Electrochmicrr A&, 1977, Vol. 22, pp. 11631187. Pergamon Press.

Rioted in Great Britain

THE EXTENDED JONESDOLE EQUATION

N~mous MARTINI%*, CRAWFORDD. SW-~ and COLIN A. VINCENT*
Departments of Chemistry* and Statisticst, University of St. Andrews, St. Andrews, Fife, Scotland

(Received 29 November 1976)

Abatrac-A number of authors have recently interpreted the relative viscosities of electrolytic solutions
in terms of an extended Jones-Dole equation containing a term DC’:
(q/q0 - 1) = Ac”z + Bc + Dc2.
The B-coefficients resulting from such analyses are not directly comparable with those arising from
the simple Jones-Dole expression because of the high negative correlation between rhe B and D terms.
There is therefore some difficultyin relating results from both types of analysis. In this paper a number
of data sets are examined statistically with a view to establishing a standard procedure for obtaining
estimates of the B-coefficients which relate exclusively to the linear term and are not influenced by
the effect of higher terms. Two methods which recognize the high correlation of B and D terms are
suggested.

INTRODUCTION B-parameter. In 1957 Kaminsky[S] used an extended

equation
Studies of the variation of viscosity of electrolytic
solutions with solute concentration have provided a q/q0 - I = AC”* + Bc + DC’ (3)
useful insight into solution structure and interactions. in order to fit the results of very precise measurements
It may be shown from the theory of interionic attrac- in aqueous systems at high concentrations. The fitted
tion[1,2] that as the concentration c of a salt tends D-parameter was a &mall number compared with the
to zero, corresponding value of B, and the latter was not
q/v,, - I = AelI= 0) greatly affected by the introduction of the extra term.
However in a study of electrolyte solutions in
where q is the viscosity of the solution, ‘lo is that N-methylformamide, Feakins and Lawrenc.e[S)]
of the solvent and A is a constant. Values of A cabu- reported that use of the extended equation (3) altered
lated for any temperature from knowledge of the di- the derived B-values by some 0.015 or 3% from the
electric constant, solvent viscosity, valency and limit- values derived using (2). Up to 0.1 mol clme3, the
ing conductance of the ions, agree closely with experi- Dc2 term contributed less than 3% to (q/q0 - l), but
mental determinations[3]. At higher concentrations this increased to around 10% for a system studied
the empirical equation of Jones and Dole[4] with concentrations up to 0.4mol duP3. Sodium
iodide solutions in dimethylsulphoxide have been
v/v* - 1 = AC”2 + Bc studied by two groups of workers. Adolph and
(2)
Siedel[lO] using a maximum concentration of ap
where B is an empirical constant, is found to hold proximately 0.08 mol chne3 fitted (3) to their data
for electrolytic solutions, in some cases up to concen- which at 25°C gave
trations of over 0.1 mol dm- 3. It has also been shown
(rl/rlo - 1) = 0.0145 c1’2 + 0.734 c + 0.685 cz.
that B is an additive.property of the ions[5] which
gives a useful measure of ionsolvent interactions {see For c = 0.1 mol dnP, the DC’ term contributed
eg[3]), and that the temperature coefficient of B gives some 8% to (V/Q, - 1). Recently, Feakins et al[ll]
further information on this subject, Evaluation of ac- have m-examined this system, and, finding no signifi-
curate B-coefficients for electrolytes and the intercom- cant deviations from (2) over the range O-0.08 mol
parison of these coefficients for a range of salts has dm- 3, evaluated their B-coefficients using the simpler
become an important area of study for both aqueous equation over this restricted concentration range.
and non-aqueous systems. They reported the viscosity-concentration function at
In 1933 it was reported[6] that for aqueous solu- 25°C as
tions of KBr and KC1 having concentrations above
0.1X0 mol dmm3, experimental results indicated the (tllrlo - I) = 0.009 cl’2 + 0.802c.
need for the addition of a further term, proportional
to the square of the concentration, to the Jones-Dole It is likely that I3 and D parameters will be highly
equation. On examining a number of salts in meth- interdependent and therefore that B-parameters deter-
anolic solution, Jones and FomwaltEfl found that mined using (2) will not be directly comparable with
while (2) adequately described experimental data up those derived using (33. Further, the physical signifi-
to a concentration of 0.01 mol dm-‘, it failed badly cance of the D-parameter is not obvious and the ques-
above this value. They therefore used only the results tion arises as to its usefulness in the interpretation
below this concentration in the evaluation of the of solution properties. We have therefore examined
1183
1184 NICHOLASMARTINUS, CRAWFORD D. SINCLAIR AND COLIN A. VINCENT

