J. Chern. Phys. 57, 5457
(1972).
P 1
qlJ - V"II+IIIJ NjlNjl
sG. M. Torrie and J. P. Valleau, J. Comput. Phys. 23, 187
(1977). and
6J • P. Valleau and G. M. Torrie, "A Guide to Monte Carlo for
P 1
Statistical Mechanics. 2. Byways," in Modern Theoretical
qJ/ NjNjll+III_1 NjlNj!
Chemistry, Vol. 5A, Statistical Mechanics: Equilibrium
Techniques, edited by B. Berne (Plenum, New York, 1977), However, the probability lTl of a configuration i for labeled
p. 169. particles will be [ef. Eq. (5)]
7W. W. Wood, in Physics of Simple Liquids, edited by H. N.
V. Temperley, G. S. Rushbrooke, and J. S. Rowlinson lTI=~ Njl~j! A'!1\"!'1 expfJ3(J.I+Nj+J.I_Nj) -IW I ]
(North-Holland, Amsterdam, 1968), Chap. 5.
8G. E. Norman and V. S. Filinov, High Temp. (USSR) 7, 216 and thus the ratio of acceptance probabilities (6) would be un-
(1969). changed. This point of view seems, however, cumbersome
9D• J. Adams, Mol. Phys. 28, 1241 (1974). and artifiCial, since no such labeling in fact takes place.
Ion. J. Adams, Mol. Phys. 29, 307 (1975). 16C. S. Hoskins and E. R. Smith (private communication).
ilL. A. Rowley, D. Nicholson, and N. G. Parsonage, J. Com- I7S. G. Brush, H. L. Sahlin, and E. Teller, J. Chern. Phys.
put. Phys. 17, 401 (1975). 45, 2102 (1966).
12L. A. Rowley, D. Nicholson, and N. G. Parsonage, Mol. 18B. P. Chasovskikh and P. N. Vorontsov-Vel'iaminov, High
Phys. 31, 365, 389 (1976); see also J. Comput. Phys. 26, Temp. (USSR) 14, 174 (1976).
66 (1978). 19G. Stell, K. C. Wu, and B. Larsen, Phys. Rev. 37, 1369
13J. M. Hammersley and D. C. Handscomb, Monte Carlo (1976).

J. Chem. Phys .• Vol. 72. No. 11.1 June 1980


This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to IP:
137.112.236.68 On: Thu, 18 Dec 2014 19:04:03

Вам также может понравиться