Table 1. Application of Method 1 to sodium iodide solutions at ZYC[lO]

Estimated
Number standard (Estimate of D)/ Residual sum of squares x lo7
Of Estimate of error (Standard error Equation Equation
points I3 a B D of D) (3) (2)

7 0.754 0.550 O.oc-42 0.059 9.26 0.767 13.90

6 0.763 0.322 0.0048 0.104 3.10 0.3 14 1.069
5 0.775 -0.233 0.0090 0.379 -0.62 0.179 0.201

statistically a series of experimental measurements for ing the F-test, it was again decided to stop at the
both aqueous and non-aqueous systems with a view same stage of the data reduction.
to determining the best way of evaluating the B- Viscosity measurements on sodium sulphate at
parameter. In all cases a theoretically established 15°C by Kaminsky[8] provided a set of extremely
value for the A-coefficient was used This procedure good data, which is analysed in Table 2. Both stop
has been shown[ll, 121 to be the best method of fit- ping criteria discussed above indicated a reduction
ting results to the Jones-Dole or extended Jones to six points, which led to an estimated B-value of
Dole equations because of the rapid increase in ex- 0.3607 + (2.78 x 0.00103), ie 0.357-0.364 for a 95%
perimental error on measuring relative viscosity at confidence interval.
very low concentrations.
METHOD 2

MhTHOD I This approach was a modification of Method 1 in

which it was recognised that the estimated values for
This method involved fitting (3) to the. data in
stages, estimating the B and D coef%icients by least the coefficients B and D were highly correlated, The
correlations in the three fits for the data of Adolph
squares, and then successively reducing the number
and Siedel[lO] shown in Table 1 were -0.959,
of data points in the analysis by omitting the point
-0.955 and -0.966, which reflected the similarity of
which was currently at the highest concentration. The
aim was to determine the stage at which equation the behaviour of c and cz over the range of concen-
tration studied. It was therefore considered more in-
(2) provided an adequate description of the data.
formative to construct a two-dimensional confidence
For example, the results of applying this process to
region for the B and D-parameters jointly. Making
the data of AdoIph and SiedelElO] for sodium iodide
in dimethyl sulphoxide at 25°C are shown in Table the usual assumptions about the error term, a 95%
1. The theoretically established value of A was confidence ellipse for the values of B and D is given
0.0119 dm3” mol-‘12*. Using as an indicator for stop- by
ping the data reduction the criterion that the value
I(B - &%3)a - 2R(B - a}
of D was not significantly different from zero, in-
volved comparing the ratio of the estimated value of x (D - b)/(SB.SD) + {(D - @/SD)2
D to its estimated standard error with the appropriate < 2[1 - P)F (4)
percentage point of the t-distribution. Using either
the 10% or 5% point, the two highest concentration where fi and 6 are the estimated values of B and
points were discarded in this case, giving a point esti- D respectively, SB and SD are their estimated Stan-
mate of B of 0.775 with an estimated standard error dard errors, R is the coefficient of correlation between
of 0.009. However the interval 0.775 + (3.18 x 0.009) B and D and F is the appropriate percentage point
or 0.7460.803 did not necessarily provide a good of the F distribution with 2 and (p - 2) degrees of
95% confidence level for B, as will be seen later. Using freedom, where p is the number of data points in
an alternative criterion of stopping--viz when the the fit. For example, in the case quoted above at the
reduction in residual sum of squares obtained by five lowest concentration points, the 95% confidence
adding the term in D was not significant-and apply- region is the interior of the ellipse
12373(B - fi),’ + 567(B - h)
* The value of 0.0176 quoted in [lo] is in error. x (D - b) + 6.962(0 - 6)’ = 1.277.

Table 2. Application of Method 1 to sodium sulphate solutions in water at lS’C[S]

Estimated
Number standard @stirnate of D)/ Residual sum of squares x 10%
of Estimate of error (Standard error Equation Equation
points B D B D of D) (3) (2)

9 0.3609 0.0422 o.ooo33 0.0029 14.51 0.278 8.653

8 0.3604 0.0487 o.oOo52 0.0062 7.82 0.227 2.547
7 0.3613 0.0332 0.00055 0.0081 4.09 0.109 0.472
6 0.3605 0.0499 o.oM)79 0.0146 3.42 0.075 0.295
5 0.3607 0.0445 0.00103 0.0224 1.99 0.072 0.167
4 0.3573 0.2480 0.00160 0.0890 2.79 0.019 0.095
 The extended Jones-Dole equation 1185

The portion of the line D = 0 which lies inside this

region is that between the roots of the quadratic
equation in B:

i2373 {B - 0.775)’ + 132.1 (B - 0.575) - 0.277 = 0

ie B = 0.760-0.780.

Figure 1 shows the elliptical confidence regions

obtained in this way based on 7, 6 and 5 data points.
Note that when the parameters were estimated using
all seven data points the ellipse did not intersect the
B-axis, confirming that the term in D was required
for a satisfactory fit of the data. As the number of
points was reduced the ellipse became larger and its
centre moved In practice, however, the segment of
the B-axis cut off by the ellipse was moderately stable
while there was still a reasonable number of points
in the fit. A practical lower limit for the number of
obseruations is seven-five more than the number of
parameters to be estimated. From Fig. 1, it is seen
that the interval 0.76C~O.780 is likely to contain the
true value of B.
Applying this method to the sodium sulphate data
of Kaminsky[8] led to the confidence regions shown
in Fig. 2. The very large negative correlation coeffi-
cients (-0.956 to -0.977) produced long thin ellipses
sloping from left to right. The portion of the B-axis
cut off by the five observation ellipse was 0.3616 to
0.3638 providing a point estimate for B of 0.363. Ag
plication of Method 2 to the results of measurements
Fig. 1. 9S”<, confidence regions as defined by Method 2 of sodium bromide in N-methylformamide at 35”C[9]
for sodium iodide solutions in DMSO at 25”C[lO]. is illustrated in Fig. 3.

Fig. 2. 95% confidence regions as defined by Method 2 Fig. 3. 95% confidence regions as defined by Method 2
for sodium sulpbate solutions in water at WC[S]. for sodium bromide solutions in NMF at 35”C[9].
1166 NICHOLAS MARTINUS, CRAWFORD D. SINCLAIR AND CBIJN A. WNCENT

MmOD 3 where the summation is over these values of i which

In this method a different approach was used. ln
order to avoid the problem of correlation between
the estimates for B and D, equation (3) was reformu-
lated as
(v/%I - 1) = AC”’ + Bc + P.Q(c)
where Q(c) was a polynominal in c”‘, c and cz whose
coefficients /+’and y were chosen so as to make Q(c)
orthogonal to c112 and c over the range of concen-
tration values, cir involved in the fit: -
Q(c) = /?c”2 + yc + 2.
Now
correspond to the observations being used in the fit
at any stage.
For example, Table 3 shows the values of Q(c) for
the data of Feakins and Lawrence discussed above,
first for all eleven observations and then for the data
reduced by the highest and by the two highest con-
(5) centrations. When the coefficients B and P were fitted
their estimated values were uncorrelated because the
variance matrix of the estimated values was
u’(XTX)-’ and we had ensured that XTX had zero
entries at all points off the diagonal, and consequently
so did (X=X)-‘.
(6) X is the matrix:

C c”‘Q(cJ = 0, and c ciQ(cJ = 0

L I

and thus
Since the coefficients /I and y changed from stage to
stage of the data reduction, the coefficient P was esti-
mating a different quantity each time and was of little
interest. Table 4 shows the estimates of B together
with their standard etiors and 95% confidence inter-
vals. It is seen that the latter are ambiguous in their
description of 5 since the fit was at first very poor
and but became reasonable when the number of observa-
tions was reduced to eight. Moreover the point esti-
Y = (T4” F G” - 4 ci C c?)/(T
GT 2
mate of B at first moved rapidly and systematically
with the reduction in observations. The width of the
confidence interval is a function both of the number
of points and the standard error of the estimated
values of B. Figure 4 shows how it decreased as the

Table 3. Orthogonal polynominal values for the appli-

cation of Method 3 to sodium bromide solutions in NMF
at 35”C[9]

i Ql, (4 Ql csd Qd3

1 0.000622 0.002916 0.001903 0.001015
2 0.002471 0.005116 0.003257 0.001662
3 oLm4759 0.006358 0.00395 1 0.001925
4 cJ.008303 O.W7289 0.004371 0.001983
5 0.01183 O.CQ7667 0.004433 0.001855
6 0.02366 * 0.007262 0.003587 O.OOO!J27
7 0.05013 0.003344 o.oooo29 -0.001612
8 0.09652 - 0.004868 -0.005973 -0.004414
9 0.2004 -0.015160 -0.008803 0.002159 0.50 0.55 0 60 0.65
10 0.3041 -0.008225 0.007028
11 0.3954 0.013840 Fig. 4. 95% confidence intervals as defined by Method 3
for sodium bromide solutions in nmfat 35”C[9].

Table 4. Application of Method 3 to sodium bromide solutions in nd at 35”C[93

Estimated 95% confidence Residual sum

Number Estimate standard error interval of squares x IO6
of points of B of B for B (equation [S])

11 0.614 0.0051 0.602-0.625 69.92

10 0.596 0.0041 0.58W.605 19.20
9 0.578 0.0029 0.5624.593 3.05 1
8 0.562 0.0018 0.558-0.566 0.258
7 0.561 0.0034 0.552-0.570 0.192
6 0.547 0.0015 0.543~.551 0.074
5 0.556 0.0054 0.539-0.573 0.021
4 0.552 0.0092 0.512-0.591 0.017
 The extended Jones-Dole equation 1187

Table 5. Comparison of results in estimating B-coefficients by Methods 1, 2 and 3

Method 1 Method 2 Method 3

Data source point interval point interval point interval

Adolph and Siedel[lO] 0.115 0.74&0.803 0.770 0.76’&0.780 0.700 0.7654.775

Kaminsky[S] 0.361 0.3570364 0.363 0.362-0.364 0.363(5) 0.363-0.364
Feakins and Lawrence[9] 0.557 0.541-0.572 0.563 O.55UJ.568 0.562 0.558X066

point estimate settled clown, and then increased again.
The interval corresponding to fitting the eight lowest
concentrations is in the narrow range OSXHL566.
However a more reliable interval contains these
values which are common to the 6, 7 and 8 point
intervals, viz 0.54LVO.570.
CONCLXJSIONS
A comparison of the results obtained using
Methods 1, 2 and 3 of data analysis on the three
sets of experimental observations discussed above is
given in Table 5.
Method 1, while simplest both in concept and in
calculation, is unlikely to provide good values for B
since it does not allow for the correlation between
estimated values of B and D. Method 2 takes this
noted that the results obtained by these two tech-
niques are comparable both for point and interval
estimates whether the observations exhibit a greater
degree of accuracy (as in the aqueous data studied)
or a lesser degree (as in the non-aqueous systems).
REFERENCES
1. H. Falkenhagen and E. L. Vernon, Phil. Mug. 14, 537
(1932).
2. L. Onsager and R. M. Fuoss, J. phys. Chem. 36, 2689
(1932).
3. R. H. Stokes and R. Mills, Viscosity of Electrolytes,
p. 31. Pergamon Press, Oxford (1965).
4. G. Jones and M. Dole, J. Am them SIX. 51, 2950
(1929)
5. W. M. Cox and .I. H. Wolfenden. fioc. R. Sot. A 145.
486 (1934).
6. G. Jones and S. K. Tallev. J. Am them. Sot. 55. 4124
_I

correlation into account and produces a narrower (1933).

range of B-values. While it is not necessary to draw
the ellipses in order to find where they cut the B-axis,
the confidence region diagram is useful for inspecting
the consistency of the reduced sets of data in estimat-
ing B. Method 3 gives apparently more precise esti-
mates at certain stages of data reduction: however
a more conservative estimate is made by choosing
values of B common to several confidence intervals.
Accordingly, the use of Methods 2 or 3 is recom-
mended when it is desired to estimate the value of
B as precisely as possible. From Table 5 it may be
7. G. Jones and H. J. Fomwalt, J. Am. &em. Sec. 58,
619 (1936).
8. M. Kaminsky, Z. phys. Chem 12, 206 (1957).
9. D. Feakins and K. G. Lawrence, J. them. Sot. A, 212
(1966).
10. W. Adolph and W. Siedel, Z. phys. Chem 93, 173
(1974).
Il. R. T. M. Bicknell, K. G. Lawrence, M. A. Se&y, D.
Feakins and L. Werblan, J. them. Sot. Faraday I 12,
307 (1976).
12. N. Martinus, C. D. Sinclair and C. A. Vincent, unpub
lished work.